{
    "pages": [
        {
            "page_number": 1,
            "text": ""
        },
        {
            "page_number": 2,
            "text": "Springer Series on \nSIGNALS AND COMMUNICATION TECHNOLOGY\n"
        },
        {
            "page_number": 3,
            "text": "SIGNALS AND COMMUNICATION TECHNOLOGY\nWireless Network Security \nY. Xiao, S. Shen, and D. Du (Eds.) \nISBN 0-387-28040-5 \nWireless Ad Hoc and Sensor Networks\nA Cross-Layer Design Perspective \nR. Jurdak \nISBN 0-387-39022-7 \nCryptographic Algorithms on Reconfigurable \nHardware\nF. Rodrgues-Henriquez, N.A. Saaqib, A. Diaz \nPerez, and C.K. Koc \nISBN 0-387-33956-6 \nMultimedia Database Retrieval\nA Human-Centered Approach \nP. Muneesawang and L. Guan \nISBN 0-387-25267-X \nBroadband Fixed Wireless Access\nA System Perspective \nM. Engels and F. Petre \nISBN 0-387-33956-6 \nDistributed Cooperative Laboratories \nNetworking, Instrumentation, and Measurements \nF. Davoli, S. Palazzo and S. Zappatore (Eds.) \nISBN 0-387-29811-8 \nThe Variational Bayes Method \nin Signal Processing \nV. Šmídl and A. Quinn \nISBN 3-540-28819-8 \nTopics in Acoustic Echo and Noise Control \nSelected Methods for the Cancellation of \nAcoustical Echoes, the Reduction of \nBackground Noise, and Speech Processing \nE. Hänsler and G. Schmidt (Eds.) \nISBN 3-540-33212-x \nEM Modeling of Antennas and RF \nComponents for Wireless Communication \nSystems \nF. Gustrau, D. Manteuffel \nISBN 3-540-28614-4 \nInteractive Video \nMethods and Applications \nR. I Hammoud (Ed.) \nISBN 3-540-33214-6 \nContinuous Time Signals \nY. Shmaliy \nISBN 1-4020-4817-3 \nVoice and Speech Quality Perception \nAssessment and Evaluation \nU. Jekosch \nISBN 3-540-24095-0 \nAdvanced ManMachine Interaction \nFundamentals and Implementation \nK.-F. Kraiss \nISBN 3-540-30618-8 \nOrthogonal Frequency Division Multiplexing \nfor Wireless Communications \nY. (Geoffrey) Li and G.L. Stüber (Eds.)\nISBN 0-387-29095-8 \nCircuits and Systems  \nBased on Delta Modulation  \nLinear, Nonlinear and Mixed Mode Processing  \nD.G. Zrilic     ISBN 3-540-23751-8 \nFunctional Structures in Networks  \nAMLn—A Language for Model Driven \nDevelopment of Telecom Systems  \nT. Muth     ISBN 3-540-22545-5  \nRadioWave Propagation  \nfor Telecommunication Applications \nH. Sizun     ISBN 3-540-40758-8 \nElectronic Noise and Interfering Signals  \nPrinciples and Applications  \nG. Vasilescu     ISBN 3-540-40741-3  \nDVB\nThe Family of International Standards  \nfor Digital Video Broadcasting, 2nd ed.  \nU. Reimers     ISBN 3-540-43545-X \nDigital Interactive TV and Metadata  \nFuture Broadcast Multimedia  \nA. Lugmayr, S. Niiranen, and S. Kalli  \nISBN 3-387-20843-7 \nAdaptive Antenna Arrays  \nTrends and Applications  \nS. Chandran (Ed.)     ISBN 3-540-20199-8 \nDigital Signal Processing  \nwith Field Programmable Gate Arrays  \nU. Meyer-Baese     ISBN 3-540-21119-5  \nNeuro-Fuzzy  and  Fuzzy  Neural  Applications  \nin  Telecommunications  \nP. Stavroulakis (Ed.)  ISBN 3-540-40759-6 \ncontinued after index \n"
        },
        {
            "page_number": 4,
            "text": "Wireless Network Security\nYANG XIAO, XUEMIN SHEN,\nand DING-ZHU DU\nSpringer\n"
        },
        {
            "page_number": 5,
            "text": "Editors: \nYang Xiao                                                         Xuemin (Sherman) Shen \nDepartment of Computer Science                     Department of Electrical & Computer Engineering \nUniversity of Alabama                                      University of Waterloo \n101 Houser Hall                                                 Waterloo, Ontario, Canada N2L 3G1 \nTuscaloosa, AL 35487 \nDing-Zhu Du \nDepartment of Computer Science & Engineering \nUniversity of Texas at Dallas \nRichardson, TX 75093 \nWireless Network Security \nLibrary of Congress Control Number:  2006922217 \nISBN-10   0-387-28040-5 \n           e-ISBN-10    0-387-33112-3 \nISBN-13   978-0-387-28040-0                e-ISBN-13    978-0-387-33112-6 \nPrinted on acid-free paper. \n© 2007 Springer Science+Business Media, LLC \nAll rights reserved. This work may not be translated or copied in whole or in part without \nthe written permission of the publisher (Springer Science+Business Media, LLC, 233 Spring \nStreet, New York, NY 10013, USA), except for brief excerpts in connection with reviews or \nscholarly analysis. Use in connection with any form of information storage and retrieval, \nelectronic adaptation, computer software, or by similar or dissimilar methodology now \nknow or hereafter developed is forbidden. \nThe use in this publication of trade names, trademarks, service marks and similar terms, \neven if they are not identified as such, is not to be taken as an expression of opinion as to \nwhether or not they are subject to proprietary rights. \n9  8  7  6  5  4  3  2  1                    \nspringer.com\n"
        },
        {
            "page_number": 6,
            "text": "CONTENTS\nPreface\nvii\nPart I: Security in General Wireless/Mobile Networks\n1\nChapter 1: High Performance Elliptic Curve Cryptographic Co-processor\n3\nJonathan Lutz and M. Anwarul Hasan\nChapter 2: An Adaptive Encryption Protocol in Mobile Computing\n43\nHanping Lufei and Weisong Shi\nPart II: Security in Ad Hoc Network\n63\nChapter 3: Pre-Authentication and Authentication Models in\nAd Hoc Networks\n65\nKatrin Hoeper and Guang Gong\nChapter 4: Promoting Identity-Based Key Management in\nWireless Ad Hoc Networks\n83\nJianping Pan, Lin Cai, and Xuemin (Sherman) Shen\nChapter 5: A Survey of Attacks and Countermeasures in\nMobile Ad Hoc Networks\n103\nBing Wu, Jianmin Chen, Jie Wu, and Mihaela Cardei\n137\nVenkata C. Giruka and Mukesh Singhal\nYang Xiao, Xuemin Shen, and Ding-Zhu Du\nChapter 6: Secure Routing in Wireless Ad-Hoc Networks\n"
        },
        {
            "page_number": 7,
            "text": "vi\nTABLE OF CONTENTS\nChapter 7: A Survey on Intrusion Detection in\nMobile Ad Hoc Networks\n159\nTiranuch Anantvalee and Jie Wu\nPart III: Security in Mobile Cellular Networks\n181\nChapter 8: Intrusion Detection in Cellular Mobile Networks\n183\nBo Sun, Yang Xiao, and Kui Wu\nChapter 9: The Spread of Epidemics on Smartphones\n211\nBo Zheng, Yongqiang Xiong, Qian Zhang, and Chuang Lin\nPart IV: Security in Wireless LANs\n243\nChapter 10: Cross-Domain Mobility-Adaptive Authentication\n245\nHahnsang Kim and Kang G. Shin\n273\nJon W. Mark, Yixin Jiang, and Chuang Lin\nChapter 12: An Experimental Study on Security Protocols in WLANs\n295\nAvesh Kumar Agarwal and Wenye Wang\nPart V: Security in Sensor Networks\n323\nChapter 13: Security Issues in Wireless Sensor Networks\nused in Clinical Information Systems\n325\nChapter 14: Key Management Schemes in Sensor Networks\n341\nChapter 15: Secure Routing in Ad Hoc and Sensor Networks\n381\nXu (Kevin) Su, Yang Xiao, and Rajendra V. Boppana\nAbout the Editors\n403\nIndex\n407\nMinghui Shi, Humphrey Rutagemwa, Xuemin (Sherman) Shen,\nChapter 11: AAA Architecture and Authentication\nfor Wireless LAN Roaming\nJelena Misic and Vojislav B. Misic\nˇ ´\nˇ ´\nVenkata Krishna Rayi, Yang Xiao, Bo Sun, Xiaojiang (James) Du, and Fei Hu\n"
        },
        {
            "page_number": 8,
            "text": "PREFACE\nWireless/mobile communications network technologies have been dramatically ad-\nvanced in recent years, inculding the third generation (3G) wireless networks, wireless\nLANs, Ultra-wideband (UWB), ad hoc and sensor networks. However, wireless net-\nwork security is still a major impediment to further deployments of the wireless/mobile\nnetworks. Security mechanisms in such networks are essential to protect data integrity\nand confidentiality, access control, authentication, quality of service, user privacy, and\ncontinuity of service. They are also critical to protect basic wireless network function-\nality.\nThis edited book covers the comprehensive research topics in wireless/mobile net-\nwork security, which include cryptographic co-processor, encryption, authentication,\nkey management, attacks and countermeasures, secure routing, secure medium access\ncontrol, intrusion detection, epidemics, security performance analysis, security issues in\napplications, etc. It can serve as a useful reference for researchers, educators, graduate\nstudents, and practitioners in the field of wireless/network network security.\nThe book contains 15 refereed chapters from prominent researchers working in\nthis area around the world. It is organized along five themes (parts) in security issues\nfor different wireless/mobile networks.\nPart I: Security in General Wireless/Mobile Networks: Chapter 1 by Lutz\nand Hasan describes a high performance and optimal elliptic curve processor as\nwell as an optimal co-processor using Lopez and Dahab’s projective coordinate\nsystem. Chapter 2 by Lufei and Shi proposes an adaptive encryption protocol to\ndynamicallychooseaproperencryptionalgorithmbasedonapplication-specific\nrequirements and device configurations.\nPartII:SecurityinAdHocNetworks: Thenextfivechaptersfocusonsecurity\nin ad hoc networks. Chapter 3 by Hoeper and Gong introduces a security\nframework for pre-authentication and authenticated models in ad hoc networks.\nChapter 4 by Pan, Cai, and Shen promotes identity-based key management in\nad hoc networks. Chapter 5 by Wu et al. provides a survey of attacks and\ncountermeasures in ad hoc networks. Chapter 6 by Giruka and Singhal presents\nseveral routing protocols for ad-hoc networks, the security issues related to\n"
        },
        {
            "page_number": 9,
            "text": "viii\nPREFACE\nrouting, and securing routing protocols in ad hoc networks.\nChapter 7 by\nAnantvalee and Wu classifies the architectures for intrusion detection systems\nin ad hoc networks.\nPart III: Security in Mobile Cellular Networks: The next two chapters dis-\ncuss security in mobile cellular networks. Chapter 8 by Sun, Xiao, and Wu\nintroduces intrusion detection systems in mobile cellular networks. Chapter 9\nby Zheng et al. proposes an epidemics spread model for smartphones.\nPart IV: Security in Wireless LANs: The next three chapters study the secu-\nrity in wireless LANs. Chapter 10 by Kim and Shin focuses on cross-domain\nauthentication over wireless local area networks, and proposes an enhanced\nprotocol called the Mobility-adjusted Authentication Protocol that performs\nmutual authentication and hierarchical key derivation. Chapter 11 by Shi et\nal. proposes Authentication, Authorization and Accounting (AAA) architec-\nture and authentication for wireless LAN roaming. Chapter 12 by Agarwal\nand Wang studies the cross-layer interactions of security protocols in wireless\nLANs, and presents an experimental study.\nPart V: Security in Sensor Networks: The last three chapters focus on security\nin sensor networks. Chapter 13 by Miˇsi´c and Miˇsi´c reviews confidentiality\nand integrity polices for clinical information systems and compares candidate\ntechnologies IEEE 802.15.1 and IEEE 802.15.4 from the aspect of resilience\nof MAC and PHY layers to jamming and denial-of-service attacks. Chapter\n14 by Rayi et al. provides a survey of key management schemes in sensor\nnetworks.\nThe last chapter by Su, Xiao, and Boppana introduces security\nattacks, and reviews the recent approaches of secure network routing protocols\nin both mobile ad hoc and sensor networks.\nAlthough the covered topics may not be an exhaustive representation of all the\nsecurity issues in wireless/mobile networks, they do represent a rich and useful sample\nof the strategies and contents.\nThis book has been made possible by the great efforts and contributions of many\npeople. First of all, we would like to thank all the contributors for putting together\nexcellent chapters that are very comprehensive and informative. Second, we would\nlike to thank all the reviewers for their valuable suggestions and comments which have\ngreatly enhanced the quality of this book. Third, we would like to thank the staff\nmembers from Springer, for putting this book together. Finally, We would like to\ndedicate this book to our families.\nYang Xiao\nTuscaloosa, Alabama, USA\nXuemin (Sherman) Shen\nWaterloo, Ontario, CANADA\nDing-Zhu Du\nRichardson, Texas, USA\n"
        },
        {
            "page_number": 10,
            "text": "Part I\nSECURITY IN GENERAL\nWIRELESS/MOBILE NETWORKS\n"
        },
        {
            "page_number": 11,
            "text": "1\nHIGH PERFORMANCE ELLIPTIC CURVE\nCRYPTOGRAPHIC CO-PROCESSOR\nJonathan Lutz\nGeneral Dynamics - C4 Systems\nScottsdale, Arizona\nE-mail: Jonathan.Lutz@gdc4s.com\nM. Anwarul Hasan\nDepartment of Electrical and Computer Engineering\nUniversity of Waterloo, Waterloo, ON, Canada\nE-mail: ahasan@ece.uwaterloo.ca\nFor an equivalent level of security, elliptic curve cryptography uses shorter key sizes and is\nconsidered to be an excellent candidate for constrained environments like wireless/mobile\ncommunications. In FIPS 186-2, NIST recommends several finite fields to be used in the\nelliptic curve digital signature algorithm (ECDSA). Of the ten recommended finite fields,\nfive are binary extension fields with degrees ranging from 163 to 571. The fundamental\nbuilding block of the ECDSA, like any ECC based protocol, is elliptic curve scalar mul-\ntiplication. This operation is also the most computationally intensive. In many situations\nit may be desirable to accelerate the elliptic curve scalar multiplication with specialized\nhardware.\nIn this chapter a high performance elliptic curve processor is described which is\noptimized for the NIST binary fields. The architecture is built from the bottom up starting\nwith the field arithmetic units. The architecture uses a field multiplier capable of performing\na field multiplication over the extension field with degree 163 in 0.060 microseconds.\nArchitectures for squaring and inversion are also presented. The co-processor uses Lopez\nand Dahab’s projective coordinate system and is optimized specifically for Koblitz curves.\nA prototype of the processor has been implemented for the binary extension field with\ndegree 163 on a Xilinx XCV2000E FPGA. The prototype runs at 66 MHz and performs an\nelliptic curve scalar multiplication in 0.233 msec on a generic curve and 0.075 msec on a\nKoblitz curve.\n1.\nINTRODUCTION\nThe use of elliptic curves in cryptographic applications was first proposed inde-\npendently in [15] and [23]. Since then several algorithms have been developed whose\n"
        },
        {
            "page_number": 12,
            "text": "4\nJONATHAN LUTZ and M. ANWARUL HASAN\nstrength relies on the difficulty of the discrete logarithm problem over a group of elliptic\ncurve points. Prominent examples include the Elliptic Curve Digital Signature Algo-\nrithm (ECDSA) [24], EC El-Gammal and EC Diffie Hellman [12]. In each case the\nunderlying cryptographic primitive is elliptic curve scalar multiplication. This opera-\ntion is by far the most computationally intensive step in each algorithm. In applications\nwhere many clients authenticate to a single server (such as a server supporting SSL\n[7, 26] or WTLS [1]), the computation of the scalar multiplication becomes the bottle\nneck which limits throughput. In a scenario such as this it may be desirable to acceler-\nate the elliptic curve scalar multiplication with specialized hardware. In doing so, the\nscalar multiplications are completed more quickly and the computational burden on the\nserver’s main processor is reduced.\nThe selection of the ECC parameters is not a trivial process and, if chosen in-\ncorrectly, may lead to an insecure system [12, 24, 22]. In response to this issue NIST\nrecommends ten finite fields, five of which are binary fields, for use in the ECDSA [24].\nThe binary fields include GF(2163), GF(2233), GF(2283), GF(2409) and GF(2571) de-\nfined by the reduction polynomials in Table 1. For each field a specific curve, along with\nTable 1. NIST Recommended Finite Fields\nField\nReduction Polynomial\nGF(2163)\nF(x) = x163 + x7 + x6 + x3 + 1\nGF(2233)\nF(x) = x233 + x74 + 1\nGF(2283)\nF(x) = x283 + x12 + x7 + x5 + 1\nGF(2409)\nF(x) = x409 + x87 + 1\nGF(2571)\nF(x) = x571 + x10 + x5 + x2 + 1\na method for generating a pseudo-random curve, are supplied. These curves have been\nintentionally selected for both cryptographic strength and efficient implementation.\nSuch a recommendation has significant implications on design choices made while\nimplementing elliptic curve cryptographic functions. In standardizing specific fields\nfor use in elliptic curve cryptography (ECC), NIST allows ECC implementations to\nbe heavily optimized for curves over a single finite field. As a result, performance of\nthe algorithm can be maximized and resource utilization, whether it be in code size for\nsoftware or logic gates for hardware, can be minimized.\nDescribed in this chapter are hardware architectures for multiplication, squaring\nand inversion over binary finite fields. Each of these architectures is optimized for a\n"
        },
        {
            "page_number": 13,
            "text": "WIRELESS NETWORK SECURITY\n5\nspecific finite field with the intent that it might be implemented for any of the five NIST\nrecommended binary curves. These finite field arithmetic units are then integrated\ntogether along with control logic to create an elliptic curve cryptographic co-processor\ncapable of computing the scalar multiple of an elliptic curve point. While the co-\nprocessor supports all curves over a single binary field, it is optimized for the special\nKoblitz curves [16].\nTo demonstrate the feasibility and efficiency of both the finite field arithmetic units\nand the elliptic curve cryptographic co-processor, the latter has been implemented in\nhardware using a field programmable gate array (FPGA). The design was synthesized,\ntimed and then demonstrated on a physical board holding an FPGA.\nThis chapter is organized as follows. Section 2 gives an overview of the basic\nmathematical concepts used in elliptic curve cryptography. This section also provides\nan introduction to the hardware/software system used to implement the elliptic curve\nscalar multiplier. Section 3 presents efficient hardware architectures for finite field\nmultiplication and squaring. A method for high speed inversion is also discussed. In\nSection 4 and Section 5 a hardware architecture of an elliptic curve scalar multiplier is\npresented. This architecture uses the multiplication, squaring and inversion methods\ndiscussed in Section 3. Finally Section 6 provides concluding remarks and a summary\nof the research contributions documented in this report.\n2.\nBACKGROUND\nThefundamentalbuildingblockforanyellipticcurve-basedcryptosystemiselliptic\ncurve scalar multiplication. It is this operation that is to be performed by the co-\nprocessor. Provided in this section is an overview of the mathematics behind elliptic\ncurve scalar multiplication, including both field arithmetic and curve arithmetic.\n2.1. Arithmetic over Binary Finite Fields\nThe elements of the binary field GF(2m) are interrelated through the operations of\naddition and multiplication. Since the additive and multiplicative inverses exist for all\nfields, the subtraction and division operations are also defined. Discussed in this section\nare basic methods for computing the sum, difference and product of two elements. Also\npresented is a method for computing the inverse of an element. The inverse, along with\na multiplication, is used to implement division.\nAddition and Subtraction:\nIf two field elements a, b ∈GF(2m) are represented as\npolynomials A(x) = am−1xm−1 + · · · + a1x + a0 and B(x) = bm−1xm−1 + · · · +\nb1x + b0 respectively, then their sum is written\nS(x) = A(x) + B(x) =\nm−1\n\u0001\ni=0\n(ai + bi)xi.\n(1)\n"
        },
        {
            "page_number": 14,
            "text": "6\nJONATHAN LUTZ and M. ANWARUL HASAN\nA field of characteristic two provides two distinct advantages. First, the bit additions\nai+bi in (1) are performed modulo 2 and translate to an exclusive-OR (XOR) operation.\nThe entire addition is computed by a component-wise XOR operation and does not\nrequire a carry chain. The second advantage is that in GF(2) the element 1 is its own\nadditive inverse (i.e. 1 + 1 = 0 or 1 = −1). Hence, addition and subtraction are\nequivalent.\nMultiplication:\nThe product of field elements a and b is written as\nP(x) = A(x) × B(x)\nmod F(x) =\nm−1\n\u0001\ni=0\nm−1\n\u0001\nj=0\naibjxi+j\nmod F(x)\nwhere F(x) is the field reduction polynomial. By expanding B(x) and distributing\nA(x) through its terms we get\nP(x) = bm−1xm−1A(x) + · · · + b1xA(x) + b0A(x)\nmod F(x).\nBy repeatedly grouping multiples of x and factoring out x we get\nP(x) = (· · · (((A(x)bm−1)x + A(x)bm−2)x + · · · + A(x)b1)x\n+ A(x)b0)\nmod F(x).\n(2)\nA bit level algorithm can be derived from (2). However, many of the faster mul-\ntiplication algorithms rely on the concept of group-level multiplication. Let g be an\ninteger less than m and let s = ⌈m/g⌉. If we define the polynomials\nBi(x) =\n⎧\n⎪\n⎪\n⎪\n⎪\n⎪\n⎪\n⎨\n⎪\n⎪\n⎪\n⎪\n⎪\n⎪\n⎩\ng−1\n\u0001\nj=0\nbig+jxj\n0 ≤i ≤s −2,\n(m\nmod g)−1\n\u0001\nj=0\nbig+jxj\ni = s −1,\nthen the product of a and b is written\nP(x) = A(x)\n\u0006\nx(s−1)gBs−1(x) + · · · + xgB1(x) + B0(x)\n\u0007\nmod F(x).\nIn the derivation of equation (2) multiples of x were repeatedly grouped then factored\nout. This same grouping and factoring procedure will now be implemented for multiples\nof xg arriving at\nP(x) = (· · · ((A(x)Bs−1(x))xg + A(x)Bs−2(x))xg + · · · )xg\n+ A(x)B0(x)\nmod F(x)\nwhich can be computed using Algorithm 1.\n"
        },
        {
            "page_number": 15,
            "text": "WIRELESS NETWORK SECURITY\n7\nAlgorithm 1.\nGroup-Level Multiplication\nInput: A(x), B(x), and F(x)\nOutput: P(x) = A(x)B(x) mod F(x)\nP(x) ←Bs−1(x)A(x) mod F(x);\nfor k = s −2 downto 0 do\nP(x) ←xgP(x);\nP(x) ←Bk(x)A(x) + P(x) mod F(x);\nInversion:\nFor any element a ∈GF(2m) the equality a2m−1 ≡1 holds. When a ̸= 0,\ndividing both sides by a results in a2m−2 ≡a−1. Using this equality the inverse, a−1,\ncan be computed through successive field squarings and multiplications. In Algorithm\n2 the inverse of an element is computed using this method.\nAlgorithm 2.\nInversion by Square and Multiply\nInput: Field element a\nOutput: b ≡a(−1)\nb ←a;\nfor i = 1 to m −2 do\nb ←b2 ∗a;\nb ←b2;\nThe primary advantage to this inversion method is the fact that it does not require\nhardware dedicated specifically to inversion. The field multiplier can be used to perform\nall required field operations.\n2.2. Arithmetic over the Elliptic Curve Group\nThe field operations discussed in the previous section are used to perform arith-\nmetic over an elliptic curve. This chapter is aimed at the elliptic curve defined by the\nnon-supersingular Weierstrass equation for binary fields. This curve is defined by the\nequation\ny2 + xy = x3 + αx2 + β\n(3)\n"
        },
        {
            "page_number": 16,
            "text": "8\nJONATHAN LUTZ and M. ANWARUL HASAN\nwhere the variables x and y are elements of the field GF(2m) as are the curve parameters\nα and β. The points on the curve, defined by the solutions, (x, y), to (3) form an additive\ngroup when combined with the “point at infinity”. This extra point is the group identity\nand is denoted by the symbol O. By definition, the addition of two elements in a group\nresults in another element of the group. As a result any point on the curve, say P, can\nbe added to itself an arbitrary number of times and the result will also be a point on the\ncurve. So for any integer k and point P adding P to itself k −1 times results in the\npoint\nkP =\nP + P + · · · + P\n\b\n\t\n\u000b .\nk times\nGiven the binary expansion k = 2l−1kl−1 + 2l−2kl−2 + · · · + 2k1 + k0 the scalar\nmultiple kP can be computed by\nQ = kP = 2l−1kl−1P + 2l−2kl−2P + · · · + 2k1P + k0P.\nBy factoring out 2, the result is\nQ = (2l−2kl−1P + 2l−3kl−2P + · · · + k1P)2 + k0P.\nBy repeating this operation it is seen that\nQ = (· · · ((kl−1P)2 + kl−2P)2 + · · · + k1P)2 + k0P\nwhich can be computed by the well known (left-to-right) double and add method for\nscalar multiplication shown in Algorithm 3.\nTwo basic operations required for elliptic curve scalar multiplication are point\nADD and point DOUBLE. The mathematical definitions for these operations are derived\nfrom the curve equation in (3). Consider the points P1 and P2 represented by the\ncoordinate pairs (x1, y1) and (x2, y2) respectively. Then the coordinates, (xa, ya), of\npoint Pa = P1 + P2 (or ADD(P1, P2)) are computed using the equations\nxa =\n\f y1 + y2\nx1 + x2\n\r2\n+ y1 + y2\nx1 + x2\n+ x1 + x2 + α\nya =\n\f y1 + y2\nx1 + x2\n\r\n(x1 + xa) + xa + y1.\nSimilarly the coordinates (xd, yd) of point Pd = 2P1 (or DOUBLE(P1)) are com-\nputed using the equations\nxd = x2\n1 +\n\f β\nx2\n1\n\r\nyd = x2\n1 +\n\f\nx1 + y1\nx1\n\r\nxd + xd.\n"
        },
        {
            "page_number": 17,
            "text": "WIRELESS NETWORK SECURITY\n9\nAlgorithm 3.\nScalar Multiplication by Double and Add Method\nInput: Integer k = (kl−1, kl−2, . . . , k1, k0)2, Point P\nOutput: Point Q = kP\nQ ←O;\nif (kl−1 == 1) then\nQ ←P;\nfor i = l −2 downto 0 do\nQ ←DOUBLE(Q);\nif (ki == 1) then\nQ ←ADD(Q, P);\nSo the addition of two points can be computed using two field multiplications, one\nfield squaring, eight field additions and one field inversion. The double of a point can\nbe computed using two field multiplications, one field squaring, six field additions and\none field inversion.\n3.\nHIGH PERFORMANCE FINITE FIELD ARITHMETIC\nIn order to optimize the curve arithmetic discussed in Section 2.2 the underlying\nfield operations must be implemented in a fast and efficient way. The required field\narithmetic operations are addition, multiplication, squaring and inversion. Each of\nthese operations have been implemented in hardware for use in the prototype discussed\nin Section 5. Generally speaking, field multiplication has the greatest effect on the\nperformance of the entire elliptic curve scalar multiplication.1 For this reason, focus\nwill be primarily on the field multiplier when discussing hardware architectures for\nfield arithmetic.\nThis section is organized as follows. Section 3.1 presents a hardware architecture\ndesigned to perform finite field multiplication. In Section 3.2 the ideas presented for\nmultiplication are extended to create a hardware architecture optimized for squaring.\nSection 3.3 gives a method for inversion due to Itoh and Tsujii. This method does not\nrequire any additional hardware but instead uses the multiplication and squaring units\ndescribed in Sections 3.1 and 3.2. Section 3.4 gives a description of a comparator/adder\n1 Inversion takes much longer than multiplication, but its effect on performance can be greatly reduced\nthrough use of projective coordinates. This is discussed in greater detail in Section 4.1.\n"
        },
        {
            "page_number": 18,
            "text": "10\nJONATHAN LUTZ and M. ANWARUL HASAN\nwhich both compares and adds finite field elements. Finally, Section 3.5 summarizes\nresults gleaned from a hardware prototype of each arithmetic unit/routine.\n3.1. Multiplication\nIn [11] a digit serial multiplier is proposed which is based on look-up tables.\nThis method was implemented in software for the field GF(2163) and reported in [14].\nTo the best of our knowledge this performance of 0.540 µ-seconds for a single field\nmultiplication is the fastest reported result for a software implementation.\nIn this\nsection the possibilities of using this look-up table-based algorithm in hardware will be\nexplored.\nFirst to be described in this section is the algorithm used for multiplication. Then\nwe present a hardware structure designed to compute R(x)W(x) mod F(x) where\nR(x) and W(x) are polynomials with degrees g −1 and m −1 respectively and\ng << m. A description of the multiplier’s data path follows. In conclusion there will\nbe a discussion behind the reasons for the choice of digit sizes.\nMultiplication Algorithm:\nThe computations of\nP(x) ←xgP(x)\nmod F(x) and\nP(x) ←Bk(x)A(x) + P(x)\nmod F(x)\nfrom the for loop of Algorithm 1 on page 7 can be broken up into the following steps.\nV1 = xg\nm−g−1\n\u0001\ni=0\npixi,\nV2 = xg\nm−1\n\u0001\ni=m−g\npixi\nmod F(x)\nV3 = Bk(x)A(x)\nmod F(x) and\nP(x) = V1 + V2 + V3\nNote that V1 is a g-bit shift of the lower m −g bits of P(x). V2 is a g-bit shift of\nthe upper g bits of P(x) followed by a modular reduction. V3 requires a polynomial\nmultiplication and reduction where the operand polynomials have degree g −1 and\nm −1. Algorithm 1 can be modified to create Algorithm 4.\nIn [11] polynomials V2 and V3 are computed with the assistance of look-up tables\nmainly for software implementation. The look-up tables used to compute V2 and V3 are\nreferred to as the M-Table and T-Table respectively. The M-Table is addressed by the\nbit string (pm−1, pm−2, . . . , pm−g) interpreted as the integer 2g−1pm−1+2g−2pm−2+\n· · · + pm−g. Similarly the T-Table is addressed by the coefficients of Bk(x), or the\ninteger Bk(x = 2). The elements of the M-Table are a function of the reduction\npolynomial F(x) and can be precomputed. The elements of the T-Table are a function\n"
        },
        {
            "page_number": 19,
            "text": "WIRELESS NETWORK SECURITY\n11\nAlgorithm 4.\nEfficient Group Level Multiplication\nInput: A(x), B(x), and F(x)\nOutput: P(x) = A(x)B(x) mod F(x)\nP(x) ←Bs−1(x)A(x) mod F(x);\nfor k = s −2 downto 0 do\nV1 ←xg \u000em−g−1\ni=0\npixi;\nV2 ←xg \u000em−1\ni=m−g pixi mod F(x);\nV3 ←Bk(x)A(x) mod F(x);\nP(x) ←V1 + V2 + V3;\nof A(x) and hence are dynamic. These values need to be computed each time a new\nA(x) is used.\nComputation of R(x)W(x) mod F(x):\nInstead of using tables, below the polyno-\nmials V2 and V3 are computed on the fly. The computation of V2 and V3 are similar\nin that they both require a multiplication of two polynomials followed by a reduction,\nwhere the first polynomial has degree g −1 and the other has degree less than m. This\nis obvious for V3 and can be shown easily for V2. Note that\nV2 = pm−1xm+g−1 + · · · + pm−g+1xm+1 + pm−gxm\nmod F(x)\n= xm \u000f\npm−1xg−1 + · · · + pm−g+1x + pm−g\n\u0010\nmod F(x).\nThe field reduction polynomial F(x) = xm + xd + · · · + 1 provides us the equality\nxm ≡xd + · · · + 1. Substituting for xm we see that\nV2 =\n\u000f\nxd + · · · + 1\n\u0010 \u000f\npm−1xg−1 + · · · + pm−g+1x + pm−g\n\u0010\nmod F(x).\nProvided d + g < m, V2 results in a polynomial of degree less than m which does\nnot need to be reduced. Since d is relatively small for all five NIST polynomials, it is\nreasonable to assume that d+g < m. For the remainder of this chapter, this assumption\nis used.\nWith this said, the following method can be used to compute both V2 and V3.\nConsider the polynomial multiplication and reduction R(x)W(x) mod F(x) where\n"
        },
        {
            "page_number": 20,
            "text": "12\nJONATHAN LUTZ and M. ANWARUL HASAN\nR(x) = \u000eg−1\ni=0 rixi and W(x) is a polynomial with degree less than m. Then\nR(x)W(x)\nmod F(x) =rg−1(xg−1W(x)\nmod F(x))\n+rg−2(xg−2W(x)\nmod F(x))\n...\n+r1(xW(x)\nmod F(x))\n+r0(W(x)\nmod F(x))\nThe value xiW(x) mod F(x) is just a shifted and reduced version of xi−1W(x)\nmod F(x). So each value xiW(x) mod F(x) can be generated sequentially starting\nwith x0W(x) as shown in Figure 1. When using a reduction polynomial with a low\nHamming weight, such as a trinomial or pentanomial, these terms can be computed\nquickly at very little cost. Once these values are determined, the final result is computed\nusing a g-input modulo 2 adder. The inputs to the adder are enabled by their corre-\nsponding coefficient ri. This is shown in Figure 2. Note that the polynomial xiW(x)\naffects the output of the adder only if the coefficient bit ri is a one. Otherwise the input\nassociated with xiW(x) is driven with zeros.\n= Shift and Reduction\nFigure 1. Generating xiW(x) mod F(x)\nEach individual output bit of the g-operand mod 2 adder is computed using g −1\nXOR gates and g AND gates. The AND gates are used to enable each input bit and the\nXOR gates compute the mod 2 addition. Figure 3 demonstrates how this is done. The\ndepth of the logic in the figure is linearly related to g.\nThis method for multiplication is implemented for computation of both V2 and V3.\nIn the case of V3, the polynomial W(x) has degree m −1 and will change for every\n"
        },
        {
            "page_number": 21,
            "text": "WIRELESS NETWORK SECURITY\n13\nFigure 2. Computing R(x)W(x) mod F(x)\nfield multiplication. For V2 the polynomial W(x) has degree d and is fixed. The value\nd is the degree of the second leading non-zero coefficient of F(x). For reasonable digit\nsizes this computation can be performed in a single clock cycle.\nMultiplier Data Path:\nThe multiplier’s data path connecting the V2 and V3 generators\nalong with the adder used to compute P(x) = V1 + V2 + V3 is shown in Figure 4.\nA buffer is inserted at the output of the V3 generator to separate its delay from the\ndelay of the adder for V1 + V2 + V3. This, in effect, increases the maximum possible\nvalue for the digit size g. If added by itself, this buffer would add a cycle of latency to\nthe multiplier’s performance time. This extra cycle is compensated for by bypassing\nthe P(x) register and driving the multiplier’s output with the output of the 3-operand\nmod2 adder. It is important to note that the delay of the 3-operand mod2 adder is being\nmerged with the delay of the bus which connects the multiplier to the rest of the design.\nIn this case the relatively relaxed bus timing has room to accommodate the delay.\nChoice of Digit Size:\nThe multiplier will complete a multiplication in ⌈m/g⌉clock\ncycles. Since this is a discrete value, the performance may not change for every value of\ng. To minimize cost of the multiplier (which increases with g) the smallest digit size g\nshould be chosen for a given performance ⌈m/g⌉. For example, the digit sizes g = 21\nand g = 22 for field size m = 163 result in the same performance, ⌈163\n21 ⌉= ⌈163\n22 ⌉= 8,\nbut g = 22 requires a larger multiplier.\nImplementation results of a prototype of this multiplier for the field GF(2163) and\nNIST polynomial for various digit sizes are shown in Table 2. For each digit size, the\ntable lists the corresponding cycle performance and resource cost. A maximum digit\n"
        },
        {
            "page_number": 22,
            "text": "14\nJONATHAN LUTZ and M. ANWARUL HASAN\nFigure 3. Computation of a Single Bit in R(x)W(x) mod F(x)\n"
        },
        {
            "page_number": 23,
            "text": "WIRELESS NETWORK SECURITY\n15\nm −g\ng\nP (x)\nV3\nBuffer\nm\nV1\nV2\ngenerate V3\nV1 + V2 + V3\ng + d\n(function of F (x))\ngenerate V2\ng-operand mod 2 adder to\ng-operand mod 2 adder to\nA(x)\nm\ng\nP (x) Register\nB(x)\nFigure 4. Multiplier Data-Path\nsize of g = 41 is a good choice for several reasons. First, as the performance cost of\nthe actual field multiplication decreases, the relative cost of loading and unloading the\nmultiplier increases. So as the digit size increases, its affect on the total performance\n(including time to load and unload the multiplier) decreases. Second, results showed\nthat g > 41 had difficulty meeting timing at the target operating frequency of 66 MHz.\n3.2. Squaring\nWhile squaring is a specific case of general multiplication and can be performed by\nthe multiplier, performance can be improved significantly by optimizing the architecture\nspecifically for the case of squaring. The square of an element a represented by A(x)\ninvolves two mathematical steps. The first is the polynomial multiplication of A(x)\nresulting in\nA2(x) = am−1x2m−2 + · · · + a2x4 + a1x2 + a0.\n"
        },
        {
            "page_number": 24,
            "text": "16\nJONATHAN LUTZ and M. ANWARUL HASAN\nTable 2. Performance/Cost Trade-off for Multiplication over GF(2163)\nDigit\nPerformance\n# LUTs\n# Flip\nSize\nin clock cycles\nFlops\ng = 1\n163\n677\n670\ng = 4\n41\n854\n670\ng = 28\n6\n3,548\n670\ng = 33\n5\n4,040\n670\ng = 41\n4\n4,728\n670\nThe second is the reduction of this polynomial modulo F(x). Assuming that m is an\nodd integer, which is the case for all five NIST recommended binary fields, if the terms\nwith degree greater than m −1 are separated and xm+1 is factored out where possible\nthe result will be A2(x) = Ah(x)xm+1 + Al(x) where\nAh(x) = am−1xm−3 + · · · + a( m+3\n2 )x2 + a( m+1\n2 )\nAl(x) = a( m−1\n2 )xm−1 + · · · + a1x2 + a0,\nThe polynomial Al(x) has degree less than m and does not need to be reduced. The\nproduct Ah(x)xm+1 may have degree as large as 2m −2. The reduction polynomial\ngives us the equality xm = xd + · · · + 1. Multiplying both sides by x, we get xm+1 =\nxd+1 + · · · + x. So\nAh(x)xm+1 = Ah(x)\n\u000f\nxd+1 + · · · + x\n\u0010\n.\nThis multiplication can be performed using a method similar to the one described in\nSection 3.1. The same architecture used to compute R(x)W(x) mod F(x) in the\nmultiplier is used here to compute xm+1Ah(x). The digit size is set to g = d + 2\nand the elements of g-operand mod 2 adder are generated from Ah(x). Ah(x) is in\nturn generated by expanding A(x) (i.e., inserting zeros between the coefficient bits of\nA(x)). Since the digit size is set to d + 2, the multiplication is completed in a single\ncycle. This method only works if d + 2 < m which is the case for each of the NIST\npolynomials. Figure 5 shows the data flow for the squaring operation. Note that the\nflow does not include any buffers and so is implemented in pure combinational logic.\n"
        },
        {
            "page_number": 25,
            "text": "WIRELESS NETWORK SECURITY\n17\nFigure 5. Data-Path of the Squaring Unit\nThe prototype of this squaring unit for field GF(2163) using the NIST reduction\npolynomial runs at 66 MHz and is capable of performing a squaring operation in a\nsingle clock cycle. This implementation requires 330 LUTs and 328 Flip Flops.\n3.3. Inversion\nThe inversion method described inAlgorithm 2 on page 7 requires m−1 squarings\nand m −2 multiplications. In order to accurately estimate the cycle performance of\nthe inversion, consideration must be given to the performance of the multiplication and\nsquaring units as well as the time required to load and unload these units. The architec-\nture of the elliptic curve scalar multiplier will be discussed in detail in Section 5. For\nnow, it is sufficient to know that the arithmetic units are loaded using two independent\nm bit data buses and unloaded using a single m bit data bus. The operands are stored\nin a dual port memory which takes two clock cycles to read from and one cycle to write\nto. These combined makes three cycles that are required to both load and unload any\narithmetic unit. Further analysis assumes that these three cycles remain constant for all\nm. If Cs and Cm denote the number of clock cycles required to complete a squaring\nand multiplication respectively, then an inversion can be completed in\n(Cs + 3)(m −1) + (Cm + 3)(m −2)\nclock cycles. For the field GF(2163) where Cs = 1 and Cm = 4, this translates to 1775\nclock cycles.\nPerformance can be improved by using Algorithm 5 due to Itoh and Tsujii [13].\nThis algorithm is derived from the equation a(−1) ≡a2m −2 ≡\n\f\n22m −1−1\n\r2\n"
        },
        {
            "page_number": 26,
            "text": "18\nJONATHAN LUTZ and M. ANWARUL HASAN\nwhich is true for any non-zero element a ∈GF(2m). From\na2t−1 ≡\n⎧\n⎪\n⎨\n⎪\n⎩\n\u0006\na2t/2−1\u00072t/2 \u0006\na2t/2−1\u0007\nfor t even,\na\n\u0006\na2t−1−1\u00072\nfor t odd,\n(4)\nthe computation required for the exponentiation 22m−1−1 can be iteratively broken\ndown. Algorithm 5 requires ⌊log2(m−1)⌋+H(m−1)−1 multiplications and m−1\nsquarings. Using the notation defined earlier, this translates to\n(Cs + 3)(m −1) + (Cm + 3)(⌊log2(m −1)⌋+ H(m −1) −1)\nclock cycles. For GF(2163) this translates to 711 clock cycles.\nAlgorithm 5.\nOptimized Inversion by Square and Multiply\nInputs:\nField element a ̸= 0,\nBinary representation of m −1 = (ml−1, . . . , m2, m0)2\nOutput:\nb ≡a(−1)\nb ←aml−1;\ne ←1;\nfor i = l −2 downto 0 do\nb ←b2eb;\ne ←2e;\nif (mi == 1) then\nb ←b2a;\ne = e + 1;\nb ←b2;\nNow, the majority of the time spent for each squaring operation is used to load and\nunload the squaring unit (three out of the four cycles). Algorithm 5 requires several\nsequences of repetitive squaring (i.e. computations of the form x2t). These repeated\nsquarings do not require intermediate values to be stored outside the squaring unit. By\nmodifying the squaring unit to support the re-square of an element, most of the memory\naccesses otherwise required to load and unload the squaring unit are eliminated. In fact,\n"
        },
        {
            "page_number": 27,
            "text": "WIRELESS NETWORK SECURITY\n19\nthe squaring unit only needs to be loaded and unloaded once for each multiplication.\nHence the number of clock cycles is reduced to\n(Cs(m −1) + 3(⌊log2(m −1)⌋+ H(m −1) −1))\n+ (Cm + 3)(⌊log2(m −1)⌋+ H(m −1) −1)\nclock cycles. For the field GF(2163) with Cs = 1 and Cm = 4, this results in 252 clock\ncycles.\nThisisacompetitivevaluesinceatypicalhardwareimplementationoftheExtended\nEuclidean Algorithm (EEA) is expected to complete an inversion in approximately 2m\nclock cycles or 326 cycles for GF(2163). This corresponds to a 60 clock cycle reduction\nor 20% performance improvement without requiring hardware dedicated specifically\nfor inversion. Table 3 lists the performance numbers of the previously mentioned\ninversion methods when implemented over the field GF(2163).\nTable 3. Comparison of Various Inversion Methods for GF(2163)\nMethod\n# Squarings\n# Multiplications\n# Cycles\nSquare & Multiply\nm −1\nm −2\n1127\nItoh & Tsujii\nm −1\n⌊log2(m −1)⌋+ H(m) −1\n711\nItoh & Tsujii w/ re-square\nm −1\n⌊log2(m −1)⌋+ H(m) −1\n252\nEEA\n-\n-\n326\nThe actual time to complete an inversion using the ECC co-processor architecture\ndiscussed in Section 5 is 259 clock cycles. The 7 extra cycles are due to control related\ninstructions executed in the micro-sequencer.\n3.4. Comparator/Adder\nThe primary purpose of the Comparator/Adder is to compute the sum of two field\nelements. This is done with an array of m exclusive OR gates. To minimize register\nusage as well as time to complete the addition, the sum of the two operands is the\nonly value stored in a register. In this way, the sum is available immediately after the\noperands are loaded into the Comparator/Adder. In other words, it takes no extra clock\ncycles to complete a finite field addition.\nInadditiontocomputingthesumoftwofinitefieldelements, theComparator/Adder\nalso acts as a comparator. The comparison is performed by taking the logical NOR of\nall the bits in the sum register. If the result is a one, then the sum is zero and the two\noperands are equal. If operand a is set to zero, then operand b can be tested for zero.\n"
        },
        {
            "page_number": 28,
            "text": "20\nJONATHAN LUTZ and M. ANWARUL HASAN\nThe logic depth for the zero detect circuitry (the m-bit NOR gate) is log2(m) and is\nregistered before being sent out of the module. Figure 6 provides a functional diagram\nof the Comparator/Adder.\nFigure 6. Data-Path of the Comparator/Adder\n3.5. Remarks\nIn this section, we have discussed hardware architectures designed to perform finite\nfield addition, multiplication and squaring. Also discussed was an efficient method for\ninversion which uses the squaring and multiplication units. The performance results\nassociated with these arithmetic units are summarized in Table 4.\n4.\nECC SCALAR MULTIPLICATION\nThe section is organized as follows. Section 4.1 introduces projective coordinates\nand discusses some of the reasons for using a projective system. Section 4.2 presents\ntwo methods for recoding the scalar. They are non-adjacent form (NAF) and τ-adic\nnon-adjacent form (τ-NAF).\n4.1. Choice of Coordinate Systems\nProjective coordinates allow the inversion required by each DOUBLE and ADD\nto be eliminated at the expense of a few extra field multiplications. The benefit is\nmeasured by the ratio of the time to complete an inversion to the time to complete a\nmultiplication. The inversion algorithm proposed by Itoh and Tsujii [13] will be used\n"
        },
        {
            "page_number": 29,
            "text": "WIRELESS NETWORK SECURITY\n21\nTable 4. Performance of Finite Field Operations\nOperation\n# Cycles\n# Cycles Including Initial and\n(g = 41)\nFinal Data Movement\nMultiplication\n4\n7\nSquaring\n1\n4\nAddition\n0\n3\nInversion\n256\n259\nand therefore, the above ratio is guaranteed to be larger than ⌊log2(m −1)⌋and could\nbe larger depending on the efficiency of the squaring operations. Therefore, projective\ncoordinates will provide us the best performance for NIST curves. Several flavors of\nprojective coordinates have been proposed over the last few years. The prominent ones\nare Standard [21], Jacobian [4, 12] and L´opez & Dahab [18] projective coordinates.\nIf the affine representation of P be denoted as (x, y) and the projective represen-\ntation of P be denoted as (X, Y, Z), then the relation between affine and projective\ncoordinates for the Standard system is\nx = X\nZ\nand\ny = Y\nZ .\nFor Jacobian projective coordinates the relation is\nx = X\nZ2\nand\ny =\nY\nZ3 .\nFinally for L´opez & Dahab’s, the relation between affine and projective coordinates is\nx = X\nZ\nand\ny =\nY\nZ2 .\nFor L´opez & Dahab’s system the projective equation of the elliptic curve in (3) then\nbecomes\nY 2 + XY Z = X3Z + αX2Z2 + βZ4.\nIt is important to note that when using the left-to-right double and add method for scalar\nmultiplication all point additions are of the form ADD(P, Q). The base point P is never\nmodified and as a result will maintain its affine representation (i.e. P = (x, y, 1)).\nThe constant Z coordinate significantly reduces the cost of point addition (from 14\nfield multiplications down to 10). The addition of two distinct points (X1, Y1, Z1) +\n(X2, Y2, 1) = (Xa, Ya, Za) using mixed coordinates (one projective point and one\n"
        },
        {
            "page_number": 30,
            "text": "22\nJONATHAN LUTZ and M. ANWARUL HASAN\naffine point) is then computed by\nA = Y2 · Z2\n1 + Y1\nB = X2 · Z1 + X1\nC = Z1 · B\nD = B2 · (C + α · Z2\n1)\nZa = C2\nE = A · C\nXa = A2 + D + E\nF = Xa + X2 · Za\nG = Xa + Y2 · Za\nYa = E · F + Za · G\n(5)\nSimilarly, the double of a point (X1, Y1, Z1) is (Xd, Yd, Zd) = 2(X1, Y1, Z1) is com-\nputed by\nZd = Z2\n1 · X2\n1\nXd = X4\n1 + β · Z4\n1\nYd = β · Z4\n1 · Zd + Xd · (α · Zd + Y 2\n1 + β · Z4\n1)\n(6)\nInTable 5, thenumberoffieldoperationsrequiredfortheaffine, Standard, Jacobean\nand L´opez & Dahab coordinate systems are provided. In the table the symbols M, S,\nA and I denote field multiplication, squaring, addition and inversion respectively.\nTable 5. Comparison of Projective Point Systems\nSystem\nPoint Addition\nPoint Doubling\nAffine\n2M + 1S + 8A + 1I\n3M + 2S + 4A + 1I\nStandard\n13M + 1S + 7A\n7M + 5S + 4A\nJacobian\n11M + 4S + 7A\n5M + 5S + 4A\nL´opez & Dahab\n10M + 4S + 8A\n5M + 5S + 4A\nThe projective coordinate system defined by L´opez and Dahab will be used since\nit offers the best performance for both point addition and point doubling.\n4.2. Scalar Multiplication using Recoded Integers\nThe binary expansion of an integer k is written as k = \u000el−1\ni=0 ki2i where ki ∈\n{0, 1}. For the case of elliptic curve scalar multiplication the length l is approximately\nequal to m, the degree of the extension field. Assuming an average Hamming weight,\na scalar multiplication will require approximately l/2 point additions and l −1 point\n"
        },
        {
            "page_number": 31,
            "text": "WIRELESS NETWORK SECURITY\n23\ndoubles. Several recoding methods have been proposed which in effect reduce the\nnumber of additions. In this section two methods are discussed, namely NAF [9, 29]\nand τ-adic NAF [16, 29].\nScalar Multiplication using Binary NAF:\nThe symbols in the binary expansion are\nselected from the set {0, 1}. If this set is increased to {0, 1, −1} the expansion is\nreferred to as signed binary (SB) representation. When using this representation, the\ndouble and add scalar multiplication method must be slightly modified to handle the\n−1 symbol (often denoted as ¯1). If the expansion k′\nl−12l−1 + · · · + k′\n12 + k′\n0 where\nk′\ni ∈{0, 1, ¯1} is denoted by (k′\nl−1, . . . , k′\n1, k′\n0)SB, thenAlgorithm 6 computes the scalar\nmultiple of point P. The negative of the point (x, y) is (x, x + y) and can be computed\nAlgorithm 6.\nScalar Multiplication for Signed Binary Representation\nInput: Integer k = (k′\nl−1, k′\nl−2, . . . , k′\n1, k′\n0)SB, Point P\nOutput: Point Q = kP\nQ ←O;\nif (k′\nl−1 ̸= 0) then\nQ ←k′\nl−1P;\nfor i = l −2 downto 0 do\nQ ←DOUBLE(Q);\nif (k′\ni ̸= 0) then\nQ ←ADD(Q, k′\niP);\nwith a single field addition. The signed binary representation is redundant in the sense\nthat any given integer has more than one possible representation. For example, 17 can\nbe represented by (1001)SB as well as (101¯1)SB.\nInterest here is in a particular form of this signed binary representation called NAF\nor non-adjacent form. A signed binary integer is said to be in NAF if there are no\nadjacent non-zero symbols. The NAF of an integer is unique and it is guaranteed to\nbe no more than one symbol longer than the corresponding binary expansion. The\nprimary advantage gained from NAF is its reduced number of non-zero symbols. The\naverage Hamming weight of a NAF is approximately l/3 [29] compared to that of the\nbinary expansion which is l/2. As a result, the running time of elliptic curve scalar\nmultiplication when using binary NAF is reduced to (l + 1)/3 point additions and l\npoint doubles. This represents a significant reduction in run time.\n"
        },
        {
            "page_number": 32,
            "text": "24\nJONATHAN LUTZ and M. ANWARUL HASAN\nIn [29], Solinas provides a straightforward method for computing the NAF of an\ninteger. This method is given here in Algorithm 7.\nAlgorithm 7.\nGeneration of Binary NAF\nInput: Positive integer k\nOutput: k′ = NAF(k)\ni ←0;\nwhile (k > 0) do\nif (k ≡1 (mod 2)) then\nk′\ni ←2 −(k mod 4);\nk ←k −k′\ni;\nelse\nk′\ni ←0;\nk ←k/2;\ni ←i + 1;\nScalar Multiplication using τ-NAF:\nAnomalous Binary Curves (ABC’s), first pro-\nposed for cryptographic use in [16], provide an efficient implementation when the scalar\nis represented as a complex algebraic number. ABC’s, often referred to as the Koblitz\ncurves, are defined by\ny2 + xy = x3 + αx2 + 1\n(7)\nwith α = 0 or α = 1. The advantage provided by the Koblitz curves is that the DOUBLE\noperation in Algorithm 6 can be replaced with a second operation, namely Frobenius\nmapping, which is easier to perform.\nIf point (x, y) is on a Koblitz curve then it can be easily checked that (x2, y2) is also\non the same curve. Moreover, these two points are related by the following Frobenius\nmapping\nτ(x, y) = (x2, y2)\nwhere τ satisfies the quadratic equation\nτ 2 + 2 = µτ.\n(8)\nIn (8), µ = (−1)1−α and α is the curve parameter in (7) and is 0 or 1 for the Koblitz\ncurves.\n"
        },
        {
            "page_number": 33,
            "text": "WIRELESS NETWORK SECURITY\n25\nThe integer k can be represented with radix τ using signed representation. In this\ncase, the expansion is written\nk = κl−1τ l−1 + · · · κ1τ + κ0,\nwhere κi ∈{0, 1, ¯1}. Using this representation,Algorithm 6 can be rewritten, replacing\nthe DOUBLE(Q) operation with τQ or a Frobenius mapping of Q.\nThe modified\nalgorithm is shown in Algorithm 8. Since τQ is computed by squaring the coordinates\nof Q, this suggests a possible speed up over the DOUBLE and ADD method.\nAlgorithm 8.\nScalar Multiplication for τ-adic Integers\nInput: Integer k = (κl−1, κl−2, . . . , κ1, κ0)τ, Point P\nOutput: Point Q = kP\nQ ←O;\nif (κl−1 ̸= 0) then\nQ ←κl−1P;\nfor i = l −2 downto 0 do\nQ ←τQ;\nif (κi ̸= 0) then\nQ ←ADD(Q, κiP);\nThis complex representation of the integer can be improved further by computing\nits non-adjacent form. Solinas proved the existence of such a representation in [29] by\nproviding an algorithm which computes the τ-adic non-adjacent form or τ-NAF of an\ninteger. This algorithm is provided here in Algorithm 9. In most cases, the input to\nAlgorithm 9 will be a binary integer, say k (i.e. r0 = k and r1 = 0). If k has length l\nthen TNAF(k) will have length 2l, roughly twice the length of NAF(k).\nThe length of the representation generated byAlgorithm 9 can be reduced by either\npreprocessing the integer k, as is done in [29], or by post processing the result. A method\nfor post processing the output of Algorithm 9 is presented here.\nRemember that τ(x, y) = (x2, y2). Since z2m = z for all z ∈GF(2m), it follows\nthat\nτ m(x, y) = (x2m, y2m) = (x, y).\nThis relation gives us the general equality\n(τ m −1)P ≡0\n"
        },
        {
            "page_number": 34,
            "text": "26\nJONATHAN LUTZ and M. ANWARUL HASAN\nAlgorithm 9.\nGeneration of τ-adic NAF\nInput: r0 + r1τ where r0, r1 ∈Z\nOutput: u =TNAF(r0 + r1τ)\ni ←0;\nwhile (r0 ̸= 0 or r1 ̸= 0) do\nif (r0 ≡1 (mod 2)) then\nui ←2 −(r0 −2r1 mod 4);\nr0 ←r0 −ui;\nelse\nui ←0;\nt ←r0;\nr0 ←r1 + µr0/2;\nr1 ←−t/2;\ni ←i + 1;\nwhere P is a point on a Koblitz curve. As a result, any integer k expressed with radix τ\ncan be reduced modulo τ m−1 without changing the scalar multiple kP. This reduction\nis performed easily with a few polynomial additions. Consider the τ-adic integer\nu = u2m−1τ 2m−1 + · · · + um+1τ m+1 + umτ m + um−1τ m−1 + · · · + u1τ + u0.\nFactoring out τ m wherever possible, the result is\nu = (u2m−1τ m−1 + · · · + um+1τ + um)τ m\n+(um−1τ m−1 + · · · + u1τ + u0)\nSubstituting τ m with 1 and combining terms results in\nu = ((u2m−1 + um−1)τ m−1 + · · · + (um+1 + u1)τ + (um + u0).\nThe output of Algorithm 9 is approximately twice the length of the input but may\nbe slightly larger. Assuming the length of the input to be approximately m symbols,\nthe reduction method must be capable of reducing τ-adic integers with length slightly\ngreater 2m. Algorithm 10 describes this method for reduction.\n"
        },
        {
            "page_number": 35,
            "text": "WIRELESS NETWORK SECURITY\n27\nAlgorithm 10.\nReduction mod τ m\nInput: u = ul−1τ l−1 + · · · + u1τ + u0 with m ≤l < 3m\nOutput: v =REDUCE TM(u)\nv ←0;\nif (l > 2m) then\nv ←(ul−1τ l−2m−1 + · · · + u2m+1τ + u2m);\nif (l > m) then\nv ←v + (u2m−1τ m−1 + · · · + um+1τ + um);\nv ←v + (um−1τ m−1 + · · · + u1τ + u0);\nNow the result of Algorithm 10 has length m but is no longer in τ-adic NAF form.\nThere may be adjacent non-zero symbols and the symbols are not restricted to the set\n{0, 1, ¯1}.\nThe input of Algorithm 9 is of the form r0 + r1τ where r0, r1 ∈Z. The output is\nthe τ-adic representation of the input. For v ∈Z[τ] we can write\nv = vm−1τ m−1 + · · · + v2τ 2 + v1τ + v0\n= vm−1τ m−1 + · · · + v2τ 2 + TNAF(v1τ + v0)\nNow the two least significant symbols of v are in τ-adic NAF. Repeating this procedure\nfor every bit in v the entire string can be converted to τ-adic NAF. This process is\ndescribed in Algorithm 11.\nThe output of Algorithm 11 is in τ-adic NAF and has a length of approximately m\nsymbols. If the result is larger than m symbols, it is possible to repeat Algorithms 10\nand 11 to further reduce the length. Algorithms 9, 10 and 11 have been implemented in\nC and were used to generate test vectors for the prototype discussed later in this section.\nDuring testing, it was found that a single pass of these algorithms generates a τ-adic\nrepresentation with average length of m and a maximum length of m + 5.\nLike radix 2 NAF the τ-adic NAF uses the symbol set {1, 0, ¯1} and has an average\nHamming weight of approximately l/3 for an l-bit integer [29]. So Algorithm 8 has a\nrunning time of l/3 point additions and l −1 Frobenius mappings.\nSummary and Analysis:\nA point addition using L´opez & Dahab’s projective coor-\ndinates requires ten field multiplications, four field squarings and eight field additions.\nA point double requires five field multiplications, five field squarings and four field\nadditions. Using this information, the run time for scalar multiplication can be written\nin terms of field operations. Typically scalar multiplication is measured in terms of field\n"
        },
        {
            "page_number": 36,
            "text": "28\nJONATHAN LUTZ and M. ANWARUL HASAN\nAlgorithm 11.\nRegeneration of τ-adic NAF\nInput: v = vm−1τ m−1 + · · · + v1τ + v0\nOutput: w =REGEN TNAF(v)\nw ←v;\ni ←0;\nwhile (wj ̸= 0 for some j ≥i) do\nif (wi == 0) then\ni ←i + 1;\nelse\nt0 ←wi;\nt1 ←wi+1;\nwi ←0;\nwi+1 ←0;\nw ←w+TNAF(t1τ + t0);\ni ←i + 1;\nmultiplications, inversions and squarings, ignoring the cost of addition. In the case of\nthis architecture, field multiplication and squaring are completed quickly enough that\nthe cost of field addition becomes significant. The run times using binary, binary NAF\nand τ-adic NAF representations are shown in Table 6. These values are based on the\ncurve addition and doubling equations defined in (5) and (6) assuming arbitrary curve\nparameters α and β and the average Hamming weights discussed in the previous sec-\ntions. For the case of τ-NAF, a Frobenius mapping is assumed to require three squaring\noperations. The symbols M, S, A and I correspond to field multiplication, squaring,\naddition and inversion respectively. In each case it is assumed that the length of the\ninteger is approximately equal to m.\n5.\nACO-PROCESSORARCHITECTUREFORECCSCALARMULTIPLICATION\nIn the recent past, several articles have proposed various hardware architectures/\naccelerators for ECC.These elliptic curve cryptographic accelerators can be categorized\ninto three functional groups. They are\n"
        },
        {
            "page_number": 37,
            "text": "WIRELESS NETWORK SECURITY\n29\nTable 6. Cost of Scalar Multiplication in terms of Field Operations\nGeneric m\nm = 163\nBinary\n(10M + 7S + 8A)m + I\n1630M + 1141S + 1304A + I\nNAF\n( 25\n3 M + 19\n3 S + 20\n3 A)m + I\n1359M + 1033S + 1087A + I\nτ-NAF\n( 10\n3 M + 13\n3 S + 8\n3A)m + I\n544M + 706S + 435A + I\n1. Accelerators which use general purpose processors to implement curve oper-\nations but implement the finite field operations using hardware. References\n[2] and [30] are examples of this. Both of these implementations support the\ncomposite field GF(2155).\n2. Accelerators which perform both the curve and field operations in hardware\nbut use a small field size such as GF(253). Architectures of this type include\nthose proposed in [28] and [8]. In [28], a processor for the field GF(2168) is\nsynthesized, but not implemented. Both works discuss methods to extend their\nimplementation to a larger field size but do not actually do so.\n3. Accelerators which perform both curve and field operations in hardware and use\nfields of cryptographic strength such as GF(2163). Processors in this category\ninclude [3, 10, 17, 25, 27].\nThe work discussed in this section falls into category three. The architectures pro-\nposed in [25] and [27] were the first reported cryptographic strength elliptic curve\nco-processors. Montgomery scalar multiplication with an LSD multiplier was used\nin [27]. In [25] a new field multiplier is developed and demonstrated in an elliptic\ncurve scalar multiplier. In both [17] and [3] parameterized module generation is dis-\ncussed. To the best of our knowledge the architecture proposed in [10] offers the fastest\nscalar multiplication using FPGA technology at 0.144 milliseconds. This architecture\nuses Montgomery scalar multiplication with L´opez and Dahab’s projective coordinates.\nThey use a shift and add field multiplier but also compare LSD and Karatsuba multi-\npliers.\nThis section describes a hardware architecture for elliptic curve scalar multiplica-\ntion. The architecture uses projective coordinates and is optimized for scalar multipli-\ncation over the Koblitz curves using the arithmetic routines discussed in Section 3 to\nperform the field arithmetic.\n5.1. Co-processor Architecture\nThe architecture, which is detailed in this section, consists of several finite field\narithmetic units, field element storage and control logic. All logic related to finite field\narithmetic is optimized for specific field size and reduction polynomial. Internal curve\ncomputations are performed using L´opez & Dahab’s projective coordinate system.\n"
        },
        {
            "page_number": 38,
            "text": "30\nJONATHAN LUTZ and M. ANWARUL HASAN\nWhile generic curves are supported, the architecture is optimized specifically for the\nspecial Koblitz curves.\nThe processor’s architecture consists of the data path and two levels of control.\nThe lower level of control is composed of a micro-sequencer which holds the routines\nrequired for curve arithmetic such as DOUBLE and ADD. The top level control is im-\nplemented using a state machine which parses the scalar and invokes the appropriate\nroutines in the lower level control. This hierarchical control is shown in Figure 7.\nFigure 7. Co-Processor’s Hierarchical Control Path\nCo-processor Data Path\nThe data path of the co-processor consists of three finite field arithmetic units as\nwell as space for operand storage. The arithmetic units include a multiplier, adder,\nand squaring unit. Each of these are optimized for a specific field and corresponding\nfield polynomial. In an attempt to minimize time lost to data movement, the adder and\nmultiplier are equipped with dual input ports which allow both operands to be loaded\nat the same time (the squaring unit requires a single operand and cannot benefit from\nan extra input bus). Similarly, the field element storage has two output ports used to\nsupply data to the finite field units. In addition to providing field element storage, the\nstorage unit provides the connection between the internal m-bit data path and the 32-bit\nexternal world. Figure 8 shows how the arithmetic units are connected to the storage\nunit.\nThe internal m-bit busses connecting the storage and arithmetic units are controlled\nto perform sequences of field operations. In this way the underlying curve operations\nDOUBLE and ADD as well as field inversion are performed.\nField Element Storage:\nThe field element storage unit provides storage for curve\npoints and parameters as well as temporary values. Parameters required to perform\n"
        },
        {
            "page_number": 39,
            "text": "WIRELESS NETWORK SECURITY\n31\nFigure 8. Co-Processor Data-Path\nelliptic curve scalar multiplication include the field elements α and β and coordinates\nof the base point P. Storage will also be required for the coordinates of the scalar\nmultiple Q. The point addition routine developed for this design also requires four\ntemporary storage locations for intermediate values. Figure 9 shows how the storage\nspace is organized.\nFigure 9. Field Element Storage\nThe top eight field element storage locations are implemented using 32-bit dual-\nport RAMs generated by the Xilinx Coregen tool and the bottom three storage locations2\n2 These locations are shaded gray in Figures 9 and 10.\n"
        },
        {
            "page_number": 40,
            "text": "32\nJONATHAN LUTZ and M. ANWARUL HASAN\nare made of register files with 32-bit register widths. The dual 32-bit/m-bit interface\nsupport is achieved by instantiating ⌈m\n32⌉dual-port storage blocks (either memories\nor register files) with 32-bit word widths as shown in Figure 10. The figure assumes\nm = 163. If the 32-bit storage locations in Figure 10 are viewed as a matrix then the\nrows of the matrix hold the m-bit field words. Each 32-bit location is accessible by\nthe 32-bit interface and each m-bit location is accessible by the m-bit interface. For\nsimplicity sake the field elements are aligned at 32 byte boundaries.\nFigure 10. 32-bit/163-bit Address Map\nComputation ofτQ:\nInadditiontoprovidingstorage, theregistersinthebottomthree\nm-bit locations are capable of squaring the resident field element. This is accomplished\nbyconnectingthelogicrequiredforsquaringdirectlytotheoutputofthestorageregister.\nThe squared result is then muxed in to the input of the storage register and is activated\nwith an enable signal. Figure 11 provides a diagram of this connection. This allows the\nsquaring operations required to compute τQ to be performed in parallel. Furthermore,\nit eliminates the data movement otherwise required if the squaring unit were to be\nloaded and unloaded for each coordinate of Q. This provides significant performance\nimprovement when using Koblitz curves.\nThe Micro-sequencer\nThe micro-sequencer controls the data movement between the field element storage\nand the finite field arithmetic units. In addition to the fundamental load and store\noperations, it supports control instructions such as jump and branch. The following list\nbriefly summarizes the instruction set supported by the micro-sequencer.\nld: Load operand(s) from storage location into specified field arithmetic unit.\nst: Store result from field arithmetic unit into specified storage location.\nj: Jump to specified address in the micro-sequencer.\n"
        },
        {
            "page_number": 41,
            "text": "WIRELESS NETWORK SECURITY\n33\nFigure 11. Efficient Frobenius Mapping\njr: Jump to specified micro-sequencer address and push current address onto\nthe program counter stack.\nret: Return to micro-sequencer address. The address is supplied by the program\ncounter stack.\nbne: Branch if the last field elements loaded into the ALU are NOT equal.\nnop: Increment program counter but do nothing.\nset: Set internal counter to specified value.\nrsq: Resquares the contents of the squaring unit.\ndbnz: Decrement internal counter and branch if the new value of the counter is\nzero. This opcode also causes the contents of the squaring unit to be resquared.\nA two-pass perl assembler was developed to generate the micro-sequencer bit\ncode. The assembler accepts multiple input files with linked addresses and merges\nthem into one file. This file is then used to generate the bit code. The multiple input file\nsupport allows different versions of the ROM code to be efficiently managed. Different\nimplementations of the same micro-sequencer routine can be stored in different files\nallowing them to be easily selected at compile time.\nMicro-sequencerRoutines:\nThemicro-sequencersupportsthecurvearithmeticprim-\nitives, field inversion as well as a few other miscellaneous routines. The list below\nprovides a summary of routines developed for use in the design.\nPOINT ADD (P, Q): Adds the elliptic curve points P and Q where P is repre-\nsented in affine coordinates and Q is represented using projective coordinates.\nThe result is given in projective coordinates.\n"
        },
        {
            "page_number": 42,
            "text": "34\nJONATHAN LUTZ and M. ANWARUL HASAN\nPOINT SUB (P, Q): Computes the difference Q −P. P is represented using\naffine coordinates and Q is represented using projective coordinates. The result\nis given in projective coordinates. This routine calls the POINT ADD routine.\nPOINT DBL (Q): Doubles the elliptic curve point Q. Both Q and the result\nare in projective coordinates.\nINVERT (X): Computes the inverse of the finite field element X.\nCONVERT (Q): Computes the affine coordinates of an elliptic curve point Q\ngiven the point’s projective coordinates. This routine calls the INVERT routine.\nCOPY P2Q (P, Q): Copies the x and y coordinates of point P to the x and y\ncoordinates of point Q. The z coordinate of point Q is set to 1.\nCOPY MP2Q (P, Q): Computes the x and y coordinates of point −P and copies\nthem to the x and y coordinates of point Q. The z coordinate of point Q is set\nto 1.\nSeveral versions of the POINT ADD routine have been developed.\nThe most\ngeneric one supports any curve over the field GF(2m). In this version, the values\nof α and β are used when computing the sum of two points. This curve also checks\nif Q ̸= P, Q ̸= −P and Q ̸= O. The second version of the point addition routine\nis optimized for a Koblitz curve by assuming α and β are equal to the NIST recom-\nmended values. The number of field multiplications required to compute the addition\nof two points is reduced from 10 to 9. The third version of the routine is optimized\nfor a Koblitz curve and also forgoes the checks of point Q. If the base point P has a\nlarge prime order and the integer k is less than this order3, it will never be the case that\nQ = ±P or Q = O. This final version of the routine is the fastest of the three routines\nand is the one used to achieve the results reported at the end of the section.\nTop Level Control\nThe routines listed above along with the POINT FRB(Q) operation are invoked\nby the top level state machine. The POINT FRB(Q) routine computes the Frobenius\nmap of the point Q. This operation is not as complex as the other operations and is not\nimplemented in the micro-sequencer. It is invoked by the top level state machine all\nthe same.\nThe state machine parses the scalar k and calls the routines as needed. Since\nintegers in NAF and τ-NAF require use of the symbol −1 (denoted ¯1), the scalar\nrequires more than just an m-bit register for storage. In the implementation given here,\neach symbol in the scalar is represented using two bits; one for the magnitude and one\nfor the sign. Table 7 provides the corresponding representation. For each bit ki in the\nscalar k the magnitude is stored in the register k(m)\ni\nand the sign is stored in register\n3 These are fair assumptions since the security of the ECC implementation relies on these properties.\n"
        },
        {
            "page_number": 43,
            "text": "WIRELESS NETWORK SECURITY\n35\nTable 7. Representation of the Scalar k\nSymbol\nMagnitude\nSign\n0\n0\n-\n1\n1\n0\n¯1\n1\n1\nk(s)\ni\n. Table 8 provides example representations for integers in binary form, NAF, and\nτ-adic NAF using m = 8.\nTable 8. Example Representations of the Scalar\nk\nk(m)\nk(s)\n(01001100)2\n(01001100)2\n(00000000)2\n(0100¯1010)NAF\n(01001010)2\n(00001000)2\n(0100¯1010)τ−NAF\n(01001010)2\n(00001000)2\nThe top level state machine is designed to support binary, NAF and τ-adic NAF\nrepresentations of the scalar. This effectively requires the state machine to perform\nAlgorithms 3, 6 and 8. By taking advantage of the similarities between these algorithms,\nthe top level state machine can perform this task with the addition of a single mode.\nThis is shown in Algorithm 12. The algorithm is written in terms of the underlying\ncurve and field primitives provided by the micro-sequencer (listed in Section 5.1).\nThe first step of Algorithm 12 is to search for the first non-zero bit in k(m). Once\nfound, either P or −P is copied to Q depending on the sign of the non-zero bit. The\nwhile loop then iterates over all the remaining bits in the scalar performing “doubles\nand adds” or “Frobenius mappings and adds” depending on the mode. Since the curve\narithmetic is performed using projective coordinates, the result must be converted to\naffine coordinates at the end of computation.\nChoice of Field Arithmetic Units\nThe use of redundant arithmetic units, specifically field multipliers, has been sug-\ngested in [3] and should be considered when designing an elliptic curve scalar multiplier.\n"
        },
        {
            "page_number": 44,
            "text": "36\nJONATHAN LUTZ and M. ANWARUL HASAN\nAlgorithm 12.\nState Machine Algorithm\nInputs:\nk(m) = (k(m)\nl−1, k(m)\nl−2, . . . , k(m)\n1\n, k(m)\n0\n)2,\nk(s) = (k(s)\nl−1, k(s)\nl−2, . . . , k(s)\n1 , k(s)\n0 )2,\nPoint P and mode (NAF or τ-NAF)\nOutput:\nPoint Q = kP\ni ←l −1;\nwhile (k(m)\ni\n== 0) do\nk ←i −1;\nif (k(s)\ni\n== 1) then\nCOPY MP2Q(P, Q);\nelse\nCOPY P2Q(P, Q);\ni ←i −1;\nwhile (i ≥0) do\nif (mode == τ-NAF) then\nQ ←POINT FRB(Q);\nelse\nQ ←POINT DBL(Q);\nif (k(m)\ni\n== 1) then\nif (k(s)\ni\n== 1) then\nQ ←POINT SUB(Q, P);\nelse\nQ ←POINT ADD(Q, P);\ni ←i −1\nQ ←CONVERT(Q);\n"
        },
        {
            "page_number": 45,
            "text": "WIRELESS NETWORK SECURITY\n37\nIt seems the advantage provided remains purely theoretical. This can be seen by exam-\nining the top performing ECC multipliers in [10] and [27], both of which use a single\nfield multiplier. Reasons for doing the same for this ECC accelerator are twofold. (1)\nOne of the limiting factors for the performance of the design is data movement. As\nshown in Figures 12 and 13 the bus usage for point addition and point doubling is very\nhigh (83% and 80% respectively). If another multiplier is added to the design there\nmay not be enough free bus cycles to capitalize on the extra computational power. For\nthe field GF(2163), the multiplier computes a product in four clock cycles and requires\nthree cycles to load and unload the unit. If a second multiplier is added, then two\nmultiplications can be completed in four cycles but six cycles are required to unload\nthe multiplier. (2) Many of the multiplications in point addition and point doubling\nare dependent on each other and must be performed in sequence. For this reason, the\nsecond multiplier may sit idle much of the time. The combination of these observations\nseems to argue against the use of multiple field multiplication units in the design.\n5.2. FPGA Prototype\nA prototype of the architecture has been implemented for the field GF(2163) using\nthe NIST recommended field polynomial. The design was coded using Verilog HDL\nand synthesized using Synopsys FPGA Compiler II. Xilinx’ Foundation software was\nused to place, route and time the netlist. The prototype was designed to run at 66 MHz\non a Xilinx’Virtex 2000E FPGA.\nThe resulting design was verified on the Rapid Prototyping Platform (RPP) pro-\nvided by Canadian Microelectronics Corporation (CMC) [5, 6]. The hardware/software\nsystemincludesanARMIntegrator/LM-XCV600E+(boardwithaVirtex2000EFPGA)\nand an ARM Integrator/ARM7TDMI (board with an ARM7 core) connected by the\nARM Integrator/AP board. The design was connected to an AHB slave interface which\nmade it directly accessible by the ARM7 core. Stimulated by compiled C-code, the\ncore read from and wrote to the prototype. The Integrator/AP’s system clock had a\nmaximum frequency of 50 MHz. In order to run our design at 66 MHz it was necessary\nto use the oscillator generated clock provided with the Integrator/LM-SCV600E+. The\ndata headed to and coming from the design was passed across the two clock domains.\n5.3. Results\nTable 9 shows the performance in clock cycles of the prototypes field and curve\noperations. These values were gathered using a field multiplier digit size of g = 41.\nNote that the multiple instantiations of the squaring logic allow for the Frobenius\nmapping of a projective point to be completed in a single cycle. This significantly\nimproves the performance of scalar multiplication when using the Koblitz curves.\nThe prototype of the scalar multiplier has been implemented using several digit\nsizes in the field multiplier.\nTable 10 reports the area consumption and resulting\nperformance of the architecture given the different digit sizes.\nTable 11\nprovides a comparison of published performance results for scalar multiplication.\n"
        },
        {
            "page_number": 46,
            "text": "38\nJONATHAN LUTZ and M. ANWARUL HASAN\nFigure 12. Utilization of Finite Field Units for Point Addition\n"
        },
        {
            "page_number": 47,
            "text": "WIRELESS NETWORK SECURITY\n39\nFigure 13. Utilization of Finite Field Units for Point Doubling\n"
        },
        {
            "page_number": 48,
            "text": "40\nJONATHAN LUTZ and M. ANWARUL HASAN\nTable 9. Performance of Field and Curve Operations\nOperation\n# Cycles\n(g = 41)\nPoint Addition\n79\nPoint Subtraction\n87\nPoint Double\n68\nFrobenius Mapping\n1\nTable 10. Performance and Cost Results for Scalar Multiplication\nMultiplier\nDigit\n# LUTs\n# FFs\nBinary\nNAF\nτ-NAF\nSize\n(ms)\n(ms)\n(ms)\ng = 4\n6,144\n1,930\n1.107\n0.939\n0.351\ng = 14\n7,362\n1,930\n0.446\n0.386\n0.135\ng = 19\n7,872\n1,930\n0.378\n0.329\n0.113\ng = 28\n8,838\n1,930\n0.309\n0.272\n0.090\ng = 33\n9,329\n1,930\n0.286\n0.252\n0.083\ng = 41\n10,017\n1,930\n0.264\n0.233\n0.075\nTable 11. Comparison of Published Results\nImplementation\nField\nFPGA\nScalar Mult. (ms)\nS. Okada et. al. [25]\nGF(2163)\nAltera EPF10K250\n45\nLeong & Leung [17]\nGF(2155)\nXilinx XCV1000\n8.3\nM. Bednara et. al. [3]\nGF(2191)\nXilinx XCV1000\n0.27\nOrlando & Paar [27]\nGF(2167)\nXilinx XCV400E\n0.210\nN. Gura et. al. [10]\nGF(2163)\nXilinx XCV2000E\n0.144\nOur design (g = 14)\nGF(2163)\nXilinx XCV2000E\n0.135\nOur design (g = 41)\nGF(2163)\nXilinx XCV2000E\n0.075\n"
        },
        {
            "page_number": 49,
            "text": "WIRELESS NETWORK SECURITY\n41\nThe performance of 0.144 ms reported in [10] is the fastest reported scalar multiplication\nusing FPGA technology. The design presented in this report provides almost double\n(0.075 ms) the performance for the specific case of Koblitz curves.\nThe co-processor discussed in this chapter requires approximately half the CLBs\nused in the co-processor of [10] using the same FPGA. It must be noted that the co-\nprocessor presented in [10] is robust in that it supports all fields up to GF(2256). In\napplications where support for a only single field size is required it is overkill to support\nelliptic curves over many fields. In scenarios such as this, this new elliptic curve co-\nprocessor offers an improved cost effective solution.\n6.\nCONCLUDING REMARKS\nIn this chapter, the development of an elliptic curve cryptographic co-processor\nhas been discussed. The co-processor takes advantage of multiplication and squaring\narithmetic units which are based on the look-up table-based multiplication algorithm\nproposed in [11]. Field elements are represented with respect to the polynomial ba-\nsis. While the base point and resulting scalar are given in affine coordinates, internal\narithmetic is performed using projective coordinates. This choice of coordinate system\nallows the scalar multiple of a point to be computed with a single field inversion allevi-\nating the need for a highly efficient inversion method. The processor was designed to\nsupport signed, unsigned and τ-NAF integer representation. All curves over a specific\nfield are supported, but the architecture is optimized specifically for the Koblitz curves.\n7.\nACKNOWLEGEMENTS\nThis work was supported in part by the Security Technology Center in the Semi-\nconductor Products Sector of Motorola. Dr. Hasan’s work was supported in part by\nNSERC. Pieces of the work were presented at SPIE 2003 [19] and ITCC 2004 [20].\n8.\nREFERENCES\n1. Wireless Application Protocol - Version 1.0, 1998.\n2. G. B. Agnew, R.C. Mullin, and S. A. Vanstone. An implementation of elliptic curve cryptosystems\nover F2155. IEEE Journal on Slected Areas in Communications, 11:804–813, June 1993.\n3. Marcus Bednara, Michael Daldrup, Joachim von zur Gathen, Jamshid Shokrollahi, and Jurgen Teich.\nImplementation of elliptic curve cryptographic coprocessor over GF(2m) on an FPGA. In International\nParallel and Distributed Processing Symposium: IPDPS Workshops, April 2002.\n4. D. Chudnovsky and G. Chudnovsky. Sequences of numbers generated by addition in formal groups\nand new primality and factoring tests. Advances in Applied Mathematics, 1987.\n5. Canadian Microelectronics Corporation. CMC Rapic-Prototyping Platform: Design Flow Guide,\n2002.\n6. Canadian Microelectronics Corporation. CMC Rapic-Prototyping Platform: Installation Guide, 2002.\n7. T. Dierks and C. Allen. The TLS Protocol - Version 1.0 IETF RFC 2246, 1999.\n"
        },
        {
            "page_number": 50,
            "text": "42\nJONATHAN LUTZ and M. ANWARUL HASAN\n8. Lijun Gao, Sarvesh Shrivastava, and Gerald E. Sobelman. Elliptic curve scalar multiplier design using\nFPGAs. In Cryptographic Hardware and Embedded Systems (CHES), 1999.\n9. Daniel M. Gordon. A survey of fast exponentiation methods. J. Algorithms, 27(1):129–146, 1998.\n10. Nils Gura, Sheueling Chang Shantz, Hans Eberle, Summit Gupta, Vipul Gupta, Daniel Finchelstein,\nEdouard Goupy, and Douglas Stebila. An end-to-end systems approach to elliptic curve cryptography.\nIn Cryptographic Hardware and Embedded Systems (CHES), 2002.\n11. M. Anwarul Hasan.\nLook-up table-based large finite field multiplication in memory constrained\ncryptosystems. IEEE Transactions on Computers, 49(7), July 2000.\n12. IEEE. P1363: Editorial Contribution to Standard for Public Key Cryptography, February 1998.\n13. T. Itoh and S. Tsujii. A fast algorithm for computing multiplicative inverses in GF(2m) using normal\nbases. Information and Computing, 78(3):171–177, 1988.\n14. Brian King. An improved implementation of elliptic curves over GF(2n) when using projective point\narithmetic. In Selected Areas in Cryptography, 2001.\n15. Neal Koblitz. Elliptic curve cryptosystems. Mathematics of Computation, 1987.\n16. Neal Koblitz. CM curves with good cryptographic properties. In Advances in Cryptography, Crypto\n’91, pages 279–287. Springer-Verilag, 1991.\n17. Philip H. W. Leong and Ivan K. H. Leung. A microcoded elliptic curve processor using FPGA\ntechnology. IEEE Transactions on VLSI Systems, 10(5), October 2002.\n18. Julio Lopez and Ricardo Dahab. Improved algorithms for elliptic curve arithmetic in GF(2n). In\nSelected Areas in Cryptography, pages 201–212, 1998.\n19. Jonathan Lutz and Anwarul Hasan. High performance finite field multiplier for cryptographic applica-\ntions. In SPIE’s Advanced Signal Processing Algorithms, Architectures, and Implemenations, Volume\n5205, pages 541-551, 2003.\n20. Jonathan Lutz and Anwarul Hasan. High performance fpga based elliptic curve cryptographic co-\nprocessor. In IEEE International Conference on Information Technology (ITCC), Volume II, pages\n486-492, 2004.\n21. Alfred Menezes. Elliptic curve public key cryptosystems. Kluwer Academic Publishers, 1993.\n22. A. Menezes, E.Teske,A.Weng. Weak Fields for ECC. Technical Report CORR 2003-15, Centre forAp-\nplied Cryptographic Research, University of Waterloo, 2003. See http://www.cacr.math.uwaterloo.ca\n23. Victor Miller. Uses of elliptic curves in cryptography. In Advances in Cryptography, Crypto ’85, 1985.\n24. NIST. FIPS 186-2 draft, Digital Signature Standard (DSS), 2000.\n25. Souichi Okada, Naoya Torii, Kouichi Itoh, and Masahiko Takenaka. Implementation of elliptic curve\ncryptographic coprocessor over GF(2m) on an FPGA. In Cryptographic Hardware and Embedded\nSystems (CHES), pages 25–40. Springer-Verlag, 2000.\n26. OpenSSL. See http://www.openssl.org.\n27. Gerardo Orlando and Christof Paar. A high-performance reconfigurable elliptic curve processor for\nGF (2m). In Cryptographic Hardware and Embedded Systems (CHES), 2000.\n28. Martin Christopher Rosner. Elliptic curve cryptosystems on reconfigurable hardware. Master’s thesis,\nWorcester Polytechnic Institute, 1998.\n29. Jerome A. Solinas. Improved algorithms for arithmetic on anomalous binary curves. In Advances in\nCryptography, Crypto ’97, 1997.\n30. S. Sutikno, R. Effendi, and A. Surya. Design and implemntation of arithmetic processor F2155 for\nelliptic curve cryptosystems. In IEEEAsia-Pacific Conference on Circuits adn Systems, pages 647–650,\nNovember 1998.\n"
        },
        {
            "page_number": 51,
            "text": "2\nAN ADAPTIVE ENCRYPTION PROTOCOL IN\nMOBILE COMPUTING\nHanping Lufei and Weisong Shi\nDepartment of Computer Science\nWayne State University\nE-mails: hlufei@wayne.edu, weisong@wayne.edu\nUse of encryption for secure communication plays an important role in building applications\nin mobile computing environments. With the emergence of more and more heterogeneous\ndevices and diverse networks, it is difficult, if not impossible, to use a one-size-fits-all en-\ncryption algorithm that always has the best performance in such a dynamic environment.\nWe envision that the only way to accelerate the deployment of encryption algorithms is\nproviding a flexible adaptation of choosing an appropriate encryption algorithm from mul-\ntiple diverse algorithms according to the characteristics of heterogeneous mobile computing\nenvironments.\nBased on the Fractal framework [1], we propose and implement an adaptive encryption\nprotocol, whichcandynamicallychooseaproperencryptionalgorithmbasedonapplication-\nspecific requirements and device configurations. Performance evaluation results show that\nin the divergent environment with different devices and applications, the adaptive encryption\nprotocol successfully selects the best encryption algorithm from the candidate algorithms,\nand minimizes the total time overhead and insures the security as well.\n1.\nINTRODUCTION\nUse of encryption for secure communication is important for building distributed\napplications. With the development of computer and communication technologies,\nmore and more heterogeneous devices, like desktops, laptops, PocketPCs, and cellular\nphones are connected to the Internet using diverse networks, like Ethernet, Wi-Fi, Blue-\ntooth, 3G/4G wireless technology. On one hand, different technologies have different\ncharacteristics. On the other hand, a heterogeneous environment makes it possible\nto dynamically change between different devices and network environments. For in-\nstance, a person uses a laptop with a cable modem at home, a cell phone with 3G/4G\nor Bluetooth on the way to the office, a desktop with Ethernet LAN in the office and a\nPDA with Wi-Fi in the meeting room. Diverse network connections and heterogeneous\n"
        },
        {
            "page_number": 52,
            "text": "44\nHANPING LUFEI and WEISONG SHI\ndevices demand the adaptation functionality in a distributed fashion because no one-\nsize-fits-all single function or protocol can perform well over all these networks and\ndevices. Although many symmetric or asymmetric encryption algorithms have been\nproposed, none of them takes the diversities of device and network into the design. It\nis difficult, if not impossible, to build a one-size-fits-all encryption protocol which can\nrun well in the dynamic environment. The only way to accelerate the deployment of\nencryption algorithms is to provide a flexible adaptation of choosing multiple diverse\nalgorithms.\nAdaptation has been considered as a general approach to address the mismatch\nproblem between clients and servers [2, 3, 4, 5]. From the perspective of adaptation lo-\ncations, some of them propose the in-network adaptation, such as CANS [2], Rover [3],\nOdyssey [4], and Active Names [5], which focus on how to do the adaptation step by\nstep across an overlay path. From the network OSI model’s point of view, some of\nthem work in the network layer [6], which adapts the TCP/IP protocol dynamically ac-\ncording to the changing situations on both ends. The Fractal framework [1], a dynamic\napplication level protocol adaptation approach, utilizes the mobile code technology\nfor protocol adaptation and leverages existing content distribution networks (CDN) for\nprotocol adaptors (mobile codes) deployment. The protocol adaptation in Fractal is\nbased on the assumption that an application protocol is composed of a series of com-\nponents, also called protocol adaptors (PAD). When a protocol needs to be adapted, the\napplication simply needs to add or remove some PADs into or from it. We will give a\nbrief introduction about the Fractal framework in Section 3.\nBased on the Fractal framework, we propose and implement an adaptive encryp-\ntion protocol, which dynamically chooses a proper encryption algorithm based on\napplication-specific requirements and device configurations. Evaluation results show\nthat the adaptive encryption protocol can choose the best encryption algorithm from\nthe candidates to minimize the total time overhead and ensure the security as well.\nThe rest of the chapter is organized as follows. After a brief introduction of back-\nground in Section 2, the Fractal framework and platform of the adaptive encryption\nprotocol are depicted in Section 3. Section 4 describes the adaptation model for the\nadaptive encryption protocol. Performance evaluation and related work are described\nin Section 5 and Section 6 respectively. We summarize the chapter in Section 7.\n2.\nBACKGROUND\nIn the design and implementation of the adaptive encryption protocol, several\nbackground topics are involved, such as: mobile code [7, 8], content distribution net-\nwork [9, 10], protocol adaptation [11, 6, 12], and encryption algorithms. In this section,\nwe explain the general background of each related research field.\n2.1. Mobile Code\nMobile code [8] is defined as the data that can be executed as a program. The code\ncan be pre-compiled for immediate execution on the recipient’s processor, compiled\n"
        },
        {
            "page_number": 53,
            "text": "WIRELESS NETWORK SECURITY\n45\nupon receipt for subsequent execution or interpreted. The mobile code system has been\nused to build a distributed processing environment that is flexible in the communication\nabstractions it provides to applications and to enhance existing distributed applications.\nFor the benefit of mobile code [7], a major asset provided by code mobility is that\nit enables service customization. The ability to request the remote execution of code\nhelps increase application server flexibility without permanently affecting the size or\ncomplexity of the server. In Fractal we implement each protocol adaptor as a mobile\ncode module, which is sent and executed remotely on the client side to build a new\nprotocol allowing the client to talk with the application server.\n2.2. Content Distribution Network\nContent Distribution Networks (CDN) [10] is an intermediate layer of infrastruc-\nture between origin servers and clients. CDN can achieve scalable content delivery by\ndistributing load among its edgeservers, by serving client requests from edgeservers\nthat are close to requests, and by bypassing congested network paths. Currently CDNs\nare only used to deliver Web-based content. In Fractal framework, CDN is used to\ndeliver protocol adaptor (PAD). If we consider the PAD as a Web-based object, most of\nthe current techniques in CDN can be leveraged to the delivery of PAD. Fractal frame-\nwork extends the utilization of CDNs from traditional Web-based content to Web-based\nobjects like mobile code and mobile agent.\n2.3. Protocol Adaptation\nChanging protocols to adapt link condition and network environment is not the\nnew idea, e.g., Reno and Vegas congestion control in TCP/IP protocol [13] is a kind of\nadaptation. More sophisticated protocol adaptation approaches, such as STP proposed\nin [6], but most of them are in the network layer which makes them hard to have\na general view of the whole system status. The problem of adapting to a changing\nnetwork environment is further complicated because changes in network conditions\nare usually transparent to higher layers of the protocol stack. When higher layers, e.g.,\napplication layer, are aware of network variation, protocol adaptation can be done more\nadaptively and intelligently. Based on these observations, Fractal works entirely in the\napplication layer to adapt the application protocol according to heterogeneous client\nenvironments.\n2.4. Three Symmetric-Key Encryption Algorithms\nMany symmetric key encryption algorithms have been proposed. DES, AES, and\nRC4 are three of the most popular shared-key encryption algorithms.\n1. DES/Triple DES [14] Data Encryption Standard is addressed in FIPS PUB 46.\nData are encrypted in 64-bit blocks using a 56-bit key. DES transforms 64-bit\ninput in a series of steps into a 64-bit output. The same steps and the same key\nare used to decrypt the data. With the development of hardware technology,\n"
        },
        {
            "page_number": 54,
            "text": "46\nHANPING LUFEI and WEISONG SHI\nDES shows potential vulnerability to a brute-force attack. Triple DES (3DES)\nis an alternative of traditional DES algorithm. Triple DES provides a security\nlevel of 2112, independent of the key size. National Institutes of Standards\nand Technology (NIST) requires all new applications should use triple DES or\nmore advanced encryption algorithms, while DES is still supported for legacy\napplications. DES can be broken by brute force attack because of the limited\nkey length. Triple DES is secure but with the computation time as three times\nslower than DES. The poor performance of triple DES triggered the call for an\nadvanced encryption standard (AES).\n2. AES [15] AES is a relatively new algorithm compared with DES. Observing\nthat DES is more and more out of date and 3DES is not a long term replacement\ncandidate for the widely used DES algorithm. NIST called a new Advanced\nEncryption Standard (AES).AES is more secure than DES. It can has key length\nas long as 256 bits. It also have high computation efficiency and flexibility to\nbe practical in a wide range of applications. The security level of AES is\n2128,192,256 depending on the used key size, where the AES block sizes are\n128, 192, and 256.\n3. RC4 Stream Cipher [16] RC4 is a contemporary variable key-size stream cipher\nwith byte-oriented operations. It is based on the use of a random permutation.\nKey length is in a range from 1 to 256 bytes. RC4 is easy to be implemented\neven on resource-constraint devices, such as Berkeley Motes and smart cards.\nAdjustment of key length can achieve a tradeoff between running speed and\nsecurity level.\nThere are several other symmetric algorithms have been proposed; however, we\nbelieve these three algorithms are diverse enough to show the basic idea of adaptive\nencryption in this case study.\n3.\nPLATFORM OF THE ADAPTIVE ENCRYPTION PROTOCOL\nThe adaptive encryption protocol is utilized between two communication parties:\napplication server and client. We assume that some clients use legacy applications,\nwhich support only old encryption algorithms, while some clients have more flexibil-\nity to choose different algorithms. Three encryption algorithms, namely DES [14],\nAES [15], RC4 [16] are the candidates of encryption algorithms. The sender side\nadopts the Fractal framework [1] to choose proper encryption algorithms based on their\ndiverse characteristics and different client applications configurations. Note that we\nfocus on how to choose different algorithms in the context of symmetric encryption.\nThe procedure to set up the symmetric key(s) is beyond the scope of this chapter. It\nis very easy to set up the symmetric keys using the Diffie-Hellman [17] key exchange\nmechanism.\nFigure 1 shows the platform of the adaptive encryption protocol including five\ncomponents: Application server, Adaptation proxy, CDN edgeservers, Protocol adap-\n"
        },
        {
            "page_number": 55,
            "text": "WIRELESS NETWORK SECURITY\n47\nAdaptation Proxy\nApplication Server\nCDN\nedgeserver\nPAD-\nAES\nCDN\nedgeserver\nCDN\nedgeserver\nPAD-\n3DES\nPAD-\nRC4\nLaptop Client\nDesktop Client\nPocket PC Client\nP4 2.0GHz\n512MB RAM\n10/100Mbps NIC\nRedHat 8.0\nP4 2.0GHz\n512MB RAM\n10/100Mbps NIC\nRedHat 8.0\nP4 2.0GHz\n512MB RAM\n10/100Mbps NIC\nRedHat 8.0\nP4 3.06GHz\n512MB RAM\n802.11b\nWireless\nFedora Core 2\nP4 2.0GHz\n512MB RAM\n10/100Mbps NIC\nRedHat 8.0\nHP iPAQ h5555\nBluetooth\nadapter\nWindows CE 4.2\nLAN\nWLAN\nBluetooth\nInternet\nFigure 1. Platform for the adaptive encryption protocol.\ntors (PADs), and Clients (e.g., desktop, laptop, PocketPC). The application server is\nthe application service provider. In order to provide the functionality to heteroge-\nneous clients in diverse environments, the application server usually communicates\nwith clients through different encryption protocols. Although the application server\ncan talk in many different encryption protocols, the client may not have the necessary\nprotocol to talk with the sender. To help the client talk with the application server, the\nPAD, which is a protocol adaptor, encapsulates the encryption protocol candidates into\na mobile code module and deploys them across the CDN edgeservers that locates on\nthe edge of the Internet. By downloading and deploying one or more PADs, the client\nis then capable of starting communication with the application server using required\nencryption protocols. On the sender side, we assume the application server has already\ndeployed all PADs in advance. An important issue for the sender is which PADs should\nbe used and where to find them. Close to the application server, an adaptation proxy is\nset up to handle the issues about PAD negotiations. Before the initialization of commu-\nnication between the sender and the client, the client has to negotiate with the adaptation\nproxy to find proper PADs. The client will be asked to provide some metadata about his\nenvironments, such as computing ability, memory space, and network configurations\nto the adaptation proxy. Having these metadata, the adaptation proxy will generate the\nmetadata of the proper PADs for the client and send the metadata of PADs back to the\nclient. Inside these metadata is enough information for the client to download the PADs\nfrom the closest edgeserver of CDNs with which the application server is associated.\nNext, we will give more details about the adaptation proxy.\n"
        },
        {
            "page_number": 56,
            "text": "48\nHANPING LUFEI and WEISONG SHI\n3.1. Adaptation Proxy\nAdaptation proxy plays an important role in the adaptive encryption protocol.\nUsually it is deployed in the same administration domain as the application server and\nis responsible for negotiation with the client. A general structure of the adaptation proxy\nis shown in Figure 2, which includes a negotiation manager module and a distribution\nmanager module. Each module is running as a daemon on the adaptation proxy. Next\nwe will explain the structure and functionality of each module respectively.\nAdaptation Proxy\nAppMeta\nApplication\nServer\nDevMeta\nNtw kMeta\nProtocolCache\nPADMeta\nNegotiation Manager\nDistribution Manager\nClient\nAdaptation Cache\nPAT\nFigure 2. Structure of the adaptation proxy.\nNegotiation Manager As shown in Figure 2, the negotiation manager is the key in the\nadaptation proxy which negotiates with the client. Some application level metadata is\nneeded to be transmitted between the adaptation proxy and the application server, and\nbetween the adaptation proxy and the client to support the negotiation function. We\ndefine these metadata formats in Figure 3. In the rest of the chapter, we will use the\nacronyms in the parentheses to refer to them.\nDevice Metadata (DevMeta) = { Operating system type, CPU type, CPU speed, memory size }\nNetwork Metadata (NtwkMeta) = { Network type, Network bandwidth }\nPAD Metadata (PADMeta) = { PAD ID, PAD size, PAD overhead, Message digest, URL, Parent link, Child link, ... , Child link }\nApplication Metadata (AppMeta) = { Application ID, PADMeta 1, ... , PADMeta n}\nFigure 3. Definitions of metadata.\n"
        },
        {
            "page_number": 57,
            "text": "WIRELESS NETWORK SECURITY\n49\nDevMeta and NtwkMeta, provided by clients, contain the hardware information\nand the network environment of the client. The application server supplies PADMeta\nto the negotiation manager, who holds the general information of each PAD. PAD ID is\na unique identification generated by the application server. PAD overhead consists of\nthe computing overhead at both the client side and server side, and corresponding traffic\noverhead in the network. Message digest is computed using the SHA-1 [18] function\nand used by clients to verify the integrity of the PAD. URL is the link to download\nthe PAD. Note that it is the CDN’s responsibility to find the closest edgeserver which\nholds the PAD, and to redirect the request to that edgeserver. Parent link and Child\nlink are used to build the protocol adaptation topology in the negotiation manager.\nAppMeta is comprised of Application ID, which marks different applications, and some\nPADMeta, which forms a protocol adaptation topology. The application server pushes\nnew AppMeta to the negotiation manager when the protocol adaptation topology is first\ncreated or changed later. Usually the protocol adaptation topology is represented by a\nprotocol adaptation tree (PAT) structure as shown in Figure 2 in the upper box located\nin the negotiation manager. We will give more details about the PAT tree in Section 4.1.\nWhen the negotiation manager receives a request from a client, it first checks its\nadaptation cache, located in the distribution manager. The cache has entries mapping\nclient side information to an array of PADMeta that the client needs. Each mapping\nentry is structured as follows:\n{ DevMeta, Application ID, NtwkMeta } ⇒{ PADMeta 1, ... ,PADMeta n }\nIf the adaptation cache does not have the entry corresponding to the client side metadata,\nthe negotiation manager then will use a path search algorithm described in Section 4.2\nto form a new entry and transfer it to the distribution manager.\nDistribution Manager The distribution manager is in charge of further processing of\nthese PADMeta received from the negotiation manager, updating the adaptation cache,\nand finally sending PADMeta back to the client. When the distribution manager receives\nthe PADMeta generated by the negotiation manager, it inserts message digest and URL\ndata into the PADMeta and hides the parent and child links since the exposure to the\nclient is unnecessary. After the negotiation procedure, which will be discussed in the\nfollowing section, the distribution manager will update the adaptation cache so that the\nnegotiationresultcanbedirectlyretrievedfromthecacheifthesameclientconfiguration\noccurs later. Finally the distribution manager will handle the network communication\ndetails and send these PADMeta back to the client. Next we will explain the interactive\nnegotiation protocol.\n3.2. Interactive Negotiation Protocol\nAn interactive negotiation protocol(INP) is proposed for the interactions among\nthese components, as shown in Figure 4. We assume both the client side and server side\nunderstand the protocol definitions. The application server has pre-deployed PADs in\nthe application context and already pushed the AppMeta to the adaptation proxy, which\n"
        },
        {
            "page_number": 58,
            "text": "50\nHANPING LUFEI and WEISONG SHI\nhas built a PAT inside the negotiation manager. The PADs have been distributed across\nthe CDNs edgeservers.\nAt the beginning of the negotiation, a client first checks its own protocol cache,\nwhich contains some PADMeta saved for previous requests. If there is an entry of the\nprotocol cache which matches the current request, the client will directly start the appli-\ncation communication with the application server. If not, the client sends INIT REQ,\nwhich contains application request in payload, to the adaptation proxy 1 to initialize\nthe protocol negotiation. Each packet has an INP header segment, which is used to\nmaintain the interactive negotiation protocol integrity, and we will omit the details in the\nINP header. The adaptation proxy then sends INIT REP as well as Cli META REQ,\nhaving empty DevMeta and NtwkMeta to be filled by the client, to acknowledge the\nrequest and ask some information about the client. After getting the reply, the client\ngets the content of DevMeta and NtwkMeta locally by probing the system using system\ncalls and sends out the Cli META REP. Based on the Cli META REP, PADMeta is\ncomputed and sent back to the client in PAD META REP by the adaptation proxy. Next,\nthe client updates his protocol cache and sends PAD DOWNLOAD REQ containing PAD\nID to the URL of the PAD. The CDN will automatically choose a close CDN edgeserver\nand send back the PAD code in PAD DOWNLOAD REP. If multiple PADs are required,\nit is not necessary that those PADs downloaded from the same edgeserver. It is up to the\nCDN to manage the delivery of PADs. After the security check and PAD(s) deployment,\nthe client sends out the APP REQ to the application server. The APP REQ contains the\napplication request as well as the negotiated protocol identifications, which notify the\napplication server to choose the proper PADs to talk with the client. From now on the\nclient and the application server continue the application session using the negotiated\nprotocol. The formats of all message types used in INP are listed on the bottom of\nFigure 4.\n4.\nADAPTATIONMODELOFTHEADAPTIVEENCRYPTIONPROTOCOL\nAdaptation is the major function of the adaptive encryption protocol. In this sec-\ntion, we will show how the adaptation model works. First, we will explain the protocol\nadaptation topology, the protocol adaptation tree (PAT), which is the main data struc-\nture in the procedure of adaptation. Then we will clarify the adaptation path search\nalgorithm.\n4.1. Protocol Adaptation Tree and Protocol Adapters\nFigure 5 shows a general example of the protocol adaptation tree (PAT), which is\nbuilt by the negotiation manager based onAppMeta received from the application server.\nEach node of PAT is a protocol adaptor. The child PAD is an auxiliary component of\nthe parent PAD. In order to run the parent PAD, one and only one of the children PADs\n1 Note that the client does not have to realize the existence of the adaptation proxy. The application server\nwill automatically redirect the request to its corresponding adaptation proxy.\n"
        },
        {
            "page_number": 59,
            "text": "WIRELESS NETWORK SECURITY\n51\nCDNs\nEdgeserver\nClient\nAdaptation\nProxy\nApplication\nServer\nINIT_REQ\nINIT_REP\nIfapplication protocol\nisin cache\nelse\nCli_M ETA_REQ\nCli_M ETA_REP\nCom pute\nPAD\nm etadata\nPAD_M ETA_REP\nUpdate\nprotocol\ncache\nPAD_DOW NLOAD_REQ\nPAD_DOW NLOAD_REP\nSecurity\ncheck and\ndeploy PAD\nAPP_REQ\nuse protocol\nto generate\nreply\nAPP_REP\nINIT_REQ\nINP header App Req\nCli_META_REQ\nINP header\nDevMeta\nNtw kMeta\nCli_META_REP\nINP header\nPAD_META_REP\nPADMeta\n...\nPAD_DOW NLOAD_REQ\nPAD ID\nApplication ID\nAPP_REQ\nApp Req\nDevMeta\nNtw kMeta\nINP header\nPADMeta\nINP header\n... PAD ID\nPAD_DOW NLOAD_REP\nPAD\nINP header\n...\nPAD\nINP header\nPAD ID\n... PAD ID\nFigure 4. The Interactive Negotiation Protocol.\nmust work together with the parent PAD. For example, in Figure 5, if PAD2 is the FTP\nprotocol, PAD7 is the TCP protocol, and PAD8 is the UDP protocol, the PAD2 can\nchoose either PAD7 or PAD8, but not both. It is possible that one PAD is needed by\nmultiple PADs, like TCP protocol is needed by both FTP and HTTP protocols. For the\npurpose of maintaining the tree structure, we use a symbolic copy of the child PAD if it\nis required by more than one parent PAD. For instance, in Figure 5, PAD6 is a symbolic\nlink of PAD7, which is needed by both PAD1 and PAD2. So in order to satisfy an\napplication protocol, a path should be found from the root application to one leaf, e.g.,\nthe path composed of PAD2 and PAD7 in the dotted line in Figure 5. The PADs along\nthe path forms the adaptive protocol. The PAT in the adaptive encryption protocol is\na one-level tree as shown in Figure 6 with each leaf is an encryption PAD. Key length\nand size of each PAD is shown in Table 1. Next, we will explain how to select the PADs\nin the adaptation path to build the adaptive encryption protocol.\n"
        },
        {
            "page_number": 60,
            "text": "52\nHANPING LUFEI and WEISONG SHI\nPAD2\nPAD1\nPAD3\nPAD5\nPAD4\nPAD6\nPAD7\nPAD8\nApplication\n8\n8\n5\n7\n4\n6\n8\n5\nFigure 5. A general protocol adaptation tree.\nAES\nRC4\n3DES\nText\nFigure 6. Protocol adaptation tree of the adaptive encryption protocol.\nTable 1. The key length and size of each PAD.\nPAD name\n3DES-64\n3DES-128\n3DES-192\nAES-128\nAES-192\nAES-256\nRC4\nKey Length\n64 bits\n128 bits\n192 bits\n128 bits\n192 bits\n256 bits\n64 bits\nSize\n24KB\n24KB\n24KB\n21KB\n21KB\n21KB\n10KB\n"
        },
        {
            "page_number": 61,
            "text": "WIRELESS NETWORK SECURITY\n53\n4.2. The Adaptation Path Search Algorithm\nThe goal of the adaptation path search algorithm is to find certain PADs from PAT\nto form an adaptation path for a client so that the overhead occurred by using the PADs\nalong the path, is reduced as much as possible. For more complicated PAT such as the\none in Figure 5, the Fractal framework [1] proposed an adaptation path search algorithm\nto find the path efficiently. For the one-level PAT in the adaptive encryption protocol the\nadaptation path search algorithm reduces to evaluate the overhead of each encryption\nPAD algorithm one by one and choose the one with the least overhead.\nPADtotal = PADdownload time+PADserver\ncomp +PADclient\ncomp +PADtraffic overhead\n(1)\nWe define the total overhead of each PAD as the sum of PAD download time, server\nside and client side computing time for unit data, i.e. RC4 encryption time for 1KB\ndata on server side and decryption time for 1KB data on client side, finally the traffic\noverhead incurred by this PAD as shown in Equation 1. Since the PAD size is very\nsmall as we can see in Table 1, large amount PAD download experiment from the three\nPlanetLab nodes shows that the average download time are as close as 1 millisecond\ndifference. Furthermore, each PAD is at most downloaded once in the whole application\nprocedure. We consider the PAD download time as a constant and eliminate it from the\nPAD total overhead evaluation. On the other hand, since the three PADs, 3DES, AES,\nand RC4 do not change the size of the input data even with different key length, the\ntraffic overhead of each PAD can also be excluded. Eventually, the PAD total overhead\nis simplified as Equation 2.\nPADtotal = PADserver\ncomp + PADclient\ncomp\n(2)\nServer side computing time of each PAD can be obtained proactively by testing each\nPAD on the application server. In order to evaluate the client side computing time of\neach PAD, running each PAD on each client configuration to get the overhead is not\na wise solution because there are so many different client configurations. Instead, we\nuse a linear model to estimate the overhead, which inspired by the observation that\nthe computing overhead of each PAD is roughly proportional to the processor speed.\nAs shown in the second part of the new total overhead equation 3, if the computing\noverhead of a PAD on a standard processor speed, Stdcpu, i.e. 500MHz Pentium IV\nin the platform, is known as the PADStd\ncomp, the computing overhead on the client side\ncan be deducted from the linear ratio of the speed of the standard processor and client\nprocessor. However, this linear model is not so accurate because other parameters\nof the system introduces error into the linear model, i.e. the operating system. We\nabstract normalized ratio parameters about two key properties: processor types as A\nand application types as B in the equation. Note that it is easy to introduce more\nparameters if necessary, e.g., the operating system types and the network types defined\nin Fractal framework [1].\n"
        },
        {
            "page_number": 62,
            "text": "54\nHANPING LUFEI and WEISONG SHI\nPADtotal = PADserver\ncomp + A × B × Stdcpu\nClicpu\n× PADStd\ncomp\n(3)\nUsually, the normalized ratios such as A and B are in the form of matrix, as shown in\nEquation 4, to measure the performance ratios of 7 PADs on 3 kinds of processor types\nand on 2 kinds of application types, legacy and new system since they have different\nencryption requirements. P, D, and L represent the Intel PXA 255 processor in Pocket\nPC, Pentium IV 2.0GHz processor in Desktop, and Pentium IV 3.06GHz processor in\nLaptop respectively. We use the following simple example to explain the normalized\nmatrix.\nWinCE\nPalmOS\nWinMedia\nKinoma\n\f\n1\n∞\n∞\n1\n\r\nThe above matrix shows the impacts of two operating systems (the top line) on two\nmultimedia players (the left most column). The values in the matrix mean the Windows\nMedia works fine in the WinCE operating system (WinCE) [19] but not in PalmOS,\nwhile Kinoma player [20] runs well in PalmOS instead of WinCE. The value of ratios\ndoes not have to be an integer. Suppose now we are about to find the better one in terms\nof the computing time from these two players on WinCE platform. We get the time\nvalue using the linear method as, for instance, 5 sec forWinMedia and 2 sec for Kinoma.\nWithout the normalized matrix, Kinoma will be chosen as the better player; however,\nthe fact is that Kinoma can not run on WinCE at all. To get the correct result, we can\nuse the first column of this normalized matrix to adjust the linear results by multiplying\n2 sec with ratio 1 for WinMedia and multiplying 5 sec with ratio ∞for Kinoma. Then\nthe computing time of Kinoma becomes ∞, which immediately disqualifies itself.\nGo back to the normalized matrix A and B, because most of the operations in these\nencryption algorithms are bit operations instead of float-point operations, they have\nalmost same running efficiency in these client CPU types. We set all values as 1.\nDifferent encryption requirements of applications are reflected in B. For example, the\nlegacy systems only use the DES algorithm while the new applications will utilize the\nnew encryption algorithms. Correspondingly in the normalized ratio matrix, we set the\nratio as 1 for 3DES algorithm and ∞for others in legacy systems. In our experimental\nplatform, we specify the client applications on desktop as a legacy system and that on\nlaptop and PocketPC as a new system. This may not be always true in reality, but just\nfor comparison purpose in this experimental platform.\n"
        },
        {
            "page_number": 63,
            "text": "WIRELESS NETWORK SECURITY\n55\nA\n=\ncpu0\n. . .\ncpua\npad0\n...\npadn\n⎛\n⎜\n⎝\nα0(0)\n. . .\nα0(a)\n...\n...\n...\nαn(0)\n. . .\nαn(a)\n⎞\n⎟\n⎠\n=\nP\nD\nL\n3DES −64\n3DES −128\n3DES −192\nAES −128\nAES −192\nAES −256\nRC4 −64\n⎛\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎝\n1\n1\n1\n1\n1\n1\n1\n1\n1\n1\n1\n1\n1\n1\n1\n1\n1\n1\n1\n1\n1\n⎞\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎠\nB\n=\napp0\n. . .\nappb\npad0\n...\npadn\n⎛\n⎜\n⎝\nβ0(0)\n. . .\nβ0(b)\n...\n...\n...\nβn(0)\n. . .\nβn(b)\n⎞\n⎟\n⎠\n=\nLegacySystem\nNewSystem\n3DES −64\n3DES −128\n3DES −192\nAES −128\nAES −192\nAES −256\nRC4 −64\n⎛\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎝\n1\n∞\n1\n∞\n1\n∞\n∞\n1\n∞\n1\n∞\n1\n∞\n1\n⎞\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎠\nNow for a specific incoming client with processor type i and application type j available\nin metadata,\nthe adaptation proxy will find the corresponding ratio vector\n\u000f α0(i)\nα1(i)\n. . .\nαn(i)\n\u0010T , and\n\u000f β0(j)\nβ1(j)\n. . .\nβn(j)\n\u0010T from A, and B\nbased on its processor and application types. Given that we have only a limited number\nof consumer-used processors, the vector will be found with high probability. Other-\nwise a similar type with close parameters will be chosen instead. After the application\nsession, the normalized matrix will be extended to include the new processor types.\nThen the normalized ratio matrix can be formed to estimate the total time overhead of\neach PAD for this new client using Equation 4. After obtaining the total time overhead\nof each PAD, The adaptive encryption protocol can be decided using the reduced adap-\ntation path search algorithm. For the comprehensive descriptions of total overhead,\n"
        },
        {
            "page_number": 64,
            "text": "56\nHANPING LUFEI and WEISONG SHI\nnormalized matrix, and adaptation path search algorithm, please refer to the Fractal\nframework [1].\n⎛\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎝\npadtotal\n3DES−64\npadtotal\n3DES−128\npadtotal\n3DES−192\npadtotal\nAES−128\npadtotal\nAES−192\npadtotal\nAES−256\npadtotal\nRC4−64\n⎞\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎠\n=\n⎛\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎝\npadsvr−comp\n3DES−64\npadsvr−comp\n3DES−128\npadsvr−comp\n3DES−192\npadsvr−comp\nAES−128\npadsvr−comp\nAES−192\npadsvr−comp\nAES−256\npadsvr−comp\nRC4−64\n⎞\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎠\n+\ncpu\nClicpu\n∗\n⎛\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎝\nα3DES−64(i)\nα3DES−128(i)\nα3DES−192(i)\nαAES−128(i)\nαAES−192(i)\nαAES−256(i)\nαRC4−64(i)\n⎞\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎠\nT\n∗I ∗\n⎛\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎝\nβ3DES−64(j)\nβ3DES−128(j)\nβ3DES−192(j)\nβAES−128(j)\nβAES−192(j)\nβAES−256(j)\nβRC4−64(j)\n⎞\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎠\nT\n∗I ∗\n⎛\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎜\n⎝\npadStd−comp\n3DES−64\npadStd−comp\n3DES−128\npadStd−comp\n3DES−192\npadStd−comp\nAES−128\npadStd−comp\nAES−192\npadStd−comp\nAES−256\npadStd−comp\nRC4−64\n⎞\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎟\n⎠\n5.\nPERFORMANCE EVALUATION AND ANALYSIS\nIn our experimental platform, as shown in Figure 1, three kinds of client hosts,\ndesktop, laptop, and Pocket PC, use different message receiver applications, to connect\nto the message sender and an adaptation proxy. The hardware and software configura-\ntions of the servers and clients are also shown in Figure 1. The message sender has 100\nmessages with size as 100K bytes. We implement three encryption algorithms, 3DES,\nAES and RC4 in C code as three protocol adaptors. The first two encryption algo-\nrithms have three different key length settings. Key length and size of each algorithm\nis shown in Table 1. We also implement an adaptation proxy connected with the ap-\nplication server in the same LAN domain. Similar to the previous section. To emulate\nthe behavior of the real content distribution network and edgeservers, we utilize three\n"
        },
        {
            "page_number": 65,
            "text": "WIRELESS NETWORK SECURITY\n57\nnodes, in Wayne State University, New York University, and University of California\nat Berkeley respectively, from PlanetLab [21] as the distributed PAD servers.\nWe test the total time overhead of each algorithm for desktop, laptop, and PocketPC\nclients, as shown in Figure 7. The x-axis lists different encryption algorithms, the y-axis\nshows the total time for each algorithm including the sender encryption time and the\nreceiver decryption time. In Figure 7(a), since the receiver application of the desktop is\na legacy application in our experimental setup, which accepts only DES algorithms, the\noutput of the adaptive path selection algorithm will set all other encryption algorithms\nexcept DES algorithms to infinite, which is denoted as N/A in the figure. However,\nfor comparison purpose, we also show their corresponding computing overhead on the\nsame figure. As a matter of fact, although AES-class algorithms have less computing\noverhead, they will not be chosen as the proper encryption algorithm for the desktop,\nwhichrunslegacyapplicationsonly. Nowonly3DESalgorithmsareeligiblecandidates.\nIt is trivial that 3DES with 64 bits key should run faster than 3DES with 128 bits or\n192 bits length key. Usually the adaptive encryption protocol will choose 3DES-64\nsince it has the fastest running speed with reasonable security enforcement. But this\ndoes not prevent application from choosing 128 or 192 bits 3DES. By introducing more\nadaptation parameters, like a normalized matrix for application security requirements,\nmore secure algorithm could be selected. We believe this is a trivial task and decide\nnot to be discussed in this chapter.\nFor the applications running on the laptop, 3DES is obviously not considered\nbecause it is out of date (and replaced by AES algorithms) for new applications. AES-\n128 which has slightly less total time overhead than other three algorithms have, as\nshown in Figure 7(b), will be selected by the adaptive encryption protocol. Note that\nsimilar to the case for desktop, other AES algorithms could also be selected for more\nsecure purpose by extending the total time overhead evaluation formula. Finally, in\nFigure 7(c), we can see that the major part of the total time overhead is contributed\nby the receiver decryption time because the hardware of PocketPC on which receiver\napplication executes is not as powerful as desktop or laptop hardware configurations.\nNot surprising, RC4-64 is selected as the most appropriate encryption algorithm, which\nis much faster than other algorithms. This is compatible with the fact that RC4 is almost\nthe default encryption algorithm for small resource-constraint devices. It is worth\nnoting that the choice made by the adaptive encryption protocol is straightforward in\nthis case study. However, our work is the first effort to make the choice making in a\nformal way. We believe that the adaptive encryption protocol will be more useful in\ncomplicated applications in the foreseeable future work, includes investigating more\nencryption algorithms in heterogeneous environments, and applying this technique to\nthe distributed computer-assistant surgery application [22].\n6.\nRELATED WORK\nThe adaptive encryption protocol shares its goals with some recent efforts that are\naimed at injecting functionality into application for adaptation. We categorize related\n"
        },
        {
            "page_number": 66,
            "text": "58\nHANPING LUFEI and WEISONG SHI\n0\n200\n400\n600\n800\n1000\n1200\n1400\n1600\n3DES-64\n3DES-128\n3DES-192\nAES-128\nAES-192\nAES-256\nRC4-64\nEncryption Algorithms\nTotal Time (ms)\nReceiver decryption Time\nSender encryption time\nN/A\nN/A\nN/A\nN/A\n(a) Message receiver on Desktop\n0\n100\n200\n300\n400\n500\n600\n700\n800\n900\n3DES-64\n3DES-128\n3DES-192\nAES-128\nAES-192\nAES-256\nRC4-64\nEncryption Algorithms\nTotal Time (ms)\nReceiver decryption Time\nSender encryption time\nN/A\nN/A\nN/A\n(b) Message receiver on Laptop\n0\n500\n1000\n1500\n2000\n2500\n3DES-64\n3DES-128\n3DES-192\nAES-128\nAES-192\nAES-256\nRC4-64\nEncryption Algorithms\nTotal Time (ms)\nReceiver decryption time\nSender encryption time\nN/A\nN/A\nN/A\n(c) Message receiver on PocketPC\nFigure 7.\nA comparison of the total time overhead for different message receivers:\n(a) Desktop, (b) Laptop, and (c) PocketPC.\n"
        },
        {
            "page_number": 67,
            "text": "WIRELESS NETWORK SECURITY\n59\nresearch into three groups as distributed adaptation, protocol adaptation, and mobile\ncode and mobile agent.\nDistributed adaptation From the Internet topology’s point of view, adaptation\nfunctionality can be introduced either at the end-points or distributed on intermediate\nnodes. Odyssey [4], Rover [3] and InfoPyramid [23] are examples of systems that\nsupport end point adaptation. Conductor [24] and CANS [2] provide an application\ntransparent adaptation framework that permits the introduction of arbitrary adaptors in\nthe data path between applications and end services. While these approaches provide an\nextremely general adaptation mechanism, significant change to existing infrastructure is\nrequired for their deployment. However, the adaptive encryption protocol does not have\nthe deployment problem for leveraging the existing CDNs technology to distributed\nprotocol adaptors, which are implemented using mobile code.\nFrom the network structure’s perspective, there are two issues: whether adapta-\ntion functionality is introduced at network layer with application-transparency or at\nthe application level with application-awareness. Systems such as transformer tun-\nnels [12] and protocol boosters [11] are examples of application-transparent adaptation\nefforts that work at the network level. Such systems can cope with localized changes\nin network conditions but cannot adapt to behaviors that differ widely from the norm.\nMoreover, their transparency hinders composability of multiple adaptations. More gen-\neral are programmable network infrastructures, such as COMET [25], which supports\nflow-based adaptation, and Active Networks [26, 27], which permit special code to\nbe executed for each packet at each visited network element. While these approaches\nare very general adaptation mechanisms, significant change to existing infrastructure\nis required for their deployment. The adaptive encryption protocol overcomes this\nshortcoming because it works entirely on the application level. Similar efforts also\nwork at the application level. The cluster-based proxies in BARWAN/ Daedalus [28],\nTACC [29], and MultiSpace[30]areexamplesofsystemswhereapplication-transparent\nadaptation happens in intermediate nodes (typically a small number) in the network.\nActive Services [31] extend these systems to a distributed setting by permitting a client\napplication to explicitly start one or more services on its behalf that can transform the\ndata it receives from an end service. Our work is different from other application level\napproaches in the following ways: first, it is not using intermediate nodes which may\noccur with deployment problems. Second it does not rely on any specific data stream\nor client conditions. On the contrary, it is designed to cope with any applications and\nclient environments as long as one has the proper protocol adaptor.\nProtocol adaptation There are some research work about the protocol adaptation.\nIn network level systems such as [6], in which communicating end hosts use untrusted\nmobile code to remotely upgrade each other with the transport protocols that they\nuse to communicate. Transformer tunnels [12] and protocol boosters [11] are doing\napplication-transparent adaptation by tuning the network protocol according to the\nchange of network situations. Such systems can deal with localized changes in network\nconditions but cannot react to changing environments outside the network layer. Since\nthe adaptive encryption protocol works at the application layer, it can maximally adapt\napplication level protocols which have no way to be completed in the network layer. It\n"
        },
        {
            "page_number": 68,
            "text": "60\nHANPING LUFEI and WEISONG SHI\nis also different from the Web browser plugins, e.g., Realplay, Flash, and so on. Plugin\nis an application component which completes part of the functionality, incapable of\ndoing protocol adaptation. Although today some Web sites provide multiple choices of\nplugins to do the similar function, they still need the client to manually select one, but\nmaybe not the best. The adaptive encryption protocol adapts the functionality by means\nof protocol adaptation which has transparency to the client and other characteristics,\nsuch as flexibility and extendibility, which plugins do not have.\nMobile code and mobile agent Mobile code is a good candidate for carrying a\nprotocol module since it has long been known as a mechanism for providing a late\nbinding of function to systems [32, 33, 34]. Mobile code and related technologies also\nhave been proposed and studied as effective means of implementing content adaptation,\nprotocol update, and program migration in distributed applications. In [35, 6] they\npropose a system in which communicating end hosts use untrusted mobile code to\nremotely upgrade each other with the transport protocols that are used to communicate.\nOur work is complimentary to their work because our proposal works in the application\nlevel. A new lightweight, component-based mobile agent system that can adapt to\ndiverse devices and features resource saving is proposed in [36]. In this system, mobile\ncode is brought in and associated execution states of an application dynamically after\nmigration. NWSLite [37] provides a sophisticated predicting tools for the remote code\nexecution offloaded from mobile client to the close server. To our best knowledge, the\nadaptive encryption protocol is the first approach to use mobile code to do encryption\nprotocol adaptation that extends the utilization of mobile code technology.\nFinally, a lot of encryption algorithms have been proposed [38], e.g., DES [14],\nAES [15], and RC4 [16], however, the focus of this paper is on selecting an appropriate\nencryption algorithm for a specific client configuration. Therefore, we envision our\nwork complements to the research of cryptography algorithms very well.\n7.\nSUMMARY\nIn this chapter, an adaptive encryption protocol is proposed to benefit the applica-\ntion from choosing the appropriate encryption algorithm according to dynamic client\ndevices and application requirements. With the emergence of more and more cryptog-\nraphy algorithms, adaptation becomes a necessity because each algorithm has distinct\ncharacteristics from others although they are for the same application purposes. For\nthe whole encryption algorithms family, some of them are very secure but require more\ncomputing power, while some of them are less secure but can run on tiny resource-\nconstraint devices. We believe that the proposed adaptive encryption protocol makes\nan initial step towards using mobile code to support the application-level encryption pro-\ntocol adaptation. The adaptive encryption protocol shows great flexibility in adapting\nencryption algorithms among multiple encryption algorithms in various client envi-\nronments to reduce the total overhead without sacrificing of security, and it is a very\npromising approach in both conventional distributed applications and ever-increasing\nresource-constraint information appliances, such as smart phones, PocketPCs, and so\n"
        },
        {
            "page_number": 69,
            "text": "WIRELESS NETWORK SECURITY\n61\non. Furthermore, we have observed that our prediction of the PAD overhead is partially\nbased on the ratio between the client CPU speed and a standard CPU speed. This may\nnot completely reflect the real situation in which client may have one or more of the\nfollowing: (1) multiple processors, (2) multiple pipelines within in a single processor,\n(3) varying cache sizes, etc. All these can affect the computation time and hence the\ntotal delay time. Thus, it is very interesting to investigate the influence of diverse CPU\narchitectures and adjust the adaptation model to reflect the variation accordingly.\n8.\nREFERENCES\n1. H. Lufei, W. Shi, Fractal: A mobile code based framework for dynamic application protocol adaptation\nin pervasive computing, in: Proc. of IPDPS’05.\n2. X. Fu, W. Shi, A. Akkerman, V. Karamcheti, CANS: Composable, Adaptive Network Services Infras-\ntructure, in: Proc. of USITS’01, pp. 135–146.\n3. A. D. Joseph, J.A.Tauber, M. F. Kasshoek, Mobile Computing with the RoverToolkit, IEEETransaction\non Computers 46 (3) (1997) 337–352.\n4. B. D. Noble, Mobile Data Access, Ph.D. thesis, School of Computer Science, Carnegie Mellon Uni-\nversity (May 1998).\n5. A.Vahdat, M. Dahlin, T.Anderson, A.Aggarwal, Active names: Felxible location and transport of wide\narea resources, in: Proc. of USITS’99.\n6. P.Patel, A.Whitaker, D.Wetherall, J.Lepreau, T.Stack, Upgrading transport protocols using untrusted\nmobile code, in: Proc. of ACM SOSP’03.\n7. A. Fuggetta, G. P. Picco, G. Vigna, Understanding code mobility, in: IEEE TRANSACTIONS ON\nSOFTWARE ENGINEERING, 1998, p. Vol.24 No. 5.\n8. D. Halls, Applying mobile code to distributed systems, Ph.D. thesis, Computer Laboratory University\nof Cambridge (1997).\n9. Akamai Technologies Inc., Edgesuite services.\nURL http://www.akamai.com/html/en/sv/edgesuite over.html\n10. B. Krishnamurthy, J. Rexford, Web Protocols and Practice: HTTP/1.1, Networking Protocols, Caching\nand Traffic Measurement, Addison-Wesley, Inc, 2001.\n11. A. Mallet, J. Chung, J. Smith, Operating System Support for Protocol Boosters, in: Proc. of HIPPARCH\nWorkshop, Uppsala Sweden, 1997.\n12. P. Sudame, B. Badrinath, Transformer Tunnels: A Framework for Providing Route-Specific Adapta-\ntions, in: Proc. of USENIX’98.\n13. N. W. Group, Tcp slow start, congestion avoidance, fast retransmit, and fast recovery algorithms.\nURL http://rfc.net/rfc2001.html\n14. Data Encryption Standard.\nURL http://www.itl.nist.gov/fipspubs/fip46-2.htm/\n15. Advanced Encryption Standard.\nURL http://csrc.nist.gov/CryptoToolkit/aes/\n16. RC4 RFC3268.\nURL http://www.faqs.org/rfcs/rfc3268.html/\n17. W. Diffie, M. Hellman, Multiuser cryptographic techniques, IEEE Transactions on Information Theory\n22 (1) (1976) 644–654.\n"
        },
        {
            "page_number": 70,
            "text": "62\nHANPING LUFEI and WEISONG SHI\n18. FIPS 180-1, Secure Hash Standard, U.S. Department of Commerce/N.I.S.T., National Technical Infor-\nmation Service, Springfield, VA, 1995.\n19. Windows CE Operating Systems.\nURL http://www.microsoft.com/windowsce/\n20. Kinoma player.\nURL http://www.kinoma.com/\n21. Planetlab.\nURL http://planet-lab.org/\n22. H. Lufei, W. Shi, L. Zamorano, Communication optimization for image transmission in computer-\nassisted surgery, in: Proceedings of 2004 Congress of Neurological Surgeons Annual Meeting (ab-\nstract)), San Francisco, CA, 2004.\n23. R. Mohan, J. R. Simth, C. Li, Adapting Multimedia Internet Content for Universal Access, IEEE\nTransactions on Multimedia 1 (1) (1999) 104–114.\n24. M.Yarvis,A.Wang,A. Rudenko, P. Reiher, G. J. Popek, Conductor: DistributedAdaptation for complex\nNetworks, in: Proc. of HotOS, 1999.\n25. A. T. Campbell, et al., A Survey of Programmable Networks, ACM SIGCOMM Computer Communi-\ncation Review 29 (2) (1999) 7–23.\n26. D. Tennenhouse, D. Wetherall, Towards an Active Network Architecture, Computer Communications\nReview 26 (2).\n27. D. J.Wethrall, J.V. Guttag, D. L. Tennenhouse,ANTS:A toolkit for building and dynamically deploying\nnetwork protocols, in: Proc. of 2nd IEEE OPENARCH.\n28. A. Fox, S. Gribble, Y. Chawathe, E. A. Brewer, Adapting to Network and Client Variation Using\nInfrastructural Proxies: Lessons and Prespectives, IEEE Personal Communication 5 (4) (1998) 10–19.\n29. A. Fox, S. Gribble, Y. Chawathe, E. A. Brewer, P. Gauthier, Cluster-based Scalable Network Services,\nin: Proc. of SOSP’97.\n30. S. D. Gribble, M. Welsh, E.A.Brewer, D. Culler, The MultiSpace: An Evolutionary Platform for\nInfrastructual Services, in: Proc. of Usenix’99.\n31. E. Amir, S. McCanne, R. Katz, An active service framework and its application to real-time multimedia\ntranscoding, in: Proc. of the SIGCOMM’98.\n32. A. Birrell, G. Nelson, S. Owicki, E. Wobber, Network objects, in: Software-Practice and Experience,\n1995, pp. 25(S4):87–130.\n33. A. D. Joseph, A. F. deLespinasse, J. Tauber, D. Gifford, M. F. Kaashoek, Rover:a toolkit for mobile\ninformation access, in: Proc. of ACM SOSP’95.\n34. E. Jul, H. Levy, N. Hutchinson, A. Black, Fine-grained mobility in the emerald system, in: ACM Tran.\non Computer Systems, 1988, pp. 6(1):109–133.\n35. P. Patel, D. Wetherall, J. Lepreau, A. Whitake, Tcp meets mobile code, in: Proc. of HTOS’03.\n36. Y. Chow, W. Zhu, C. Wang, F. C. Lau, The state-on-demand execution for adaptive component-based\nmobile agent systems, in: Proc. of ICPADS’04.\n37. S. Gurun, C. Krintz, R. Wolski, Nwslite: A light-weight prediction utility for mobile devices, in: Proc.\nof MobiSys’04, Boston, MA, 2004.\n38. B. Schneier (Ed.), Applied Cryptography, Second Edition, Protocols, Algorthms, and Source Code in\nC, John Wiley & Sons, Inc., 1996.\n"
        },
        {
            "page_number": 71,
            "text": "Part I\nSECURITY IN\nAD HOC NETWORK\n"
        },
        {
            "page_number": 72,
            "text": "3\nPRE-AUTHENTICATION AND AUTHENTICATION\nMODELS IN AD HOC NETWORKS\nKatrin Hoeper\nDepartment of Electrical and Computer Engineering\nUniversity of Waterloo,\n200 University Avenue West, Waterloo, Ontario, N2L\n3G1, Canada\nE-mail: khoeper@engmail.uwaterloo.ca\nGuang Gong\nDepartment of Electrical and Computer Engineering\nUniversity of Waterloo,\n200 University Avenue West, Waterloo, Ontario, N2L\n3G1, Canada\nE-mail: ggong@calliope.uwaterloo.ca\nProviding entity authentication and authenticated key exchange among nodes are both target\nobjectives in securing ad hoc networks. In this chapter, a security framework for authenti-\ncation and authenticated key exchange in ad hoc networks is introduced. The framework is\napplicable to general ad hoc networks and formalizes network phases, protocol stages, and\ndesign goals. To cope with the diversity of ad hoc networks, many configuration parameters\nthat are crucial to the security of ad hoc networks are discussed. Special attention is paid to\nthe initial exchange of keys between pairs of nodes (pre-authentication) and the availability\nof a trusted third party in the network. Next, several pre-authentication and authentication\nmodels for ad hoc networks are discussed. The models can be implemented as a part of\nthe proposed security framework and correspond to the wide range of ad hoc network ap-\nplications. Advantages and disadvantages of the models are analyzed and suitable existing\nauthentication and key exchange protocols are identified for each model.\n1.\nINTRODUCTION\nThe number of applications that involve wireless communications among mobile\ndevices is rapidly growing. Many of these applications require the wireless network\n"
        },
        {
            "page_number": 73,
            "text": "66\nKATRIN HOEPER and GUANG GONG\nto be spontaneously formed by the participating mobile devices themselves. We call\nsuch networks ad hoc networks. The idea of ad hoc networks is to enable connectivity\namong any arbitrary group of mobile devices everywhere, at any time. We distinguish\ntwo categories of ad hoc networks, mobile ad hoc networks (MANETs) and smart sensor\nnetworks. Typical devices of MANETs are PDAs, laptops, cell phones, etc., and the\ndevices of smart sensor networks are sensors. MANETs are used at business meetings\nand conferences to confidentially exchange data, at the library to access the Internet\nwith a laptop, and at hospitals to transfer confidential data from a medical device to a\ndoctor’s PDA. Sensor networks can be used for data collection, rescue missions, law\nenforcement and emergency scenarios. Many more applications exist already or are\nimaginable in the near future. Caused by the widespread applications, a general security\nmodel and protocol framework for authentication and authenticated key establishment\nin ad hoc networks have not been defined yet.\n1.1. Ad Hoc Network Properties\nTo achieve the ambitious goal of providing ubiquitous connectivity, ad hoc net-\nworks have special properties that distinguish them from other networks. We briefly\ndiscuss those properties in the following.\nAd hoc networksare temporary networksbecausetheyareformedtofulfillaspecial\npurpose and cease to exist after fulfilling this purpose. Mobile devices might arbitrarily\njoin or leave the network at any time, thus ad hoc networks have a dynamic infrastruc-\nture. Most mobile devices use radio or infrared frequencies for their communications\nwhich leads to a very limited transmission range. Usually the transmission range is\nincreased by using multi-hop routing paths. In that case a device sends its packets to\nits neighbor devices, i.e. devices that are in transmission range. Those neighbor nodes\nthen forward the packets to their neighbors until the packets reach their destination. The\nmost distinguishing property of ad hoc networks is that the networks are self-organized.\nAll network interactions have to be executable in absence of a trusted third party (TTP),\nsuch as the establishment of a secure channel between nodes and the initialization of\nnewly joining nodes. Hence, in contrast to wireless networks, ad hoc networks do\nnot rely on a fixed infrastructure and the accessibility of a TTP. The self-organizing\nproperty is unique to ad hoc networks and makes implementing security protocols a\nvery challenging task. Another characteristics of ad hoc networks are the constrained\nnetwork devices. The constraints of ad hoc network devices are a small CPU, small\nmemory, small bandwidth, weak physical protection and limited battery power, as first\nsummarized in [23]. In most ad hoc networks all devices have similar constraints.\nThis property distinguishes the architecture of an ad hoc network from a client-server\nstructure.\n1.2. Security Challenges\nThe special properties of ad hoc networks enable all the neat features such net-\nworks have to offer, but at the same time, those properties make implementing security\n"
        },
        {
            "page_number": 74,
            "text": "WIRELESS NETWORK SECURITY\n67\nprotocols a very challenging task. There are four main security problems that need to\nbe dealt with in ad hoc networks: (1) the authentication of devices that wish to talk\nto each other; (2) the secure key establishment of a session key among authenticated\ndevices; (3) the secure routing in multi-hop networks; and (4) the secure storage of (key)\ndata in the devices. Note, that once (1) and (2) are achieved, providing confidentiality\nis easy. In the remainder of this article, we will focus on entity authentication and\nauthenticated key establishment (AAKE) protocols and their implementation issues in\nad hoc networks. Note that most security problems related to such protocols occur in\nthe bootstrapping phase, i.e. at the time nodes wish to securely communicate for the\nvery first time. We refer to this phase as the pre-authentication phase, and we define\nand discuss this stage in great detail later in this chapter.\n1.3. Outline\nAs said earlier, due to the wide range of ad hoc network applications, no general\nsecurity framework has been introduced yet. In this chapter, we introduce a security\nframework for authentication and authenticated key exchange in ad hoc networks. The\nframework is applicable to general ad hoc networks and formalizes network phases,\nprotocol stages, and design goals. To cope with the diversity of ad hoc networks,\nwe discuss many configuration parameters that are crucial to the security of ad hoc\nnetworks. We pay special attention to the initial key exchange between pairs of nodes\n(pre-authentication) and the availability of a TTP in the network. We then categorize\nseveral pre-authentication and authentication models that can be implemented as a part\nof the proposed security framework. The models correspond to the wide range of ad hoc\nnetwork applications and we analyze their advantages and disadvantages and identify\nsuitable existing authentication and key exchange protocols for each model.\nThe rest of this chapter is organized as follows. In Section 2, we introduce a\nsecurity framework for ad hoc networks, including network and authentication phases,\nprotocol stages and design goals.\nIn Section 3, we identify some security related\nconfiguration problems that are crucial for protocol implementations in many ad hoc\nnetwork applications.\nTaking all previous results into account, we categorize and\nanalyze a number of pre-authentication and authentication models in Section 4 and 5,\nrespectively. Finally, in Section 6, conclusions are drawn.\n2.\nSECURITY FRAMEWORK\nIn this section, we first discuss the different network phases that occur in the\nlifecycle of an ad hoc network. Then, we introduce the two authentication phases of\ncommunicating nodes in such networks. Next, we define the protocol stages of general\nAAKE protocols in ad hoc networks. At the end of this section, we summarize the\ndesign goals all protocols that are designed for ad hoc networks should meet. All\nthese definitions combined form a security framework for general ad hoc networks.\nThe framework helps designing security solutions for ad hoc networks. In particular,\nwhen proposing protocols for ad hoc networks, all network and authentication phases,\n"
        },
        {
            "page_number": 75,
            "text": "68\nKATRIN HOEPER and GUANG GONG\nprotocol stages and design goals as defined in this security framework need to be\naddressed.\n2.1. Network Phases\nWe distinguish two network phases in ad hoc networks, namely the network ini-\ntialization phase and the running system phase. In the first phase, the network is set\nup. All nodes that are present at the network initialization phase, i.e. during the time\nthe network is formed, are initialized. The self-organization property of the network\nis sometimes not required in this phase. For instance, a TTP might be available in the\ninitialization phase in order to initialize all present nodes with required data, such as\nsystem parameters and cryptographic keys. After the initialization phase, nodes can\nfreely join or leave the network at any time. We refer to this as running system phase.\nAd hoc networks are generally self-organized in this phase. This follows that no TTP\nor other fixed infrastructure is longer available. Consequently, current network nodes\nare responsible to initialize newly joining nodes with required key material, cope with\nleaving nodes and execute all other necessary administrative tasks in a self-organized\nmanner.\n2.2. Authentication Phases\nWedistinguishtwoauthenticationphasesforauthenticationsamongnetworknodes.\nThe first phase consist of the initial exchange of data and cryptographic key material\namong a group of two or more nodes. The data can include secret or public keys,\nfor example. The same data is used to identify each other in all later authentications\namong the same nodes. The described initial authentication phase is called imprinting\nin the resurrecting duckling model [23], and initialization in the Bluetooth protocol [4].\nHenceforth, we will adopt the term pre-authentication from [2]. The data that is ex-\nchanged in the pre-authentication phase needs to be sent over a secure channel, where\nsecure refers to an authentic and confidential channel for exchanging symmetric key\ndata, and to an authentic channel for exchanging public keys in asymmetric schemes.\nPre-authentication is not limited to the devices present at the time of the network initial-\nization phase, it also needs to be provided to subsequently joining nodes in the running\nsystem phase. All nodes that subsequently join the ad hoc network need to be able\nto securely obtain shared data and required key material from all potential commu-\nnication partners. The main challenge is to provide pre-authentication in the running\nsystem phase, even though the network environment might have changed and a TTP\nis not accessible any longer. During the second phase, the authentication phase, the\nnodes identify each other by using the authentic data that was exchanged in the pre-\nauthentication phase. These authentications are executed over an insecure channel and\nneed to be secured by the key material exchanged during pre-authentication.\n2.3. Protocol Stages\nWe now consider the protocol stages of a two party AAKE protocol. The desired\nAAKE protocol should first provide pairwise pre-authentication, then mutual authenti-\n"
        },
        {
            "page_number": 76,
            "text": "WIRELESS NETWORK SECURITY\n69\ncation between the same two nodes, and lastly, a secure establishment of a session key\nshared between the nodes. All AAKE protocols can be executed in the running system\nin an ad hoc network, i.e. after the network initialization phase. A suitable AAKE\nprotocol should take all ad hoc network properties and constraints into account. Note,\nthat the protocol design goals are defined in the next section.\nA typical AAKE protocol in our security framework for ad hoc networks consists\nof the following three stages:\n1. Pre-Authentication\nThe first stage is the pre-authentication between two devices that wish to communicate\nwith each other at this or a later time. In this phase either a secret key or an authentic\ncopy of a public key are securely shared between the devices. Keys can be shared during\npre-authentication using one of the pre-authentication models that we will introduce\nin Section 4. The best suited model needs to be chosen according to the particular\napplication.\nThe key data that has been exchanged or established during pre-authentication is\nused in all subsequent authentications between the same nodes. Hence, the next time\nthe same nodes wish to securely communicate, i.e. to execute an AAKE protocol, the\nnodes can skip the pre-authentication stage and directly start with the authentication.\nPre-authentication needs only to be repeated if keys are revoked or expired.\n2. Authentication\nIn the second stage, the authentication stage, the participants mutually authenticate each\nother using the key data from the pre-authentication phase. A suited authentication\nprotocol can be chosen out of the authentication models introduced in Section 5. The\nbest suited protocol needs to be chosen according to the respective application. If the\nauthentication of one node fails, the protocol stops and further countermeasures might\nbe taken, for example revoking the key of the rejected node.\n3. Session Key Establishment\nUpon successful mutual authentication, the nodes start establishing a session key in\nthe third protocol stage. Note that all session keys need to be established over an\nauthentic channel. Otherwise, Oscar could take over Alice’s role after her successful\nauthentication to Bob. To overcome this attack, the session key establishment stage\ncan be combined with the previous authentication stage. Again, for suitable AAKE\nprotocols please refer to Section 5.\n2.4. Design Goals\nAfter discussing the special properties and needs of ad hoc networks and several\nof the issues that occur when implementing protocols in such networks, we now derive\nthe design goals that all ad hoc network protocols should meet in order to be suitable.\nAll protocols should only require few computational steps due to the limited battery\npower of all ad hoc devices. Too many computational steps would drain the battery. For\nthe same reason protocols should only require few message flows. Caused by the nature\nof wireless networks, the communication bandwidth is very small. If messages are too\n"
        },
        {
            "page_number": 77,
            "text": "70\nKATRIN HOEPER and GUANG GONG\nlarge, they will be split into several packets. Sending many packets contradicts with the\nprevious design goal, therefore small data packages are desirable. Due to the limited\ncomputational power of ad hoc devices, preferable protocols should mainly require\ncheap computations. As a general trend, the processors of most ad hoc devices, such\nas PDAs, are becoming more and more powerful, and therefore heavy computations,\nsuch as modular exponentiations, are becoming feasible. However, heavy computa-\ntions require more battery power, and thus, it is important to restrict the number of\nheavier computations. Based on the assumption that all ad hoc network devices have\nsimilar constraints, suited protocol should be balanced, i.e. all devices need to perform\napproximately the same number of equally heavy computations. Considering the very\nlimited memory space of all devices, protocols should neither require much memory\nspace for the protocol code itself nor for the storage of parameters and key material. As\na consequence, short code, short keys and short system parameters are desirable. When\ndesigning protocols the consequences of data disclosure should be very restricted be-\ncause ad hoc network devices and especially sensors provide only a low level of physical\nprotection. Once an attacker gains access to the device, he/she is usually able to obtain\nthe stored data, including the key material. Note that this attack is quite reasonable\nsince such devices cannot be protected as some servers that are locked away in secure\nrooms, for instance. The protocol should be designed in a way that the disclosure of\nthe stored data does not compromise the entire system. Also the delectability of such\ndisclosures within the system needs to be examined when designing a protocol.\nIn addition to the previous design objectives, protocol designed for sensor networks\nshould be scalable to cope with the large number of sensors in the network and be fault\ntolerant because sensors are very prone to failures.\n3.\nSYSTEM CONFIGURATIONS\nIn this section, we identify the problems that one might encounter when imple-\nmenting AAKE protocols in ad hoc networks. Therefore, we consider several system\nsettings that occur in different applications, such as the availability of a TTP, the se-\ncurity of the communication channels, the constraints of the devices, the number of\nparticipating domains, etc..\n3.1. Availability of Trusted Third Party (TTP)\nThe availability of a TTP is crucial for a protocol implementation and one of the\nnew challenges of ad hoc networks. A TTP can play several roles in a network, for\ninstance, the TTP could be responsible to initialize devices with secret keys, issue\nand distribute public keys and certificates, distribute session keys to devices that wish\nto securely communicate, or help to verify the validity of certificates by providing\ncertificate revocation lists (CRLs). We distinguish among four different settings for the\navailability of a TTP, described in the following paragraphs and illustrated in Figure 1.\nThe four rows in the figure correspond to the four settings, where the first column\ndescribes the network initialization phase, the second column the event of a joining\n"
        },
        {
            "page_number": 78,
            "text": "WIRELESS NETWORK SECURITY\n71\nTTP\nTTP\nTTP\n2. Running System\na) new node joins\nb) execute AAKE protocol \n1. Network Initialization\nTTP\nTTP\nTTP\nAV-1\nAV-2\nAV-3\nAV-4\nSecure channel\nInsecure channel\nTTP\nNode joining network\nTrusted Third Party\nAd hoc network\nLegend\nFigure 1. Four scenarios of TTP availabilities AV-1 – AV-4, as described in Section 3.1: (1)\nduring the network initialization; and (2) in the running system when (a) new nodes join or\n(b) present nodes establish a secure channel, i.e execute an AAKE protocol.\nnode in the running system phase and the third column the event of present nodes\nestablishing a secure channel, i.e executing an AAKE protocol.\nAV-1: TTP is always available\nThe case that a TTP is accessible by all network nodes at any time is generally not\nconsidered as an option in ad hoc networks, because ad hoc networks should be self-\norganized after their initialization phase. However, in the future it might be reasonable\nto assume an Internet connection in some as hoc network applications, for example via\nan access point. In that case, we could adopt WLAN solutions and modify them to\ncope with the resource constraints and mobility of ad hoc network devices.\n"
        },
        {
            "page_number": 79,
            "text": "72\nKATRIN HOEPER and GUANG GONG\nAV-2: TTP is available at network initialization phase and every time a node joins\nThe second option comprises all scenarios where a TTP is available at the network\ninitialization phase and, in addition, the TTP is accessible by all nodes that subsequently\njoin the network. This assumption is not as restrictive as it might seem, because the\nTTP does not need to be accessible by all network nodes every time a new node joins\na network. For instance, there could be applications in which nodes contact a TTP\nto receive the required system parameters and keys before joining the network. The\nnetwork itself is still self-organized and the present nodes have no access to a TTP.\nAV-3: TTP is available at network initialization phase\nIn this scenario only the nodes that were present at the time of the network initialization\nphase are initialized by the TTP. Usually this is called self-organization property of the\nnetwork. The present network nodes are responsible to take over the tasks of the TTP,\nsuch as issuing and distributing keys and/or certificates to subsequently joining nodes.\nAV-4: No TTP is available at any network phase\nIn this scenario network nodes need to take over the tasks of the TTP during the network\ninitialization phase and in the running system phase. If no TTP is available at any time,\nimplementing security protocols such as AAKE protocols is very challenging. If we\nwant to implement a symmetric scheme we would need to develop a security model\nin which devices can securely exchange their common keys. Whereas implementing a\npublic key encryption schemes would require an authentic channel to exchange public\nkeys without the aid of a TTP that issues keys or key credentials.\n3.2. Other Configuration Parameter\nThere are many other implementation issues that depend on the particular ad hoc\nnetwork application. We will discuss some of those issues that could affect the imple-\nmentations of security protocols.\nFirst of all, the security of the communication channels is a crucial parameter in ad\nhoc network applications. We distinguish two communication channels. One channel\nto exchange the data during the pre-authentication phase and another channel for the\nauthentication and key establishment phases. As discussed earlier, pre-authentication\nrequires a secure channel among the devices to securely exchange authentic public\nkey data or authentic and confidential secret key data. Upon pre-authentication, all\ncommunications can be executed over an insecure channel where the communication\nis secured by the key material that was exchanged during pre-authentication. How a\nsecure pre-authentication or authentication channels can be established is discussed in\nSection 4 and 5, respectively.\nAnother implementation issue is the level of resource constraints. Depending on\nthe computational constraints of the network devices it might be feasible or infeasible\nto execute protocols requiring heavy or many computations, as required in most public\nkey schemes. In addition to the computational constraints, we have to consider the\ncommunication and power constraints when designing or implementing a protocol.\nGenerally sensors are too constrained for implementing public key protocols.\n"
        },
        {
            "page_number": 80,
            "text": "WIRELESS NETWORK SECURITY\n73\nHierarchical ad hoc networks haven been proposed as alternative to flat ad hoc\ntopologies to overcome some limitations of the latter, as for instance described in [5].\nHierarchical ad hoc networks have several layers, where each layer consists of a set of\nsimilar devices. For instance, the lowest layer consists of the least powerful devices,\ne.g. sensors, and each higher level consists of some more powerful devices, where the\ntop level could be the Internet. In this way, all heavy computations could be shifted from\nthe very constrained devices to the more powerful ones and thus asymmetric schemes\ncould become feasible. For this reason, the model is attractive for sensor networks. It\nneeds to be analyzed for particular applications if it is reasonable to assume that higher\nlayers can be accessed by all sensor networks at any time.\nWhen Stajano and Anderson [23] were among the first to consider the special prop-\nerties of ad hoc networks, they assumed a controller (mother duck) and several devices\nthat are controlled (ducklings) in all ad hoc networks. In the proposed resurrecting\nduckling model, the mother duck imprints their ducklings, who, from then on, follow\ntheir mother. In another more recent paper Messerges et. al [21] described some ap-\nplications that require a controller, e.g. sensor networks used for industrial control and\nbuilding automation. In networks without a controller all nodes have similar roles and\nare assumed to have similar resource constraints. Whether we have an ad hoc network\nwith or without controller depends on the application.\nIn some scenarios devices might be aware of their location and are able to provide\ninformation about their location, such as their geographical coordinates. A simple\nsolution for providing the present location of mobile devices is to embed an additional\nintegrated chip, such as a GPS chip, in all devices. For instance, some high-end PDAs\nare already equipped with GPS chips. However, there are many different systems\nthat provide location coordinates depending on the network range and location. The\nmost commonly known systems for tracking down devices are: (1) satellite navigation\nsystems, such as GPS, or the European equivalent Galileo; (2) systems for locating\ndevices inside a building using visual, ultra sonic, radio, or infrared channels; and\n(3) network based positioning system, such as GSM, and WLAN. If a user knows\nthe location of its communication partner the data could be used to build an authentic\nchannel, e.g. for authentication or public key exchange. However, special equipment\nfor tracking devices is unnecessary if the location of devices is predictable. For instance,\nin some sensor networks, the sensors have an expected location. This knowledge is\nused in a location-based pairwise key establishment protocol [18], for instance.\nThe last system property we consider is the number of domains in our network.\nAll devices in one domain share the same domain parameters, such as shared keys,\nthat has been distributed during the network initialization, a certificate issued by the\ndomain’s certification authority (CA), or system parameters required for some compu-\ntations. In most sensor networks, it is reasonable to assume one domain. However, in\nmany MANETs, devices are from different domains. Providing authentication in those\nscenarios is harder to implement. Communicating parties need mechanisms to verify\nthe trustworthiness of devices outside their own domain and to securely agree on some\ncommon system parameters. These compatibility issues have to be considered when\nimplementing an AAKE protocol.\n"
        },
        {
            "page_number": 81,
            "text": "74\nKATRIN HOEPER and GUANG GONG\n4.\nPRE-AUTHENTICATION MODELS\nIn this section we discuss several symmetric and asymmetric pre-authentication\nmodels (PAMs) for providing pre-authentication in ad hoc networks. We summarize\nall models for better comparison in Table 1. We reference some papers that introduced\nprotocols in the respective models in the second column and summarize the advantages\nand disadvantages of each model in the right column.\n4.1. Symmetric Solutions\nWhen using symmetric encryption a secret must be shared among the devices\nthat wish to communicate. The secrets are established during the network initializa-\ntion phase and the pre-authentication phase of the devices. Clearly, an authentic and\nconfidential channel needs to be established to ensure secure pre-authentication. The\nfollowing models describe how such a secure channel can be established.\nPAM-S1. Secure Side Channel Model\nIn this model the secret information is exchanged over a secure side-channel during the\nnetwork initialization phase and the pre-authentication phase of the devices. How this\nsecure channel is established is not further specified in the model and left to be done\nby the users or the administrators that implement the protocol. For instance, the IEEE\nstandard for wireless local area networks (WLAN) IEEE 802.11 [14] does not provide\nany recommendations and information of how pre-authentication can be achieved and\nassumes that pre-authentication has taken place before devices start communicating\nwith each other. Hence, IEEE 802.11 is a protocol standard proposed in the discussed\nmodel.\nPAM-S2. PIN Model\nProtocols in this model require that passwords, PINs, or keys are manually entered in\nall devices that wish to securely communicate. This can be done by an administrator\nduring the network initialization phase or by users as pre-authentication of their devices.\nSolutions in this model do not scale well because the secret needs to be entered manually\nin each device. An example for a protocol in this model is the Bluetooth protocol that\nwas introduced by the Bluetooth Special Interest Group (SIG) [4]. The protocol is\nstandardized as IEEE 802.15 [14] for wireless personal area networks (WPANs).\nPAM-S3. Physical Contact Model\nIn this model the symmetric keys are exchanged by physical contact among the devices.\nNote that the physical contact provides an authentic and confidential channel. The\nrequirement of physical contact among all communicating devices is be too restrictive\nin some applications. A protocol in this model is introduced in [23].\n"
        },
        {
            "page_number": 82,
            "text": "WIRELESS NETWORK SECURITY\n75\nTable 1. Pre-authentication models for ad hoc networks\nModel\nImplementation\nComments∗\nPAM-S1.\nSecure Side-\nChannel\nKeysexchangedoversecureside-\nchannel, e.g. IEEE 802.11 [14]\n−secure channel not pro-\nvided by system itself\nPAM-S2.\nPIN\nPIN manually entered in all de-\nvices, e.g. Bluetooth [4]\n−does not scale well\nPAM-S3.\nPhysical Contact\nKey exchanged by physical con-\ntact, e.g.\nresurrecting duckling\nprotocol [23]\n−requires proximity of the\ndevices\nPAM-S4.\nPairwise Key Pre-\nDistribution\nSensors initialized with\nsubset of key pool before de-\nployed, e.g. [8]\n−only one domain\n−requires TTP for every\ninitialization\nPAM-A1.\nLocation-Limited\nPublic key directly exchanged,\ne.g. [2, 6]\n−requires proximity of de-\nvices\nPAM-A2.\nID-Based\nIdentity used as self-\nauthenticated public key, e.g. [15,\n12]\n+ implicit\npre-authentication\n−KGC is key escrow\nPAM-A3.\nSelf-Certified Pub-\nlic Key\nCertificate embedded in public\nkey, e.g. [9]\n+ implicit\npre-authentication\n−no AAKE protocols\nPAM-A4.\nDistributed CA\nCA represented by n nodes using\nthreshold scheme [27, 16, 19]\n+ self-organized\n−not efficient†\n−requires many nodes\nPAM-A5.\nTrusted Path\nPGP-like; find trusted path be-\ntween two nodes, e.g. [11, 7]\n+ self-organized\n−not efficient\n−requires many nodes\n∗“+”/“−” denote advantages and disadvantages of the model, respectively.\n† Efficiency with respect to computation and communication cost.\n"
        },
        {
            "page_number": 83,
            "text": "76\nKATRIN HOEPER and GUANG GONG\nPAM-S4. Pairwise Key Pre-Distribution Model\nPublic key cryptography is not feasible in sensor networks and therefore only symmetric\nschemes are applicable. The approach that all sensors share the same secret key is not\nsuited because once a single key is compromised the entire sensor network would be\ncompromised as well. Due to the weak physical protection of sensors, compromising\na single sensor and thus its stored key material is very likely. For this reason, sharing\nkeys in a pair-wise fashion seems to be a more reasonable approach. Since sensors\nhave very constrained memory, they cannot store symmetric keys of every other sensor\nin the network. To overcome this constraint, the pairwise key pre-distribution model is\nintroduced, in which each sensor is initialized with a subset of all network keys. Note\nthat all sensors need to belong to the same domain. However, in most sensor network\napplications, it can be assumed that a trusted authority can set-up all sensors before\nthey are deployed. An example of a protocol in this model is in [8].\n4.2. Asymmetric Solutions\nWe describe several asymmetric pre-authentication models (PAM-As) in this sec-\ntion. Each model provides a method to obtain an authentic copy of the public key of\na communication partner. The lack of a central CA is the main problem when imple-\nmenting asymmetric protocols in ad hoc networks. We distinguish four categories of\nPAM-As: (1) with CA and use of certificates; (2) with CA and no use of certificates; (3)\nwithout CA and use of certificates; and (4) without CA and no use of certificates. The\nfirst category includes the distributed CA model; the second one includes the identity-\nbased model and the self-certified public key; the third category contains the trusted\npath model; and the fourth contains the location-limited model.\nPAM-A1. Location-Limited Model\nIf proximity of the ad hoc network devices is given, a secure pre-authentication channel\ncanbeestablishedbyvisualorphysicalcontactamongthecommunicatingdevices. This\nsecure pre-authentication channel enables the devices to directly exchange their public\nkeys, i.e. without the necessity of a CA and public key certificates. This model is\nbased on two assumptions, (1) all participants are located in the same room; and (2)\nall participants trust each other a priori. The model is well suited in all scenarios that\nmeet those two assumptions and not applicable in any other scenario. Protocols in this\nmodel are introduced in [2, 6]. Note that in most cases where devices can perform\nphysical contact implementing the physical contact model (PAM-S3) seems to be more\nreasonable.\nPAM-A2. Identity-Based Model\nIdentity (ID)-based schemes, introduced in [22], do not require any key exchange prior\nto the actual authentication, because common information, such as names and email\naddresses, is used as public key. Since public keys are self-authenticating, certificates\nare redundant in this model. Pre-authentication is implicitly provided by the system\nbecause the (authentic) public keys of all network devices are known prior communi-\ncating. As a consequence, protocols in the identity-based model do not require any\n"
        },
        {
            "page_number": 84,
            "text": "WIRELESS NETWORK SECURITY\n77\nsecure pre-authentication channel. This feature makes the ID-based model attractive\nfor ad hoc networks. ID-based schemes require a TTP that serves as key generation\ncenter (KGC) in the network initialization phase in order to generate and distribute the\npersonal secret keys to all users. Drawbacks of this model are: (1) the KGC knows\nthe secret keys of all users; and (2) a confidential and authentic channel between the\nCA and each network device is required for the securely distribution of the secret keys.\nThe latter problem can be eliminated by using a blinding technique as shown in [17]\nand the first drawback is shown to have low impact in ad hoc networks in [13]. The\nfirst protocol in this model is in [15], but the authors do not provide an actual AAKE\nprotocol and many open questions for a protocol implementation remain. Some AAKE\nprotocols in this model are in [12].\nPAM-A3. Self-Certified Public Key Model\nIn this model the certificates are embedded in the public keys themselves. So-called\nself-certified public keys are introduced in [9] and, other than in ID-based schemes, the\nidentity itself is not directly used as a key. In fact the identity is a part of the user’s\npublic key and signed by a CA and the users themselves. Hence, the public keys are\nunpredictable and need to be exchanged prior to the communication. The authenticity\nof the public keys is provided by the keys themselves, and thus we do not need a secure\npre-authentication channel. In addition, this approach helps to save some bandwidth\nand memory space, because certificates do not need to be transmitted and stored. A\nCA is required to generate the self-certified public keys using the devices’ public keys,\nidentifiers, and the CA’s master secret key as input. Protocols in this model are an\nauthentication protocol without key agreement and a static DH-like key agreement [9].\nPAM-A4. Distributed CA Model\nIn the distributed CA model the power and the tasks of a CA are distributed to t network\nnodes by implementing a (t, n)-threshold scheme. The idea is based on the fact that a\nCA should not be represented by a single node, because nodes can be relatively easily\ncompromised by an adversary. In this model a group of t nodes can jointly issue and\ndistribute certificates. Protocols in this model might support certificate renewing and\nrevocation. We distinguish two cases, (1) a distributed CA with special nodes and (2)\na distributed CA without special nodes. In the first case t special nodes, that have\nmore computational power and that were present at the network initialization phase,\nrepresent the CA. The special role of the server nodes contradicts with the property of\nsimilar constrained devices as stated in our design goals. An example of a protocol that\nhas been proposed in this model is [27]. In the distributed CA model without special\nnodes, any t network node represent the CA and can thus issue certificates. Protocols\nthat have been proposed in this model are [16, 19].\nPAM-A5. Trusted Path Model\nThe trusted path model emphasizes the self-organization property which is a unique\nand challenging feature of ad hoc networks. Network nodes issue and distribute their\nown certificates and sign other certificates. The model assumes the existence of trust\nbetween some nodes and generates trust between nodes in a PGP manner, i.e. by\n"
        },
        {
            "page_number": 85,
            "text": "78\nKATRIN HOEPER and GUANG GONG\nfinding a so-called trusted path consisting of certificates between the communicating\nnodes. The performance of pre-authentication highly depends on the length of the\ntrusted path, which is generally hard to predict. This approach is very efficient in the\nset-up phase and does not require any heavy computation steps from any parties other\nthan the communicating ones. However, a node probably needs to verify more than\none certificate for pre-authentication. An example of a proposed protocol in this model\nis [11]. This model is also applied to a group case, in which trusted subgroups search\nfor intersections to create a trusted path [10].\n5.\nAUTHENTICATION MODELS\nAfter the pre-authentication phase, the exchanged key material can be used to\nenable authentication and key establishment in any of the authentication models (AMs)\ndescribed in this section. We briefly discuss some symmetric, hybrid, and asymmetric\nauthentication models (AM-S, AM-H, AM-A) and summarize the models in Table 2,\nwherewereferenceproposedprotocolsinthesecondcolumnandsummarizeadvantages\nand disadvantagesofthemodelsintherightcolumn. Pleasenotethatmanymoremodels\nexist and we only present a small subset, where we limit our focus to models suitable\nto ad hoc networks.\n5.1. Symmetric Solutions\nAfter successful pre-authentication has taken place in any of the previously de-\nscribed symmetric pre-authentication models, we can run any symmetric AAKE pro-\ntocol.\nAM-S1. Challenge-response Using Symmetric Schemes\nThe devices can use their shared key in a challenge-response type protocol [20], in\nwhich devices authenticate each other by demonstrating knowledge of the shared key\nby encrypting a challenge.\n5.2. Hybrid Solutions\nSomeadhocnetworksolutionscombinesymmetricandasymmetriccryptoschemes\ntoprovideentityauthenticationandoptionallykeyestablishmentafterpre-authentication\nphase has taken place in any of the presented pre-authentication models.\nAM-H1. Password Model\nDepending on the available memory size and the way the secret is exchanged, it might\nbe desirable to share a short password instead of a long secret key. Note that such pass-\nwords are weak secret keys. Due to their shortness, passwords are prone to brute-force\nattacks, where user-friendly passwords are also prone to off-line dictionary-attacks.\nThe password needs to be securely exchanged by one of the PAM-S discussed in Sec-\ntion 4.1. AAKE protocols in the password model combine a weak password and an\nasymmetric scheme to obtain a strong shared key and are called password-authenticated\n"
        },
        {
            "page_number": 86,
            "text": "WIRELESS NETWORK SECURITY\n79\nTable 2. Authentication models for ad hoc networks\nModel\nImplementation\nComments∗\nAM-S1.\nChallenge-\nResponse\nDevices demonstrate knowledge of\nshared key by encrypting a\nchallenge [20]\n+ efficient†\n−requires long and secure\nshared secret\nAM-H1.\nPassword\nShared password is used for encrypting\npublic keys, e.g. [3, 1]\n+ requires short (memoriz-\nable) password\n−not efficient\nAM-A1.\nChallenge-\nResponse\nDevices either decrypt a challenge that is\nencrypted under their public key or sign\na challenge [20]\n−not efficient\nAM-A2.\nKey Chain\nAnchor x0 of hash chain is private key,\nxn public key [24, 25, 26]\n+ very efficient\n−no key agreement\n∗“+”/“−” denote advantages and disadvantages of the model, respectively.\n† Efficiency with respect to computation cost.\nkey exchange (PAKE) protocols [3]. Due to the use of asymmetric crypto schemes,\nPAKE protocols require some heavy computational steps, and are thus only applicable\nto ad hoc networks consisting of powerful devices that have sufficient computation\npower. Examples of protocols in this model are [3] for the two-party case and [1] for\nthe multi-party case.\n5.3. Asymmetric Solutions\nWe can implement any asymmetric AAKE protocol after pre-authentication has\ntaken place in one of the PAM-As discussed in Section 4.2.\nAM-A1. Challenge-Response Using Asymmetric Schemes\nOnce the devices share authentic copies of their public keys they can use either these\npublic keys or their own private keys to prove their identities. A common method are\nchallenge-response type protocols [20]. The establishment of an encryption key for the\ncurrent session may be a part of the protocol as well.\nAM-A2. Key Chain Model\nHash chains [20] are an asymmetric approach that is attractive for ad hoc network due\nto the excellent performance. In hash chain schemes, a hash function h(·) is applied\nn times to a random value x. The initial value x0 = x is the so-called anchor which\nserves as the private key, whereas the last value of the hash chain xn = hn(x) serves\n"
        },
        {
            "page_number": 87,
            "text": "80\nKATRIN HOEPER and GUANG GONG\nas public key. Each device first computes its own hash chain, also called key chain,\nthen authentically exchanges xn with its communication partners in one of the pre-\nauthentication models described in Section 4.2. The value x0 is kept secret. A device\nthat is challenged by a value xi from its key chain can prove its identity by responding\nwith the previous value xi−1 of the chain. Only a device that knows the anchor x0 is able\nto compute the required response. This scheme requires only the computation of hash\nvalues which can be implemented very efficiently. Note that schemes implementing\nkey chains provide only unidirectional authentication and no key is established during\nthe protocol execution. Examples of the protocols in this model are [24, 25]. Note that\nthe protocol in [24] is broken and fixed in [26].\n6.\nCONCLUSIONS\nAuthentication and authenticated key exchange are both identified as primary secu-\nrity objectives in ad hoc networks. In this chapter, we introduced a security framework\nfor general ad hoc networks to achieve these two goals. As part of the security frame-\nwork we defined network phases, protocol stages, and design goals. Next, we coped\nwith the diversity of ad hoc network applications. Therefore, we identified crucial\nconfiguration parameters of particular applications, that need to be taken into account\nwhen implementing AAKE protocols in these scenarios. Here, our special focus was\non the availability of a TTP, but many other security related configuration parameters\nwere discussed as well. Considering all special network and device constraints, we\nderived a set of design objectives for general ad hoc network protocols. Taking all\nprevious results into account, we finally categorized a number of pre-authentication\nand authentication models based on symmetric, hybrid, and asymmetric cryptographic\nschemes. The models can be implemented as a part of the security framework and they\ncorrespond to the diversity of ad hoc network applications. We analyzed the models,\npointed out their advantages and disadvantages, and showed for which application par-\nticular models are best suited. Furthermore, we identified several previously proposed\nAAKE protocols that are suitable for each model. Our results can be used as a toolbox\nfor designing and analyzing AAKE protocols as well as a guideline for choosing the\nbest suited protocol for particular ad hoc network applications.\nWe conclude from our analysis that some commercial ad hoc network applications\ncan be securely and efficiently implemented by existing symmetric solutions. The\nPIN model is applicable to all PANs, in which a user can set up all of his/her devices\nwith one PIN or password, or an administrator is able to set up all authorized devices\nin order to share network resources. The physical contact model is suitable for all\napplications where people or devices, who already trust each other, are located in a\nsmall area. Protocols in the pre-distribution scheme are suited in sensor networks\nin which all sensors belong to one domain. An asymmetric approach which seems\nto be suitable for mobile device-terminal connections is the exchange of public keys\nover a location-limited channel. This approach could be implemented in some civil\napplications, such as virtual classrooms, internet access points, and all communications\n"
        },
        {
            "page_number": 88,
            "text": "WIRELESS NETWORK SECURITY\n81\nbetween PDAs and laptops of different users. The approach is limited to networks with\na small number of devices that provide moderate computational power. All approaches\nin the distributed CA and trusted path model are only suitable for networks with a large\nnumber of nodes. Furthermore, we believe that these two models are not efficient in\nterms of the computational and communication overhead. The identity-based and the\nself-certified public key model are both promising pre-authentication models because\nthey do not require a secure channel. However, those models need to be further studied\nand protocols have to be proposed.\n7.\nREFERENCES\n1. N. Asokan and P. Ginzboorg, Key Agreement in Ad Hoc Networks, Computer Communications, Vol.\n23, No. 17, 2000, pp. 1627-1637.\n2. D. Balfanz, D.K. Smetters, P. Stewart, and H. Chi Wong, Talking to Strangers: Authentication in Ad-\nHoc Wireless Networks, Proceedings of Network and Distributed System Security Symposium 2002\n(NDSS ’02), 2002.\n3. S.M. Bellovin and M. Merritt, Encrypted Key Exchange: Password-Based Protocols Secure Against\nDictionary Attacks, Proceedings of the 1992 IEEE Symposium on Security and Privacy, IEEE Computer\nSociety, ISBN: 0-8186-2825-1, 1992, pp. 72-84.\n4. Bluetooth SIG, Specification of the Bluetooth System, Version 1.1; February 22, 2001, available at\nhttps://www.bluetooth.com.\n5. M. Bohge and W. Trappe, An authentication framework for hierarchical ad hoc sensor networks, Pro-\nceedings of the 2003 ACM workshop on Wireless security, ISBN:1-58113-769-9, ACM Press, 2003,\npp.79-87.\n6. M. Cagalj, S. Capkun and J.P. Hubaux, Key agreement in peer-to-peer wireless networks, to appear in\nProceedings of IEEE, Special Issue on Security and Cryptography, 2005.\n7. S. ˇCapkun, J.-P. Hubaux, and L. Butty´an, Self-Organized Public-Key Management for Mobile Ad Hoc\nNetworks, IEEE Transactions on Mobile Computing, Vol. 2, No. 1, 2003, pp. 52-64.\n8. L. Eschenauer and V.D. Gligor, A Key-Management Scheme for Distributed Sensor Networks, 9th ACM\nconference on Computer and Communications Security, ISBN:1-58113-612-9, ACM Press, 2002, pp.\n41-47.\n9. M. Girault, Self-Certified Public Keys, Advances in Cryptology- EUROCRYPT ’91, LNCS 547,\nSpringer, 1991, pp. 490-497.\n10. S. Gokhale and P. Dasgupta, Distributed Authentication for Peer-to-Peer Networks, Symposium on\nApplications and the Internet Workshops 2003 (SAINT’03 Workshops), IEEE Computer Society 2003,\nISBN 0-7695-1873-7, 2003, pp. 347-353.\n11. J.P. Hubaux, L. Butty´an and S. ˇCapkun, The Quest for Security in Mobile Ad Hoc Networks, ACM\nSymposium on Mobile Networking and Computing –MobiHOC 2001, 2001.\n12. K. Hoeper and G. Gong, Identity-Based Key Exchange Protocol for Ad Hoc Networks, Canadian\nWorkshop of Information Theory -CWIT ’05, 2005.\n13. K. Hoeper and G. Gong, Short Paper: Limitations of Key Escrow in Identity-Based Schemes in Ad\nHoc Networks, Security and Privacy for Emerging Areas in Communication Networks –SecureComm\n‘05, 2005.\n14. IEEE\n802.11,\nStandard\nSpecifications\nfor\nWireless\nLocal\nArea\nNetworks,\nhttp://standards.ieee.org/wireless/.\n"
        },
        {
            "page_number": 89,
            "text": "82\nKATRIN HOEPER and GUANG GONG\n15. A. Khalili, J. Katz, and W. Arbaugh, Toward Secure Key Distribution in Truly Ad-Hoc Networks, 2003\nSymposium on Applications and the Internet Workshops (SAINT 2003), IEEE Computer Society, ISBN\n0-7695-1873-7, 2003, pp. 342-346.\n16. J. Kong, P. Zerfos, H. Luo, S. Lu, and L. Zhang, Providing Robust and Ubiquitous Security Support for\nMobile Ad-Hoc Networks, International Conference on Network Protocols (ICNP) 2001, 2001.\n17. B. Lee, C. Boyd, E. Dawson, K. Kim, J. Yang, and S. Yoo, Secure key issuing in ID-based cryptography,\nCRPIT ’04: Proceedings of the second workshop on Australasian information security, Data Mining\nand Web Intelligence, and Software Internationalisation, Australian Computer Society, Inc., 2004, pp.\n69-74.\n18. D. Liu and P. Ning, Location-Based Pairwise Key Establishments for Static Sensor Networks, 1st ACM\nWorkshop Security of Ad Hoc and Sensor Networks (SASN) ’03, ISBN:1-58113-783-4, ACM Press,\n2003, pp. 72-82.\n19. H. Luo, P. Zerfos, J. Kong, S. Lu, and L. Zhang, Self-Securing Ad Hoc Wireless Networks, Seventh\nIEEE Symposium on Computers and Communications (ISCC ’02), 2002.\n20. A.J. Menezes, P.C. von Orschot, and S.A. Vanstone, Handbook of Applied Cryptography, 1997 by CRC\npress LLC.\n21. T.S. Messerges, J. Cukier, T.A.M. Kevenaar, L. Puhl, R. Struik, and E. Callaway, A security design for\na general purpose, self-organizing, multihop ad hoc wireless network, 1st ACM workshop on Security\nof ad hoc and sensor networks (SASN) ’03, ISBN:1-58113-783-4, ACM Press, 2003, pp. 1-11.\n22. A. Shamir, Identity-based Cryptosystems and Signature Schemes, Advances in Cryptology- CRYPTO\n’84, LNCS 196, Springer, 1984, pp. 47-53.\n23. F.StajanoandR.Anderson, TheResurrectingDuckling: SecurityIssuesforAd-HocWirelessNetworks,\nIn Proceedings of the 7th International Workshop on Security Protocols, LNCS 1796, Springer, pp.\n172-194, 1999.\n24. A. Weimerskirch and D. Westhoff. Zero Common-Knowledge Authentication for Pervasive Networks,\nTenth Annual International Workshop on Selected Areas in Cryptography (SAC 2003), 2003.\n25. A. Weimerskirch and D. Westhoff, Identity Certified Authentication for Ad-hoc Networks, Proceedings\nof the 1st ACM workshop on Security of ad hoc and sensor networks (SASN), 2003, ACM Press, ISBN:1-\n58113-783-4, 2003, pp. 33-40.\n26. S. Lucks, E. Zenner, A. Weimerskirch, and D. Westhoff, How to Recognise a Stranger - Efficient and\nSecure Entity Recognition for Low-End Devices, submitted for publication.\n27. L. Zhou and Z.J. Haas, Securing Ad Hoc Networks, IEEE Network Journal, Vol. 13, No. 6, 1999, pp.\n24-30.\n"
        },
        {
            "page_number": 90,
            "text": "4\nPROMOTING IDENTITY-BASED KEY\nMANAGEMENT IN WIRELESS AD HOC NETWORKS\nJianping Pan\nDept. of Computer Science\nUniversity of Victoria, BC, Canada\nE-mail: pan@uvic.ca\nLin Cai\nDept. of Electrical & Computer Engineering\nUniversity of Victoria, BC, Canada\nE-mail: cai@uvic.ca\nXuemin (Sherman) Shen\nDept. of Electrical & Computer Engineering\nUniversity of Waterloo, ON, Canada\nE-mail: xshen@bbcr.uwaterloo.ca\nIn wireless ad hoc networks, mobile peers communicate with other peers over wireless\nlinks, without the support of preexisting infrastructures, which is an attractive form of\npeer communications for certain applications. Although many enabling technologies have\nprogressed significantly in recent years, the highly-anticipated deployment of large-scale,\nheterogeneouswirelessadhocnetworksstillfacesconsiderabletechnicalchallenges, among\nwhich achieving secure, trustworthy and dependable peer communications is a major one.\nIn this chapter, we promote identity-based key management, which serves as a prerequisite\nfor various security procedures. We first identify that peer identity plays an irreplaceable\nrole in wireless ad hoc networks, where autonomous peers can join or leave such systems\nand change their location in these systems at any time. Next, we show that identity-based\nkey management schemes are effective and efficient for bootstrapping any chosen security\nprocedures, especially in wireless ad hoc networks where both over-the-air communica-\ntion and on-board computing resources can be severely constrained. Finally, we illustrate\nidentity-based secure communication schemes with a security enhancement to the Dynamic\nSource Routing protocol. We find that identity-based schemes are intrinsically suitable for\nand practically capable of securing wireless ad hoc networks and may have great impact on\ndealing with other network security issues.\n"
        },
        {
            "page_number": 91,
            "text": "84\nJIANPING PAN, et al.\n1.\nINTRODUCTION\nWith the rapid advance of miniaturized computers and radio communication tech-\nnologies, wireless ad hoc networks have attracted a lot of attention from both research\ncommunities and the industry in recent years [1, 2, 3, 4]: without relying on any preex-\nisting communication and computing infrastructures, autonomous peers are envisioned\nto communicate with other peers over wireless links, or to assist communications among\nothers when necessary. Also, mobile peers can join or leave such systems at any time;\nwhen peers are in these systems, they can change their location at any time. This\nself-organizing and adaptive form of peer communications is particularly attractive in\ncertain scenarios, where communication or computing infrastructures are either too\nexpensive to build or too fragile to maintain. Wireless ad hoc networks have found\nmany applications in military, commercial and consumer domains; they also have other\nvariants (e.g., wireless sensor networks) with various similarities.\nHowever, thehighly-anticipateddeploymentoflarge-scale, heterogeneouswireless\nad hoc networks still faces considerable technical challenges. Among them, achieving\nsecure, trustworthy and dependable peer communications is a major one, which can hin-\nder the further development of these systems. Due to the absence of properly-protected\nmedia and well-trusted infrastructures, and due to the reliance on unknown third-parties\nto relay data, peer communications in these systems are intrinsically vulnerable to var-\nious passive and active attacks [5], which can compromise the confidentiality, integrity\nand authenticity of information exchange among peers. Also, in some wireless ad hoc\nnetworks, peers can become selfish, greedy and even tampered by adversaries, which\nbrings more challenges to secure the already vulnerable peer communications in these\nsystems.\nMany efforts have been devoted to securing peer communications in wireless ad\nhoc networks, and most of them are based on either symmetric-key (SKC) or public-\nkey cryptography (PKC) systems (see [5, 6] and the references therein). Although\nthese systems have successfully demonstrated their capability in securing information\ninfrastructures in other contexts (e.g., the Internet), many of them are found inade-\nquate for wireless ad hoc networks, either due to severe communication or computing\nconstraints, or due to the lack of infrastructure support in such networks. One issue,\nkey management, is of the greatest interest [7], since it is a prerequisite for any secu-\nrity procedures of publicly-known cryptographic algorithms. For example, in SKC,\nshared keys or preshared secrets should be arranged for involved peers before they can\ncommunicate; in PKC, information senders should obtain the public-key of receivers\nand verify it with trusted third-parties. Pairwise keying is cumbersome in wireless ad\nhoc networks of many peers with dynamic membership; public-key verification usually\nrelies on centralized key directories or hierarchical certificate authorities, which may\nnot be always available in wireless ad hoc networks. In addition, voluntary public-\nkey verifications may introduce a risk of denial-of-service (DoS) attacks due to the\namount of computing and communication resources involved even before the regular\ncommunications among peers can happen.\n"
        },
        {
            "page_number": 92,
            "text": "WIRELESS NETWORK SECURITY\n85\nIn this chapter, based on the latest advances in identity-based cryptography (IBC),\nwe prompt identity-based key management in wireless ad hoc networks. IBC is a special\nform of PKC [8]. In regular PKC, an entity (or a peer in ad hoc networks) of known\nidentity generates a pair of public-key and private-key or obtains it from public-key\ninfrastructures (PKIs). The binding between the peer identity and its public-key should\nbe certified by trusted third-parties; otherwise, a peer can easily impersonate others by\nforging their public-keys and compromise communications intended for those peers.\nIn IBC, such binding and verifying are unnecessary, since the public-key of a peer is\nexactly its identity (or a known transformation of the identity). As far as a peer can\ncommunicate with others by their identity, the peer can apply any security procedures\nbootstrapped from identities to secure its communications with those peers. We find\nthat the unique features offered by IBC make identity-based key management a strong\ncandidate for securing peer communications in wireless ad hoc networks.\nThe contributions of this chapter are twofold. First, we present identity-based\nkey management schemes designed for bootstrapping various security procedures in\nwireless ad hoc networks. We show that these schemes not only accomplish their\ngoals without the support of communication and security infrastructures, but also ac-\ncommodate dynamic peer membership for potentially a large number of mobile peers.\nAlso, these schemes are effective and efficient. For example, a sender-only peer has\nno security overhead in terms of verifying the public-key of others or obtaining its\nown private-key; a peer can send another peer some information only accessible by\nthe latter in the future; a compromised peer can be easily identified and excluded from\nsuch systems. Second, we illustrate identity-based secure communication schemes\nwith a security enhancement to the Dynamic Source Routing (DSR) protocol, in order\nto demonstrate that these schemes are intrinsically suitable for and practically capable\nof securing wireless ad hoc networks. We also expect that such schemes have great\nimpact on dealing with other network security issues. An IBC and threshold-based\nkey distribution scheme is independently proposed in [9]; in contrast to a conceptual\nsketch in [9], here we give a concrete design of all necessary building blocks. Although\nIBC has been explored in other contexts such as IPsec, personal area networks, IPv6\nneighbor discovery and grid computing [10, 11, 12, 13], our goal in this chapter is not\nonly to show that IBC-based schemes can support confidentiality, integrity and authen-\nticity, but also to reveal that these security properties can be achieved more effectively\nand efficiently with IBC-based schemes due to the irreplaceable role of peer identity in\nwireless ad hoc networks.\nThe remainder of this chapter is organized as follows. In Section 2, we present\na model of wireless ad hoc networks and their security requirements; we also briefly\noverview identity-based cryptography and its latest advances. In Section 3, we in-\ntroduce identity-based key management schemes for bootstrapping and managing any\nchosen security procedures in wireless ad hoc networks. In Section 4, we illustrate\nidentity-based secure communication schemes to ensure the confidentiality, integrity\nand authenticity of information exchange among autonomous peers in these systems;\nwe also design a security enhancement to DSR, with focus on its route discovery and\nmaintenance procedures and its resistance against various attacks. Section 5 offers\n"
        },
        {
            "page_number": 93,
            "text": "86\nJIANPING PAN, et al.\nd1\nd2\ni\nj\nk\nd\npark\nmobile\nactive\nstationary\nidle\nbooth\nFigure 1. A wireless ad hoc network at a recreation park.\nfurther discussion, and Section 6 reviews related work. Section 7 concludes the chapter\nwith directions of our future work.\n2.\nPRELIMINARIES\n2.1. Network Model\nWireless ad hoc networks are fully-distributed systems of self-organizing peers that\nwant to exchange information over wireless links but do not rely on any preexisting\ninfrastructures [1, 2, 3, 4]. Fig. 1 shows such networks in a generic format. Mobile\npeers (e.g., laptop computers with wireless interfaces as filled or unfilled dots) can join\nor leave such systems (depicted by a large dashed circle, e.g., a recreation park) at any\ntime. Only peers require keying have to pass by an offline authority regularly (e.g., a\nticketing booth within a small dotted circle). However, there are no physical barriers\naround the vicinity, and peers can join or leave systems at any locations (e.g., a sender-\nonly peer without keying). While peers are in the system, they can remain stationary or\nchange their location, and keep idle or communicate with others. Also, peers can assist\ncommunications among others if they choose to do so. Without any centralized online\nauthorities, peers communicate in uni- or bi-direction, single- or multi-hop, single- or\nmulti-path, and single- or multi-point form, or any combinations of these forms.\nFor a given information exchange between two peers, e.g., transferring a bulk data\nof b unit amount from peer i to k that is d unit distance away in Fig. 1 (zoomed in a\ndotted ellipse), i has two strategies. With the first one, i transmits b to k directly, and\nconsumes energy\net\ni(b, d) = (t1 + t2dn)b,\n(1)\nwhere 2 ≤n ≤6 is the path loss exponent, and t1 and t2 are the coefficients of distance-\nindependent and distance-related energy consumption, respectively. Some facts may\nprevent i from adopting this strategy: i) when d > D, where D is the maximum\ntransmission range of i; ii) direct wireless communications of i and k may impose\nstrong interference on peers between i and j. With the second strategy, when there is\na third peer j that lies in between i and k, i may save energy by requesting j to relay b\n"
        },
        {
            "page_number": 94,
            "text": "WIRELESS NETWORK SECURITY\n87\nto k. Without loss of generality, assume j is d1 away from i and d2 from k. If d1 < d,\nrelaying b through j is preferable for i, while j has to volunteer er\nj(b) = rb to receive\nb, et\nj(b, d2) = (t1 + t2dn\n2)b to transmit b to k, and eo\nj(b) to cover its local expenses. If\net\ni(b, d) −et\ni(b, d1) > er\nj(b) + et\nj(b, d2) + eo\nj(b),\n(2)\nrelaying b through j is also preferable for the entire system, since, overall, it takes less\nenergy to move the same b from i to k. This relaying strategy can be applied recursively\nto peers in the vicinity of i, j and k.\nTo enable such relayed communications, peers need to identify other peers of their\ninterest. There are many different naming schemes; e.g., on the Internet, nodes are\nidentified by their IP address or host name. Public IP addresses are location-dependent\nwith regard to the attachment point of addressed nodes to the global Internet routing\nfabric, which is not available in wireless ad hoc networks. Although host names can\nbe location-invariant, they have to be mapped to IP addresses with the assistance of\na hierarchical Domain Name System (DNS), which may not be always available in\nwireless ad hoc networks. Therefore, mobile peers can only be identified by their own\nidentity of spatial and temporal invariance. For example, peers propose their identity\nwhen joining such systems (sender-only peers can have no identity and remain anony-\nmous). To keep collision-free in the identity space, the offline authority can append a\ntimestamp or sequence number to the identity proposed by peers when they request key-\ning. To find a multi-hop path from one peer to another, the source peer initiates a route\ndiscovery procedure with the source and destination peer identities. Route requests are\nforwarded by neighboring peers after their identities have been appended to the request.\nThis process is recursive until the request reaches the destination peer, where a route\nreply is sent back to the source peer by reversing the forward path identified by the\nidentities of forwarding peers. As we can see, peer identity plays an irreplaceable role\nin enabling multi-hop communications in wireless ad hoc networks; in the next section,\nwe will see it also plays an important role in key management.\n2.2. Security Model\nWith relaying, peers no longer always communicate with intended peers directly, so\nthey should be assisted with additional security procedures to ensure the confidentiality,\nintegrity and authenticity of their information exchange with intended peers. Without\nany preexisting communication and security infrastructures, peers may have to deal\nwith unknown relaying peers without the preestablished trustworthiness.\nManysecuritythreatsappearinadhocnetworks[5]. Inadditiontopoorly-protected\ncommunication channels open to various passive and active attacks, pairwise trust-\nworthiness among all involved peers is unpractical to build and difficult to maintain,\nespecially when there is a large number of mobile peers joining or leaving such sys-\ntems without notice. Selfish peers have the motive and excuse to corrupt relayed data\n(no matter intentionally or not). Relaying peers and neighboring non-relaying peers\nhave the incentive to eavesdrop relayed data. Malicious or compromised peers can\n"
        },
        {
            "page_number": 95,
            "text": "88\nJIANPING PAN, et al.\nimpersonate others to steal genuine information or inject false information into these\nsystems. Besides data plane attacks, there are control plan attacks (e.g., black/gray\nholes [14, 15], replay attacks [16], network partitions) for specific routing protocols.\nWhen collaborative relaying becomes profitable, greedy relaying or non-relaying peers\nhave a strong motive to boost their wealth improperly, by trying to cheat source, desti-\nnation or other relaying peers. When there is a certain number of malicious peers, they\nmay collude with each other and attempt to beat the entire system (e.g., wormholes [17],\nrushing attacks [18]). Our focus in this chapter is not on individual or new data plane\nor control plane attacks, but on the identity-based key management schemes that can be\nused to bootstrap various security procedures to defend against these and other possible\nattacks.\nTraditional cryptographic techniques have been used to provide certain security\nproperties in networks with trusted infrastructures; similar efforts were attempted in\nwireless ad hoc networks. For example, source and destination peers should authenti-\ncate to each other before information exchange. Also, information should be encrypted\nby source peers to keep confidentiality, and be verified by destination peers to preserve\nintegrity. These procedures rely on either certified public-keys in PKC-based systems,\nor pairwise shared-keys in SKC-based systems. If there is a trusted infrastructure (e.g.,\na generic PKI in corporate networks or a base-station in multi-hop cellular networks),\nsuch requirements can be satisfied accordingly.\nHowever, these techniques may not be readily applicable to wireless ad hoc net-\nworks. First, there is no generic PKI or central online authority in these systems that\ncan always be involved in communications between any pairs of mobile peers. Sec-\nond, most end-to-end communications in these systems occur in a hop-by-hop manner,\nwhereby unknown third-parties are required to relay packets; i.e., security proprieties\nshould be achieved not only at the end-to-end level, but also at the per-hop level. Finally,\nsome existing security procedures (e.g., electronic payment) either rely on an online\ninteractive authority (e.g., a bank), or are too heavy (in terms of communication and\ncomputing complexity) for wireless ad hoc networks, within which on-board energy\nconstraints are normally the foremost concern.\nIn summary, security procedures bootstrapped by effective and efficient key man-\nagement schemes, identity-based ones as we advocate, are highly desirable to ensure the\nconfidentiality, integrity and authenticity of information exchange among autonomous\npeers in wireless ad hoc networks.\n2.3. Identity-Based Cryptography\nMotivated by these observations, we approach this challenge from a novel angle\nand with a new tool — IBC, a special form of PKC. As shown in Fig. 2(a), in regular\nPKC,thepublic-keyshouldbecertified, sincetherearenointrinsicbindingsbetweenthe\npublic-key and the identity of an entity. Otherwise, any entities can impersonate others\nwith a forged public-key. To facilitate public-key certification, hierarchical certificate\nauthorities (CAs) are introduced, and the root CA should be trusted by everyone. This\n"
        },
        {
            "page_number": 96,
            "text": "WIRELESS NETWORK SECURITY\n89\nentity\nidentity\nprivate key\npublic key\nown\nbinding\ncertify\nown\nown\ntrust\nroot CA\nbinding\nidentity\nentity\nprivate key\n(public key)\nown\ninquire\nextract\nPKG\n(a) regular PKC\n(b) identity-based PKC (IBC)\nFigure 2. Two forms of public-key cryptography systems.\nmodel may not be applied to wireless ad hoc networks, where neither a PKI nor a CA\nhierarchy is easy to build or maintain in practice.\nUnlike regular PKC, in which an entity generates its public-key and private-key\n(or obtains them from PKI) and has the public-key certified by CA, in IBC, the entity\nproposes a unique identity (e.g., a@b.com), which is also its public-key. A private-key\ngenerator (PKG) extracts a corresponding private-key from the public system parame-\nters and the master-key that is only known to the PKG. The procedure is shown in\nFig. 2(b). For example, when a peer i wants to send a message m to another peer k\n(see Fig. 1), m is encrypted with k’s identity idk and the system parameters; only k\ncan decrypt the encrypted message with its private-key pkk and the system parameters.\nWhen k signs the receipt of m, the receipt is manipulated with pkk, and is verifiable\nby everyone knowing idk. i has to know idk when communicating with k, and no one\nelse can compromise these procedures without knowing pkk. Also, IBC can bootstrap\nsymmetric cryptographic procedures by establishing a shared-key ski,k for i and k.\nThe concept of IBC was first introduced by Shamir in 1984 [19], and several\nefficient IBC-based signature schemes had been found subsequently. However, non-\nmediated IBC-based encryption (IBE) has proved to be much more challenging, and\nit is relatively recent that practical IBE schemes were found [8]. The first efficient\nand secure IBE scheme was given by Boneh and Franklin in 2001, which employs\nWeil pairing on elliptic curves and is considered more efficient than using regular\nRSA-based counterparts [20].\nIts security is based on the bilinear Diffie-Hellman\nproblem (BDHP), which is considered secure in the random oracle model (ROM) [21].\nThe Boneh-Franklin (BF-IBE) scheme is semantically secure against chosen ciphertext\nattacks, even when an adversary has the private-key of any entities other than the one\nbeing attacked. Lynn extended the BF-IBE scheme to provide message authenticity\nwithout extra computation cost; i.e., receivers can verify the identity of senders and\nwhether the received messages have already been tampered, even without resorting to\ndigital signatures [22].\nBased on the latest advances in IBC and related techniques, in the next section,\nwe will design key management schemes to bootstrap secure communications among\nidentifiable peers in wireless ad hoc networks, without PKIs, CAs, key directories,\nalways online authorities, or manually-arranged pairwise preshared secrets among all\ninvolved peers.\n"
        },
        {
            "page_number": 97,
            "text": "90\nJIANPING PAN, et al.\n3.\nKEY MANAGEMENT\n3.1. System Setup\nBefore an IBC-powered wireless ad hoc network becomes fully functional (i.e.,\nallowing peers to join the system and request keying), an offline PKG first picks a\nrandom master-key x ∈Zq (q is a prime and Zq is an algebraic field) and a bilinear\nmapping f : G × G →Zq. f is defined on the points of an elliptic curve (as a group\nG), and has the following property that for any P, Q ∈G and for any integer a and b,\nf(aP, bQ) = f(P, bQ)a = f(aP, Q)b = f(P, Q)ab.\n(3)\nThe PKG then picks a random generator P, and publishes P, xP, f and four chosen\ncryptographic hash functions as the public system parameters. These hash functions,\nwhich will be explained shortly, are used to hash an arbitrary identity (e.g., any ASCII\nstrings)toapointontheellipticcurve(H1), toachievesecurityagainstchosenciphertext\nattacks (H2 and H3), and to encrypt plaintext (H4), respectively. The PKG should keep\nx secret, and no one else can derive x even when they have both P and xP.\nA lot of offline entities (e.g., the ticketing booth of a recreation park) can assume\nthe role of PKG, as long as they can keep the master-key secret and extract private-keys\nfrom the master-key for peers joining the system and requesting to be keyed. Once the\nprivate-key is extracted, a peer has no need to communicate with the PKG (nor to keep\nthe PKG online), unless the peer wants to propose a new identity. Also, the offline\nPKG can key peers in batch (e.g., only during normal business hours), since peers can\nreceive regular, encrypted information even before they request keying. Compared\nwith an online PKI, the offline PKG has many advantages in wireless ad hoc networks.\nWith a PKI, whenever a peer k joins a system, the PKI should verify the binding of\nthe public-key of k and its identity, and broadcast the authenticated public-key to all\nexisting peers, or keep the public-key in a central directory for queries from other peers.\nNo matter when another peer i wants to communicate with k, i has to obtain both the\nidentity and the public-key of k, and i should have a way of verifying the public-key.\nThe complexity of obtaining, verifying and managing public-keys creates considerable\noverhead in energy-constrained systems that rely on radio technologies to exchange\nidentities, keys and data.\n3.2. Peer Keying\nWhen a peer k joins an IBC-powered wireless ad hoc network, k proposes a system-\nwide unique identity idk (or the PKG appends a timestamp or sequence number to\npeer identity). The PKG obtains a corresponding point Q = H1(idk) on the elliptic\ncurve by hashing idk, and extracts k’s private-key pkk = xQ from the master-key x.\nidk can be the email address of k, concatenated with temporal or spatial properties\n(e.g., a@b.com@date@site). Identity ownership should be easily verified, e.g., by\nshort-range encounters [23] when peers passing by the PKG or by sending a request-\nto-confirm email to a@b.com. pkk is conveyed back to k in a secure, out-of-band side\n"
        },
        {
            "page_number": 98,
            "text": "WIRELESS NETWORK SECURITY\n91\nchannel (e.g., through the ticketing process at a recreation park); the system parameters\nare periodically broadcasted by the PKG (e.g., through public announcement). To\nfight against identity theft or spoofing, the PKG should not extract private-keys more\nthan once for the same identity even claimed by the same entity; instead, by using\ntimestamp or sequence number, the entire identity space is always collision-free and\nforward-secure.\nThe security of the entire system relies on the master-key x kept by the PKG, since\nthe private-key of all peers in IBC-based wireless ad hoc networks can be derived from\nx. To reduce the risk of total-exposure even if the PKG is compromised and to address\nthe concern of key escrow for peers with a new PKG, x can be distributed in a t-of-n\nmanner to a group of n PKGs by applying threshold cryptography (TC) techniques [24].\nWith TC, k thereby derives pkk alone by combining pkt\nk obtained from any t PKGt.\nUnless there are more than t unknowingly-compromised or bogus PKGs, the secrecy\nof all peers and their private-key are still preserved.\nTo support a large entity population, Gentry and Silverberg extended the BF-IBE\nschemewithahierarchicalPKGstructure(GS-HIBC),wherealower-levelPKGinherits\nthe identities of its ancestors and obtains its master-key from the parent PKG [25]. In\nHIBC-powered systems, peers are identified by a tuple of identities, corresponding to\ntheir location in the PKG hierarchy, which is also their localized public-key. With\nHIBC, a peer can easily roam from one ad hoc network to another, and communicate\nwith peers in other networks, by just knowing their identities and the system parameters\nof the root PKG (not the PKG of correspondent peers). For simplicity, here we focus\non keying with a single PKG; our schemes can be extended for t-of-n or hierarchical\nPKGs as well.\n3.3. Key Maintenance\nIn identity-based schemes, the public-key of a peer is exactly its identity or a known\ntransformation of the identity. Hence, a peer can receive regular information encrypted\nwith its identity from other peers even before the peer has obtained its private-key from\nthe PKG. This unique feature allows asynchronous communications in wireless ad hoc\nnetworks, where autonomous peers can be in active, idle or sleep state periodically\nwithout global synchronization to conserve energy. Also, this feature reduces the cost\nof operating the offline PKG, since peers can request keying in batch only after they\nare actively and willingly involved in receiving information from other peers and when\nthe PKG goes online according to its own schedule. In contrast, in SKC or regular\nPKC systems, peers have to establish pairwise shared-keys or obtain public-key and\nprivate-key pairs way before any secure communications can happen; i.e., keying is\nalways mandatory and proactive for all peers, even if they eventually have no secure\ncommunications throughout the validity of their keys in these systems.\nOnce a peer obtains its private-key extracted from its identity and the system para-\nmeters, the peer can decrypt received information encrypted with its identity, authenti-\ncate itself to other peers, and sign outgoing messages. We will present these procedures\nin detail in the next section. Also, peers can bootstrap shared-keys or derive session-\n"
        },
        {
            "page_number": 99,
            "text": "92\nJIANPING PAN, et al.\nkeys from their identity-based private-keys for symmetric security procedures. Once\nbootstrapped, symmetric procedures have much less overhead than their asymmetric\ncounterparts. Depending on the definition of peer identity, a peer, as well as the PKG,\ncan determine the lifetime of its private-key. For example, a peer can propose the same\nidentity (e.g., username) to systems with different parameters (i.e., the peer will have\ndifferent private-keys in different systems); even if its private-key is compromised in\none system, the information exposure is confined to that system. A peer can propose\nan ephemeral identity (e.g., user@time); even if its private-key is compromised at a\ncertain time, the peer can request a new private-key with a partially-updated identity in\ntime portion, without totally losing its identity or forcedly leaving the system. When\nnecessary, a peer can proactively refresh its identity (e.g., user@date) with the PKG\nand remain forward-secure even if its current private-key is captured and compromised\nby adversaries. To deal with an unknown PKG, a peer can propose a temporary identity\n(e.g., user@site) to a newly-encountered system, while maintaining credentials with\nother well-known systems. As we mentioned, a peer can request keying with multiple\nor hierarchical PKGs to reduce its exposure due to compromised PKGs, and to ease its\nconcern of key escrow by untrusted PKGs.\nThe PKG, on the other hand, can also control the validity of peer identities and\nextracted private-keys. For example, a peer should have a way of proving its identity\nownership (e.g., a@b.com) or accept assigned identities (e.g., prepaid personal identi-\nfication number, PIN). A peer is uniquely identified by its identity, which can be both\ntime and location invariant within the system. No matter how the peer changes its lo-\ncation and status in the system, it solely relies on its identity to receive information and\ncommunicate with other peers. In addition, its identity is related to its reputation (e.g.,\ncooperativeness in relaying) and wealth (e.g., collected credits for its cooperation) in\nthe system. If a peer is found greedy and always fails to relay for other peers, this fact\ncan be taken into account when the peer is in need of relaying by other peers. If a peer is\nfound malicious, either persistently or opportunistically, the peer can be excluded from\nthe system by identity blacklisting or key expiring (e.g., the PKG enforces an identity\nupgrade and refuses to key compromised peers). The PKG can have differentiated\npolicies, e.g., extracting keys of user@month for well-established or reputable peers\n(e.g., a monthly pass to a recreation park) and of user@day for new or ill-behaving\npeers (e.g., a one-time ticket). Certainly, the PKG can enforce a system-wide rekeying\nafter a long time-period by updating the master-key and the system parameters, and\npeers will need to contact the PKG again to extract their new private-key.\nThe irreplaceable role of peer identity in wireless ad hoc networks leads to the\npromotion of identity-based key management schemes in these systems. These key\nmanagement schemes can effectively and efficiently bootstrap security procedures pro-\nposed in Section 4 to ensure the confidentiality, integrity and authenticity of information\nexchange among peers.\n"
        },
        {
            "page_number": 100,
            "text": "WIRELESS NETWORK SECURITY\n93\nm\nΗ3\nσ\nΗ2\nf\nidk\nxP\nP\nrP\nσ   Η2(gr)\nm   H4(  )\nσ\n.\nΗ4\nΗ1\nsystem\nparameter\nQ\nr\ng\noutput\ninput\nU\nf\nV\nW\nΗ2\nΗ4\nm’\nΗ3\nP\npkk\n.\n=U?\nreject m’\nσ\nyes\n’\noutput\nr’\ngr’\nno\noutput m’\ninput\nsystem parameter\n(a) encryption\n(b) decryption\nFigure 3. IBC-based encryption and decryption flows.\n4.\nSECURE COMMUNICATIONS\n4.1. Information Exchange\nEncryption and decryption\nSuppose that peer i wants to send a message m to peer k (see Fig. 1). i first picks\na random number σ, and obtains r = H3(σ, m). i then employs gr = f(xP, Q)r as a\nsession-key for m, where Q = H1(idk), and sends rP to k. Consequently, k has\ngr′ = f(rP, pkk) = f(rP, xQ) = f(P, xQ)r = gr,\n(4)\nsince f(P, xQ) = f(xP, Q) according to the bilinear pairing property of f in (3).\nWith this procedure, both i and k derive the same session key gr, without knowing the\nsecrecy of their counterpart. Other peers can learn about rP and Q, as well as P and\nxP, but they cannot obtain r or x; in other words, there is no way for these peers to\nobtain gr, nor can they recover the encrypted version of message m.\nFig. 3 gives a detailed illustration of the BF-IBE encryption and decryption flows.\nBesides f, only hash functions and XOR operations are used, allowing these procedures\nto be efficiently implemented in resource-constrained peers. Also, there are consider-\nable efforts to implement pairing (e.g., Tate pairing) more efficiently in software and\nhardware [26]. For a plaintext m, the ciphertext has three parts {rP, σ ⊕H2(gr), m ⊕\nH4(σ)}. When k receives a ciphertext {U, V, W}, k first recovers gr′ from U, with\nits private-key pkk, and then recovers σ′ from V , with the hashed H2(gr′). m′ is re-\ncovered from W, with the hashed H4(σ′). Finally, r′ = H3(σ′, m′) is recovered. To\nverify message integrity, k compares r′P with U. If r′P == U, m′ is accepted as m;\notherwise, m′ is rejected by k.\n"
        },
        {
            "page_number": 101,
            "text": "94\nJIANPING PAN, et al.\nAuthenticated encryption\nIf i obtains g = f(pki, H1(idk)) = f(xQi, Qk), k then has\ng′ = f(H1(idi), pkk) = f(Qi, xQk) = g,\n(5)\nsince f(xQi, Qk) = f(Qi, xQk) according to (3). With this procedure, both i and k\nhave derived the same shared-key g, even without having any physical communications\nbetween them. Also, k knows that only i can create such keys with its own pki. Thus, k\ncan be assured that the shared-key g and ciphertext W are indeed created and encrypted\nby i, respectively. This scheme (IBAE) achieves authenticated encryption for messages\nbetween i and k without relying on the digital signature of i or k on each message, which\nis another advantage for energy-constrained wireless ad hoc networks, since signing\ndigital signatures is an expensive procedure in general.\nSigned encryption\nAlthough BF-IBE can verify whether the recovered plaintext should be accepted\nor rejected, message integrity can be significantly strengthened by applying keyed-hash\nmessageauthenticationcode(HMAC)withsharedsecret(e.g., theauthenticatedshared-\nkey g), or by applying signed encryption (i.e., signcryption) in asymmetric procedures.\nLibert and Quisquarter further extended the BF-IBE scheme, and proposed an identity-\nbased signcryption (LQ-IBSC) scheme, by combining the functionality of signature\nand encryption (but with much less cost than that of a sign-then-encrypt procedure)\nand offering confidentiality, authenticity, integrity and non-repudiation seamlessly [27].\nMessage integrity is then verified by applying the same HMAC function with the shared-\nkey derived from the identity of senders and the private-key of receivers, or by applying\nthe unsigncryption procedure in IBSC. When message confidentiality is not a concern,\nBoneh, Lynn and Shacham proposed an IBC-based short signature scheme (BLS-IBS)\nthat is also based on Weil pairing [28]. With efficient elliptic curve cryptography\n(ECC) primitives, a BLS-IBS-based signature is only about half the size of a DSA-\nbased signature, but still offers a similar level of security and protection, which is\nalso very attractive for energy-constrained wireless ad hoc networks, where shorter\nsignatures are always preferred.\n4.2. Message Routing\nWith achieved secure information exchange, we can further secure the underlying\nrouting protocol in wireless ad hoc networks. Now, we assume that peers are collabora-\ntive once they choose to do so. Designing schemes to stimulate peers to be collaborative\nand compensate them if they indeed are is one of our future work items.\n"
        },
        {
            "page_number": 102,
            "text": "WIRELESS NETWORK SECURITY\n95\nRoute discovery\nHere, we want to secure a DSR-like reactive ad hoc routing protocol [29] with\nidentity-based key management. It is feasible to secure other routing protocols with the\ndesigned security procedures [30, 31]. In DSR, when a peer i wants to send a message\nm to another peer k and has no known routes to k in its route cache, i initiates a route\ndiscovery procedure by broadcasting a route request message RREQ{idk, rn, idi}\nwith the identities of i and k, and a sequence number rn to suppress broadcast loops.\nIf a neighboring peer j has a valid route to k in its cache (e.g., {j +1, · · · , k −1, k}), j\ncan respond a route reply message RREP{idk, rn, idi, idj, idj+1, · · · , idk} to i with\nits identity idj and the cached route; otherwise, j appends idj to i’s request message\nand broadcasts the updated RREQ{idk, rn, idi, idj}. This process is recursive, until\nthe request message reaches k, where a route reply message will be generated and sent\nback to i by reversing the forward path. With rn, peers never react on duplicated or\noutdated routing messages, but can learn from bypassed messages.\nObviously, DSR-like routing schemes rely on voluntary peer collaborations, and\nare highly vulnerable to false routing information corrupted or injected by malicious\npeers. To fight against these attacks, routing messages should be authenticated by\ntheir initiators and verified by their recipients. First, i authenticates its routing request\nmessage with its private key pki, which is verifiable by all other peers knowing idi’s\nidentity. Similar procedures are required for peers that forward the appended request\nmessage. When a routing reply message is generated by j or k, the initiator should\nalso authenticate itself and the route information. Finally, when i receives the reply\nmessage, the authenticity of peers (e.g., j and k) among the discovered route is verified\nby their identities (idj and idk). The BLS-IBS-based routing request message arriving\nat k then has the format\nRREQ{{{idk, rn, idi}pki, idj}pkj, ··, idk−1}pkk −1,\n(6)\nwhere {·}pkj implies that the message has been authenticated by j’s private-key pkj,\nand is verifiable by j’s identity idj. En-route peer identity has to appear in routing\nmessages no matter whether systems are powered by IBC. The construction in (6) is\nsimilar to that with a regular PKC; however, with IBC, there is no need to obtain the\npublic-key of a peer to verify its messages, which is very attractive for wireless ad hoc\nnetworks. The routing reply message generated by k and arriving at i has the format\nRREP{{{idk, rn, idi}pki, idj}pkj, · · · , idk}pkk.\n(7)\nIf the reply message is generated by j according to its route cache, it has the following\nformat instead\nRREP{{idk, rn, idi}pki, idj, idj+1, · · · , idk}pkj.\n(8)\nWith (7) and (8), i can tell whether a route to k is actually certified by k or just endorsed\nby j. Hence, no peer can corrupt a relayed routing message without altering message\nauthenticity or revealing its identity.\n"
        },
        {
            "page_number": 103,
            "text": "96\nJIANPING PAN, et al.\nIn the above route discovery procedure, all involved peers have authenticated them-\nselves, so the discovered route is cacheable by all peers, which can reduce the commu-\nnication cost if these peers also want to obtain the route to k or other downstream peers.\nHowever, this procedure requires every en-route peer to sign the hash of the appended\nrequest message, which may impose a non-negligible computing overhead in dense ad\nhoc networks. An alternative is to have all en-route peers authenticate themselves only\nto the request initiator i, by using a keyed hash of the appended request message with\nthe pairwise shared-key derived from their private-key and i’s identity. Accordingly,\n(6) can be redefined in the following format\nRREQ{{{idk, rn, idi}pki, idj}ski,j, · · · }ski,k−1,\n(9)\nwhere {·}ski,j implies that the message is protected by an HMAC with the shared-key\nski,j defined by (5). Similarly, (7) then is redefined in the following format\nRREP{{{idk, rn, idi}pki, idj}ski,j, ··, idk}ski,k.\n(10)\nBy doing so, only i can verify the authenticity of all peers that appear in the discovered\nroute, which suggests that the route is not cacheable by peers other than i, unless they\nhave already established trustworthiness with downstream peers. These peers have\nto initiate their own route discovery if necessary, although some heuristics can help\nthem verify route validity (e.g., i and k exchange data packets successfully after route\ndiscovery).\nRoute Maintenance\nAnother procedure in DSR is route maintenance. If a peer finds that a route is\nbroken, it notifies the source peer with a route error message, and the source peer\ninitiates another route discovery if there are no alternative routes in its route cache.\nApparently, a malicious peer can abuse error report messages and mount DoS attacks.\nTherefore, report messages should be authenticated by the report initiator with its\nprivate-key and be verifiable for everyone knowing its identity (which is included in the\nmessage, along with the sequence number and the reversed forward path). The route\nerror message generated by j has the format\nRERR{idk, rn, idi, · · · , idj−1, idj}pkj,\n(11)\nor alternatively with HMAC,\nRERR{idk, rn, idi, · · · , idj−1, idj}ski,j.\n(12)\nThe source peer and other upstream peers should verify the authenticity and integrity\nof the message, and update their route cache accordingly. Again, the reporting peer\nhas to trade off computing and communication overhead, by choosing either to sign\n"
        },
        {
            "page_number": 104,
            "text": "WIRELESS NETWORK SECURITY\n97\nthe hash of the error report message, or to apply a keyed hash on the message to only\nauthenticate with the source peer by using their shared-key.\nWith identity-based key management schemes, securing information exchange and\nmessage routing in wireless ad hoc networks becomes feasible with either asymmetric\nor symmetric procedures. On the other hand, the irreplaceable role of peer identity in\nthese systems justifies the need and applicability of identity-based key management.\n5.\nFURTHER DISCUSSION\n5.1. Practical extensions\nIdentity-based key management schemes offer many attractive features that are\nhighly desirable in wireless ad hoc networks, in which peer identity usually is the\nonly means to identify autonomous and mobile peers. String-based identity can have\nvery rich semantics (e.g., along with the date and location information). The location-\naware identity (e.g., grid-based one) can assist location-aware routing in wireless ad\nhoc networks: when a peer sends a message to another peer, the routing path for the\nmessage is implicitly suggested by their identities. Also, a peer can propose its identity\nindicating the services (e.g., email@adhoc.net) or content (movie trailer title) provided\nby itself to assist resource discovery in wireless ad hoc networks. When a peer wants to\nobtain a specific service or content, it securely solicits the peer identified by the service\ndescription or the content hash.\nIBC-based schemes with pairing are also very attractive for energy-constrained\nwireless ad hoc networks. For example, the BF-IBE and follow-on schemes employ\nbilinear pairings on elliptic curves in ECC, an approach considered much more efficient\n(in terms of key size and computation complexity) than regular RSA-based PKC proce-\ndures. Most operations in these schemes mainly involve hashing and bitwise XOR, and\nmore efficient pairing implementations in software and hardware are appearing as well.\nIn our secure communication schemes, we provide both asymmetric procedures (e.g.,\nBLS-IBS signature) and their symmetric counterparts (HMAC), so that peers can trade\noff computing and communication overhead properly. Also, IBC-based schemes allow\npeers to authentically establish shared-key and bootstrap even more efficient symmet-\nric operations without having any physical communications beforehand. In addition,\nBoyen gave a multipurpose IBC-based signcryption (IBSE) scheme (a.k.a. swiss army\nknife, since it can be flexibly used for encryption, signing and sign-and-encrypt proce-\ndures) with even stronger security properties (i.e., confidentiality, authenticity, integrity,\nnon-repudiation, anonymity and unlinkability) and better runtime efficiency (less ci-\nphertext expansion and fewer high-cost operations) [32]. The Boyen scheme is also\nbased on bilinear pairings, and can be introduced in our identity-based key management\nschemes. Further, secure IBE schemes without ROM are proposed recently [33], which\ngives more assurance on adopting them in wireless ad hoc networks.\n"
        },
        {
            "page_number": 105,
            "text": "98\nJIANPING PAN, et al.\n5.2. Known limitations\nIn our identity-based key management schemes, peers obtain their private-key from\nthe PKG that oversees the entire system. Therefore, the PKG has total-control over the\nsecrecy and wealth of individual peers. This is not a concern when peers can trust the\nPKG (e.g., the PKG is the administrator of a managed-open wireless ad hoc network).\nHowever, some peers, especially foreign peers, may be concerned about a compromised\nPKG or an unknown PKG that decrypts messages with their private-key extracted by\nthe PKG, impersonates their identity, and collects their wealth during their tenure in the\nsystem. Nevertheless, these concerns also apply to any regular PKC-based systems,\nin which compromised CAs can always issue false certificates to malicious peers, or\nbogus PKIs can later reveal public-key and private-key pairs assigned to genuine peers.\nThere are some identity-based approaches that can alleviate these concerns to\nsome extent. First, the master-key can be distributed to several PKGs that are not\nunder any single administration (e.g., t-of-n PKGs). Therefore, unless the number of\ncompromised or bogus PKGs exceeds a certain threshold, peer secrecy and wealth are\nstill well-preserved. With this approach, peers have to derive their private-key from\nmultiple PKGs, which unavoidably increases their computing cost. Alternatively, peers\ncan resort to hierarchical PKGs when they roam across different systems frequently.\nSecond, the PKG can be required to refresh its master-key and system parameters\nperiodically. Therefore, the vulnerability of a certain master-key and the potential\ndamage of a compromised master-key are limited. With this approach, peers have to\ninquire the PKG periodically as well to extract their private-key from the latest master-\nkey and system parameters, which increases their communication cost. We argue that\nthe PKG of a wireless ad hoc network usually is the entity, often offline, that enables the\nsystem by providing other resources (e.g., the PKG is the ticketing booth of a recreation\npark), and that peers should have a certain degree of trustworthiness on the PKG while\nthey are willingly in these systems. A visiting peer can propose a PKG-dependent\nidentity to an unknown system, while still maintaining credentials with trusted PKGs\nin other systems, until the peer has developed trustworthiness with the new PKG.\n6.\nRELATED WORK\nWireless ad hoc networks have attracted intensive research attention in recent\nyears [1, 2, 3, 4, 5, 7]. Their intrinsic vulnerabilities due to the lack of infrastructure,\nunsecured media, untrusted peers, reliance on relaying, and high system dynamics (e.g.,\npeer membership, working mode and network topology) have geared a considerable\namount of research effort toward securing peer communications in these systems [5, 7].\nIn this section, we briefly review two research topics closely related to our work, and\ncompare reported work with our approach.\nInformation exchange — Schemes proposed to secure information exchange in\nwireless ad hoc networks are based on either SKC or PKC. With SKC, pairwise shared-\nkeys, derived from preshared secret or bootstrapped by other means, should be estab-\nlished for all peer pairs beforehand, which is very impractical to achieve for mobile\n"
        },
        {
            "page_number": 106,
            "text": "WIRELESS NETWORK SECURITY\n99\npeers. Also, the total number of shared-keys is in the order of N 2, where N is the\nnumber of all potential peers and can be very large even in small systems with high\nmembership dynamic. SKC procedures are efficient in general to achieve security\nproperties, but have higher overhead with regard to key management.\nNormally, RSA-based PKC procedures are less efficient than those in SKC in\nachieving the same level of security, but the key management in PKC is has less over-\nhead than SKC, if a PKI or a CA hierarchy has already been built and well-maintained.\nA distributed certification service is proposed in [24], in which the system private-key\nused to sign peer public-key certificates is distributed to multiple servers with threshold\ncryptography. It strengthens the security and reliability of public-key certification, but\ndoes not reduce the associated overhead. A self-organizing PGP-like key management\nis proposed in [34], in which peers probabilistically obtain a certificate chain to other\npeers by merging their local certificate repositories; however, a roaming peer has dif-\nficulty in building its local repository shortly after it joins a foreign system. Random\nkey predistribution has also been attempted in wireless sensor networks [35].\nCryptographically-generated identity [36] is an approach closest to ours. With this\napproach, peers derive their statistically-unique identity from their public-key (e.g., by\nhashing), so that the binding between the identity and the public-key of an entity is\nself-verifiable, which also eliminates the need for public-key certification. However,\nsuch identities cannot have any easy-to-understand semantics for its owner and other\npeers, and additional infrastructures (similar to DNS mapping host name to IP address)\nmay be required to enable distributed applications.\nIn IBC-based schemes, peers only propose their identity, which is also their public-\nkey, and can potentially have very rich semantics. Therefore, the binding of identity\nand public-key is intrinsic, and the name-to-identity mapping is unnecessary. This fact\nreduces the communication and computing overhead for resource-constrained peers in\nwireless ad hoc networks. For example, sender-only peers have no keying requirement,\nand peers can request keying even after regular, encrypted information is received.\nAlso, these IBC-based schemes are based on ECC primitives, which are considered\nmore efficient than RSA-based primitives [10, 11, 12, 13]. As we mentioned, BLS-IBS\nsignatures achieve the similar level of security to DSA signatures with a size half of the\nlatter. Further, IBC-based schemes can authentically bootstrap symmetric procedures\neven without having any physical communications beforehand. All these features are\nvery attractive to resource-constrained peers in wireless ad hoc networks.\nMessage routing — Many wireless ad hoc routing schemes, no matter reactive or\nproactive ones, are found vulnerable to corrupted or false routing information. Several\nsecurity patches have been proposed, which are based on either SKC or regular PKC\nsystems. Broadcast operations often occur in route discovery, while traditional security\nassociations are often based on a point-to-point model. Ariadne is a DSR-like routing\nscheme, in which message authenticity can be protected by digital signature, preshared\nsecret, or a timed-release hash-chain to allow a group of recipients to verify messages\nwith the same symmetric key (i.e., Tesla keys), without allowing them to forge extra\nmessages [37]. In Ariadne, all peers require loose time synchronization to release\nkey gradually. SRP is another DSR-like routing scheme, where intermediate peers\n"
        },
        {
            "page_number": 107,
            "text": "100\nJIANPING PAN, et al.\ndo not perform cryptographic operations and have no a priori associations with end-\npeers [38]; but source and destination peers should have security associations. SAR [39]\nand SAODV [40] attempt to secure AODV, another on-demand ad hoc routing protocol.\nSEAD is a DSDV-like routing scheme that employs one-way hash function to protect\nroute update without any asymmetric cryptographic operations [41], but SEAD has\nto rely on other means to distribute and authenticate the final value (i.e., image) of a\nhash-chain. ARAN employs PKC to guarantee message authenticity, integrity and non-\nrepudiation, and to prevent modification, impersonation and fabrication attacks [42].\nIn contrast, IBC-based schemes can be seamlessly integrated with wireless ad\nhoc routing protocols, and achieve the same level of security more effectively than\nSKC-based schemes and more efficiently than regular PKC-based schemes. There are\nother security schemes proposed to defense against more sophisticated attacks such as\nblackhole, wormhole, rushing and replay attacks in ad hoc networks [14, 17, 18, 16,\n15], which are orthogonal to our effort. Further, the identity-based key management\nschemes proposed in this chapter can help reduce the risk of certain sophisticated attacks\nassociated with forged identities (e.g., Sybil attacks [43]), since malicious peers cannot\nalways request keying from the PKG arbitrarily and then freely spoof their identities to\ncheat other peers.\n7.\nCONCLUSION\nAchieving secure, trustworthy and dependable peer communications imposes a\nmajor challenge in the highly-anticipated deployment of large-scale, heterogeneous\nwireless ad hoc networks. In this chapter, after identifying the irreplaceable role of\npeer identity in these networks, we promoted identity-based key management schemes,\nwhich can effectively and efficiently bootstrap any chosen security procedures in wire-\nless ad hoc networks. In addition, we illustrated secure communication schemes with\na security enhancement to a reactive ad hoc routing protocol, and demonstrated that\nidentity-based schemes are intrinsically suitable for and practically capable of ensuring\nthe confidentiality, integrity and authenticity of information exchange among peers.\nIn this chapter, we assumed that autonomous peers are always collaborative in\nrelaying once they have chosen to do so. Designing accounting and rewarding schemes\nto stimulate selfish peers to become collaborative and to compensate them if they do\nso is one of our future work items.\n8.\nREFERENCES\n1. C. Perkins (ed). Ad hoc networking. Addison-Wesley, 2001.\n2. Z. Haas, J. Deng, B. Liang, P. Papadimitatos, and S. Sajama. Wireless ad hoc networks. in J. Proakis\n(ed) Encyclopedia of Telecommunications, 2002.\n3. R. Ramanathan and J. Redi. A brief overview of ad hoc networks: challenges and directions. IEEE\nCommunications, 40(5):20–22, 2002.\n"
        },
        {
            "page_number": 108,
            "text": "WIRELESS NETWORK SECURITY\n101\n4. Z. Haas, M. Gerla, D. Johnson, C. Perkins, M. Pursley, M. Steenstrup, and C.-K. Toh (eds). Special\nissue on wireless ad hoc networks. IEEE J. on Selected Areas in Communications, 17(8), 1999.\n5. L. Buttyaen and J.-P. Hubaux (eds). Report on a working session on security in wireless ad hoc networks.\nMobile Computing and Communications Review, 7(1), 2003.\n6. S. Capkun and J.-P. Hubaux. BISS: building secure routing out of an incomplete set of secure associa-\ntions. Proc. of 2nd ACM Wireless Security (WiSe’03), pp. 21–29, 2003.\n7. J.-P. Hubaux. What could we submit next year to WiSe? Research challenges in wireless security.\nInvited Presentation at 2nd ACM Wireless Security (WiSe’03), 2003.\n8. M. Gagnee. Identity-based encryption: a survey. RSA Laboratories Cryptobytes, 6(1):10–19, 2003.\n9. A. Khalili, J. Katz, and W. Arbaugh. Toward secure key distribution in truly ad-hoc networks. Proc.\nof IEEE Security and Assurance in Ad-Hoc Networks at Int’l Symp. on Applications and the Internet\n(SAINT’03), pp. 342–346, 2003.\n10. G. Appenzeller and B. Lynn. Minimal-overhead IP security using identity based encryption. Available\nat http://rooster.stanford.edu/∼ben/pubs/ipibe.pdf, 2002.\n11. T. Garefalakis and C. Mitchell. Securing personal area networks. Proc. of 13th IEEE Personal, Indoor\nand Mobile Radio Communications (PIMRC’02), pp. 1257–1259, 2002.\n12. J. Arkko, T. Aura, J. Kempf, V. Mantyla, P. Nikander, and M. Roe. Securing IPv6 neighbor and router\ndiscovery. Proc. 1st ACM Wireless Security (WiSe’01), pp. 77–86, 2002.\n13. T. Stading. Secure communication in a distributed system using identity based encryption. Proc. of 3rd\nIEEE/ACM Cluster Computing and Grid (CCGRID’03), pp. 414–420, 2003.\n14. H. Deng, W. Li, and D. Agrawal. Routing security in wireless ad hoc networks. IEEE Communications,\n40(10):70–75, 2002.\n15. B. Awerbuch, D. Holmer, C. Nita-Rotaru, and H. Rubens. An on-demand secure routing protocol\nresilient to byzantine failures. Proc. of 1st ACM Wireless Security (WiSe’02), pp. 21–30, 2002.\n16. J. Zhen and S. Srinivas. Preventing replay attacks for secure routing in ad hoc networks. Proc. of 2nd\nAd Hoc Networks & Wireless (ADHOC-NOW’03), pp. 140–150, 2003.\n17. Y.-C. Hu, A. Perrig, and D. Johnson. Packet leashes: a defense against wormhole attacks in wireless\nnetworks. Proc. of 22nd IEEE Infocom (Infocom’03), pp. 1976–1986, 2003.\n18. Y. Hu, A. Perrig, and D. Johnson. Rushing attacks and defense in wireless ad hoc network routing\nprotocols. Proc. of 2nd ACM Wireless Security (WiSe’03), pp. 30–40, 2003.\n19. A. Shamir. Identity-based cryptosystems and signature schemes. Proc. of 4th IACR Cryptology\n(Crypto’84), pp. 47–53, 1984.\n20. D. Boneh and M. Franklin. Identity-based encryption from the Weil pairing. Proc. of 21st IACR Cryp-\ntology (Crypto’01), pp. 213–229, 2001.\n21. M. Bellare and P. Rogaway. Random oracle models are practical: a paradigm for designing efficient\nprotocols. Proc. of 1st ACM Computer and Communications Security (CCS’93), pp. 62–73, 1993.\n22. B. Lynn. Authenticated identity-based encryption. Cryptology ePrint Archive, 2002/072, 2002.\n23. S. Capkun, J.-P. Hubaux, and L. Buttyan. Mobility helps security in ad hoc networks. Proc. of 4th ACM\nMobile Ad Hoc Networking and Computing (MobiHoc’03), pp. 46–56, 2003.\n24. L. Zhou and Z. Haas. Securing ad hoc networks. IEEE Network, 13(6):24–30, 1999.\n25. C. Gentry and A. Silverberg. Hierarchical ID-based cryptography. Proc. of 8th IACR AsiaCrypt (Asi-\naCrypt’02), pp. 548–566, 2002.\n26. P. Grabher and D. Page. Hardware acceleration of the Tate pairing in characteristic three. Proc. of 7th\nIACR Cryptographic Hardware and Embedded Systems (CHES’05), pp. 398–411, 2005.\n"
        },
        {
            "page_number": 109,
            "text": "102\nJIANPING PAN, et al.\n27. B. Libert and J.-J.Quisquarter. New identity based signcryption schemes based on pairings. Cryptology\nePrint Archive, 2003/023, 2003.\n28. D. Boneh, B. Lynn, and H. Shacham. Short signature from the Weil pairing. Proc. of 7th AsiaCrypt\n(AsiaCrypt’01), pp. 514–532, 2001.\n29. D. Johnson. Routing in ad hoc networks of mobile hosts. Proc. of 1st IEEE Workshop on Mobile\nComputing Systems and Applications (WMCSA’94), pp. 158–163, 1994.\n30. E. Royer and C.-K. Toh. A review of current routing protocols for ad hoc mobile wireless networks.\nIEEE Personal Communications, 4(2):46–55, 1999.\n31. M.Abolhasan, T.Wysocki, andE.Dutkiewicz.Areviewofroutingprotocolsformobileadhocnetworks.\nAd Hoc Networks, 2:1–22, 2004.\n32. X. Boyen. Multipurpose identity-based signcryption: a swiss army knife for identity-based cryptogra-\nphy. Proc. of 23rd IACR Cryptology (Crypto’03), pp. 383–399, 2003.\n33. D. Boneh and X. Boyen. Secure identity based encryption without random oracles. Proc. of 24th IACR\nCryptology (Crypto’04), 2004.\n34. J.-P. Hubaux, L. Buttyaen, and S. Capkun. The quest for security in mobile ad hoc networks. Proc. of\n2nd ACM Mobile Ad Hoc Networking and Computing (MobiHoc’01), pp. 146–155, 2001.\n35. H. Chan, A. Perrig, and D. Song. Random key predistribution schemes for sensor networks. Proc. of\n24th IEEE Security & Privacy (S&P’03), pp. 197–215, 2003.\n36. G. Montenegro and C. Castelluccia. Statistically unique and cryptographically verifiable (SUCV) iden-\ntifiers and addresses. Proc. of 9th ISOC Network and Distributed Systems Security (NDSS’02), 2002.\n37. Y.-C. Hu, A. Perrig, and D. Johnson. Ariadne: a secure on-demand routing protocol for ad hoc networks.\nProc. of 8th ACM Mobile Computing and Networking (MobiCom’02), pp. 12–23, 2002\n38. P. Papadimitratos and Z. Haas. Secure routing for mobile ad hoc networks. Proc. of 7th SCS Commu-\nnication Networks and Distributed Systems Modeling and Simulation (CNDS’02), 2002.\n39. S. Yi, P. Naldurg, and R. Kravets. Security-aware ad hoc routing for wireless networks. Proc. of 2nd\nACM Mobile Ad Hoc Networking and Computing (MobiHoc’01), pp. 299–302, 2001.\n40. M. Zapata and N. Asokan. Securing ad hoc routing protocols. Proc. of 1st ACM Wireless Security\n(WiSe’01), pp. 1–10, 2002.\n41. Y.-C. Hu, D. Johnson, and A. Perrig. SEAD: secure efficient distance vector routing in mobile wire-\nless ad hoc networks. Proc. of 4th IEEE Workshop on Mobile Computing Systems and Applications\n(WMCSA’02), pp. 3–13, 2002.\n42. K. Sanzgiri, B. Dahill, B. Levine, C. Shields, and E. Belding-Royer. A secure routing protocol for ad\nhoc networks. Proc. of 10th IEEE Int’l Conf. on Network Protocols (ICNP’02), pp. 78–89, 2002.\n43. J. Newsome, E. Shi, D. Song, and A. Perrig. The Sybil attack in sensor networks: analysis & defenses.\nProc. of 3rd IEEE/ACM Information Processing in Sensor Networks (IPSN’04), pp. 259–268, 2004.\n"
        },
        {
            "page_number": 110,
            "text": "5\nA SURVEY OF ATTACKS AND\nCOUNTERMEASURES IN\nMOBILE AD HOC NETWORKS\nBing Wu, Jianmin Chen, Jie Wu, Mihaela Cardei\nDepartment of Computer Science and Engineering\nFlorida Atlantic University\nE-mail: {bwu, jchen8}@fau.edu, {jie,\nmihaela}@cse.fau.edu\nSecurity is an essential service for wired and wireless network communications.\nThe\nsuccess of mobile ad hoc network (MANET) will depend on people’s confidence in its\nsecurity. However, the characteristics of MANET pose both challenges and opportunities\nin achieving security goals, such as confidentiality, authentication, integrity, availability,\naccess control, and non-repudiation. We provide a survey of attacks and countermeasures\nin MANET in this chapter. The countermeasures are features or functions that reduce\nor eliminate security vulnerabilities and attacks. First, we give an overview of attacks\naccording to the protocol layers, and to security attributes and mechanisms. Then we\npresent preventive approaches following the order of the layered protocol layers. We also\nput forward an overview of MANET intrusion detection systems (IDS), which are reactive\napproaches to thwart attacks and used as a second line of defense.\n1.\nINTRODUCTION\nA MANET is referred to as a network without infrastructure because the mobile\nnodes in the network dynamically set up temporary paths among themselves to transmit\npackets. In a MANET, a collection of mobile hosts with wireless network interfaces\nform a temporary network without the aid of any fixed infrastructure or centralized ad-\nministration. Nodes within each other’s wireless transmission ranges can communicate\ndirectly; however, nodes outside each other’s range have to rely on some other nodes\nto relay messages [22]. Thus, a multi-hop scenario occurs, where several intermediate\nhosts relay the packets sent by the source host before they reach the destination host.\nEvery node functions as a router. The success of communication highly depends on\n"
        },
        {
            "page_number": 111,
            "text": "104\nBING WU et al.\nother nodes’ cooperation. At a given time, the system can be viewed as a random\ngraph due to the movement of the nodes, their transmitter/receiver coverage patterns,\nthe transmission power levels, and the co-channel interference levels. The network\ntopology may change with time as the nodes move or adjust their transmission and\nreception parameters. Thus, a MANET has several salient characteristics [21]:\nDynamic topology\nResource constraints\nNo infrastructure\nLimited physical security\nIn 1996, The Internet Engineering Task Force(IETF) created a MANET working\ngroup with the goal to standardize IP routing protocol functionality suitable for wireless\nrouting applications within both static and dynamic topologies.\nPossible applications of MANET include: soldiers relaying information for situ-\national awareness on the battlefield, business associates sharing information during a\nmeeting, attendees using laptop computers to participate in an interactive conference,\nand emergency disaster relief personnel coordinating efforts after a fire, hurricane or\nearthquake. Other possible applications [22] include personal area and home network-\ning, location-based services, and sensor networks.\nSecurity is an essential service for wired and wireless network communications.\nThe success of MANET strongly depends on whether its security can be trusted. How-\never, the characteristics of MANET pose both challenges and opportunities in achieving\nthe security goals, such as confidentiality, authentication, integrity, availability, access\ncontrol, and non-repudiation.\nThere are a wide variety of attacks that target the weakness of MANET. For exam-\nple, routing messages are an essential component of mobile network communications,\nas each packet needs to be passed quickly through intermediate nodes, which the packet\nmust traverse from a source to the destination. Malicious routing attacks can target the\nrouting discovery or maintenance phase by not following the specifications of the rout-\ning protocols. There are also attacks that target some particular routing protocols, such\nas DSR, or AODV [10] [20]. More sophisticated and subtle routing attacks have been\nidentified in recent published papers, such as the blackhole (or sinkhole) [35], Byzan-\ntine [17], and wormhole [15] [32] attacks. Currently routing security is one of the\nhottest research areas in MANET.\nThe mobile hosts forming a MANET are normally mobile devices with limited\nphysical protection and resources. Security modules, such as tokens and smart cards,\ncan be used to protect against physical attacks. Cryptographic tools are widely used to\nprovide powerful security services, such as confidentiality, authentication, integrity, and\nnon-repudiation. Unfortunately, cryptography cannot guarantee availability; for exam-\nple, it cannot prevent radio jamming. Meanwhile, strong cryptography often demands\n"
        },
        {
            "page_number": 112,
            "text": "WIRELESS NETWORK SECURITY\n105\nTable 1. Security Attacks Classification\nPassive Attacks\nEavesdropping, traffic analysis, monitoring\nActive Attacks\nJamming, spoofing, modification, replaying, DoS\na heavy computation overhead and requires the auxiliary complicated key distribu-\ntion and trust management services, which mostly are restricted by the capabilities of\nphysical devices (e.g. CPU or battery).\nThe characteristics and nature of MANET require the strict cooperation of partic-\nipating mobile hosts. A number of security techniques have been invented and a list of\nsecurity protocols have been proposed to enforce cooperation and prevent misbehavior,\nsuch as 802.11 WEP [47], SEAD [11], ARAN [32], SSL [51], etc. However, none of\nthose preventive approaches is perfect or capable to defend against all attacks. A sec-\nond line of defense called intrusion detection systems (IDS) is proposed and applied in\nMANET. IDS are some of the latest security tools in the battle against attacks. Distrib-\nuted IDS were introduced in MANET to monitor either the misbehavior or selfishness\nof mobile hosts. Subsequent actions can be taken based on the information collected\nby IDS.\nThis chapter is structured as follows. In Section 3, we describe the attacks on each\nlayer of the Internet model: application, transport, network, data link, and physical\nlayer. In Section 4, we overview attack countermeasures, including intrusion detection\nand co-operation enforcement at different layers of the Internet model. In Section 5,\nwe briefly discuss open challenges and future directions.\n2.\nSECURITY ATTACKS\nAvarietyofattacksarepossibleinMANET.Someattacksapplytogeneralnetwork,\nsome apply to wireless network and some are specific to MANETs. These security\nattacks can be classified according to different criteria, such as the domain of the\nattackers, or the techniques used in attacks. These security attacks in MANET and all\nother networks can be roughly classified by the following criteria: passive or active,\ninternal or external, different protocol layer, stealthy or non-stealthy, cryptography or\nnon-cryptography related.\nPassive vs. active attacks: The attacks in MANET can roughly be classified\ninto two major categories, namely passive attacks and active attacks [9][23].\nA passive attack obtains data exchanged in the network without disrupting\nthe operation of the communications, while an active attack involves informa-\ntion interruption, modification, or fabrication, thereby disrupting the normal\nfunctionality of a MANET. Table 1 shows the general taxonomy of security\nattacks against MANET. Examples of passive attacks are eavesdropping, traffic\n"
        },
        {
            "page_number": 113,
            "text": "106\nBING WU et al.\nTable 2. Security Attacks on each layer of the Internet Model\nLayer\nAttacks\nApplication layer\nRepudiation, data corruption\nTransport layer\nSession hijacking, SYN flooding\nNetwork layer\nWormhole, blackhole, Byzantine, flooding,\nresource consumption, location disclosure attacks\nData link layer\nTraffic analysis, monitoring, disruption MAC (802.11),\nWEP weakness\nPhysical layer\nJamming, interceptions, eavesdropping\nMulti-layer attacks\nDoS, impersonation, replay, man-in-the-middle\nanalysis, and traffic monitoring. Examples of active attacks include jamming,\nimpersonating, modification, denial of service (DoS), and message replay.\nInternal vs. external attacks: The attacks can also be classified into external\nattacks and internal attacks, according the domain of the attacks. Some papers\nrefer to outsider and insider attacks [39]. External attacks are carried out by\nnodes that do not belong to the domain of the network. Internal attacks are from\ncompromised nodes, which are actually part of the network. Internal attacks\nare more severe when compared with outside attacks since the insider knows\nvaluable and secret information, and possesses privileged access rights.\nAttacks on different layers of the Internet model: The attacks can be further\nclassified according to the five layers of the Internet model. Table 2 presents\na classification of various security attacks on each layer of the Internet model.\nSome attacks can be launched at multiple layers.\nStealthy vs. non-stealthy attacks: Some security attacks use stealth [34],\nwhereby the attackers try to hide their actions from either an individual who is\nmonitoring the system or an intrusion detection system (IDS). But other attacks\nsuch as DoS cannot be made stealthy.\nCryptography vs. non-cryptography related attacks: Some attacks are non-\ncryptography related, and others are cryptographic primitive attacks. Table 3\nshows cryptographic primitive attacks and the examples.\nFortherestofthesection, wepresentasurveyofsecurityattacksinMANEToneach\nlayer of the Internet model. Physical layer attacks are discussed in Section 3.1, followed\n"
        },
        {
            "page_number": 114,
            "text": "WIRELESS NETWORK SECURITY\n107\nTable 3. Cryptographic Primitive Attacks\nCryptographic Primitive Attacks\nExamples\nPseudorandom number attack\nNonce, timestamp,\ninitialization vector (IV)\nDigital signature attack\nRSA signature, ElGamal signature,\ndigital signature standard (DSS)\nHash collision attack\nSHA-0, MD4, MD5,\nHAVAL-128, RIPEMD\nby link layer attacks in Section 3.2; and network layer attacks in Section 3.3. Transport\nlayer attacks are discussed in Section 3.4, application layer attacks are discussed in Sec-\ntion 3.5, and multi-layer attacks are discussed in Section 3.6. Cryptographic primitive\nattacks are discussed in Section 3.7.\n2.1. Physical layer attacks\nWireless communication is broadcast by nature. A common radio signal is easy\nto jam or intercept. An attacker could overhear or disrupt the service of a wireless\nnetwork physically.\nEavesdropping: Eavesdropping is the intercepting and reading of messages\nand conversations by unintended receivers. The mobile hosts in mobile ad hoc\nnetworks share a wireless medium. The majorities of wireless communications\nuse the RF spectrum and broadcast by nature. Signals broadcast over airwaves\ncan be easily intercepted with receivers tuned to the proper frequency [47]\n[48]. Thus, messages transmitted can be overheard, and fake messages can be\ninjected into network.\nInterference and Jamming: Radio signals can be jammed or interfered with,\nwhich causes the message to be corrupted or lost [47] [48]. If the attacker has\na powerful transmitter, a signal can be generated that will be strong enough to\noverwhelmthetargetedsignalsanddisruptcommunications. Themostcommon\ntypes of this form of signal jamming are random noise and pulse. Jamming\nequipment is readily available. In addition, jamming attacks can be mounted\nfrom a location remote to the target networks.\n"
        },
        {
            "page_number": 115,
            "text": "108\nBING WU et al.\n2.2. Link layer attacks\nThe MANET is an open multipoint peer-to-peer network architecture. Specifically,\none-hop connectivity among neighbors is maintained by the link layer protocols, and the\nnetwork layer protocols extend the connectivity to other nodes in the network. Attacks\nmay target the link layer by disrupting the cooperation of the layer’s protocols.\nWireless medium access control (MAC) protocols have to coordinate the trans-\nmissions of the nodes on the common transmission medium. Because a token-passing\nbus MAC protocol is not suitable for controlling a radio channel, IEEE 802.11 pro-\ntocol is specifically devoted to wireless LANs. The IEEE 802.11 MAC protocol uses\ndistributed contention resolution mechanisms for sharing the wireless channel. The\nIEEE 802.11 working group proposed two algorithms for contention resolution. One is\na fully distributed access protocol called the distributed coordination function (DCF).\nThe other is a centralized access protocol called the point coordination function (PCF).\nPCF requires a central decision maker such as a base station. DCF uses a carrier\nsense multiple access/collision avoidance protocol (CSMA/CA) for resolving channel\ncontention among multiple wireless hosts.\nThree values for interframe space (IFS) are defined to provide priority-based access\nto the radio channel [27]. SIFS is the shortest interframe space and is used for ACK,\nCTS and poll response frames. DIFS is the longest IFS and is used as the minimum\ndelay for asynchronous frames contending for access. PIFS is the middle IFS and is\nused for issuing polls by the centralized controller in the PCF scheme. In case there\nis a collision, the sender waits a random unit of time, based on the binary exponential\nbackoff algorithm, before retransmitting. In Figure 1, node Na and node Nc contend to\ncommunicate with node Nb. First node Na gets access and reserves the channel, and\nthen Nc succeeds and reserves the channel while node Na has to back off [30].\nDisruption on MAC DCF and backoff mechanism\nCurrent wireless MAC protocols assume cooperative behaviors among all nodes.\nObviously the malicious or selfish nodes are not forced to follow the normal operation\nof the protocols. In the link layer, a selfish or malicious node could interrupt either\ncontention-based or reservation-based MAC protocols.\nA malicious neighbor of either the sender or the receiver could intentionally not\nfollow the protocol specifications. For example, the attacker may corrupt the frames\neasily by introducing some bits or ignoring the ongoing transmission. It could also\njust wait SIFS or exploit its binary exponential backoff scheme to launch DoS attacks\nin IEEE 802.11 MAC. The binary exponential scheme favors the last winner amongst\nthe contending nodes. This leads to what is called the capture effect [21]. Nodes that\nare heavily loaded tend to capture the channel by continually transmitting data, thereby\ncausing lightly loaded neighbors to backoff endlessly. Malicious nodes could take\nadvantage of this capture effect vulnerability. Moreover, a backoff at the link layer can\ncause a chain reaction in any upper layer protocols that use a backoff scheme, like TCP\nwindow management.\n"
        },
        {
            "page_number": 116,
            "text": "WIRELESS NETWORK SECURITY\n109\na\nNb\nand\ncommunicate\nNb\nNc\nNa\nNb\nNc\nDIFS\nRTS\nCTS\nDATA\nSIFS\nSIFS\nSIFS\nACK\nNAV (RTS)\nDIFS\nSIFS\nSIFS\nDATA\nACK\nNAV (RTS)\nand\ncommunicate\nDIFS\nRTS\nCTS\nSIFS\nN\nFigure 1. Illustration of Channel Contention in 802.11 MAC\n"
        },
        {
            "page_number": 117,
            "text": "110\nBING WU et al.\nThe network allocation vector (NAV) field carried in RTS/CTS frames exposes\nanother vulnerability to DoS attacks in the link layer [21] [29]. Initially the NAV field\nwas proposed to mitigate the hidden terminal problem in the carrier sense mechanism.\nDuring the RTS/CTS handshake the sender first sends a small RTS frame containing the\ntime needed to complete the CTS, data, and ACK frames. Each neighbor of the sender\nand receiver will update the NAV field and defer their transmission for the duration of\nthe future transaction according to the time that they overheard. An attacker may also\noverhear the NAV information and then intentionally corrupt the link layer frame by\ninterfering with the ongoing transmission.\nWeakness of 802.11 WEP\nIEEE 802.11 WEP incorporates wired equivalent privacy (WEP) to provide WLAN\nsystems a modest level of privacy by encrypting radio signals. 802.11 WEP standards\nsupport WEP cryptographic keys of 40 bits, though some vendors have implemented\n104 bits and even 128 bits. It is well known that WEP is broken and WEP is replaced\nby AES in 802.11i. Some of the weaknesses 802.11 WEP are listed below [27] [28]\n[47],\nWEP protocol does not specify key management.\nThe initialization vector (IV) is a 24-bit field sent in clear and is part of the\nRC4 encryption key. The reuse of IV and the weakness of RC4 lead to analytic\nattacks.\nThe combined use of a non-cryptographic integrity algorithm, CRC 32, with\nthe stream cipher is a security risk.\n2.3. Network layer attacks\nNetwork layer protocols extend connectivity from neighboring 1-hops nodes to\nall other nodes in MANET. The connectivity between mobile hosts over a potentially\nmulti-hopwirelesslinkreliesheavilyoncooperativereactionsamongallnetworknodes.\nA variety of attacks targeting the network layer have been identified and heavily\nstudied in research papers. By attacking the routing protocols, attackers can absorb\nnetwork traffic, inject themselves into the path between the source and destination, and\nthus control the network traffic flow, as shown in Figure 2 (a) and (b), where a malicious\nnode M can inject itself into the routing path between sender S and receiver D.\nThe traffic packets could be forwarded to a non-optimal path, which could intro-\nduce significant delay. In addition, the packets could be forwarded to a nonexistent path\nand get lost. The attackers can create routing loops, introduce severe network conges-\ntion, and channel contention into certain areas. Multiple colluding attackers may even\nprevent a source node from finding any route to the destination, causing the network\nto partition, which triggers excessive network control traffic, and further intensifies\nnetwork congestion and performance degradation.\n"
        },
        {
            "page_number": 118,
            "text": "WIRELESS NETWORK SECURITY\n111\nS\nY\nD\nM\n(a)\nX\nS\nM\nY\nD\n(b)\nX\nFigure 2. Illustration of Routing Attack\nAttacks at the routing discovery phase\nThere are malicious routing attacks that target the routing discovery or maintenance\nphase by not following the specifications of the routing protocols. Routing message\nflooding attacks, such as hello flooding, RREQ flooding, acknowledgement flooding,\nrouting table overflow, routing cache poisoning, and routing loop are simple examples\nof routing attacks targeting the route discovery phase [6] [35]. Proactive routing algo-\nrithms, such as DSDV [22] and OLSR [45], attempt to discover routing information\nbefore it is needed, while reactive algorithms, such as DSR [22] and AODV [22], create\nroutes only when they are needed. Thus, proactive algorithms performs worse than on-\ndemand schemes because they do not accommodate the dynamic of MANETs, clearly\nproactive algorithms require many costly broadcasts. Proactive algorithms are more\nvulnerable to routing table overflow attacks. Some of these attacks are listed below.\nRouting table overflow attack: A malicious node advertises routes that go\nto non-existent nodes to the authorized nodes present in the network. It usu-\nally happens in proactive routing algorithms, which update routing information\nperiodically. The attacker tries to create enough routes to prevent new routes\nfrom being created. The proactive routing algorithms are more vulnerable to\ntable overflow attacks because proactive routing algorithms attempt to discover\nrouting information before it is actually needed. An attacker can simply send\nexcessive route advertisements to overflow the victim’s routing table.\nRouting cache poisoning attack: In route cache poisoning attacks, attackers\ntake advantage of the promiscuous mode of routing table updating, where a node\noverhearing any packet may add the routing information contained in that packet\nheader to its own route cache, even if that node is not on the path. Suppose\na malicious node M wants to poison routes to node X. M could broadcast\nspoofed packets with source route to X via M itself; thus, neighboring nodes\nthat overhear the packet may add the route to their route caches.\n"
        },
        {
            "page_number": 119,
            "text": "112\nBING WU et al.\nAttacks at the routing maintenance phase\nThere are attacks that target the route maintenance phase by broadcasting false\ncontrol messages, such as link-broken error messages, which cause the invocation of\nthe costly route maintenance or repairing operation. For example, AODV and DSR\nimplement path maintenance procedures to recover broken paths when nodes move. If\nthe destination node or an intermediate node along an active path moves, the upstream\nnodeofthebrokenlinkbroadcastsarouteerrormessagetoallactiveupstreamneighbors.\nThe node also invalidates the route for this destination in its routing table. Attackers\ncould take advantage of this mechanism to launch attacks by sending false route error\nmessages.\nAttacks at data forwarding phase\nSome attacks also target data packet forwarding functionality in the network layer.\nIn this scenario the malicious nodes participate cooperatively in the routing protocol\nrouting discovery and maintenance phases, but in the data forwarding phase [18] [33]\nthey do not forward data packets consistently according to the routing table. Malicious\nnodes simply drop data packets quietly, modify data content, replay, or flood data\npackets; they can also delay forwarding time-sensitive data packets selectively or inject\njunk packets.\nAttacks on particular routing protocols\nThere are attacks that target some particular routing protocols. In DSR, the attacker\nmay modify the source route listed in the RREQ or RREP packets. It can delete a node\nfrom the list, switch the order, or append a new node into the list. In AODV, the\nattacker may advertise a route with a smaller distance metric than the actual distance,\nor advertise a routing update with a large sequence number and invalidate all routing\nupdates from other nodes.\nOther advanced attacks\nMoresophisticatedandsubtleroutingattackshavebeenidentifiedinrecentresearch\npapers. The blackhole (or sinkhole), Byzantine, and wormhole attacks are the typical\nexamples, which are described in detail below.\nWormhole attack: An attacker records packets at one location in the network\nand tunnels them to another location. Routing can be disrupted when routing\ncontrol messages are tunneled. This tunnel between two colluding attackers\nis referred as a wormhole [8] [32]. Wormhole attacks are severe threats to\nMANET routing protocols. For example, when a wormhole attack is used\nagainst an on-demand routing protocol such as DSR or AODV, the attack could\nprevent the discovery of any routes other than through the wormhole.\n"
        },
        {
            "page_number": 120,
            "text": "WIRELESS NETWORK SECURITY\n113\nBlackhole attack: The blackhole attack has two properties. First, the node\nexploits the mobile ad hoc routing protocol, such as AODV, to advertise itself\nas having a valid route to a destination node, even though the route is spurious,\nwith the intention of intercepting packets. Second, the attacker consumes the\nintercepted packets without any forwarding. However, the attacker runs the risk\nthat neighboring nodes will monitor and expose the ongoing attacks. There is a\nmore subtle form of these attacks when an attacker selectively forwards packets.\nAn attacker suppresses or modifies packets originating from some nodes, while\nleaving the data from the other nodes unaffected, which limits the suspicion of\nits wrongdoing.\nByzantine attack: A compromised intermediate node works alone, or a set\nof compromised intermediate nodes works in collusion and carry out attacks\nsuch as creating routing loops, forwarding packets through non-optimal paths,\nor selectively dropping packets, which results in disruption or degradation of\nthe routing services [17].\nRushing attack: Two colluded attackers use the tunnel procedure to form a\nwormhole. If a fast transmission path (e.g. a dedicated channel shared by\nattackers) exists between the two ends of the wormhole, the tunneled packets\ncan propagate faster than those through a normal multi-hop route. This forms\nthe rushing attack [19].\nThe rushing attack can act as an effective denial-\nof-service attack against all currently proposed on-demand MANET routing\nprotocols, including protocols that were designed to be secure, such as ARAN\nand Ariadne [20].\nResource consumption attack: This is also known as the sleep deprivation\nattack. An attacker or a compromised node can attempt to consume battery life\nby requesting excessive route discovery, or by forwarding unnecessary packets\nto the victim node.\nLocation disclosure attack: An attacker reveals information regarding the lo-\ncation of nodes or the structure of the network. It gathers the node location\ninformation, such as a route map, and then plans further attack scenarios. Traf-\nfic analysis, one of the subtlest security attacks against MANET, is unsolved.\nAdversaries try to figure out the identities of communication parties and ana-\nlyze traffic to learn the network traffic pattern and track changes in the traffic\npattern. The leakage of such information is devastating in security-sensitive\nscenarios.\n2.4. Transport layer attacks\nThe objectives of TCP-like Transport layer protocols in MANET include setting\nup of end-to-end connection, end-to-end reliable delivery of packets, flow control,\ncongestion control, and clearing of end-to-end connection. Similar to TCP protocols\nin the Internet, the mobile node is vulnerable to the classic SYN flooding attack or\n"
        },
        {
            "page_number": 121,
            "text": "114\nBING WU et al.\nNode A\nNode B\nSYN, Sequence Number X\nSYN/ACK, Sequence Number P,\nACK, Acknowledgment\nAcknowledgment Number X+1\nNumber P+1\nFigure 3. TCP Three-way Handshake\nsession hijacking attacks. However, a MANET has a higher channel error rate when\ncompared with wired networks. Because TCP does not have any mechanism to dis-\ntinguish whether a loss was caused by congestion, random error, or malicious attacks,\nTCP multiplicatively decreases its congestion window upon experiencing losses, which\ndegrades network performance significantly [49].\nSYN flooding attack: The SYN flooding attack is a denial-of-service attack.\nThe attacker creates a large number of half-opened TCP connections with a\nvictim node, but never completes the handshake to fully open the connection.\nFor two nodes to communicate using TCP, they must first establish a TCP\nconnection using a three-way handshake. The three messages exchanged dur-\ning the handshake, illustrated in Figure 3, allow both nodes to learn that the\nother is ready to communicate and to agree on initial sequence numbers for the\nconversation.\nDuring the attack, a malicious node sends a large amount of SYN packets to a\nvictim node, spoofing the return addresses of the SYN packets. The SYN-ACK\npackets are sent out from the victim right after it receives the SYN packets\nfrom the attacker and then the victim waits for the response of ACK packet.\nWithout receiving the ACK packets, the half-open data structure remains in the\nvictim node. If the victim node stores these half-opened connections in a fixed-\nsize table while it awaits the acknowledgement of the three-way handshake, all\nof these pending connections could overflow the buffer, and the victim node\nwould not be able to accept any other legitimate attempts to open a connec-\ntion. Normally there is a time-out associated with a pending connection, so the\nhalf-open connections will eventually expire and the victim node will recover.\nHowever, malicious nodes can simply continue sending packets that request\nnew connections faster than the expiration of pending connections.\nSession hijacking: Session hijacking takes advantage of the fact that most\ncommunications are protected (by providing credentials) at session setup, but\nnot thereafter. In the TCP session hijacking attack, the attacker spoofs the\nvictim’s IP address, determines the correct sequence number that is expected\nby the target, and then performs a DoS attack on the victim. Thus the attacker\nimpersonates the victim node and continues the session with the target.\nThe TCP ACK storm problem, illustrated in Figure 4, could be created when an\nattacker launches a TCP session hijacking attack. The attacker sends injected\n"
        },
        {
            "page_number": 122,
            "text": "WIRELESS NETWORK SECURITY\n115\nNode B\n2. Acknowledges data with ACK packet\n3. Confused B sends its last ACK\nto try to resynchronize\n2 and 3 repeat over and over\n1. Inject data\ninto session\nAttacker\nNode A\nFigure 4. TCP ACK Storm\nsession data, and node A will acknowledge the receipt of the data by sending an\nACK packet to node B. This packet will not contain a sequence number that node\nB is expecting, so when node B receives this packet, it will try to resynchronize\nthe TCP session with node A by sending it an ACK packet with the sequence\nnumber that it is expecting. The cycle goes on and on, and the ACK packets\npassing back and forth create an ACK storm. Hijacking a session over UDP is\nthe same as over TCP, except that UDP attackers do not have to worry about the\noverhead of managing sequence numbers and other TCP mechanisms. Since\nUDP is connectionless, edging into a session without being detected is much\neasier than the TCP session attacks.\n2.5. Application layer attacks\nThe application layer communication is also vulnerable in terms of security com-\npared with other layers. The application layer contains user data, and it normally\nsupports many protocols such as HTTP, SMTP, TELNET, and FTP, which provide\nmany vulnerabilities and access points for attackers. The application layer attacks are\nattractive to attackers because the information they seek ultimately resides within the\napplication and it is direct for them to make an impact and reach their goals.\nMalicious code attacks: Malicious code, such as viruses, worms, spywares,\nand Trojan Horses, can attack both operating systems and user applications.\nThese malicious programs usually can spread themselves through the network\nand cause the computer system and networks to slow down or even damaged.\nIn MANET, an attacker can produce similar attacks to the mobile system of the\nad hoc network.\nRepudiation attacks: In the network layer, firewalls can be installed to keep\npackets in or keep packets out. In the transport layer, entire connections can be\nencrypted, end-to-end. But these solutions do not solve the authentication or\nnon-repudiation problems in general. Repudiation refers to a denial of partici-\npation in all or part of the communication. For example, a selfish person could\ndeny conducting an operation on a credit card purchase, or deny any on-line\nbank transaction, which is the prototypical repudiation attack on a commercial\nsystem.\n"
        },
        {
            "page_number": 123,
            "text": "116\nBING WU et al.\n2.6. Multi-layer attacks\nSome security attacks can be launched from multiple layers instead of a particular\nlayer. Examples of multi-layer attacks are denial of service (DoS), man-in-the-middle,\nand impersonation attacks.\nDenial of service: Denial of service (DoS) attacks could be launched from\nseveral layers. An attacker can employ signal jamming at the physical layer,\nwhich disrupts normal communications. At the link layer, malicious nodes\ncan occupy channels through the capture effect, which takes advantage of the\nbinary exponential scheme in MAC protocols and prevents other nodes from\nchannel access. At the network layer, the routing process can be interrupted\nthrough routing control packet modification, selective dropping, table overflow,\nor poisoning. At the transport and application layers, SYN flooding, session\nhijacking, and malicious programs can cause DoS attacks.\nImpersonation attacks: Impersonation attacks are launched by using other\nnode’s identity, such as MAC or IP address. Impersonation attacks sometimes\nare the first step for most attacks, and are used to launch further, more sophis-\nticated attacks.\nMan-in-the-middle attacks: An attacker sits between the sender and the re-\nceiver and sniffs any information being sent between two ends. In some cases\nthe attacker may impersonate the sender to communicate with the receiver, or\nimpersonate the receiver to reply to the sender.\n2.7. Cryptographic primitive attacks\nCryptography is an important and powerful security tool. It provides security\nservices, such as authentication, confidentiality, integrity, and non-repudiation. In all\nlikelihood, there exist attacks on many cryptographic primitives that have not yet been\ndiscovered. There could be new attacks designed and developed for hash functions,\ndigital signatures, both block and stream ciphers. Most security holes are due to poor\nimplementation, i.e.\nweakness in security protocols.\nFor example, authentication\nprotocols and key exchange protocols are often the target of malicious attacks. Crypto-\ngraphic primitives are considered to be secure, however, recently some problems were\ndiscovered, such as collision attacks on hash function, e.g. SHA-1 [46]. Pseudorandom\nnumber attacks [51], digital signature attacks [14], and hash collision attacks [46] are\ndiscussed as following.\nPseudorandom number attacks: To make packets fresh, a timestamp or ran-\ndom number (nonce) is used to prevent a replay attack [51]. The session key\nis often generated from a random number. In the public key infrastructure\nthe shared secret key can be generated from a random number too. The con-\nventional random number generators in most programming languages are de-\nsigned for statistical randomness, not to resist prediction by cryptanalysts. In\n"
        },
        {
            "page_number": 124,
            "text": "WIRELESS NETWORK SECURITY\n117\nthe optimal case, random numbers are generated based on physical sources of\nrandomness that cannot be predicted. The noise from an electronic device or\nthe position of a pointer device is a source of such randomness. However, true\nrandom numbers are difficult to generate. When true physical randomness is\nnot available, pseudorandom numbers must be used. Cryptographic pseudoran-\ndom generators typically have a large pool (seed value) containing randomness.\nNew environmental noise should be mixed into the pool to prevent others from\ndetermining previous or future values. The design and implementation of cryp-\ntographic pseudorandom generators could easily become the weakest point of\nthe system.\nDigital signature attacks: The RSA public key algorithm can be used to\ngenerate a digital signature. The signature scheme has one problem: it could\nsuffer the blind signature attack. The user can get the signature of a message\nand use the signature and the message to fake another message’s signature.\nThe ElGamal signature is based on the difficulty in breaking the discrete log\nproblem.\nDigital Signature Algorithm (DSA) is an updated version of the\nElGamal digital signature scheme published in 1994 by FIPS, and was chosen\nas the digital signature standard (DSS) [14]. The attack models for digital\nsignature can be classified into known-message, chosen-message, and key-only\nattacks. In the known-message attack, the attacker knows a list of messages\npreviously signed by the victim. In the chosen-message attack, the attacker can\nchoose a specific message that it wants the victim to sign. But in the key-only\nattack, the adversary only knows the verification algorithm, which is public.\nVery often the digital signature algorithm is used in combination with a hash\nfunction. The hash function needs to be collision resistant.\nHash collision attacks: The goal of a collision attack is to find two messages\nwith the same hash, but the attacker cannot pick what the hash will be. Collision\nattacks were announced in SHA-0, MD4, MD5, HAVAL-128, and RIPEMD.\nThe collisions against MD4, MD5, HAVAL-128, and RIPEMD were found\nrecently. A successful attack against SHA-1 [46] was found, and the collisions\nin SHA-1 can be found with an estimated effort of 269 hash computations .\nNormally all major digital signature techniques (including DSA and RSA) in-\nvolve first hashing the data and then signing the hash value.\nThe original\nmessage data is not signed directly by the digital signature algorithm for both\nperformance and security reasons. Collision attacks could be used to tam-\nper with existing certificates. An adversary might be able to construct a valid\ncertificate corresponding to the hash collision.\nKey management vulnerability: Key management protocols deal with the key\ngeneration, storage, distribution, updating, revocation, and certificate service.\nAttackers can launch attacks to disclose the cryptographic key at the local host\nor during the key distribution procedure. The lack of a central trusted entity\nin MANET makes it more vulnerable to key management attacks [5] [7] [9]\n"
        },
        {
            "page_number": 125,
            "text": "118\nBING WU et al.\n[24]. For example, the man-in-the-middle attack is a design pitfall of the Diffie-\nHellman (DH) key exchange protocol. For key management protocols that rely\non a trusted key distribution center or certificate authority, the trusted central\nentity becomes the focus of attacks.\n3.\nSECURITY ATTACK COUNTERMEASURES\nSecurity is essential for the widespread of MANET. However, the characteristics\nof MANET pose both challenges and opportunities in achieving the security goals,\nsuch as confidentiality, authentication, integrity, availability, access control, and non-\nrepudiation.\nThe attacks countermeasures presentation is as follows. An overview of security\nattributes and security mechanisms is presented in Sections 3.1 and 3.2, respectively.\nWe describe the attack countermeasures by different network layers. Physical layer\ndefense is discussed in Section 3.3, link layer defense is discussed in Section 3.4,\nand network layer defense is discussed in Section 3.5. Transport layer defense and\napplication layer defense are discussed in Section 3.6 and Section 3.7 respectively.\nMulti-layer defense is in Section 3.8. Defense against key management attacks is in\nSection 3.9, and MANET intrusion detection systems are discussed in 3.10.\n3.1. Security attributes\nSecurity is the combination of processes, procedures, and systems used to en-\nsure confidentiality, authentication, integrity, availability, access control, and non-\nrepudiation.\nConfidentiality: The goal of confidentiality is to keep the information sent\nunreadable to unauthorized users or nodes. MANET uses an open medium,\nso usually all nodes within the direct transmission range can obtain the data.\nOne way to keep information confidential is to encrypt the data, and another\ntechnique is to use directional antennas.\nAuthentication: The goal of authentication is to be able to identify a node\nor a user, and to be able to prevent impersonation. In wired networks and\ninfrastructure-based wireless networks, it is possible to implement a central\nauthority at a point such as a router, base station, or access point. But there is\nno central authority in MANET, and it is much more difficult to authenticate an\nentity. Authentication can be achieved by using message authentication code\n(MAC) [62].\nIntegrity: The goal of integrity is to be able to keep the message sent from\nbeing illegally altered or destroyed in the transmission. When the data is sent\nthrough the wireless medium, the data can be modified or deleted by malicious\nattackers. The malicious attackers can also resend it, which is called a replay\nattack. The integrity can be achieved by hash functions.\n"
        },
        {
            "page_number": 126,
            "text": "WIRELESS NETWORK SECURITY\n119\nNon-repudiation: The goal of non-repudiation is related to a fact that if an\nentity sends a message, the entity cannot deny that the message was sent by\nit. By producing a signature for the message, the entity cannot later deny the\nmessage. In public key cryptography, a node A signs the message using its\nprivate key. All other nodes can verify the signed message by using A’s public\nkey, and A cannot deny that its signature is attached to the message.\nAvailability: The goal of availability is to keep the network service or resources\navailable to legitimate users. It ensures the survivability of the network despite\nmalicious incidents.\nAccess control: The goal of access control is to prevent unauthorized use\nof network services and system resources. Obviously, access control is tied\nto authentication attributes. In general, access control is the most commonly\nthought of service in both network communications and individual computer\nsystems.\n3.2. Security mechanisms\nA variety of security mechanisms have been invented to counter malicious attacks.\nThe conventional approaches such as authentication, access control, encryption, and\ndigital signature provide a first line of defense. As a second line of defense, intrusion\ndetection systems and cooperation enforcement mechanisms implemented in MANET\ncan also help to defend against attacks or enforce cooperation, reducing selfish node\nbehavior.\nPreventivemechanism: Theconventionalauthenticationandencryptionschemes\nare based on cryptography, which includes asymmetric and symmetric cryp-\ntography. Cryptographic primitives such as hash values (message digests) are\nsufficient in providing data integrity in transmission as well. Threshold cryp-\ntography can be used to hide data by dividing it into a number of shares. Digital\nsignatures can also be used to achieve data integrity and authentication services.\nIt is also necessary to consider the physical safety of mobile devices, since\nthe hosts are normally small devices, which are physically vulnerable. For\nexample, a device could easily be stolen, lost, or damaged. In the battlefield\nthey are at risk of being hijacked. The protection of the sensitive data on a\nphysical device can be enforced by some security modules, such as tokens or a\nsmart card that is accessible through PIN, passphrases, or biometrics.\nReactive mechanism: A number of malicious attacks could bypass the preven-\ntive mechanisms due to its design, implementation, or restrictions. An intrusion\ndetection system provides a second line of defense. There are widely used to\ndetect misuse and anomalies. A misuse detection system attempts to define\nimproper behavior based on the patterns of well-known attacks, but it lacks\nthe ability to detect any attacks that were not considered during the creation of\n"
        },
        {
            "page_number": 127,
            "text": "120\nBING WU et al.\nf1\nf2\nf3\nf4\nf5\nf6\nf7\nf8\nFrequency\nTime\nFigure 5. Illustration of Frequency Hopping Spread Spectrum\nthe patterns; Anomaly detection attempts to define normal or expected behavior\nstatistically. It collects data from legitimate user behavior over a period of time,\nand then statistical tests are applied to determine anomalous behavior with a\nhigh level of confidence. In practice, both approaches can be combined to be\nmore effective against attacks. Some intrusion detection systems for MANET\nhave been proposed in recent research papers.\n3.3. Physical layer defense\nSpread spectrum technology, such as frequency hopping (FHSS) [27] or direct\nsequence (DSSS) [27], can make it difficult to detect or jam signals. It changes fre-\nquency in a random fashion to make signal capture difficult or spreads the energy to a\nwider spectrum so the transmission power is hidden behind the noise level. Directional\nantennas can also be deployed due to the fact that the communication techniques can\nbe designed to spread the signal energy in space.\nFHSS: The signal is modulated with a seemingly random series of radio fre-\nquencies, which hops from frequency to frequency at fixed intervals.\nThe\nreceiver uses the same spreading code, which is synchronized with the trans-\nmitter, to recombine the spread signals into their original form. Figure 5 shows\nan example of a frequency-hopping signal.\nWith the transmitter and the receiver synchronized properly, data is transmitted\nover a single channel. However, the signal appears to be unintelligible duration\nimpulse noise for the eavesdroppers. Meanwhile, interference is minimized as\nthe signal is spread across multiple frequencies.\nDSSS: Each data bit in the original signal is represented by multiple bits in\nthe transmitted signal, using a spreading code. The spreading code spreads the\n"
        },
        {
            "page_number": 128,
            "text": "WIRELESS NETWORK SECURITY\n121\n01001011\nSpreading code\nData input\nTransmitted signal after spreading\n01100110011010111010001110110110\n01101001011010110101001101001001\nFigure 6. Illustration of Direct Sequence Spread Spectrum\nsignal across a wider frequency band in direct proportion to the number of bits\nused. The receiver can use the spreading code with the signal to recover the\noriginal data. Figure 6 illustrates that each original bit of data is represented by\n4 bits in the transmitted signal. The first bit of data, a 0 is transmitted as 0110\nwhich is first 4 bits of spreading code. The second bit, 1, is transmitted as 0110\nwhich is bit-wise complement of the second 4 bits of spreading code. In turn,\neach input bit is combined, using exclusive-or, with four bits of the spreading\ncode.\nBoth FHSS and DSSS pose difficulties for outsiders attempting to intercept the\nradio signals. The eavesdropper must know the frequency band, spreading code, and\nmodulation techniques in order to accurately read the transmitted signals. The property\nthat spread spectrum technologies do not interoperate with each other further adds dif-\nficulties to the eavesdropper. Spread spectrum technology also minimizes the potential\nfor interference from other radios and electromagnetic devices. Despite the capability\nof spread spectrum technology, it is secure only when the hopping pattern or spreading\ncode is unknown to the eavesdroppers.\n3.4. Link layer defense\nThere are malicious attacks that target the link layer by disrupting the cooperative\nnature of link layer protocols. Link layer protocols help to discover 1-hop neighbors,\nhandle fair channel access, frame error control, and maintain neighbor connections.\nSelfish nodes could disobey the channel access rule, manipulate the NAV field, cheat\nbackoff values in order to maximize their own throughput. Neighbors should monitor\nthese misbehaviors. Although it is still an open challenge to prevent selfishness, some\nschemes have been proposed, such as ERA-802.11 [12], where detection algorithms\nare proposed. Traffic analysis is prevented by encryption at data link layer.\nWEP encryption scheme defined in the IEEE 802.11 wireless LAN standard uses\nlink encryption to hide the end-to-end traffic flow information. However, WEP has\nbeen widely criticized for its weaknesses [28] [47]. Some secure link layer protocols\nhave been proposed in recent research, such as LLSP.\n"
        },
        {
            "page_number": 129,
            "text": "122\nBING WU et al.\nIn MANET, some papers propose to create a security cloud, construct a traffic cover\nmode or dynamic mix method, or use traditional traffic padding and traffic rerouting\ntechniques to prevent traffic analysis. A security cloud means that each node under the\nsecurity cloud is identical in terms of traffic generation. A traffic cover mode hides\nthe changes of an end-to-end flow traffic pattern, because certain tactical information\nmight be inferred from the unusual changes in the traffic pattern. A dynamic mix\nmethod is used to hide the source and destination information during message delivery\nvia a cryptographic method and to “mix\" nodes in the network.\n3.5. Network layer defense\nThe passive attack on routing information can be countered with the same methods\nthat protect data traffic. Some active attacks, such as illegal modification of routing\nmessages, can be prevented by mechanisms such as source authentication and message\nintegrity. DoS attacks on a routing protocol could take many forms. DoS attacks can be\nlimited by preventing the attacker from inserting routing loops, enforcing the maximum\nroute length that a packet should travel, or using some other active approaches. The\nwormhole attack can be detected by an unalterable and independent physical metric,\nsuch as time delay or geographical location. For example, packet leashes are used to\ncombat wormhole attacks [15].\nIn general, some kind of authentication and integrity mechanism, either the hop-by-\nhop or the end-to-end approach, is used to ensure the correctness of routing information.\nFor instance, digital signature, one-way hash function, hash chain, message authenti-\ncation code (MAC), and hashed message authentication code\n(HMAC) are widely used for this purpose. IPsec and ESP are standards of security\nprotocols on the network layer used in the Internet that could also be used in MANET,\nin certain circumstances, to provide network layer data packet authentication, and a cer-\ntain level of confidentiality; in addition, some protocols are designed to defend against\nselfish nodes, which intend to save resources and avoid network cooperation. Some\nsecure routing protocols have been proposed in MANET in recent papers. We outline\nthose defense techniques at below sections.\nSection 4.5.1 describes the proposed defense against wormhole attacks. Section\n4.5.2 outlines the defense against blackhole attacks. Section 4.5.3 presents the defense\nagainst impersonation and repudiation attacks. Section 4.5.4 talks about the defense\nagainst modification attacks.\nDefense against wormhole attacks\nA packet leash protocol [15] is designed as a countermeasure to the wormhole\nattack. The SECTOR mechanism [52] is proposed to detect wormholes without the\nneed of clock synchronization. Directional antennas [42] are also proposed to prevent\nwormhole attacks.\nIn the wormhole attack, an attacker receives packets at one point in the network,\ntunnels them to another point in the network, and then replays them into the network\n"
        },
        {
            "page_number": 130,
            "text": "WIRELESS NETWORK SECURITY\n123\nfrom that point. To defend against wormhole attacks, some efforts have been put into\nhardware design and signal processing techniques. If data bits are transferred in some\nspecial modulating method known only to the neighbor nodes, they are resistant to\nclosed wormholes. Another potential solution is to integrate the prevention methods\ninto intrusion detection systems. However, it is difficult to isolate the attacker with a\nsoftware-only approach, since the packets sent by the wormhole are identical to the\npackets sent by legitimate nodes.\nPacket leashes [15]: The Packet leashes are proposed to detect wormhole at-\ntacks. A leash is the information added into a packet to restrict its transmission\ndistance. A temporal packet leash sets a bound on the lifetime of a packet,\nwhich adds a constraint to its travel distance. A sender includes the trans-\nmission time and location in the message. The receiver checks whether the\npacket has traveled the distance between the sender and itself within the time\nframe between its reception and transmission. Temporal packet leashes require\ntightly synchronized clocks and precise location knowledge. In geographical\nleashes, location information and loosely synchronized clocks together verify\nthe neighbor relation.\nSECTOR [52]: The SECTOR mechanism is based primarily on distance-\nbounding techniques, one-way hash chains, and the Merkle hash tree. SECTOR\ncan be used to prevent wormhole attacks in MANET without requiring any\nclock synchronization or location information. SECTOR can also be used to\nhelp secure routing protocols in MANET using last encounters, and to help\ndetect cheating by means of topology tracking.\nDirectional antennas [42]: Directional antennas are also proposed as a coun-\ntermeasure against wormhole attacks. This approach does not require either\nlocation information or clock synchronization, and is more efficient with en-\nergy.\nDefense against blackhole attacks\nSome secure routing protocols, such as the security-aware ad hoc routing protocol\n(SAR) [54], can be used to defend against blackhole attacks. The security-aware ad hoc\nrouting protocol is based on on-demand protocols, such as AODV or DSR. In SAR, a\nsecurity metric is added into the RREQ packet, and a different route discovery procedure\nis used. Intermediate nodes receive an RREQ packet with a particular security metric\nor trust level. At intermediate nodes, if the security metric or trust level is satisfied,\nthe node will process the RREQ packet, and it will propagate to its neighbors using\ncontrolled flooding. Otherwise, the RREQ is dropped. If an end-to-end path with the\nrequired security attributes can be found, the destination will generate a RREP packet\nwith the specific security metric. If the destination node fails to find a route with the\nrequired security metric or trust level, it sends a notification to the sender and allows\nthe sender to adjust the security level in order to find a route.\n"
        },
        {
            "page_number": 131,
            "text": "124\nBING WU et al.\nTo implement SAR, it is necessary to bind the identity of a user with an associated\ntrust level.\nTo prevent identity theft, stronger access control mechanisms such as\nauthentication and authorization are required. In SAR, a simple shared secret is used to\ngenerate a symmetric encryption/decryption key per trust level. Packets are encrypted\nusing the key associated with the trust level; nodes belonging to different levels cannot\nread the RREQ or RREP packets. It is assumed that an outsider cannot obtain the key.\nIn SAR, a malicious node that interrupts the flow of packets by altering the security\nmetric to a higher or lower level cannot cause serious damage because the legitimate\nintermediate or destination node is supposed to drop the packet, and the attacker is not\nable to decrypt the packet. SAR provides a suite of cryptographic techniques, such as\ndigital signature and encryption, which can be incorporated on a need-to-use basis to\nprevent modification.\nDefense against impersonation and repudiation attacks\nARAN [32] is one example to provide authentication and non-repudiation, how-\never this does not need to be part of a routing protocol. There are several other solutions,\neach with its own weaknesses. Here routing protocol ARAN is used as a case study\nto defend against impersonation and repudiation attacks at network layer. ARAN pro-\nvides authentication and non-repudiation services using predetermined cryptographic\ncertificates for end-to-end authentication. In ARAN, each node requests a certificate\nfrom a trusted certificate server. Route discovery is accomplished by broadcasting a\nroute discovery message RDP from the source node. The reply message REP is\nunicast from the destination to the source. The routing messages are authenticated at\neach intermediate hop in both directions.\nRouting discovery authentication at each hop is illustrated in Figure 7. The RDP\npacket includes [RDP, IPX, CertA, NA, t]KA−, where RDP is a packet identifier,\nA is the source node, IPX is the destination node X’s IP address, NA is a nonce,\nCertA is A’s certificate, t is the current time, and KA−after the packet RDP, IPX,\nCertA, NA, t means the packet was signed with A’s private key. If the intermediate\nnode B is the first hop from node A, after validating A’s signature and checking its\ncertificate for expiration, it will decide to sign the packet by adding its own signature and\ncertificate, and then it will forward [[RDP, IPX, CertA, NA, t]KA−]KB−, CertB\nto all its neighbors. Each hop verifies the signature of the previous hop and replaces it\nwith its own. The destination node X unicasts a REP packet [REP, IPA, CertX,\nNA, t]KX−back to source A.\nBecause RDPs do not contain a hop count or specific recorded source route, and\nbecause messages are signed at each hop, malicious nodes have no chance to form a\nrouting loop by redirecting traffic or using impersonation to instantiate routes. The\ndisadvantage of ARAN is that it uses hop-by-hop authentication, which incurs a large\ncomputation overhead. Meanwhile, each node needs to maintain one table entry per\nsource-destination pair that is currently active.\n"
        },
        {
            "page_number": 132,
            "text": "WIRELESS NETWORK SECURITY\n125\nB\nC\nX\nA\nA\nA, N , t]K\nA\nA−\nA\nB\nA\nC\nx\nx\nx, cert\nbroadcast: [[RDP, IP\nbroadcast: [[RDP, IP\nbroadcast: [RDP, IP\nA, N , t]K\nA\nA−\n, cert\nA, N , t]KA−\n, cert\n] K , certB\n] KC−, certC\nB−\nFigure 7. Illustration of ARAN Routing Discovery Authentication at Each Hop\nDefense against modification attacks\nThe security protocol SEAD [11] is used here as an example of a defense against\nmodification attacks at network layer. Similar to a packet leash [15], the SEAD protocol\nutilizes a one-way hash chain to prevent malicious nodes from increasing the sequence\nnumber or decreasing the hop count in routing advertisement packets. In SEAD, nodes\nneed to authenticate neighbors by using TESLA [12] broadcast authentication or a\nsymmetric cryptographic mechanism. Specifically, in SEAD, a node generates a hash\nchain and organizes the chain into segments of m elements as (h0, h1, ..., hm−1), ...,\n(hkm, hkm+1, ..., hkm+m−1), ..., hn, where k = n\nm - i, m is the maximum network\ndiameter, and i is the sequence number.\nIllustrated in table 4, the network diameter is 5, the length of hash chain n’s value is\n20, i is the sequence number, and j is the metric, which is number of hops to destination.\nBecause hi=H(hi−1), given hi it is easy to verify the authenticity of hj, as long as j<i.\nGiven hi, hj cannot be derived for j<i, but hj can be derived for j>i. Because different\nhash function is used for different i and j and used by the order showed in table 4, the\nattacker can never forge lower metric value, or greater sequence value. Because, after\nreceiving a routing update in routing protocol DSDV, a node updates its advertised\nrouting table when the sequence number is greater or when the sequence number is the\nsame but the metric is lower, SEAD prevents malicious nodes from decreasing the hop\ncount value or increasing the sequence number based on the design of DSDV.\n3.6. Transport layer defense\nIn MANET, like TCP protocols in the Internet, nodes are vulnerable to the classic\nSYN flooding attack, or session hijacking attack.\nPoint-to-point or end-to-end encryption provides message confidentiality at or\nabove the transport layer in two end systems. TCP is a connection-oriented reliable\ntransport layer protocol. Because TCP does not perform well in MANET, TCP feedback\n(TCP-F) [49], TCP explicit failure notification (TCP-ELFN) [49], ad hoc transmission\ncontrol protocol (ATCP) [49], and ad hoc transport protocol (ATP) [49] have been\ninvented, but none of these protocols are designed with security in mind.\n"
        },
        {
            "page_number": 133,
            "text": "126\nBING WU et al.\nTable 4. SEAD Example: Hash Function used for Message Authentication, i is sequence\nnumber, j is metric, the network diameter (m) is 5, the length of hash chain (n) is 20\nj=0\n1\n2\n3\n4\ni=1\nh15\nh16\nh17\nh18\nh19\n2\nh10\nh11\nh12\nh13\nh14\n3\nh5\nh6\nh7\nh8\nh9\n4\nh0\nh1\nh2\nh3\nh4\nSecure Socket Layer (SSL) [51], Transport Layer Security (TLS) [51], and Private\nCommunications Transport (PCT) [51] protocols were designed for secure commu-\nnications and are based on public key cryptography. TLS/SSL can help secure data\ntransmission. It can also help to protect against masquerade attacks, man-in-the-middle\n(or bucket brigade) attacks, rollback attacks, and replay attacks. TLS/SSL is based\non public key cryptography, which is CPU-intensive and requires comprehensive ad-\nministrative configuration. Therefore, the application of these schemes in MANET\nis restricted. TLS/SSL has to be modified in order to address the special needs of\nMANET. Some firewall at a higher level can be configured to defend against SYN\nflooding attacks.\n3.7. Application layer defense\nLike the other protocol layers, the application layer also needs to be secured.\nIn a network with a firewall installed, the firewall can provide access control, user\nauthentication, packet filtering, and a logging and accounting service. Application\nlayer firewalls can effectively prevent many attacks, and application-specific modules,\nfor example, spyware detection software, have also been developed to guard mission-\ncritical services. However, a firewall is mostly restricted to basic access control and is\nnot able to solve all security problems. For example, it is not effective against attacks\nfrom insiders. Because of MANET’s lack of infrastructure, a firewall is not particularly\nuseful.\nIn MANET, an Intrusion Detection System (IDS) can be used as a second line of\ndefense. Intrusion detection can be installed at the network layer, but in the application\nlayer it is not only feasible, but also necessary. Certain attacks, such as an attack that\ntries to gain unauthorized access to a service, may seem legitimate to the lower layers,\nsuch as the MAC protocols. Also some attacks may be more obvious in the application\nlayer. For instance, the application layer can detect a DoS attack more quickly than\nthe lower layers when a large number of incoming service connections have no actual\noperations, since low layers need more time to recognize it.\n"
        },
        {
            "page_number": 134,
            "text": "WIRELESS NETWORK SECURITY\n127\n3.8. Defense against multi-layer attacks\nThe DoS attacks, impersonation attacks, man-in-the-middle attacks, and many\nother attacks can target multiple layers. The countermeasures for these attacks need to\nbe implemented at different layers. For example, directional antennas [52] are used at\nthe media access layer to defend against wormhole attacks, and packet leashes [15] are\nused as a network layer defense against wormhole attacks. The countermeasures for\nmulti-layer attacks can also be implemented in an integrated scheme. For example, if\na node detects a local intrusion at a higher layer, lower layers are notified to do further\ninvestigation.\nAs an example, we give a detailed description about the defense against DoS\nattacks.\nDefense against DoS attacks: In MANET, two types of DoS attacks [55] are\nquite common. One is at the routing layer, and another is at the MAC layer.\nAttacks at the routing layer could consist of but is not limited to the following:\n1. The malicious node participates in a route but simply drops some of the\ndata packets.\n2. The malicious node transmits falsified route updates.\n3. The malicious node could potentially replay stale updates.\n4. The malicious node reduces the TTL (time-to-live) field in the IP header\nso that the packet never reaches the destination.\nIf end-to-end authentication is enforced, attacks by independent malicious node\nof types (2) and (3) may be thwarted. An attack of type (1) may be handled by\nassigning confidence levels to nodes and using routes that provide the highest\nlevel of confidence. An attack of type (4) may be countered by making it\nmandatory that a relay node ensures that the TTL field is set to a value greater\nthan the hop count to the intended destinations.\nIf nodes collude, the authentication mechanisms fail and it is an open problem\nto provide protection against such routing attacks.\nAt the MAC layer DoS attacks could include, among others, the following\nmisbehaviors:\n1. Keeping the channel busy in the vicinity of a node leads to a denial of\nservice attack at that node.\n2. By using a particular node to continually relay spurious data, the battery\nlife of that node may be drained.\nEnd-to-end authentication may prevent the above two cases from succeeding.\nIf the node does not have a certificate of authentication, it may be prevented\nfrom accessing the channel. Usually the nodes are outsiders. However, if nodes\ncollude, and the colluding nodes include the sending node and the destination,\nMAC layer attacks are very feasible.\n"
        },
        {
            "page_number": 135,
            "text": "128\nBING WU et al.\n3.9. Defense against key management attacks\nCryptographic algorithms are security primitives, which are widely used for the\npurposes of authentication, confidentiality, integrity, and non-repudiation. Most cryp-\ntographic systems include the underlining secure, robust, and efficient key management\nsystem. Key management is in the central part of any secure communication, and is the\nweak point of system security and protocol design. A key is a piece of input informa-\ntion for cryptographic algorithms. If the key were released, the encrypted information\nwould be disclosed. The secrecy of the symmetric key and private key must be assured\nlocally. The Key Encryption Key (KEK) approach [62] could be used at local hosts to\nbuild a line of defense.\nKey distribution and key agreement over an insecure channel are at high risk and\nsuffer from potential attacks. In the traditional digital envelop approach, a session\nkey is generated at one side and is encrypted by the public-key algorithm. Then it is\ndelivered and recovered at the other end. In the Diffie-Hellman (DH) scheme [62], the\ncommunication parties at both sides exchange some public information and generate a\nsession key on both ends. Several enhanced DH schemes have been invented to counter\nman-in-the-middle attacks. In addition, a multi-way challenge response protocol, such\nas Needham-Schroeder [62], can also be used. Kerberos [62], which is based on a\nvariant of Needham-Schroeder, is an authentication protocol used in many real systems\nincluding Microsoft Windows.\nKey integrity and ownership should be protected from advanced key attacks. Dig-\nital signature, hash function, and hash function based on message authentication code\n(HMAC) [62] are techniques used for data authentication or integrity purposes. Sim-\nilarly, public key is protected by the public-key certificate, in which a trusted entity\ncalled the certification authority (CA) in PKI [62] vouches for the binding of the public\nkey with the owner’s identity. In systems lacking a trusted third party (TTP) [62], the\npublic-key certificate is vouched for by peer nodes in a distributed manner, such as\npretty good privacy (PGP) [62]. In some distributed approaches, the system secret is\ndistributed to a subset or all of the network hosts based on threshold cryptography.\nObviously, a certificate cannot prove whether an entity is “good” or “bad”, but can\nprove ownership of a key. Mainly it is for key authentication.\nA cryptographic key could be compromised or disclosed after a certain period of\nusage. Since the key should no longer be useable after its disclosure, some mechanism\nis required to enforce this rule. In PKI, this can be done implicitly or explicitly. The\ncertificate contains the lifetime of validity-it is not useful after expiration.\nBut in\nsome cases, the private key could be disclosed during the valid period, in which case\ncertificationauthority(CA)needstorevokeacertificateexplicitlyandnotifythenetwork\nby posting it onto the certificate revocation list (CRL) to prevent its usage.\nCurrently there are three types of key management on MANET: the first one is\nvirtual CA approach [3], the second one is certificate chaining [57], and the third one\nis composite key management, which combines the first two [9].\n"
        },
        {
            "page_number": 136,
            "text": "WIRELESS NETWORK SECURITY\n129\nlocal response\nlocal\nsecure\nglobal response\ncooperative\ndetection engine\nlocal\ndetection engine\ndata collection\nother traces, ...\nsystem calls activities\nIDS agents\nneighboring\nIDS agent\ncommunication\ncommunication activities\nFigure 8. A Conceptual Model for an IDS Agent in MANET\n3.10. MANET intrusion detection systems (IDS)\nBecause MANET has features such as an open medium, dynamic changing topol-\nogy, and the lack of a centralized monitoring and management point, many of the in-\ntrusion detection techniques developed for a fixed wired network are not applicable in\nMANET. Zhang [37] gives a specific design of intrusion detection and response mech-\nanisms for MANET. Marti [36] proposes two mechanisms: watchdog and pathrater,\nwhich improve throughput in MANET in the presence of nodes that agree to forward\npackets but fail to do so. In MANET, cooperation is very important to support the basic\nfunctions of the network so the token-based mechanism, the credit-based mechanism,\nand the reputation-based mechanism were developed to enforce cooperation. Each\nmechanism is discussed in this chapter.\nMANET IDS agent conceptual architecture\nThe basic approach in MANET [36] is that each mobile node runs an IDS agent in-\ndependently. It has to observe the behavior of neighboring nodes, detect local intrusion,\ncooperate with neighboring nodes, and, if needed, make decisions and take actions. An\nIDS agent has data collection, a local detection engine, local response, a cooperative\ndetection engine, global response, and secure communication with neighboring IDS\nagents. Figure 8 is a conceptual model of an IDS agent.\nApproaches to detect routing misbehavior\nWatchdog and pathrater [36] are proposed for the DSR routing protocol. It is\nassumed that wireless links are bi-directional; wireless interfaces support promiscuous\nmode operation, which means that if a node A is within the transmission range of B,\nit can overhear communications to and from B even if those communications do not\ndirectly involve A.\nThe watchdog methods detect misbehaving nodes. A node may measure a neigh-\nboring node’s frequency of dropping or misrouting packets, or its frequency of invalid\nrouting information advertisements. The implementation of a watchdog maintains a\n"
        },
        {
            "page_number": 137,
            "text": "130\nBING WU et al.\nbuffer of recently sent packets and compares each overheard packet with the packets\nin the buffer to see if there is a match. If there is a match, the node removes the packet\nfrom the buffer; otherwise if a packet has remained in the buffer for longer than a\ncertain timeout, the watchdog increments a failure tally for the neighboring node. If\nthe tally exceeds a certain threshold bandwidth, it sends a message to the source noti-\nfying it of the misbehaving node. The weaknesses of watchdog are that it might not\ndetect a misbehaving node because of ambiguous collisions, receiver collisions, limited\ntransmission power, false behavior, collusion, and partial dropping.\nIn another scheme, pathrater is run by each node. Each node keeps track of the\ntrustworthiness rating of every known node, including calculating path metrics by av-\neraging the node ratings in the path to each known node. If there are multiple paths\nto the same destination, then according to standard DSR routing protocol the shortest\npath in the route cache is chosen, but when using pathrater the path with the highest\nmetric is chosen.\nCooperation enforcement\nGenerally, there are two kinds of misbehaving nodes: one is the selfish node, and\nthe other is the malicious node. Selfish nodes don’t cooperate for selfish reasons, such\nas saving power. Even though the selfish nodes do not intend to damage other nodes,\nthe main threat from selfish nodes is the dropping of packets, which may affect the\nperformance of the network severely. Malicious nodes have the intention to damage\nother nodes, and battery saving is not a priority. Without any incentive for cooperating,\nnetwork performance can be severely degraded. The mechanisms to enforce cooper-\nating are currently split into three research areas: token-based, micro-payment, and\nreputation-based. Yang [58] proposed a token-based scheme. Buttyan [59] proposed\nthe nuglets scheme. The nuglets scheme is micro-payment scheme. Buchegger’s CON-\nFIDANT [41], Michiard’s CORE [60], and Bansel’s OCEAN [61] are reputation-based\nschemes.\nToken-based mechanism: The token-based scheme [58] is a unified network-\nlayer security solution in MANET based on the AODV protocol. In this scheme,\neach node carries a token in order to participate network operations, and its lo-\ncal neighbors collaboratively monitor any misbehavior in routing or packet for-\nwarding services. The approach is different from a watchdog, which monitors\nneighbors alone, not collaboratively.\nNodeswithoutavalidtokenareisolatedinthenetwork, andalloftheirlegitimate\nneighbors will not interact with them in routing and forwarding services. Upon\nexpiration of the token, each node renews its token via its neighbors. The\nlifetime of a token is related to the node’s behavior. A well-behaving node with\na good record needs to renew its token less often.\nThis approach uses asymmetric cryptographic primitives such as RSA. There\nis a global secret key and public key pair. Each legitimate node carries a token\n"
        },
        {
            "page_number": 138,
            "text": "WIRELESS NETWORK SECURITY\n131\nstamped with an expiration time and marked with a signature. The design is\nbased on several assumptions to simplify the mechanism:\n1. Any two nodes within wireless transmission range may monitor each\nother.\n2. The approach is only based on network-layer security, not physical-layer\nor link-layer issues.\n3. Only the secure route for data forwarding between the source and desti-\nnation is discussed, not data packet confidentiality and integrity.\n4. Each node has a unique ID.\n5. Multiple attackers are possible, but there is a limit to attackers in any\nneighborhood.\n6. Every legitimate node has a token signed with the private key, which can\nbe verified by its neighbors.\nCredit-based mechanism: The nuglets scheme [59] is an approach analogous\nto virtual currency. A node that consumes a service must pay the nodes that pro-\nvide the service in nuglets. The combination of watchdog and pathrater cannot\nhold any misbehaving nodes accountable, and misbehaving nodes are still able\nto send and receive packets. However, in the nuglets scheme, a misbehaving\nnode will be locked out by its neighbors. That is much better in fairness.\nNuglets are designed to simulate packet forwarding. The nuglets are related to\nthe counters in the nodes. The counter is maintained by a trusted and tamper-\nresistant hardware module at each node. A packet purse holds nuglets, which\nare contained in the packet. The packet purse is protected from unauthorized\nmodification and detachment from the original packet by cryptographic mech-\nanisms. The packet forward protocol is designed on fixed per hop charges.\nReputation-based mechanism: CONFIDANT [41] presents an extension to\nthe routing protocol in order to detect and isolate misbehaving nodes. The\nprotocol is designed to be able to make cooperation fair. With CONFIDANT,\neach node has four components: a monitor, a reputation system, a trust manager,\nand a path manager.\nThe CONFIDANT approach copes with MANET security, robustness, and fair-\nness by retaliating for malicious behavior and warning affiliated nodes to avoid\nbad experiences. Nodes learn not only from their own experience, but also from\nobserving the neighborhood and from the experience of their friends.\n4.\nOPEN CHALLENGES AND FUTURE DIRECTIONS\nSecurity is such an important feature that it could determine the success and wide\ndeployment of MANET. A variety of attacks have been identified. Security coun-\ntermeasures either currently used in wired or wireless networking or newly designed\n"
        },
        {
            "page_number": 139,
            "text": "132\nBING WU et al.\nspecifically for MANET are presented in the above sections. Security must be en-\nsured for the entire system at all levels since overall security level is determined by the\nsystem’s weakest point.\nThe research on MANET is still in an early stage. Existing papers are typically\nbased on one specific attack. They could work well in the presence of designated\nattacks, but there are many unanticipated or combined attacks that remain undiscovered.\nResearch is still being performed and will result in the discovery of new threats as\nwell as the creation of new countermeasures. More research is needed on robust key\nmanagement system, trust-based protocols, integrated approaches to routing security,\nand data security at different layers. Here are some research topics and future work in\nthe area:\nKey management: Cryptography is the fundamental security technique used in\nalmostallaspectsofsecurity. Thestrengthofanycryptographicsystemdepends\non proper key management. The public-key cryptography approach relies on\nthe centralized CA entity, which is a security weak point in MANET. Some\npapers propose to distribute CA functionality to multiple or all network entities\nbased on a secret sharing scheme, while some suggest a fully distributed trust\nmodel, in the style of PGP. Symmetric cryptography has computation efficiency,\nyet it suffers from potential attacks on key agreement or key distribution. Many\ncomplicated key exchange or distribution protocols have been designed, but for\nMANET, they are restricted by a node’s available resources, dynamic network\ntopology, and limited bandwidth. Efficient key agreement and distribution in\nMANET is an ongoing research area.\nTrust-based system: Most of the current work is on preventive methods with\nintrusion detection as the second line of defense. One interesting research issue\nis to build a trust-based system so that the level of security enforcement is de-\npendant on the trust level. Building a sound trust-based system and integrating\nit into the current preventive methods can be done in future research.\nMulti-fencesolution: Sincemostattacksareunpredictable, aresiliency-oriented\nsecurity solution will be more useful, which depends on a multi-fence security\nsolution. Cryptography-based methods offer a subset of solutions. Other solu-\ntions will be in future research.\n5.\nACKNOWLEDGEMENT\nThis work was supported in part by NSF grants CCR 0329741, CNS 0422762,\nCNS 0434533, ANI 0073736, EIA 0130806, and by a federal earmark project on\nSecure Telecommunication Networks.\n"
        },
        {
            "page_number": 140,
            "text": "WIRELESS NETWORK SECURITY\n133\n6.\nREFERENCES\n1. A. Salomaa, Public-Key Cryptography, Springer-Verlag, 1996.\n2. A. Tanenbaum, Computer Networks, PH PTR, 2003.\n3. L. Zhou and Z. Haas, Securing Ad Hoc Networks, IEEE Network Magazine Vol.13 No.6 (1999) pp.\n24-30.\n4. S. Yi, P. Naldurg, and R. Kravets, Security Aware Ad hoc Routing for Wireless Networks. Report\nNo.UIUCDCS-R-2002-2290, UIUC, 2002.\n5. H. Luo and S. Lu, URSA: Ubiquitous and Robust Access Control for Mobile Ad-Hoc Networks,\nIEEE/ACM Transactions on Networking Vol.12 No.6 (2004) pp. 1049-1063.\n6. W. Lou and Y. Fang, A Survey of Wireless Security in Mobile Ad Hoc Networks: Challenges and\nAvailable Solutions. Ad Hoc Wireless Networks, edited by X. Chen, X. Huang and D. Du. Kluwer\nAcademic Publishers, pp. 319-364, 2003.\n7. S. Burnett and S. Paine, RSA Security’s Official Guide to Cryptography, RSA Press, 2001.\n8. M. Ilyas, The Handbook of Ad Hoc Wireless Networks, CRC Press, 2003.\n9. S. Yi and R. Kravets, Composite Key Management for Ad Hoc Networks. Proc. of the 1st Annual Inter-\nnational Conference on Mobile and Ubiquitous Systems: Networking and Services (MobiQuitous’04),\npp. 52-61, 2004.\n10. M. Zapata, Secure Ad Hoc On-Demand Distance Vector (SAODV). Internet draft, draft-guerrero-manet-\nsaodv-01.txt, 2002.\n11. Y. Hu, D. Johnson, and A. Perrig, SEAD: Secure Efficient Distance Vector Routing in Mobile Wireless\nAd-Hoc Networks. Proc. of the 4th IEEE Workshop on Mobile Computing Systems and Applications\n(WMCSA’02), pp. 3-13, 2002.\n12. A. Perrig, R. Canetti, J. Tygar, and D. Song, The TESLA Broadcast Authentication Protocol. Internet\nDraft, 2000.\n13. P. Papadimitratos and Z. Haas, Secure Routing for Mobile Ad Hoc Networks. Proc. of the SCS Com-\nmunication Networks and Distributed Systems Modeling and Simulation Conference (CNDS 2002),\n2002.\n14. W. Mehuron, Digital Signature Standard (DSS). U.S. Department of Commerce, National Institute of\nStandards and Technology (NIST), Information Technology Laboratory (ITL). FIPS PEB 186, 1994.\n15. Y. Hu, A. Perrig, and D. Johnson, Packet Leashes: A Defense Against Wormhole Attacks in Wireless\nAd Hoc Networks. Proc. of IEEE INFORCOM, 2002.\n16. H. Deng, W. Li, and D. P. Agrawal, Routing Security in Wireless Ad Hoc Networks. IEEE Communi-\ncations Magazine, vol. 40, no. 10, 2002.\n17. B. Awerbuch, D. Holmer, C. Nita-Rotaru, and H. Rubens, An On-demand Secure Routing Protocol\nResilient to Byzantine Failures. Proceedings of the ACM Workshop on Wireless Security, pp. 21-30,\n2002.\n18. P. Papadimitratos and Z. Haas, Secure Data Transmission in Mobile Ad Hoc Networks. Proc. of the\n2003 ACM Workshop on Wireless Security, pp. 41-50, 2003.\n19. Y. Hu, A. Perrig, and D. Johnson, Rushing Attacks and Defense in Wireless Ad Hoc Network Routing\nProtocols. Proc. of the ACM Workshop on Wireless Security (WiSe), pp. 30-40, 2003.\n20. Y. Hu, A. Perrig, and D. Johnson, Ariadne: A Secure On-Demand Routing for Ad Hoc Networks. Proc.\nof MobiCom 2002, Atlanta, 2002.\n"
        },
        {
            "page_number": 141,
            "text": "134\nBING WU et al.\n21. H. Yang, H. Luo, F. Ye, S. Lu, and L. Zhang, Security in Mobile Ad Hoc Networks: Challenges and\nSolutions. IEEE Wireless Communications, pp. 38-47, 2004.\n22. C. Perkins, Ad Hoc Networks, Addison-Wesley, 2001.\n23. R. Oppliger, Internet and Intranet Security, Artech House, 1998.\n24. B. Wu, J. Wu, E. Fernandez, S. Magliveras, and M. Ilyas, Secure and Efficient Key Management\nin Mobile Ad Hoc Networks. Proc. of 19th IEEE International Parallel & Distributed Processing\nSymposium, Denver, 2005.\n25. L. Buttyan and J. Hubaux, Report on Working Session on Security in Wireless Ad Hoc Networks.\nMobile Computing and Communications Review, vol. 6, 2002.\n26. S. Ravi, A. Raghunathan, and N. Potlapally, Secure Wireless Data: System Architecture Challenges.\nProc. of International Conference on System Synthesis, 2002.\n27. W. Stallings, Wireless Communication and Networks, Pearson Education, 2002.\n28. N. Borisov, I. Goldberg and D.Wagner, Interception Mobile Communications: The Insecurity of 802.11.\nConference of Mobile Computing and Networking, 2001.\n29. P. Kyasanur and N. Vaidya, Detection and Handling of MAC Layer Misbehavior in Wireless Networks.\nProc. of the International Conference on Dependable Systems and Networks, pp. 173-182, 2003.\n30. A. Crdenas, S. Radosavac, and J. Baras, Detection and Prevention of MAC layer Misbehavior in Ad\nHoc Networks. Proc. of the 2nd ACM Workshop on Security of Ad Hoc and Sensor Networks, pp. 17-22,\n2004.\n31. C. Murthy and B. Manoj, Ad Hoc Wireless Networks: Architectures and Protocols, Prentice Hall PTR,\n2005.\n32. K. Sanzgiri, B. Dahill, B. Levine, C. Shields, and E. Belding-Royer, A Secure Routing Protocol for Ad\nHoc Networks. Proc. of IEEE International Conference on Network Protocols (ICNP), pp. 78-87, 2002\n33. K. Ng and W. Seah, Routing Security and Data Confidentiality for Mobile Ad Hoc Networks. Proc. of\nVehicular Technology Conference(VTC), Jeju, Korea, 2003.\n34. M. Jakobsson, S. Wetzel, and B. Yener, Stealth Attacks on Ad Hoc Wireless Networks. Proc. of IEEE\nVehicular Technology Conference (VTC), 2003.\n35. Y. Hu and A. Perrig, A Survey of Secure Wireless Ad Hoc Routing. IEEE Security & Privacy, pp.\n28-39, 2004.\n36. S. Marti, T. Giuli, K. Lai, and M. Baker, Mitigating Routing Misbehavior in Mobile Ad Hoc Networks,\nProc. of the Sixth Annual International Conference on Mobile Computing and Networking (MOBI-\nCOM), Boston, 2000.\n37. Y. Zhang and W. Lee, Intrusion Detection in Wireless Ad-hoc Networks, Proc. of the Sixth Annual\nInternational Conference on Mobile Computing and Networking (MOBICOM), Boston, 2000.\n38. P. Kyasanur and N. Vaidya, Detection and Handling of MAC Layer Misbehavior in Wireless Net-\nworks, Proc. of Dependable Computing and Communications Symposium (DCC) at the International\nConference on Dependable Systems and Networks (DSN), 2003.\n39. A. Cardenas, N. Benammar, G. Papageorgiou, and J. Baras, Cross-Layered Security Analysis of Wire-\nless Ad Hoc Networks, Proc. of 24th Army Science Conference, 2004.\n40. H. Yang, X. Meng, and S. Lu, Self-Organized Network-Layer Security in Mobile Ad Hoc Networks.\nProc. of ACM MOBICOM Wireless Security Workshop (WiSe’02), Atlanta, 2002.\n41. S. Buchegger and J. Boudec, Nodes Bearing Grudges: Towards Routing Security, Fairness, and Ro-\nbustness in Mobile Ad Hoc Networks, Proc. of the 10th Euromicro Workshop on Parallel, Distributed\nand Network-based Processing, Canary Islands, Spain, 2002.\n"
        },
        {
            "page_number": 142,
            "text": "WIRELESS NETWORK SECURITY\n135\n42. L. Hu and D. Evans, Using Directional Antennas to Prevent Wormhole Attacks. Proc. of Networks and\nDistributed System Security Symposium (NDSS), 2004.\n43. P. Ning and K. Sun, How to Misuse AODV: A Case Study of Inside Attacks against Mobile Ad-Hoc\nRouting Protocols, Proceedings of the 2003 IEEE Workshop on Information Assurance, United States\nMilitary Academy, West Point, NY, 2003.\n44. V. Park and S. Corson, Temporally-Ordered Routing Algorithm (TORA) Ver. 1 Functional Specification,\nIETF draft, 2001.\n45. T. Clausen and P. Jacquet, Optimized Link State Routing Protocol (OLSR) Project, Hipercom, INRIA,\nwww.ietf.org/rfc/rfc3626.txt, RFC-3626, 2003.\n46. X. Wang, D. Feng, X. Lai, and H. Yu, Collisions for Hash Functions MD4, MD5, HAVAL-128 and\nRIPEMD, Cryptology ePrint Archive, Report 2004/199, http://eprint.iacr.org/, 2004.\n47. T. Karygiannis and L. Owens, Wireless Network Security-802.11, Bluetooth and Handheld Devices.\nNational Institute of Standards and Technology. Technology Administration, U.S Department of Com-\nmerce, Special Publication 800-848, 2002.\n48. R. Nichols and P. Lekkas, Wireless Security-Models, Threats, and Solutions, McGraw-Hill, Chapter 7,\n2002.\n49. H. Hsieh and R. Sivakumar, Transport Over Wireless Networks. Handbook of Wireless Networks and\nMobile Computing, Edited by Ivan Stojmenovic. John Wiley and Sons, Inc., 2002.\n50. N. Weaver, V. Paxson, S. Staniford, and R. Cunningham, \"A Taxonomy of Computer Worms\", First\nWorkshop on Rapid Malcode (WORM), 2003.\n51. C. Kaufman, R. Perlman, and M. Speciner, Network Security Private Communication in a Public World,\nPrentice Hall PTR, A division of Pearson Education, Inc., 2002\n52. S. Capkun, L. Buttyan, and J. Hubaux, Sector: Secure Tracking of Node Encounters in Multi-hop\nWireless Networks. Proc. of the ACM Workshop on Security of Ad Hoc and Sensor Networks, 2003.\n53. W. Wang, B. Bhargava, Y. Lu, and X. Wu, Defending Against Wormhole Attacks in Mobile Ad Hoc\nNetworks, under review at Wiley Journal Wireless Communication and Mobile Computing (WCMC).\n54. S. Yi, P. Naldurg, and R. Kravets, Security-Aware Ad-hoc Routing for Wireless Networks. Report\nNo.UIUCDCS-R-2002-2290, UIUC, 2002.\n55. V. Gupta, S. V. Krishnamurthy, and M. Faloutsos, Denial of Service Attacks at the MAC Layer in\nWireless Ad Hoc Networks. In Proc. of MILCOM, 2002.\n56. I. Aad, J. Hubaux, and E. W. Knightly, Denial of Service Resilience in Ad Hoc Networks, In Proc. of\n10th Ann. Int’l Conf. Mobile Computing and Networking (MobiCom 2004), pp. 202 - 215, ACM Press,\n2004.\n57. J. Hubaux, L. Buttyan, and S. Capkun, The Quest for Security in Mobile Ad Hoc Networks, In Proc.\nof the ACM Symposium on Mobile Ad Hoc Networking & Computing (MobiHoc 2001), Long Beach,\nCA, Oct. 2001.\n58. H. Yang, X. Meng, and S. Lu, Self-organized Network Layer Security in Mobile Ad Hoc Networks,\nACM MOBICOM Wireless Security Workshop (WiSe’02).\n59. L. Buttyan and J. Hubaux, Nuglets: A Virtual Currency to Simulate Cooperation in Self-organized\nAd Hoc Networks. Technial Report DSC/2001/001, Swiss Federal Institute of Technology - Lausanne,\n2001.\n60. P. Michiardi and R. Molva, Core: A Collaborative Reputation Mechanism to Enforce Node Cooperation\nin Mobile Ad Hoc Networks, IFIP-Communication and Multimedia Security Conference 2002.\n61. S. Bansal and M. Baker, Observation-based Cooperation Enforcement in Ad Hoc Networks,\nhttp://arxiv.rog/pdf/cs.NI/0307012, July 2003.\n62. A. Menezes, P. Oorschot, and S. Vanstone, Handbook of Applied Cryptography, CRC Press, 1996.\n"
        },
        {
            "page_number": 143,
            "text": "6\nSECURE ROUTING IN WIRELESS AD-HOC\nNETWORKS\nVenkata C. Giruka\nDepartment of Computer Science\nUniversity of Kentucky, Lexington KY, 40506 USA.\nE-mail: venkata@cs.uky.edu\nMukesh Singhal\nDepartment of Computer Science\nUniversity of Kentucky, Lexington KY 40506 USA.\nE-mail: singhal@cs.uky.edu\nRouting in wireless ad-hoc networks is one of the fundamental tasks which helps nodes send\nand receive packets. Traditionally, routing protocols for wireless ad-hoc networks assume\na non-adversarial and a cooperative network setting. In practice, there may be malicious\nnodes that may attempt to disrupt the network communication by launching attacks on the\nnetwork or the routing protocol itself. In this chapter, we present several routing protocols\nfor ad-hoc networks, the security issues related to routing, and securing routing protocols\nfor mobile wireless ad-hoc networks.\n1.\nINTRODUCTION\nWireless ad-hoc networks are rapidly deployable networks in which nodes with\nwireless radios form a network on-the-fly without the need of any fixed infrastructure.\nTwo main features of ad-hoc networks, namely, connecting without cables and user\nmobility, provide a powerful combination that enables networking in situations where\nit is not feasible to establish and maintain a network. Ad-hoc networks were of primary\ninterest in military communications and disaster relief because of their “infrastructure-\nless\" nature. However, over the past decade these networks gained popularity in the\nform of personal-area networks [6] and civilian networks [23].\nOne of the basic functions of an ad-hoc network is routing, which enables nodes\nto send and receive packets. Due to the limited transmission range (typically 250m or\n"
        },
        {
            "page_number": 144,
            "text": "138\nVENKATA C. GIRUKA and MUKESH SINGHAL\nless) of nodes, routing in ad-hoc networks generally involves multiple hops. Thus, each\nnode acts as a router as well as an end-node to relay or receive packets in the network.\nRouting in ad-hoc networks is challenging because of node mobility, lack of predefined\ninfrastructure, peer-to-peer mode of communication and limited radio range. Several\nprotocols [3, 9, 10, 11, 12, 13, 17, 20] have been proposed in the literature for routing in\nad-hoc networks, each with its own niche of applicability. At the core, all these routing\nprotocols try to find a ‘good path’ from the source to the destination, assuming that\nnodes in the network are ‘friendly’ and cooperative. If we relax the assumption of node\ncooperation for routing and take into account the presence of malicious nodes, then it\nadds a new dimension to the problem, viz., security.\nSecurity in ad-hoc networks is an essential component that safeguards the proper\nfunctioning of the network and underlying protocols.\nIn general, securing ad-hoc\nnetworks is a nontrivial task due to lack of a pre-existing infrastructure, wireless nature\nof communication links, and frequently changing network topology. Unlike wired\nnetworks where the attacker needs to gain access to the physical medium to launch any\nkind of attack, in ad-hoc networks, an intruder can easily eavesdrop on the on-going\ntraffic. Further, lack of infrastructural support impedes the use of well known (those\nused in wired-networks) security architectures/protocols to detect and thwart intruders\nin the context of multi-hop wireless networks. However, there are several protocols for\nwireless networks in the literature that enforce/implement security at different layers\nand at different levels in ad-hoc networks. In this chapter, we focus on secure-routing\nprotocols for ad-hoc networks.\nAn enthusiastic reader may ask: “Is a secure-routing protocol a new routing pro-\ntocol that is designed from the scratch with security as one of its goal or is it a secure\nextension of an existing routing protocol?\" Our answer is: It can be either. The former\napproach still makes sense, since there is no standard routing protocol for ad-hoc net-\nworks as of the time of writing this book. However, authors believe that one may end-up\nre-inventing something similar to one of the existing routing techniques plus security\nextensions, in designing a new secure-routing protocol. The later approach is motivated\nby the fact that routing in ad-hoc networks, which is challenging in itself, has some of\napproaches like AODV [20], DSR [11], and OLSR [10], which IETF MANET group\nis considering for standardization. Thus, securing such a routing protocol requires\nassessing attacks specific to that protocol and securing them accordingly.\nIn the rest of the chapter, we concentrate on secure versions existing single path\nrouting protocols. To this end, we present a classification and a brief review of few\nwell known routing protocols for ad-hoc networks in Section 2. We discuss possible\nattacks on routing protocols in Section 3. Section 4 presents secure topology-based\nrouting protocols for ad-hoc networks. Section 5 presents security issues and counter\nmeasures in position-based routing, and we summarize the chapter in Section 6.\n"
        },
        {
            "page_number": 145,
            "text": "WIRELESS NETWORK SECURITY\n139\n2.\nROUTING IN AD-HOC NETWORKS\nRouting in ad-hoc networks involves finding a path from the source to the desti-\nnation, and delivering packets to the destination nodes while nodes in the network are\nmoving freely. Due to node mobility, a path established by a source may not exist after\na short interval of time. To cope with node mobility nodes need to maintain routes in\nthe network. Depending on how nodes establish and maintain paths, routing protocols\nfor ad-hoc networks broadly fall into pro-active [17], reactive [11, 20], hybrid [9], and\nlocation-based [3, 12, 13] categories.\n2.1. Proactive Routing Protocols\nPro-active routing protocols are table-driven protocols that maintain up-to-date\nrouting table using the routing information learnt from the neighbors on a continu-\nous basis. Routing in such protocols involves selecting a path form the source to the\ndestination, where the source node and each intermediate node selects a next hop, by\nrouting table look up, and forwarding the packet to next hop until destination receives\nthe packet. A drawback of such protocols is the proactive overhead due to route main-\ntenance and frequent route updates to cope with node mobility. An example of this\nclass is the DSDV [17].\nDSDV: The Destination-Sequenced Distance-Vector Routing protocol (DSDV) is an\nenhanced version of distributed Bellman-Ford algorithm, for mobile ad-hoc networks.\nIn this protocol, each node maintains a routing table that contains an entry for every node\nin the network. Each entry in the routing table consists of the destination ID, the next\nhop ID, a hop count, and a sequence number for that destination. The sequence number\nhelps nodes maintain a fresh route to the destination(s) and avoid routing loops. To\ncope with frequently changing network topology, nodes periodically broadcast routing\ntable updates thought-out the network.\nWhen a node receives a route-update packet, it changes its routing table entries if\nthe sequence number of the destination in the update packet is higher (fresh) than the\none in its routing table. If the sequences numbers are the same, then the node selects\na route with smaller metric (hop count). To reduce the network traffic due to large\nupdate packets, DSDV employs two types of updates –full dump and incremental. A\nfull dump packet generated by a node contains all entries in its routing table. Whereas\nan incremental packet contains only the routing table entries that are changed by the\nnode since the last full dump. A node triggers an update when either the metric for a\ndestination changes or when the sequence number changes. In the later case, it is called\nDSDV-SQ.\n2.2. Reactive Routing Protocols\nReactive routing protocols are demand-driven protocols that find path on-the-fly\nas and when necessary. In such protocols, establishing a new route involves a route\ndiscovery phase consisting of route request (flooding) and a route reply (by the des-\n"
        },
        {
            "page_number": 146,
            "text": "140\nVENKATA C. GIRUKA and MUKESH SINGHAL\ntination node). Nodes maintain only the active routes until a desired period or until\ndestination becomes inaccessible along every path from the source node. A drawback\nof such protocols is the delay due to route discovery on-the-fly. We briefly discuss the\nAODV and DSR protocols next.\nAODV: In Ad-hoc On-demand Distance Vector Routing (AODV), a node discovers and\nmaintains a route to the destination as and when necessary. Nodes maintain a routing\ntable containing routes towards source(s)-destination(s) that are actively communicat-\ning with each other. Each entry in the routing table consists of the destination ID, the\nnext hop ID, a hop count, and a sequence number for that destination (the same as one in\nDSDV). The sequence number helps nodes maintain a fresh route to the destination(s)\nand avoid routing loops. Thus, each node maintains a sequence number for itself and\nthe respective source(s) and destination(s). A node increments its sequence number if\nit initiates a new route request or if it detects a link-break with one of its neighbors.\nTo establish a path to the destination, a source node broadcasts a route request\n(RREQ) packet. The RREQ packet contains the source ID, the destination ID, sequence\nnumber of the source, and the latest sequence number of the destination node that is\nknown to the source node. When a node receives a RREQ packet, it makes an entry\nfor the route request in the route-request cache, and stores the address of the node\nfrom which it received the request as the next hop towards the source in its routing\ntable. If receiving node is the destination or it has a fresh route to that destination1,\nthen it responds with a route reply (RREP). Otherwise, it rebroadcasts the RREQ to its\nneighbors. When a node receives a RREP, it stores the address of the node from which\nit received RREP as the next hop towards the destination in its routing table and unicast\nthe RREP to the next hop towards the source node.\nOnce the source receives the RREP packet, it starts transmitting data packets along\nthe path traced by the RREP packet. Due to the node mobility, path(s) established\nby a source node may break. A node detects a path break if it attempts to forward\na data packet and receives a packet-drop notification from the media access control\n(MAC) layer. When a node detects a path-break, it drops the packet for the destination\nand generates a route error (RERR) packet for the destination and sends the RERR\nto the source. Upon receiving a RERR, the source node buffers data packets for the\ndestination and tries to re-establish a path to the destination.\nDSR: Dynamic Source Routing (DSR) [11] was one of the first reactive routing pro-\ntocols for ad-hoc networks. In DSR, nodes use RREQ, RREP, and RERR packets to\nestablish and maintain paths to the destination. However, unlike AODV, RREQ packet\naccumulates a list of node IDs along the path from the source to the destination and\nthe corresponding RREP packet carries this list of IDs back to the source. Once the\nsource node receives RREP packet, it starts transmitting data packets to the destination\n1 A node determines the freshness of its route table entry (provided such an entry exists) for that destination\nby comparing the destination sequence number in the RREQ with that of its route table entry.\n"
        },
        {
            "page_number": 147,
            "text": "WIRELESS NETWORK SECURITY\n141\nby embedding the route from the source to the destination in the packet header. The\npath in the data packet header is referred to as the “source route\".\nEvery node in the network stores route to other nodes in the network by maintaining\na dynamic route cache. A node learns routes to other nodes when it initiates a RREQ\nto a particular destination or when the node lies on an active path to that destination.\nIn addition to these, a node may also learn a route by overhearing transmissions (in the\npromiscuous mode) along the routes of which it is not a part.\n2.3. Hybrid Routing Protocols\nHybrid protocols combine the advantages of various approaches of routing pro-\ntocols into a single protocol. The Zone Routing Protocol (ZRP) [9], is one such hy-\nbrid protocol that combines both the proactive and reactive routing approaches. ZRP\ntakes advantage of pro-active discovery within a node’s local neighborhood, and uses\na reactive protocol for communication between these neighborhoods. The local neigh-\nborhoods are called Zones, and each node may be within multiple overlapping zones.\nZRP is motivated by the fact that “ the most communication takes place between nodes\nclose to each other. Changes in the topology are most important in the vicinity of\na node - the addition or the removal of a node on the other side of the network has\nonly limited impact on the local neighborhoods\". The performance of ZRP depends\non choosing a radius, which decides the transition from pro-active to reactive behavior.\nWith a carefully chosen radius, ZRP can achieve better efficiency and scalability over\nboth pro-active and reactive routing protocols.\n2.4. Position-based Routing Protocols\nPosition-based routing protocols utilize position of nodes in the network and make\nthe least use of the topology information. Routing protocols using such a scheme elimi-\nnate drawbacks due to frequently changing network topology. DREAM [3], GPSR [12],\nand LAR [13] are some of the examples of position-based routing protocols.\nIn Position-based routing protocols nodes maintain local (one or two hop) topol-\nogy information with the help of a hello protocol. To route a packet to the destination,\nthe source node uses a greedy-forwarding to select a next hop towards the destina-\ntion. In greedy-forwarding, a node selects a next-hop towards the destination that is\ngeographically closest to the destination among its neighboring nodes. Since there is\nno pre-established route from a source to the destination, each packet may follow a\ndifferent path depending on the network topology.\nThere are two parts to position-based routing: (a) given the position of the source,\nthe position of the destination, and a local neighbor table of each node, delivering\npackets from the source to the destination, and (b) given that each node can determine\nits own position, using some positioning system like GPS, obtaining the position of\nany other node in the system. The former part is the position-based routing, examples\ninclude GFG [5], GPSR [12]. Position-based routing is typically greedy-forwarding\nalong with a recovery mechanism to circumvent local optima due to greedy-forwarding,\n"
        },
        {
            "page_number": 148,
            "text": "142\nVENKATA C. GIRUKA and MUKESH SINGHAL\na condition where there is no node close to an intermediate node in its neighborhood\nthan the node itself. The later part is called the location service. Some of the examples\nof location-service protocols are GLS [15], DLM [27], and RLS [22]. Interestingly,\nmost location-service protocols including GLS and DLM, rely on the underlying greedy\nforwarding algorithm (although there are few other variants of greedy forwarding [26]\nexists) to send and receive control packets like location updates and location queries.\nThe advantage of these protocols is that nodes need not establish, maintain routes,\nand these protocols are more scalable compared to reactive and pro-active routing\nprotocols.\n3.\nPOSSIBLE ATTACKS ON ROUTING PROTOCOLS\nHaving explained functioning of some routing protocols in the previous sections,\nwe now present possible attacks on routing protocols. Attacks on routing protocols can\nbe both active and passive. In passive attacks an attacker does not actively participate in\nbringing the network down. Attackers are typically involved in unauthorized listening\nto routing packets. An attacker just eavesdrops on the network traffic as to determine\nwhich nodes are trying to establish routes to which other nodes, which nodes are the\ncenter of the network and so on. A major advantage for the attacker is that passive\nattacks are usually impossible to detect and hence makes defending against such attacks\nextremely difficult. Further, routing information can reveal relationships between nodes\nor disclose their addresses. If a route to a particular node is requested more often than\nto other nodes, the attacker might expect that the node is important for the functioning\nof the network, and disabling it could bring the entire network down. Such attacks can\nbe prevented mostly by applying cryptographic techniques on messages, to protect the\nmessage contents from being exposed to the attacker.\nActive attacks involves modification, fabrication of messages, or preventing the\nnetwork from functioning properly. Further, active attacks can be due to an external\nattacker(s) and an internal attacker(s). External attackers are unauthorized nodes with-\nout a shared cryptography key in the network. Internal attackers are authorized but\ncompromised nodes and are more dangerous and hard to detect as they are in the net-\nwork and own the necessary cryptography keys. Active attacks can be classified into\npacket-dropping, modification, fabrication, and other miscellaneous attacks.\n3.1. Packet Dropping\nMalicious nodes may ensure that certain messages are not transmitted by simply\nforwarding few packets and dropping the remaining one. By dropping packets, an\nattacker succeeds in disrupting the network operation. Such misbehavior can be hard\nto detect as valid nodes may, from time to time, drop packets due to congestion/collision.\nDepending on the strategy of dropping packets, there are two types of attacks:\nBlack holes: The attacker injects falsified routing packets to attract traffic. The attacker\nintercepts or drops control as well as data packets to deny services to authentic nodes.\n"
        },
        {
            "page_number": 149,
            "text": "WIRELESS NETWORK SECURITY\n143\nThis attack can be prevented by establishing routes free of such nodes or by removing\nthem from existing routes.\nGray holes: The attacker drops data packets but not control packets. This attack\nis difficult to detect. A promiscuous mode operation within the routing protocol is\nrequired to detect such an attack.\n3.2. Modification\nMost routing protocols assume that nodes do not alter fields of the protocol mes-\nsages. The protocol messages, or control packets, carry important routing information\nthat governs the behavior of their transmission. Since the level of trust in a traditional\nad-hoc network cannot be measured or enforced, malicious nodes may participate di-\nrectly in route discovery and may intercept and disrupt communication. They can easily\ncause redirection of network traffic and denial of service attacks by simply altering fields\nin protocol messages. These attacks can be classified as follows [24]:\nRemote redirection with modified route sequence number: A malicious node\nuses the routing protocol to advertise itself as having the shortest path to destina-\ntion whose packets it wants to intercept. Typically, routing protocols maintain routes\nusing monotonically increasing sequence numbers for each destination. A malicious\nnode may divert traffic through itself by advertising a route to a node with a destination\nsequence number greater than the authentic value.\nRedirection with modified hop count: In some protocols such as AODV, the route\nlength is represented in the message by a hop count field. A malicious node can succeed\nin diverting all the traffic to a particular destination through itself by advertising a\nshortest route (with a very low hop count) to that destination.\nDenial of service with modified source routes: DSR routing protocol explicitly states\nroutes in data packets called the source route. In the absence of any integrity checks\non the source route, a malicious node can modify this source route and hence succeed\nin creating loops in the network or launching a simple denial of service attack.\n3.3. Fabrication\nFabrication of messages means generating false routing messages. Such attacks\nare difficult to detect. There are three types of such attacks.\nFalsifying route error messages: AODV and DSR have measures to handle broken\nroutes when constituent nodes move or fail. If the destination node or an intermediate\nnode along an active path moves or fails, the node, which precedes the broken link,\nbroadcasts a route error message to all active neighbors which precede the broken\nlink. The nodes then invalidate the route for this destination in their routing tables. A\nmalicious node can succeed in launching a denial of service attack against a benign\nnode by sending false route error messages against this benign node.\n"
        },
        {
            "page_number": 150,
            "text": "144\nVENKATA C. GIRUKA and MUKESH SINGHAL\nRoute cache poisoning: In DSR, a node can learn routing information by overhearing\ntransmissions on routes of which it is not a part. The node then adds this information\nto its own cache. An attacker can easily exploit this method of learning and poison\nroute caches. If a malicious node, M, wants to launch a denial of service attack on\nnode X, it can simply broadcast spoofed packets with source routes to X via itself. Any\nneighboring nodes that overhear the packet transmission may add the route to their\nroute cache.\nRouting table overflow attack: A malicious node may attempt to overwhelm the\nprotocol by initiating route discovery to non-existent nodes. The logic behind this is\nto create so many routes that no further routes could be created as the routing tables of\nnodes are already overflowing.\n3.4. Other Attacks\nImpersonation: A malicious node masquerades as another node.\nIt does this by\nmisrepresenting its identity by changing its own IP or MAC address to that of some other\nnode, thereby masquerading as that node. Using stronger authentication procedures can\nprevent this type of attack.\nSybil attack: In the Sybil attack, an adversary presents multiple identities to other\nnodes in the network. This attack disrupts routing protocols by causing nodes to appear\nto be “in more than one place at once\" [14]. This reduces the diversity of routes available\nin the network. It also diminishes the effectiveness of fault-tolerant schemes such as\ndistributed storage, disparity, multi-path routing, and topology maintenance.\nWormhole attacks: The attacker receives packets at one point in the network and\ntunnels them to another part of the network. It then replays them into the network\nfrom that point onwards. This kind of attack does not require the attacker to have any\nknowledge of cryptographic keys. Using packet leashes can prevent these attacks [25].\nLocation spoofing attacks:\nApart from the usual attacks on routing protocols,\nposition-based protocols face a new attack, viz., the position spoofing attack. In the\nposition spoofing attack, a malicious node aims to disrupt the normal functioning of\ngreedy forwarding by fabricating its position information in favor of itself. A selfish\nnode may declare a selected position (e.g., away from the destinations’ position) to stay\naway from forwarding data packets. On the other hand, a malicious node can declare a\nfalse position (e.g., closest to the destination) to attract the traffic so that it can launch\nattacks. An attack combining Sybil attack and position spoofing can construct a wall\naround a node and control all traffic from that node.\n4.\nSECURE TOPOLOGY-BASED ROUTING PROTOCOLS\nIn this section we present secure proactive and reactive routing protocols. The\nSEAD protocol that is described in the next subsection is a secure extension of DSDV\n"
        },
        {
            "page_number": 151,
            "text": "WIRELESS NETWORK SECURITY\n145\nand it is a proactive secure routing protocol. The rest of the protocols in this section\nare reactive protocols, and are secure extensions of DSR or AODV routing protocols.\n4.1. SEAD\nTheSecureEfficientAdhocDistancevectorroutingprotocol(SEAD)[7]isasecure\nrouting protocol based on the DSDV-SQ protocol described in Section 2.1. Recall that\nin DSDV-SQ, nodes send both periodic routing updates as well as triggered routing\nupdates. These updates can be either the whole routing table (full dump) or only those\ntable entries that correspond to the destinations for which route has changed since the\nlast full dump. A node sends a triggered update, if the node receives a new metric\nvalue for the destination or if it receives a new sequence number for the destination, to\ncommunicate such changes to the other nodes.\nA malicious node may send updates advertising lower hop count for certain desti-\nnations to its neighboring nodes. The neighbors would be fooled into believing that this\nmalicious node has the shortest path to those destinations, and so they would make this\nmalicious node as the next hop for routes to those destinations. Thus, this malicious\nnode would be able to launch denial of service attacks against those destinations by\nhaving all routes to go through itself. It can then selectively drop packets and wreck\nhavoc in the network. In SEAD protocol, nodes prevent such modification attacks by\nauthenticating the routing update packets.\nThe SEAD protocol assumes that the network diameter has a value of at most m−1,\nwhere m is a positive integer. Thus all metrics in any routing update are less than m.\nSEAD uses one-way hash chains, which are computationally efficient as compared\nto public key cryptography or secret key cryptography paradigms, for authenticating\nupdate packets from a given node. To generate a hash chain, a node picks a ρ bit\nlong random number x, and generates the values h0, h1, h2 · · · , hn, where h0 = x,\nhi = H(hi−1) 1 ≤i ≤n, n is divisible by m, H is a one-way hash function like SHA-\n1 [16] or MD5 [21]. Given the authentic element hn, a node can authenticate hn−3 by\ncomputing H(H(H(hn−3))) and comparing with hn. If they are equal then message\ncarrying hn−3 is authentic, else it is not authentic. SEAD assumes some mechanism\nfor a node to distribute its authentic hn element of the hash chain that can be used by\nother nodes to authenticate all other elements of the hash chain of that node.\nThere are two parts for authenticating an entry in a routing update, the sequence\nnumber and the metric value for that entry. SEAD uses a single element from the\nhash chain of the node, corresponding an entry in the route update to authenticate the\nroute update entry. Note that for a given sequence number i, the corresponding metric\nvalue can be a number j, 0 ≤j < m. For this reason, a node generates a hash\nchain h0, h1, h2 · · · , hn, such that n is divisible by m. The number n/m represents the\nmaximum value of sequence number for a node. For each sequence number there is a\ngroup of m elements in the hash chain, one for each metric value. A node X releases\nhash values in the reverse order of their generation for the purpose of authentication.\nFor the routing update entry with a sequence number i and a metric value of j, X uses\nhkm+j to authenticate itself, where k = n/m −i.\n"
        },
        {
            "page_number": 152,
            "text": "146\nVENKATA C. GIRUKA and MUKESH SINGHAL\nWhen a node sends a routing update, the node includes one hash value with every\nentry in that update. If a node lists an entry for some other destination in its update, it\nsets the destination address to that nodes’ address, the metric and the sequence number\nto the corresponding values in its routing table for that destination node and the hash\nvalue is set to the hash of the hash value of the routing update entry from which it learned\nthe route to that destination. If the node lists an entry for itself in that update, it sets the\naddress in that entry to its own node address, the metric to 0, the sequence number to its\nown next sequence number, and a hash value its own hash chain elements corresponding\nto that sequence number and metric as explained before. The role played by sequence\nnumber and metric in selecting the hash value for routing update entry prevents any\nnode from advertising a route to a destination claiming a greater sequence number than\nthat destination’s current sequence number due to one-way hash functions.\nSuppose an attacker receives a routing update having metric j for a particular\nentry. The attacker decides to decrease the metric for that entry to say j −1, then\nthe attacker will have to authenticate the entry with hash chain element hkm+j−1.\nHowever, this chain element cannot be calculated from hkm+j as the hash function\ncannot be inverted. Hence, any attempt to decrease the metric of a particular routing\ntable entry would be thwarted as the attacker cannot generate the necessary hash chain\nelement to authenticate the resulting metric.\nWhen a node receives a routing update, depending upon the sequence number and\nmetric in the received entry and the sequence number and metric of the prior authentic\nhash value for that destination, it decides how many times the hash value in the newly\nreceived update entry needs to be hashed so that it should be the same as the prior\nauthentic hash value. If the two hash values are found to be equal, the entry is authentic\nand the node processes the update, else it drops the update packet.\nSEAD, however, cannot prevent the same distance attack where a node receives\nan advertisement for a particular sequence number and metric and then it re-advertises\nthe same sequence number and metric. This is because SEAD only secures the lower\nbound on the metric ensuring the node does not reduce the metric.\n4.2. SRP\nThe Secure Routing Protocol (SRP) [19] is an extension to reactive routing pro-\ntocols like Dynamic Source Routing (DSR), which helps nodes defend against attacks\nthat disrupt the route establishment phase. SRP attempts to guarantee that the node\ninitiating the route discovery will be able to differentiate between the legitimate replies\nand the replies meant to provide false topological information and can discard such\nmalevolent replies. The protocol assumes that there is a security association (SA) be-\ntween the source node S and the destination T. By using the SA, a source/destination\npair that participated in the route establishment verify each other. The source and\ndestination share a secret key KST , which is negotiated by the SA.\nIn SRP, a source node adds an additional header, called SRP header, to the un-\nderlying routing protocol packet. The SRP header contains a query sequence number,\na random query identifier, and a Message Authentication Code (MAC), called SRP\n"
        },
        {
            "page_number": 153,
            "text": "WIRELESS NETWORK SECURITY\n147\nMAC, generated by the source using the shared key KST . The Query Sequence Num-\nber, QSEQ, is a monotonically increasing 32 bit sequence number maintained by the\nsource node S for each destination T it has a security association with. QSEQ increases\nmonotonically for every route request generated by S for T, thus allowing T to detect\noutdated/replayed requests. QSEQ is initialized at the establishment of the SA and is\ngenerally not allowed to wrap around. The Query Identifier QID is a random 32 bit\nidentifier generated by S and is used by the intermediate nodes as a means to identify\nthe request. Since QID is an output of a secure pseudo-random number generator and\nis unpredictable by an adversary, it provides protection against attackers who fabricate\nrequests only to cause subsequent requests to be dropped. SRP MAC is a 96 bit value\ncalculated using the shared key KST over IP addresses of the source S and target T and\nthe two identifiers QSEQ and QID. It not only validates the integrity of the request\nbut also authenticates the origin of the packet to the target, as the MAC could have\nbeen calculated only by the source or the destination node which have the knowledge\nof KST .\nWhen an intermediate node receives a route request, and if an SRP header is not\npresent in the route request packet, it drops the packet. Otherwise, the node extracts the\nIP address of the source and destination as well as the QID from the request and creates\nan entry for the request in the query table. If an entry already exists for that source\ndestination pair with the same QID, the request is dropped by the node. Otherwise,\nthe node appends its IP address to the request and rebroadcasts the request. Thus IP\naddresses of the intermediate nodes keep on accumulating on the route request.\nThe above situation warrants that the QID should be sufficiently random and an\nadversary with finite computation capacity should not be able to predict it. Otherwise,\nthe attacker can prevent route from being established between the given source and the\ndestination pair, as it would fabricate request packets with this QID and the intermediate\nnodes will not forward the legitimate requests, as an entry already exists in the query\ntable for that particular QID.\nWhen the destination T receives this request packet, it verifies that the packet\noriginated at the node with which it has SA. The destination compares the QSEQ with\nSMAX, the maximum query sequence number received from S. If QSEQ ≤SMAX, the\nrequest is outdated/replayed and the destination discards the packet. Else, it calculates\nthe keyed hash of the request field and matches against the SRP MAC. The equality\nvalidates the integrity of the request as well as the authenticity of the sender.\nThe destination broadcasts a route reply to its one-hop neighbors in order to thwart a\npotentially malicious neighbor from controlling multiple replies. For each valid request,\nthe destination puts the accumulated route in the form of IP addresses of intermediate\nnodes into the route reply packet. The QSEQ and QID fields from the route request are\ncopied into the corresponding fields of the reply packet. MAC is calculated to preserve\nthe integrity of the packet in transit. The QSEQ and QID fields verify the freshness of\nthe packet to the source.\nWhen the source S receives the route reply packet, it checks source and destination\naddresses, QID and QSEQ and discards the reply if it does not correspond to the\ncurrently pending query. Otherwise, it compares the reply IP source-route with the\n"
        },
        {
            "page_number": 154,
            "text": "148\nVENKATA C. GIRUKA and MUKESH SINGHAL\nreverse of the route carried in the reply payload. If the two routes match, MAC is\ncalculated using the replied route, the SRP header fields, and KST . The successful\nverification confirms that the request did indeed reach the intended destination T and\nthe reply was not corrupted on the way back from T to S. Furthermore, since the reply\npacket has been routed and successfully received over the reverse of the route it carries,\nthe routing information has not been compromised during the request propagation.\nIntermediatenodesalsomeasurethefrequencyofqueriesreceivedfromtheirneigh-\nbors. Intermediate nodes maintain a priority ranking of their neighbors - highest priority\nto nodes generating requests at the lowest rate and the lowest rating for nodes generating\nrequests with highest rate. In case two packets arrive at the same time, the neighbor\nwhose ranking is higher, is given priority in routing over the one with the lower ranking.\nThe secure routing protocol guarantees the discovery of a correct route, even in the\npresence of malicious nodes. The protocol obviates the need of a certification authority,\nthereby suiting itself to the ad-hoc paradigm. The protocol does not necessitate the\nknowledge of keys of all member nodes. The only requirement of this protocol is\nthat there should be a prior security association between the source and the destination\nnodes. This kind of a security association is realized through shared secret keys between\nany two pair of nodes. However, when malicious nodes succeed in subverting benign\nnodes, the malicious nodes could easily gain access to the shared secret keys. The\nmalicious node can then masquerade as the subverted node and initiate communication\nwith other good nodes with whom the subverted node has a security association.\n4.3. ARIADNE\nAriadne [8] is an on-demand secure routing protocol based on the DSR protocol.\nAriadne prevents attackers or compromised nodes from tampering with uncompromised\nroutes consisting of benign nodes. It is based on efficient symmetric cryptographic\nprimitives and prevents several types of denial of service attacks. Unlike SRP, Ariadne\nuses a broadcast authentication protocols TESLA [18], which enables a node to verify\nthat a broadcast packet (like RREQ) received by the node is indeed generated by the\ninitiator of the message. Such a broadcast authentication is essential in defending\nagainst impersonation and denial of service attacks. The basic idea of the Ariadne\nprotocol is to insure that the destination node can authenticate the source node, the\nsource node can authenticate every intermediate node on the path from the source to\nthe destination (received by the source in RREP), and malicious nodes cannot tamper\nwith routes in RREQ or RREP by inserting dummy IDs or removing benign node IDs.\nThe idea behind the TESLA protocol is to have a random initial key (kn) for each\nnode from which each node generates a one-way key chain by repeated computation\nof a one-way hash function (H) such that kn−1 = H(kn) and in general for any j < i,\nkj = Hi−j(ki). A node discloses each key of its one-way key chain in an order that is\nexactly the reverse of the order in which the node generates the keys. Further, a node\npublishes its key ki at a time T0+i∗t, where T0 is the time at which k0 is published, and\nt is the key publication interval. The rationale behind having a reverse key disclosure\nschedule is that using a previously known hash chain element, like kj, any other node\n"
        },
        {
            "page_number": 155,
            "text": "WIRELESS NETWORK SECURITY\n149\ncan authenticate subsequent elements, ki, i > j, from a nodes hash chain by using the\nequation kj = Hi−j(ki). However, other nodes cannot generate ki due to the one-way\nproperty of the hash function.\nFor broadcast authentication using TESLA, a node generates a broadcast packet,\nadds a Message Authentication Code (MAC) of the packet generated by the node using\nits future (next in its schedule) TESLA key and then releases the key used in MAC at\na later time. A node receiving the packet verifies the TESLA security condition that\nthe key ki used to authenticate the packet has not yet been released by the nodes2. If\nthe condition holds, then the receiving node waits for the TESLA key to be released\nby the sender and verifies the key (using the one-way hash function) and the MAC of\nthe packet. If they are authentic, then the receiver accepts the packet, else it drops the\npacket.\nThe Protocol: Ariadne protocol assumes that the source and the destination share a\nsecret key KST that allows them to authenticate each other. To establish a secure route\nto the destination, the source node floods a RREQ packet that has eight fields <ROUTE\nREQUEST, initiator, target, id, time interval, hash chain, node list, MAC list>. The\ninitiator and the target are set to the source ID and the destination ID, respectively. The\n‘id’ is an identifier that has not been recently used in route discovery. The ‘time interval’\nis set to TESLA time interval at the pessimistic arrival time of the request at the target,\nwith maximum possible clock offset/skew and maximum transmission delay. The hash\nchain field is initialized by the initiator to the MAC calculated over initiator, target, id,\ntime interval, using the key KST (MACKST (initiator, target, id, time interval)). The\nnode list and MAC list are empty initially and will be filled by the intermediate and\ntarget nodes.\nWhen an intermediate node, A, receives a RREQ, the node checks its local table\nfor the (initiator, id) entry. If it finds an entry for the same route discovery, it discards\nthe RREQ, else the node verifies the time interval of the RREQ. If the time interval is\ntoo much in the future or the key corresponding to it has been disclosed, the RREQ is\ndiscarded. Otherwise, the node appends its address to the node list in the RREQ packet,\nand replaces the hash chain field with H(A, oldhashchain). The node then appends\na MAC of the entire request to the MAC list, where the MAC is calculated using key\nki corresponding to the time interval in the RREQ. The node then rebroadcasts the\nmodified RREQ.\nWhen the destination node receives the RREQ, it determines whether the keys\ncorresponding to the time interval mentioned in the RREQ have not been disclosed yet,\nand the hash chain field is equal to\nH(In, H(In−1, H(· · · , H(I1, MACKST (initiator, target, id, time interval)) · · ·))),\nwhere Ii is the intermediate node at position i and n is the number of nodes in the node\nlist. If both the conditions hold, then the destination is assured that the RREQ is valid,\nand it constructs a RREP packet.\n2 This is because if the key is released it is also known to malicious nodes.\n"
        },
        {
            "page_number": 156,
            "text": "150\nVENKATA C. GIRUKA and MUKESH SINGHAL\nThe RREP packet consists of target, initiator, time interval, node list, MAC list\n(which correspond to fields from the corresponding RREQ), target MAC and key list.\nTarget MAC is a MAC calculated by the destination over first five fields with the key\nKST . Key list is left empty to be initialized by the intermediate nodes, along the reverse\nroute in the RREQ. The destination sends the RREP to the initiator along the source\nroute which is the reverse of the sequence of hops in the node list in the RREQ. The node\nforwarding the route RREP waits until it is able to disclose the key for the specified time\ninterval. The node then appends the key to the key list field in the RREP and forwards\nthe RREP to the next hop towards the source. The waiting delays do not add significant\ncomputation overhead but adds to storage overheads. When the initiator receives the\nRREP, it checks if the keys in the key list are valid, target MAC is valid and each MAC\nin the MAC list is valid. If all these are valid only then will it accept the RREP.\nOne-way hash chain in RREQ/RREP ensures that no hop is omitted by some\nmaliciousnode. TochangeorremoveaprevioushopformtheRREQ/RREP,theattacker\nmust be able to invert the one-way hash function, which is computationally infeasible.\nHowever, a malicious node might succeed in removing the address of any previous node\nfrom the node list, but won’t be able to remove that node’s address from the hash chain\nfield. Such a fabrication would be easily detected by the destination/source, since the\ncomputed hash chain field won’t be the same as the hash chain in the received packet\nand hence the RREQ/RREP would be discarded.\nWhen an intermediate node detects a route break, i.e., it is unable to deliver the\npacket to the next hop after a fixed finite number of retransmissions, it generates a\nroute error RERR and sends it to the source node. To deal with false RERR messages,\nthe protocol requires the source to authenticate the RERR messages using TESLA. If\nthe authentication succeeds, then the source tries to reestablish a route to the destina-\ntion, else it drops the RERR. However, the protocol does not guard against attackers\nintentionally dropping genuine RERR messages.\n4.4. ARAN\nAuthenticated Routing for Ad-hoc Networks (ARAN) [24] detects and protects\nagainst malicious actions by third parties and peers in an ad-hoc environment. ARAN\nassumes a managed-open environment, meaning that there is an opportunity for pre-\ndeployment of certain security infrastructure. Using such an infrastructure helps nodes\nexchange initialization parameters before hand through a trusted third party like a\ncertification authority. With the help of initialization parameters, like a certificate from\na trusted server, ARAN provides authentication, message integrity, and non-repudiation\nin an ad-hoc environment. Table 1 presents the notations used in the rest of this section.\nARAN protocol assumes a trusted certification server T, whose public key is known\nto all the valid nodes in the networks. The protocol consists of three stages, the prelim-\ninary certification stage, the authenticated route discovery phase, and the authenticated\nroute setup phase. In the preliminary certification stage, each node obtains a certificate,\ncertA, from the server T. The certificate of a node, certA = [IPA, KA+, t, e]KT −,\ncontains the IP address of A, the public key of A, a timestamp t of the time the certifi-\n"
        },
        {
            "page_number": 157,
            "text": "WIRELESS NETWORK SECURITY\n151\nTable 1. Notations used in ARAN\nKA+\nPublic-key of node A.\nKA−\nPrivate-key of node A.\n{d}KA+\nEncryption of data d with key KA+.\n{d}KA−\nData d digitally signed by node A\ncertA\nCertificate belonging to node A.\nNa\nNonce issued by node A.\nIPA\nIP address of node A.\nRDP\nRoute Discovery Packet identifier.\nREP\nREPly packet identifier.\nERR\nERRor packet identifier.\ncate was generated by the server, and a time e at which the certificate expires. These\nvariables are concatenated and signed by the server. Nodes maintain a fresh certificate\nissued to them by the trusted server, which helps them authenticate themselves to the\nother nodes during the exchange of control messages.\nThe authenticated route discovery phase provides end-to-end authentication, in\nwhich the source node verifies that the intended destination is reached. The source\nnode, A, initiates a route discovery for the destination X, by broadcasting a route dis-\ncovery packet (RDP). The broadcast message, [RDP, IPX, NA]KA−, certA, includes\na packet type identifier (RDP), the IP address of the destination (IPX), A’s certificate\n(certA), and a monotonically increasing nonce NA, the all signed with A’s private key.\nUpon receiving an RDP packet, an intermediate node stores (IPA, NA) of the RDP\npacket. If an intermediate node has already seen the (IPA, NA) tuple, it drops the RDP\npacket. Otherwise, it keeps track of the predecessor node from which it received the\nRDP packet, validates the signature with the given certificate, removes A’s certificate\nfrom the RDP, and rebroadcasts the RDP packet by signing it. For instance, if B is a\nneighbor of A, then B broadcasts\n[[RDP, IPX, NA]KA−]KB−, certA, certB.\nSuch message signing prevents spoofing attacks that may alter the route or form loops.\nWhen node C receives the broadcast packet, C validates signatures of A and B using\ntheir respective certificates in the RDP packet. C removes B’s signature and certificate,\nrecords B as its predecessor node, signs the contents [[RDP, IPX, NA]KA−] with its\nprivate key and appends its own certificate, and broadcasts the RDP.\nC →broadcast : [[RDP, IPX, NA]KA−]KC−, certA, certC.\n"
        },
        {
            "page_number": 158,
            "text": "152\nVENKATA C. GIRUKA and MUKESH SINGHAL\nEach intermediate node repeats the same process as node C, and eventually the desti-\nnation receives the RDP packet.\nThe destination replies to the first RDP packet it receives. Note that such an RDP\npacket may not have traversed the shortest path due to network congestion, or due\nto the presence of malicious nodes. The rationale behind choosing such a path is to\nprefer them over a congested least-hop path that reduces the end-to-end delay. As a\nresponse to the RDP, the destination generate a Reply packet (REP), and unicasts it\nalong the reverse path to the source node. If D is the next node towards the source\nfrom the destination X, then X unicasts a REP ([REP, IPA, NA]KX−, certX) packet\nto D, where REP in the packet is a packet-type identifier. Since nodes keep track of\npredecessor nodes during the RDP phase, an intermediate node forwards the REP to the\npredecessor. Each intermediate node along the reverse path signs the REP and appends\nits own certificate before forwarding the REP. For instance, if C is be the predecessor\nof D, then D unicasts\n[[REP, IPA, NA]KX−]KD−, certX, certD.\nto C. Node C validates D’s signature on the received message, removes the signature\nand certificate, then signs the contents of the message and appends its own certificate\nbefore unicasting the REP to its predecessor. This avoids impersonation and replay of\nthe message sent by X. When the source receives the REP, it verifies the destination’s\nsignature and the nonce returned by the destination. If they are valid, then it starts\ntransmitting the date along the established path.\nRoutemaintenance: Inad-hocnetworks, routesmaynotbeusedactivelybythesource\nnode for a long time or they may break due to the node mobility. In ARAN, nodes purge\nroute table entries that are not used by the source-destination for a predetermined time\nperiod (route’s lifetime). An intermediate node generates an error message (ERR) if\nthere is no active route towards the destination in its route table, or if the node finds a link\nbreak due to node mobility. If a node B finds a route break, then it generates and sends\na ERR ([ERR, IPA, IPX, NB]KB−, certB,) message to its predecessor node C. The\nERR message is forwarded along the path towards the source without modification. The\nnonce NB ensures the ERR message is fresh. Because messages are signed, malicious\nnodes cannot generate ERR messages for other nodes. Non-repudiation provided by\nthe signed ERR message allows a node to be verified as the source of each ERR message\nthat it sends. A node which transmits a large number of ERR messages, whether the\nERR messages are valid or fabricated, should be avoided.\nKey revocation:\nARAN attempts a best effort key revocation that is backed with\nlimited time certificates. In the event of a certificate revocation, the trusted certificate\nserver, T, sends a broadcast message ([revoke, CertR]KT −) to the ad-hoc group that\nannounces the revocation. Any node receiving this message re-broadcasts it and stores\nthe message until the revoked certificate expires normally. Neighbors of the node with\nthe revoked certificate need to reform routes as necessary to avoid transmission through\nsuch nodes.\n"
        },
        {
            "page_number": 159,
            "text": "WIRELESS NETWORK SECURITY\n153\nIf an untrusted node whose certificate is being revoked, is the only link between\ntwo partitions of an ad-hoc network, it may not propagate the revocation message to\nthe other part - leading to a partitioned network. Such a partition may last until the\nuntrusted node is no longer the sole connection between the two partitions. Thus,\nthis method is not fail-safe. To detect such situations and to hasten the propagation\nof revocation notices, nodes exchange a summary of its revocation notices with new\nneighbors (as and when discovered). If these summaries do not match, then nodes that\ndetect inconsistency rebroadcasts signed notices to restart propagation of the notice.\n4.5. Coping with Byzantine Failures\nThe secure routing protocols described so far assume that nodes in the network\ndo not collude to attack the network. However, in realistic networks attacks can be\ndue to an individual malicious node or due to colluding malicious nodes. Baruch\net al.[1] proposed an on-demand protocol to provide resilience to Byzantine failures\ncaused by individual or colluding nodes.\nIn this protocol [1], the emphasis is on\nsurvivability of routes under situations where an intermediate node or group of nodes\nare known to be malicious and may attempt ‘Byzantine’ attacks such as creation of\nrouting loops, misrouting of packets along non-optimal (unnecessarily long) paths or\nselective dropping of packets (black or gray holes).\nInstead of laying the blame of a route failure on a single misbehaving node, the\nprotocol [1] takes into account a pair of nodes that share a link in the network. Such an\napproachcanameliorateroutingmisbehaviorswhereintwoadjacentnodesarecolluding\nwith each other and dropping packets. Each link between two adjacent nodes has certain\nweight associated with it. When a node detects a link to be faulty, it increases the weight\nassociated with multiplicatively. When multiple routes are discovered for a particular\ndestination, the initiating node selects the route that has the least sum of link weights.\nThe least sum of link weights of a route implies that the route has the least likelihood\nof having a faulty link on it. The protocol consists of the following three phases. (a)\nRoute Discovery with fault avoidance, (b) Byzantine fault detection, and (c) link weight\nmanagement.\nRoute Discovery with Fault Avoidance: A source node initiates a Route Discovery\nby generating a route REQUEST packet, digitally signing it using its private key, and\nflooding the REQUEST packet in the network. The request consists of the source\nID, the destination ID, a sequence number and link weight list. Digital signature helps\nintermediate nodes to authenticate the source, and to safeguard against malicious nodes\ntrying to initiate route discovery and consume valuable network resources.\nWhen an intermediate node receives a route request, it checks its valid request\nlist to see if there is a matching request in the list for the same source. If there is no\nmatching request and the source’s signature is valid, it rebroadcasts the request, else\nthe request is dropped. When the destination receives a request from the source for\nthe first time, it checks the source signature on the request. If the signature is valid, it\ngenerates and signs the response consisting of source, destination, a response sequence\n"
        },
        {
            "page_number": 160,
            "text": "154\nVENKATA C. GIRUKA and MUKESH SINGHAL\nnumber and the weight list from the request packet. Unlike DSR, intermediate nodes\ndo not cache routes and respond to the source node.\nWhen an intermediate node receives a response, it computes the total weight of\nthe path by summing weights of all the links, which constitute the path. If the total\nweight is less than any of the previous responses for that particular request, it checks\nthe signature on the response header and every hop listed on the packet. If each element\nof the packet is verified, the node appends its identifier to the end of the packet, signs\nthe new packet, and broadcasts it.\nWhen the source receives the response, it verifies the digital signature of interme-\ndiate nodes. If the path is better than the best path received so far, the source updates\nthe route used to send packets to the particular destination. This type of route discovery\nattempts to find the route having lowest sum of link weights, thereby selecting a route\nwhich is least likely to have a faulty link on it. Faulty links have more link weight and\nget automatically precluded from route discovery.\nIn spite of this fault avoiding route discovery, there may still be a faulty link along\na route because no alternate routes with lower link weights were discovered. To detect\na faulty link, nodes invoke a Byzantine fault detection mechanism that uses an adaptive\nprobing technique.\nStep 1\nStep 2\nStep 3\nS\nT\nFaulty Link\nS\nT\nS\nT\nB\nA’\nA\nB’\nA\nA’\nB\nB\nA\nFigure 1. Fault Detection.\nByzantine Fault Detection: For the purpose of Byzantine fault detection, the protocol\nrequires the destination to return an ACK-message to the source for every successfully\nreceived data packet. If the source node does not receives valid ACKs during the\ntimeout, it assumes that the packets were lost in transit due to the presence of malicious\nnodes or because the destination is unreachable due to a network partition. For each\ndestination, the source node selects an ACK loss rate less than a fixed threshold as\n"
        },
        {
            "page_number": 161,
            "text": "WIRELESS NETWORK SECURITY\n155\ntolerable, and this may vary with every route. The source keeps track of number of\nlosses on a path. If this number exceeds the threshold, the source node initiates a binary\nsearch on the path, assuming a faulty link exists on the source-destination route, in an\nattempt to locate the faulty link.\nThe fault detection mechanism is best explained by an example shown in Figure 1.\nThe source specifies two random intermediate nodes, A and B, on the route called\nprobes, each of which must send an ack for the successfully received packet. The\nprobes divide the route into non-overlapping continuous segments. In the example,\nprobes A and B divide the path into SA, AB, and BD. Due to the presence of the faulty\nlink, S does not receives an ack from node B. Thus S determines a fault on the segment\nAB. S inserts a new probe A’ in between that segment. The probe insertion and interval\nsubdivision continues until the faulty interval narrows down to a single faulty link. in\nthe example it is the link A’B’. Due to binary search, the source detects a faulty link\nafter log(L) steps, L being the total number of nodes on the route.\nLink Weight Management: When a node detects a faulty link, it uses a multiplicative\nincrease scheme to double its weight. The higher the weight, the lower the probability\nof that link being on any further routes.\nThus using these techniques, route discovery with fault avoidance, Byzantine fault\ndetection, and link weight management, nodes establish routes that are free of nodes\nknown to be malicious and may attempt ‘Byzantine’ attacks.\n5.\nSECURE POSITION-BASED ROUTING\nSecurity in position-based routing is a relatively new area, and to the best of authors\nknowledge, there is no secure position-based routing protocol in the literature so far.\nThus, to keep the presentation simple, we discuss security issues related to position-\nbased greedy forwarding and some possible counter measures to fight attacks in greedy-\nforwarding. Fundamental to greedy-forwarding is a neighbor discovery or a hello\nprotocol using which nodes exchange their ID and position information periodically.\nHowever, malicious node may not follow the protocol properly, and may try to spoof\ntheir ID or location as explained in Section 3.\nTo prevent external attacks, nodes may employ an authentication mechanism like\nTESLA broadcast authentication as explained in Section 4.3, along with digital signa-\ntures to avoid attacks due to unauthorized external nodes. On the other hand, compro-\nmised internal nodes can pose severe threats to the greedy-forwarding. Zhou et al. [29]\nidentified location spoofing, traffic abusing and forwarding misbehavior as three main\ninternal attacks, and proposed the following counter measures.\nDefense against location spoofing: A possible way to defend against location spoof-\ning is to use the Time of Flight (ToF) of the message and the speed of signal to estimate\nthe distance between the two nodes. Precisely, if t is the round-trip time and s is the\nspeed of the signal, then the distance d between two communicating nodes should be\nless than (t×s)/2, i.e., d ≤(t×s)/2. However, this method does not provide an upper\n"
        },
        {
            "page_number": 162,
            "text": "156\nVENKATA C. GIRUKA and MUKESH SINGHAL\nbound on the distance, as a malicious node can hold the probe message for a arbitrary\ntime to increase the ToF value. By doing this, a malicious node succeeds in claiming a\nfarther position than its true position.\nTo mitigate this problem, the basic distance estimation method described above\ncan be augmented by using a neighbor monitoring scheme along with voting. The idea\ndepends on the fact that a false-position reported by a node tends to be inconsistent\namong neighbors. However, the success of this method depends on the ability of the\nvoting system to cope with false accusations.\nDefense against traffic abusing:\nTraffic abusing may range from dropping packets\nto flooding the network with junk or meaningful data at high-rate. By doing this, an\nattacker may attempt to exhaust network resources or overwhelm a node to do lot of\npacket-processing. To mitigate this problem, one can use the following observation:\nwhen an attacker abuses a node X with traffic, neighboring nodes of X experience\nanomalous traffic even before X. Thus, neighboring nodes may choose to drop such\npackets to save the attacked node.\nFurther, nodes can choose an upper bound and lower bound on the traffic intensity\nto detect anomalous traffic behavior. If a node experiences a traffic intensity above a\npreset lower bound, then the node may simply stop processing packets. This method\nworks even if a node is surrounded by a group of colluding malicious nodes.\nDefense against forwarding misbehaviors:\nAnother common problem in secure-\nrouting is to deal with forwarding misbehavior. Forwarding misbehaviors are more\nserious due to compromised internal nodes or due to ‘selfish’ or malicious nodes. Such\nnodes may want to gain services from network, but may not want to ‘give’ services to\nsave their limiting resources like battery. Note that a ‘selfish node’ may be not malicious\nbecause a selfish node may not harm the network. To keep up with our discussion, we\nconsider malicious nodes for forwarding misbehaviors. However, readers interested in\ndealing with selfish nodes are referred to [4, 28] for more details.\nA simple way to work around forwarding misbehaviors is to use multiple paths.\nMulti-path approach mitigates packet delivery failure, but incurs control overhead to\nhave multiple paths. Another approach is to maintain two-hop neighbor table, in con-\ntrast to one-hop neighbor table that is maintained in most position-base protocols, at\neach node, and employ a neighbor monitoring mechanism to verify the next hop trans-\nmission. For this approach nodes need to work in the promiscuous mode. In the\npromiscuous mode a node can overhear transmission for other nodes within its radio\nrange. When a node A selects a next hop B using greedy forwarding, it starts a timer to\ncheck if B forwarded the packet correctly to one of its neighbors C selected using the\ngreedy forwarding. If the timer expires before A hears a transmission from B, then A\nsuspects B and takes necessary actions (like flooding an accusation message). Else, if A\nhears a transmission from B, it checks if B selected a proper next hop. Since A, as well\nas other nodes, maintains a two-hop neighbor table, it can verify the next hop selection\nof B. However, the neighbor monitoring in promiscuous mode is prone to error, and\nsometime malicious node may attempt to falsely accuse benign nodes. Thus, protocols\nthat deal with such errors and false accusations [2] may help mitigate the problem.\n"
        },
        {
            "page_number": 163,
            "text": "WIRELESS NETWORK SECURITY\n157\n6.\nSUMMARY\nAd-hoc networks are potential enablers of networking any-where and any-time\nconcept, which is the current trend in this information-sharing age. While these net-\nworks are rapidly deployable and do not need an infrastructure to operate, they are very\nvulnerable to attacks from both inside and outside of the network. As explained in\nthis chapter, even the fundamental task of routing becomes non-trivial in presence of\nmalicious node. Especially, when the number of malicious nodes cross beyond certain\nthreshold, routing becomes impossible. Another extreme is a case where there is a\nsingle malicious node that connects two part of the network. In such cases, excluding\nmalicious node renders the network partitioned in to two or parts. In this chapter, we\npresented a brief description of routing protocols for ad-hoc networks, possible attacks\nof routing protocols, and various secure routing protocols that establish secure paths\nfrom a source to the destination. Further, we discussed some security counter measures\nfor position-based routing. Secure routing in ad-hoc network, as of now, is an active\narea of research. Coming up with an efficient and secure routing protocol under a robust\nsecurity model with provable security is still an open problem.\n7.\nREFERENCES\n1. Baruch Awerbuch, David Holmer, Cristina Nita Rotaru and Herbert Rubens. An On Demand Secure\nRouting Protocol Resilient to Byzantine Failures. In ACM Workshop on Wireless Security (WiSe),\nAtlanta, Georgia, September 28 2002.\n2. Sonja Buchegger and Jean-Yves Le Boudec. Performance Analysis of the CONFIDANT Protocol\n(Cooperation Of Nodes: Fairness In Dynamic Ad-hoc NeTworks). In proceedings of the 3rd ACM\ninternational symposium on Mobile ad hoc networking & computing (MOBIHOC) 2002. Lausanne,\nSwitzerland.\n3. S. Basagni, I. Chlamtac, V. Syroutik, and B. Woodward. A distance effect routing algorithm for mobility\n(DREAM). In proceedings of the 4th annual ACM/IEEE Int. Conf. on Mobile Computing and networking\n(MOBICOM), pages 76-84, Dallas, TX, USA, 1998.\n4. Levente Butty&#225;n and Jean-Pierre Hubaux. Stimulating cooperation in self-organizing mobile ad\nhoc networks.Mob. Netw. Appl., 8(5), 579–592, Kluwer Academic Publishers 2003.\n5. Prosenjit Bose, Pat Morin, Ivan Stojmenovic, and Jorge Urrutia. Routing with Guaranteed Delivery in\nAd Hoc Wireless Networks. Wireless Networksi 7, 609-616, Kluwer Academic Publishers 2001.\n6. http://www.bluetooth.com\n7. Yih-Chun Hu, David B. Johnson and Adrian Perrig, SEAD: Secure Efficient Distance Vector Routing\nin Mobile Wireless Ad-hoc Networks. Fourth IEEE Workshop on Mobile Computing Systems and\nApplications WMCSA ’02.\n8. Yih-Chun Hu, Adrian Perrig and David B. Johnson. Ariadne: A secure on-demand routing protocol\nfor ad-hoc networks. The 8th ACM International Conference on Mobile Computing and Networking,\nSeptember 2002.\n9. Z. Haas and M. Pearlman. The performance of query control scheme for the zone routing protocol.\nACM/IEEE Transactions on Networking, 9(4) pages 427-438, August 2001.\n10. P. Jacquet, P. Muhlethaler, T. Clausen, A. Laouiti, A. Qayyum, L. Viennot. Optimized Link State\nRouting Protocol. IEEE INMIC Pakistan 2001.\n"
        },
        {
            "page_number": 164,
            "text": "158\nVENKATA C. GIRUKA and MUKESH SINGHAL\n11. D. Johnson and D. Maltz. Mobile Computing. chapter 5. Dynamic Source Routing, pages 153-181.\nKulwer Academic Publishers, 1996.\n12. Karp, B., and Kung. H. T. GPSR: Greedy Perimeter Stateless Routing for Wireless Networks. Proc. 6th\nAnnual International Conference on Mobile Computing and Networking (MOBICOM 2000), 243-254.\n13. Young-Bae Ko , Nitin H. Vaidya. Location-aided routing (LAR) in mobile ad-hoc networks.\nACM/Blatzer Wireless Networks journal, 6(4) pages 307-321, 2000.\n14. Chris Karlof and David Wagner. Secure routing in sensor networks: Attacks and countermeasures.\nIn Proceedings of the IEEE International Workshop on Sensor Network Protocols and Applications\n(SNPA-03), May 2003.\n15. Jinyang Li, John Jannotti, Douglas S. J. De Couto, David R. Karger, Robert Morris. A Scalable Location\nService for Geographic Ad Hoc Routing. Proceedings of 6th ACM International Conference on Mobile\nComputing and Networking (MOBICOM) 2000.\n16. National Institute of Standards and Technology (NIST). Secure Hash Standard, May 1993. Federal\nInformation Processing Standards (FIPS) Publication 180-1.\n17. C.Perkins and P. Bhagwat. Highly dynamic destination sequenced distance-vector routing for mobile\ncomputers. Computer Communication Review, pages 234-244, October 94.\n18. Adrian Perrig, Ran Canetti, Dawn Song, and J. D. Tygar. Efficient and Secure Source Authentication for\nMulticast. In Network and Distributed System Security Symposium, NDSS 01, pages 35-46, February\n2001.\n19. P.Papadimitratos and Z.J. Haas. Secure Routing for Mobile Ad hoc Networks. In the proceedings of\nSCS Communication Networks and Distributed Systems Modeling And Simulation Conference (CNDS\n2002), San Antonio, TX, January 27-31, 2002.\n20. C. Perkins and E. Royer. Ad-hoc on-demand distance vector routing. In Proc. Of the 2nd IEEE Workshop\non Mobile Computing Systems and Applications, pages 90-100, Feb 1999.\n21. Ronald L. Rivest. The MD5 Message-Digest Algorithm. RFC 1321, April 1992.\n22. Michael Käsemann, Holger Füßler, Hannes Hartenstein, Martin Mauve. A Reactive Location Service\nfor Mobile Ad Hoc Networks. TR-14-2002, Department of Computer Science, University of Mannheim,\nNovember 2002.\n23. http://www.terminodes.org\n24. K.Sanzgiri, B.Dahill, B.N.Levine, C.Shields, E.M.Belding-Royer.AnAuthenticatedRoutingProtocol\nfor Secure Ad Hoc Networks. IEEE Journal on Selected Areas in Communication, special issue on\nWireless Ad hoc Networks, 23(3) pages 598-610, March 2005.\n25. Ning Song, Lijun Qian, Xiangfang Li. Wormhole Attacks Detection in Wireless Ad Hoc Networks: A\nStatistical Analysis Approach. 19th IEEE International Parallel and Distributed Processing Symposium\n(IPDPS’05) - Workshop 17, 2005.\n26. Ivan Stojmenovic. Position based routing in ad-hoc networks. IEEE Communications Magazine,40(7),\npages 128-134, 2002.\n27. Y Xue, B Li and K Nahrstedt. A scalable location management scheme in mobile ad-hoc networks.\n26th Annual IEEE Conference on Local Computer Networks (LCN 2001).\n28. Yongwei Wang, Venkata C. Giruka and Mukesh Singhal. A Fair Distributed Solution for Selfish Nodes\nProblem in Wireless Ad Hoc Networks. In Proceedings of Third International Conference, ADHOC-\nNOW 2004, Vancouver, Canada, July 22-24, 2004. Proceedings.\n29. Zhi Zhou and Kin Choong Yow. Geographic Ad Hoc Routing Security: Attacks and Countermeasures.\nAd Hoc & Sensor Wireless Networks, vol.1, number 1 pp 235-253, 2005.\n"
        },
        {
            "page_number": 165,
            "text": "7\nA SURVEY ON INTRUSION DETECTION IN\nMOBILE AD HOC NETWORKS\nTiranuch Anantvalee\nDepartment of Computer Science and Engineering\nFlorida Atlantic University, Boca Raton, FL 33428\nE-mail: tanantva@fau.edu\nJie Wu\nDepartment of Computer Science and Engineering\nFlorida Atlantic University, Boca Raton, FL 33428\nE-mail: jie@cse.fau.edu\nIn recent years, the use of mobile ad hoc networks (MANETs) has been widespread in\nmany applications, including some mission critical applications, and as such security has\nbecome one of the major concerns in MANETs. Due to some unique characteristics of\nMANETs, prevention methods alone are not sufficient to make them secure; therefore,\ndetection should be added as another defense before an attacker can breach the system. In\ngeneral, the intrusion detection techniques for traditional wireless networks are not well\nsuited for MANETs. In this paper, we classify the architectures for intrusion detection\nsystems (IDS) that have been introduced for MANETs. Current IDS’s corresponding to\nthose architectures are also reviewed and compared. We then provide some directions for\nfuture research.\n1.\nINTRODUCTION\nA mobile ad hoc network (MANET) is a self-configuring network that is formed\nautomatically by a collection of mobile nodes without the help of a fixed infrastructure\nor centralized management. Each node is equipped with a wireless transmitter and\nreceiver, which allow it to communicate with other nodes in its radio communication\nrange. In order for a node to forward a packet to a node that is out of its radio range,\nthe cooperation of other nodes in the network is needed; this is known as multi-hop\ncommunication. Therefore, each node must act as both a host and a router at the same\n"
        },
        {
            "page_number": 166,
            "text": "160\nTIRANUCH ANANTVALEE and JIE WU\ntime. The network topology frequently changes due to the mobility of mobile nodes as\nthey move within, move into, or move out of the network.\nA MANET with the characteristics described above was originally developed for\nmilitary purposes, asnodesarescatteredacrossabattlefieldandthereisnoinfrastructure\nto help them form a network. In recent years, MANETs have been developing rapidly\nand are increasingly being used in many applications, ranging from military to civilian\nand commercial uses, since setting up such networks can be done without the help\nof any infrastructure or interaction with a human. Some examples are: search-and-\nrescue missions, data collection, and virtual classrooms and conferences where laptops,\nPDA or other mobile devices share wireless medium and communicate to each other.\nAs MANETs become widely used, the security issue has become one of the primary\nconcerns. For example, most of the routing protocols proposed for MANETs assume\nthat every node in the network is cooperative and not malicious [1]. Therefore, only\none compromised node can cause the failure of the entire network.\nThere are both passive and active attacks in MANETs. For passive attacks, packets\ncontaining secret information might be eavesdropped, which violates confidentiality.\nActive attacks, including injecting packets to invalid destinations into the network,\ndeleting packets, modifying the contents of packets, and impersonating other nodes vi-\nolate availability, integrity, authentication, and non-repudiation. Proactive approaches\nsuch as cryptography and authentication [10, 11, 12, 13] were first brought into consid-\neration, and many techniques have been proposed and implemented. However, these\napplications are not sufficient. If we have the ability to detect the attack once it comes\ninto the network, we can stop it from doing any damage to the system or any data. Here\nis where the intrusion detection system comes in.\nIntrusion detection can be defined as a process of monitoring activities in a system,\nwhich can be a computer or network system. The mechanism by which this is achieved\nis called an intrusion detection system (IDS). An IDS collects activity information and\nthen analyzes it to determine whether there are any activities that violate the security\nrules. Once an IDS determines that an unusual activity or an activity that is known to\nbe an attack occurs, it then generates an alarm to alert the security administrator. In\naddition, IDS can also initiate a proper response to the malicious activity.\nAlthough there are several intrusion detection techniques developed for wired net-\nworks today, they are not suitable for wireless networks due to the differences in their\ncharacteristics. Therefore, those techniques must be modified or new techniques must\nbe developed to make intrusion detection work effectively in MANETs.\nIn this paper, we classify the architectures for IDS in MANETs, each of which\nis suitable for different network infrastructures. Current intrusion detection systems\ncorresponding to those architectures are reviewed and compared.\nThe rest of the paper is structured as follows. Section 2 describes the background\non intrusion detection systems. Intrusion detection in MANETs - how it differs from\nintrusion detection in wired networks - is also presented in this section. In Section 3,\narchitectures that have been introduced for IDS in MANETs are presented. Some of\ncurrent intrusion detection systems for MANETs are given in Section 4. Then, some\n"
        },
        {
            "page_number": 167,
            "text": "WIRELESS NETWORK SECURITY\n161\nof the intrusion detection techniques for node cooperation are reviewed and compared\nin Section 5. Finally, the conclusion and future directions are given in Section 6.\n2.\nBACKGROUND\n2.1. Intrusion Detection System (IDS)\nMany historical events have shown that intrusion prevention techniques alone,\nsuch as encryption and authentication, which are usually a first line of defense, are\nnot sufficient. As the system become more complex, there are also more weaknesses,\nwhich lead to more security problems. Intrusion detection can be used as a second wall\nof defense to protect the network from such problems. If the intrusion is detected, a\nresponse can be initiated to prevent or minimize damage to the system.\nSome assumptions are made in order for intrusion detection systems to work [1].\nThe first assumption is that user and program activities are observable. The second\nassumption, which is more important, is that normal and intrusive activities must have\ndistinct behaviors, as intrusion detection must capture and analyze system activity to\ndetermine if the system is under attack.\nIntrusion detection can be classified based on audit data as either host-based or\nnetwork-based. A network-based IDS captures and analyzes packets from network\ntraffic while a host-based IDS uses operating system or application logs in its analysis.\nBased on detection techniques, IDS can also be classified into three categories as follows\n[2].\nAnomaly detection systems: The normal profiles (or normal behaviors) of users\nare kept in the system. The system compares the captured data with these\nprofiles, and then treats any activity that deviates from the baseline as a possible\nintrusion by informing system administrators or initializing a proper response.\nMisuse detection systems: The system keeps patterns (or signatures) of known\nattacks and uses them to compare with the captured data. Any matched pattern\nis treated as an intrusion. Like a virus detection system, it cannot detect new\nkinds of attacks.\nSpecification-based detection: The system defines a set of constraints that de-\nscribe the correct operation of a program or protocol. Then, it monitors the\nexecution of the program with respect to the defined constraints.\n2.2. Intrusion Detection in MANETs\nMany intrusion detection systems have been proposed in traditional wired net-\nworks, where all traffic must go through switches, routers, or gateways. Hence, IDS\ncan be added to and implemented in these devices easily [17, 18]. On the other hand,\nMANETs do not have such devices. Moreover, the medium is wide open, so both\nlegitimate and malicious users can access it. Furthermore, there is no clear separation\nbetween normal and unusual activities in a mobile environment. Since nodes can move\n"
        },
        {
            "page_number": 168,
            "text": "162\nTIRANUCH ANANTVALEE and JIE WU\narbitrarily, false routing information could be from a compromised node or a node that\nhas outdated information. Thus, the current IDS techniques on wired networks cannot\nbe applied directly to MANETs. Many intrusion detection systems have been proposed\nto suit the characteristics of MANETs, some of which will be discussed in the next\nsections.\n3.\nARCHITECTURES FOR IDS IN MANETS\nThe network infrastructures that MANETs can be configured to are either flat or\nmulti-layer, depending on the applications. Therefore, the optimal IDS architecture\nfor a MANET may depend on the network infrastructure itself [9]. In a flat network\ninfrastructure, all nodes are considered equal, thus it may be suitable for applications\nsuch as virtual classrooms or conferences. On the contrary, some nodes are considered\ndifferent in the multi-layered network infrastructure. Nodes may be partitioned into\nclusters with one clusterhead for each cluster. To communicate within the cluster,\nnodes can communicate directly. However, communication across the clusters must\nbe done through the clusterhead. This infrastructure might be well suited for military\napplications.\n3.1. Stand-alone Intrusion Detection Systems\nIn this architecture, an intrusion detection system is run on each node independently\nto determine intrusions. Every decision made is based only on information collected\nat its own node, since there is no cooperation among nodes in the network. Therefore,\nno data is exchanged. Besides, nodes in the same network do not know anything\nabout the situation on other nodes in the network as no alert information is passed.\nAlthough this architecture is not effective due to its limitations, it may be suitable in a\nnetwork where not all nodes are capable of running an IDS or have an IDS installed.\nThis architecture is also more suitable for flat network infrastructure than for multi-\nlayered network infrastructure. Since information on each individual node might not\nbe enough to detect intrusions, this architecture has not been chosen in most of the IDS\nfor MANETs.\n3.2. Distributed and Cooperative Intrusion Detection Systems\nSince the nature of MANETs is distributed and requires cooperation of other nodes,\nZhang and Lee [1] have proposed that the intrusion detection and response system in\nMANETs should also be both distributed and cooperative as shown in Figure 1. Every\nnode participates in intrusion detection and response by having an IDS agent running\non them. An IDS agent is responsible for detecting and collecting local events and data\nto identify possible intrusions, as well as initiating a response independently. However,\nneighboring IDS agents cooperatively participate in global intrusion detection actions\nwhen the evidence is inconclusive. Similarly to stand-alone IDS architecture, this\narchitecture is more suitable for flat network infrastructure, not multi-layered one.\n"
        },
        {
            "page_number": 169,
            "text": "WIRELESS NETWORK SECURITY\n163\nIDS \nIDS \nIDS \nIDS \nIDS \nIDS \nIDS \nintrusion detection state, \nintrusion response \nFigure 1. Distributed and Cooperative IDS in MANETs proposed by Zhang and Lee [1]\n3.3. Hierarchical Intrusion Detection Systems\nHierarchical IDS architectures extend the distributed and cooperative IDS archi-\ntectures and have been proposed for multi-layered network infrastructures where the\nnetwork is divided into clusters. Clusterheads of each cluster usually have more func-\ntionality than other members in the clusters, for example routing packets across clus-\nters. Thus, these clusterheads, in some sense, act as control points which are similar to\nswitches, routers, or gateways in wired networks. The same concept of multi-layering is\napplied to intrusion detection systems where hierarchical IDS architecture is proposed.\nEach IDS agent is run on every member node and is responsible locally for its node, i.e.,\nmonitoring and deciding on locally detected intrusions. A clusterhead is responsible\nlocally for its node as well as globally for its cluster, e.g. monitoring network packets\nand initiating a global response when network intrusion is detected.\n3.4. Mobile Agent for Intrusion Detection Systems\nA concept of mobile agents has been used in several techniques for intrusion\ndetection systems in MANETs. Due to its ability to move through the large network,\neach mobile agent is assigned to perform only one specific task, and then one or more\nmobile agents are distributed into each node in the network. This allows the distribution\nof the intrusion detection tasks.\nThere are several advantages for using mobile agents [2]. Some functions are not\nassigned to every node; thus, it helps to reduce the consumption of power, which is\nscarceinmobileadhocnetworks. Italsoprovidesfaulttolerancesuchthatifthenetwork\nis partitioned or some agents are destroyed, they are still able to work. Moreover, they\nare scalable in large and varied system environments, as mobile agents tend to be\n"
        },
        {
            "page_number": 170,
            "text": "164\nTIRANUCH ANANTVALEE and JIE WU\nsecure \ncommunication \ncooperative \ndetection engine \nglobal \nresponse \nlocal data \ncollection \nlocal \ndetection engine \nlocal \nresponse \nsystem calls activities, \ncommunication activities, \nother traces … \nIDS agent \nneighboring \nIDS agents \nFigure 2. A Model for an IDS Agent [1]\nindependent of platform architectures. However, these systems would require a secure\nmodule where mobile agents can be stationed to. Additionally, mobile agents must be\nable to protect themselves from the secure modules on remote hosts as well.\nMobile-agent-based IDS can be considered as a distributed and cooperative intru-\nsion detection technique as described in Section 3.2. Moreover, some techniques also\nuse mobile agents combined with hierarchical IDS, for example, what will be described\nin Section 4.3.\n4.\nSAMPLE INTRUSION DETECTION SYSTEMS FOR MANETS\nSince the IDS for traditional wired systems are not well-suited to MANETs, many\nresearchers have proposed several IDS especially for MANETs, which some of them\nwill be reviewed in this section.\n4.1. Distributed and Cooperative IDS\nAs described in Section 3.2, Zhang and Lee also proposed the model for a distrib-\nuted and cooperative IDS as shown in Figure 2 [1].\nThe model foranIDSagentisstructuredintosixmodules. Thelocaldatacollection\nmodule collects real-time audit data, which includes system and user activities within\nits radio range. This collected data will be analyzed by the local detection engine\nmodule for evidence of anomalies. If an anomaly is detected with strong evidence, the\nIDS agent can determine independently that the system is under attack and initiate a\nresponse through the local response module (i.e., alerting the local user) or the global\nresponse module (i.e., deciding on an action), depending on the type of intrusion, the\n"
        },
        {
            "page_number": 171,
            "text": "WIRELESS NETWORK SECURITY\n165\ntype of network protocols and applications, and the certainty of the evidence. If an\nanomaly is detected with weak or inconclusive evidence, the IDS agent can request the\ncooperation of neighboring IDS agents through a cooperative detection engine module,\nwhich communicates to other agents through a secure communication module.\n4.2. Local Intrusion Detection System (LIDS)\nAlbers et al. [3] proposed a distributed and collaborative architecture of IDS by\nusing mobile agents. A Local Intrusion Detection System (LIDS) is implemented on\nevery node for local concern, which can be extended for global concern by cooperating\nwith other LIDS. Two types of data are exchanged among LIDS: security data (to obtain\ncomplementary information from collaborating nodes) and intrusion alerts (to inform\nothers of locally detected intrusion). In order to analyze the possible intrusion, data\nmust be obtained from what the LIDS detects, along with additional information from\nother nodes. Other LIDS might be run on different operating systems or use data from\ndifferent activities such as system, application, or network activities; therefore, the\nformat of this raw data might be different, which makes it hard for LIDS to analyze.\nHowever, such difficulties can be solved by using SNMP (Simple Network Management\nProtocol) data located in MIBs (Management Information Base) as an audit data source.\nSuch a data source not only eliminates those difficulties, but also reduces the increase\nin using additional resources to collect audit data if an SNMP agent is already run on\neach node.\nTo obtain additional information from other nodes, the authors proposed mobile\nagents to be used to transport SNMP requests to other nodes. In another words, to\ndistribute the intrusion detection tasks. The idea differs from traditional SNMP in that\nthe traditional approach transfers data to the requesting node for computation while\nthis approach brings the code to the data on the requested node. This is motivated\nby the unreliability of UDP messages used in SNMP and the dynamic topology of\nMANETs. As a result, the amount of exchanged data is tremendously reduced. Each\nmobile agent can be assigned a specific task which will be achieved in an autonomous\nand asynchronous fashion without any help from its LIDS.\nThe LIDS architecture is shown in Figure 3, which consists of\nCommunication Framework: To facilitate for both internal and external com-\nmunication with a LIDS.\nLocal LIDS Agent: To be responsible for local intrusion detection and local\nresponse. Also, it reacts to intrusion alerts sent from other nodes to protect\nitself against this intrusion.\nLocal MIB Agent: To provide a means of collecting MIB variables for either\nmobile agents or the Local LIDS Agent. Local MIB Agent acts as an interface\nwith SNMP agent, if SNMP exists and runs on the node, or with a tailor-made\nagent developed specifically to allow updates and retrievals of the MIB variables\nused by intrusion detection, if none exists.\n"
        },
        {
            "page_number": 172,
            "text": "166\nTIRANUCH ANANTVALEE and JIE WU\nCommunication Framework \nMobile \nAgents \nPlace \nMA \nMA \nMA \nMA \nLocal \nLIDS \nAgent \nSNMP \nAgent \nLocal \nMIB \nAgent \nMANET \nAudit Source \n(MIB) \nLIDS \nFigure 3. LIDS Architecture in A Mobile Node [3]\nMobile Agents (MA): They are distributed from its LID to collect and process\ndata on other nodes. The results from their evaluation are then either sent back\nto their LIDS or sent to another node for further investigation.\nMobile Agents Place: To provide a security control to mobile agents.\nFor the methodology of detection, Local IDS Agent can use either anomaly or misuse\ndetection. However, the combination of two mechanisms will offer the better model.\nOnce the local intrusion is detected, the LIDS initiates a response and informs the other\nnodes in the network. Upon receiving an alert, the LIDS can protect itself against the\nintrusion.\n4.3. Distributed Intrusion Detection System Using Multiple Sensors\nKachirski and Guha [4] proposed a multi-sensor intrusion detection system based\non mobile agent technology. The system can be divided into three main modules, each\nof which represents a mobile agent with certain functionality: monitoring, decision-\nmaking or initiating a response. By separating functional tasks into categories and\nassigning each task to a different agent, the workload is distributed which is suitable\nfor the characteristics of MANETs. In addition, the hierarchical structure of agents is\nalso developed in this intrusion detection system as shown in Figure 4.\nMonitoring agent: Two functions are carried out at this class of agent: network\nmonitoring and host monitoring. A host-based monitor agent hosting system-\nlevel sensors and user-activity sensors is run on every node to monitor within\n"
        },
        {
            "page_number": 173,
            "text": "WIRELESS NETWORK SECURITY\n167\nPacket-level\nAction \nDecision \nMonitoring \nSystem-level \nUser-level \nFigure 4. Layered Mobile Agent Architecture proposed by Kachirski and Guha [4]\nthe node, while a monitor agent with a network monitoring sensor is run only\non some selected nodes to monitor at packet-level to capture packets going\nthrough the network within its radio ranges.\nAction agent: Every node also hosts this action agent. Since every node hosts\na host-based monitoring agent, it can determine if there is any suspicious or\nunusual activities on the host node based on anomaly detection. When there is\nstrong evidence supporting the anomaly detected, this action agent can initiate\na response, such as terminating the process or blocking a user from the network.\nDecision agent: The decision agent is run only on certain nodes, mostly those\nnodes that run network monitoring agents. These nodes collect all packets\nwithin its radio range and analyze them to determine whether the network is\nunder attack. Moreover, from the previous paragraph, if the local detection\nagent cannot make a decision on its own due to insufficient evidence, its local\ndetection agent reports to this decision agent in order to investigate further.\nThis is done by using packet-monitoring results that comes from the network-\nmonitoring sensor that is running locally. If the decision agent concludes that\nthe node is malicious, the action module of the agent running on that node as\ndescribed above will carry out the response.\nThe network is logically divided into clusters with a single clusterhead for each\ncluster. This clusterhead will monitor the packets within the cluster and only packets\nwhose originators are in the same cluster are captured and investigated. This means\nthat the network monitoring agent (with network monitoring sensor) and the decision\nagent are run on the clusterhead.\nIn this mechanism, the decision agent performs the decision-making based on its\nown collected information from its network-monitoring sensor; thus, other nodes have\nno influence on its decision. This way, spoofing attacks and false accusations can be\nprevented.\n"
        },
        {
            "page_number": 174,
            "text": "168\nTIRANUCH ANANTVALEE and JIE WU\ndetected data and/or report \naggregated data/results \ndirectives, signature updates, etc. \n2\n1\n1\n2\n3\n1\nFigure 5. Dynamic Intrusion Detection Hierarchy [16]\n4.4. Dynamic Hierarchical Intrusion Detection Architecture\nSince nodes move arbitrarily across the network, a static hierarchy is not suitable\nfor such dynamic network topology. Sterne et al. [16] proposed a dynamic intrusion\ndetection hierarchy that is potentially scalable to large networks by using clustering\nlike those in Section 4.3 and 5.5. However, it can be structured in more than two levels\nas shown in Figure 5. Nodes labeled “1” are the first level clusterheads while nodes\nlabeled “2” are the second level clusterheads and so on. Members of the first level of\nthe cluster are called leaf nodes.\nEvery node has the responsibilities of monitoring (by accumulating counts and\nstatistics), logging, analyzing (i.e., attack signature matching or checking on packet\nheaders and payloads), responding to intrusions detected if there is enough evidence,\nand alerting or reporting to clusterheads. Clusterheads, in addition, must also perform:\nData fusion/integration and data reduction: Clusterheads aggregate and cor-\nrelate reports from members of the cluster and data of their own. Data reduction\nmay be involved to avoid conflicting data, bogus data and overlapping reports.\nBesides, clusterheads may send the requests to their children for additional\ninformation in order to correlate reports correctly.\nIntrusion detection computations: Since different attacks require different\nsets of detected data, data on a single node might not be able to detect the\nattack, e.g., DDoS attack, and thus clusterheads also analyze the consolidated\ndata before passing to upper levels.\nSecurity Management: The uppermost levels of the hierarchy have the author-\nity and responsibility for managing the detection and response capabilities of\nthe clusters and clusterheads below them. They may send the signatures update,\n"
        },
        {
            "page_number": 175,
            "text": "WIRELESS NETWORK SECURITY\n169\nor directives and policies to alter the configurations for intrusion detection and\nresponse. These update and directives will flow from the top of the hierarchy\nto the bottom.\nTo form the hierarchical structure, every node uses clustering, which is typically\nused in MANETs to construct routes, to self-organize into local neighborhoods (first\nlevel clusters) and then select neighborhood representatives (clusterheads). These rep-\nresentatives then use clustering to organize themselves into the second level and select\nthe representatives. This process continues until all nodes in the network are part of\nthe hierarchy. The authors also suggested criteria on selecting clusterheads. Some of\nthese criteria are:\nConnectivity: the number of nodes within one hop\nProximity: members should be within one hop of its clusterhead\nResistance to compromise (hardening): the probability that the node will not\nbe compromised. This is very important for the upper level clusterheads.\nProcessing power, storage capacity, energy remaining, bandwidth capabilities\nAdditionally, this proposed architecture does not rely solely on promiscuous node\nmonitoring like many proposed architectures, due to its unreliability as described in\n[5]. Therefore, this architecture also supports direct periodic reporting where packet\ncounts and statistics are sent to monitoring nodes periodically.\n4.5. Zone-Based Intrusion Detection System (ZBIDS)\nSun et al. [24] has proposed an anomaly-based two-level nonoverlapping Zone-\nBased Intrusion Detection System (ZBIDS). By dividing the network in Figure 6 into\nnonoverlapping zones (zone A to zone I), nodes can be categorized into two types: the\nintrazone node and the interzone node (or a gateway node). Considering only zone\nE, node 5, 9, 10 and 11 are intrazone nodes, while node 2, 3, 6, and 8 are interzone\nnodes which have physical connections to nodes in other zones. The formation and\nmaintenance of zones requires each node to know its own physical location and to map\nits location to a zone map, which requires prior design setup.\nEach node has an IDS agent run on it which the model of the agent is shown\nin Figure 7. Similar to an IDS agent proposed by Zhang and Lee (Figure 2), the data\ncollection module and the detection engine are responsible for collecting local audit data\n(for instance, system call activities, and system log files) and analyzing collected data\nfor any sign of intrusion respectively. In addition, there may be more than one for each\nof these modules which allows collecting data from various sources and using different\ndetection techniques to improve the detection performance. The local aggregation\nand correlation (LACE) module is responsible for combining the results of these local\ndetection engines and generating alerts if any abnormal behavior is detected. These\nalerts are broadcasted to other nodes within the same zone. However, for the global\n"
        },
        {
            "page_number": 176,
            "text": "170\nTIRANUCH ANANTVALEE and JIE WU\n8\n5\n3\n6\n7\n12\n4\n1\n11 \n10\n9\n2\nA \nB C \nD\nG\nE\nH\nF\nI\nFigure 6. ZBIDS for MANETs [24]\nIDS agent\nAudit \nData \nData Collection \nModule \nData Collection \nModule \nData Collection \nModule \nDetection \nEngine \nDetection \nEngine \nDetection \nEngine \nLocal Aggregation \nAnd Correlation  \n(LACE) \nIntrusion \nResponse \nGlobal Aggregation \nAnd Correlation  \n(GACE) \ninterzone nodes: receive from intrazone nodes \nand the neighboring interzone nodes \nintrazone nodes: send to the interzone nodes \ninterzone nodes: send to the neighboring \ninterzone nodes \nFigure 7. An IDS agent in ZBIDS [24]\naggregation and correlation (GACE), its functionality depends on the type of the node.\nAs described in Figure 7, if the node is an intrazone node, it only sends the generated\nalerts to the interzone nodes. Whereas, if the node is an interzone node, it receives alerts\nfrom other intrazone nodes, aggregates and correlates those alerts with its own alerts,\nand then generates alarms. Moreover, the GACE also cooperates with the GACEs of the\nneighboring interzone nodes to have more accurate information to detect the intrusion.\nLastly, the intrusion response module is responsible for handling the alarms generated\nfrom the GACE.\nThe local aggregation and correlation algorithm used in ZBIDS is based on a\nlocal Markov chain anomaly detection. An IDS agent first creates a normal profile by\n"
        },
        {
            "page_number": 177,
            "text": "WIRELESS NETWORK SECURITY\n171\nS\nA\nC\nD\nB\nFigure 8. How watchdog works: Although node B intends to transmit a packet to node C,\nnode A could overhear this transmission\nconstructing a Markov chain from the routing cache. A valid change in the routing cache\ncan be characterized by the Markov chain detection model with probabilities, otherwise,\nit’s considered abnormal, and the alert will be generated. For the global aggregation\nand correlation algorithm, it’s based on information provided in the received alerts\ncontaining the type, the time, and the source of the attacks.\n5.\nINTRUSIONDETECTIONTECHNIQUESFORNODECOOPERATION\nIN MANETS\nSince there is no infrastructure in mobile ad hoc networks, each node must rely\non other nodes for cooperation in routing and forwarding packets to the destination.\nIntermediate nodes might agree to forward the packets but actually drop or modify them\nbecause they are misbehaving. The simulations in [5] show that only a few misbehaving\nnodes can degrade the performance of the entire system. There are several proposed\ntechniques and protocols to detect such misbehavior in order to avoid those nodes, and\nsome schemes also propose punishment as well [6, 7].\n5.1. Watchdog and Pathrater\nTwo techniques were proposed by Marti, Giuli, and Baker [5], watchdog and\npathrater, to be added on top of the standard routing protocol in ad hoc networks. The\nstandard is Dynamic Source Routing protocol (DSR) [8]. A watchdog identifies the\nmisbehaving nodes by eavesdropping on the transmission of the next hop. A pathrater\nthen helps to find the routes that do not contain those nodes.\nIn DSR, the routing information is defined at the source node. This routing infor-\nmation is passed together with the message through intermediate nodes until it reaches\nthe destination. Therefore, each intermediate node in the path should know who the\nnext hop node is. In addition, listening to the next hop’s transmission is possible be-\ncause of the characteristic of wireless networks - if node A is within range of node B,\nA can overhear communication to and from B.\nFigure 8 shows how the watchdog works. Assume that node S wants to send a\npacket to node D, which there exists a path from S to D through nodes A, B, and C.\nConsider now that A has already received a packet from S destined to D. The packet\ncontains a message and routing information. When A forwards this packet to B, A\nalso keeps a copy of the packet in its buffer. Then, it promiscuously listens to the\ntransmission of B to make sure that B forwards to C. If the packet overheard from B\n(represented by a dashed line) matches that stored in the buffer, it means that B really\n"
        },
        {
            "page_number": 178,
            "text": "172\nTIRANUCH ANANTVALEE and JIE WU\nforwards to the next hop (represented as a solid line). It then removes the packet from\nthe buffer. However, if there’s no matched packet after a certain time, the watchdog\nincrements the failures counter for node B. If this counter exceeds the threshold, A\nconcludes that B is misbehaving and reports to the source node S.\nPathrater performs the calculation of the“path metric” for each path. By keeping\nthe rating of every node in the network that it knows, the path metric can be calculated\nby combining the node rating together with link reliability, which is collected from\npast experience. Obtaining the path metric for all available paths, the pathrater can\nchoose the path with the highest metric. In addition, if there is no such link reliability\ninformation, the path metric enables the pathrater to select the shortest path too. As a\nresult, paths containing misbehaving nodes will be avoided.\nFrom the result of the simulation, the system with these two techniques is quite\neffective for choosing paths to avoid misbehaving nodes. However, those misbehaving\nnodes are not punished. In contrast, they even benefit from the network. In another\nword, they can use resources of the network - other nodes forward packets for them,\nwhile they forward packets for no one, which save their own resources. Therefore,\nmisbehaving nodes are encouraged to continue their behaviors.\n5.2. CONFIDANT\nBuchegger and LeBoudec [6] proposed an extension to DSR protocol called CON-\nFIDANT (Cooperation Of Nodes, Fairness In Dynamic Ad-hoc NeTworks), which is\nsimilar to Watchdog and Pathrater. Each node observes the behaviors of neighbor nodes\nwithin its radio range and learns from them. This system also solves the problem of\nWatchdog and Pathrater such that misbehavior nodes are punished by not including\nthem in routing and not helping them on forwarding packets. Moreover, when a node\nexperiences a misbehaving node, it will send a warning message to other nodes in the\nnetwork, defined as friends, which is based on trusted relationship.\nFigure 9 shows the components of the CONFIDANT protocol, which are the Mon-\nitor, the Trust Manager, the Reputation System, and the Path Manager. The process of\nhow they work can be divided into two parts: the process to handle its own observations\nand the process to handle reports from trusted nodes.\nFrom observations: The monitor uses a “neighborhood watch” to detect any\nmalicious behaviors with in its radio range, i.e., no forwarding, unusually fre-\nquent route update, etc. (This is similar to the watchdog in the previous scheme)\nIf a suspicious event is detected, the monitor then reports to the reputation sys-\ntem. At this point, the reputation system performs several checks and updates\nthe rating of the reported node in the reputation table. If the rating result is un-\nacceptable, it passes the information to the path manager, which then removes\nall paths containing the misbehavior node. An ALARM message is also sent\nby the trust manager to warn other nodes that it considers as friends.\nFrom trusted nodes: When the monitor receives an ALARM message from\nits friends, the message will first be evaluated by the trust manager for the\n"
        },
        {
            "page_number": 179,
            "text": "WIRELESS NETWORK SECURITY\n173\nUpdating \nALARM \nEvaluating \nalarm\nUpdating \nevent count\nManaging path \nInitial state \nEvaluating \ntrust\nSending \nALARM \nMonitor \nPath Manager\nTrust Manager \nReputation System \nMonitoring \nALARM \nreceived \ntrusted \nenough evidence \nevent  \ndetected \nnot\nsignificant \nbelow threshold \nthreshold \nexceeded \nRating \ntolerance exceeded \nwithin\ntolerance \nnot trusted \nsignificant  \nevent \nFigure 9. Components and State Diagram of CONFIDANT Protocol [6]\ntrustworthiness of the source node. If the message is trustworthy, this ALARM\nmessage, together with the level of trust, will be stored in the alarm table. All\nALARM messages of the reported node will then be combined to see if there\nis enough evidence to identify that it is malicious. If so, the information will\nbe sent to the reputation system, which then performs the same functions as\ndescribed in the previous paragraph.\nSince this protocol allows nodes in the network to send alarm messages to each\nother, it could give more opportunities for attackers to send false alarm messages that\na node is misbehaving while it’s actually not. This is one form of denial of service\nattacks.\n5.3. CORE\nMichiardi and Molva [7] presented a technique to detect a specific type of mis-\nbehaving nodes, which are selfish nodes, and also force them to cooperate. Similar\nto those in Section 5.1 and 5.2, this technique is based on a monitoring system and a\nreputation system, which includes both direct and indirect reputation from the system\nas will be described shortly.\nAs nodes sometimes do not intentionally misbehave, i.e., battery condition is low,\nthese nodes should not be considered as misbehaving nodes and excluded from the\nnetwork. To do this, the reputation should be rated based on past reputation, which\n"
        },
        {
            "page_number": 180,
            "text": "174\nTIRANUCH ANANTVALEE and JIE WU\nis zero (neutral) at the beginning. In addition, participation in the network can be\ncategorized into several functions such as routing discovery (in DSR) or forwarding\npackets.\nEach of these activities has different level of effects to the network; for\nexample, forwarding packets has more effect on the performance of the system than\nthat of routing discovery. Therefore, significance weight of functions should be used\nin the calculation of the reputation.\nLike CONFIDANT, each node can receive a report from other nodes. However,\nthe difference is CORE allows only positive reports to be passed while negative reports\nare passed in CONFIDANT. In another word, CORE prevents false accusation, thus,\nit also prevents a denial of service attack, which cannot be done in CONFIDANT. The\nnegative rating is given to a node only from the direct observation when the node does\nnot cooperate, which results in the decreased reputation for that node. The positive\nrating, in contrast, is given from both direct observation and positive reports from other\nnodes, which results in the increased reputation.\nCORE can then be said to have two components, the watchdog system and the\nreputation system. The watchdog modules, one for each function, work the same way\nas in the previous two schemes above. For the reputation system, it maintains several\nreputation tables, one for each function and one for accumulated values for each node.\nTherefore, if there is a request from a bad reputation node (the overall reputation is\nnegative), the node will be rejected and not be able to use the network.\n5.4. OCEAN\nBansal and Baker [19] also proposed an extension on top of the DSR protocol called\nOCEAN (Observation-based Cooperation Enforcement in Ad hoc Networks). OCEAN\nalso uses a monitoring system and a reputation system. However, in contrast to the\nprevious approaches above, OCEAN relies only on its own observation to avoid the new\nvulnerability of false accusation from second-hand reputation exchanges. Therefore,\nOCEAN can be considered as a stand-alone architecture.\nOCEAN categorizes routing misbehavior into two types: misleading and selfish.\nIf a node has participated in the route discovery but not packet forwarding, this is\nconsidered to be misleading as it misleads other nodes to route packets through it. But\nif a node does not even participate in the route discovery, it is considered to be selfish.\nIn order to detect and mitigate the misleading routing behaviors, after a node\nforwards a packet to a neighbor, it buffers the packet checksum and monitors if the\nneighbor attempts to forward the packet within a given time. Then, a negative or\npositive event is given as the result of the monitoring to update the neighbor rating.\nIf the rating falls below the faulty threshold, that neighbor node is added to a faulty\nlist which will be added in the RREQ as an avoid-list. In addition, all traffic from the\nfaulty neighbor node will be rejected. Nonetheless, the faulty timeout is used to allow\nthe faulty node to join back to the network in case that it might be false accused or it\nbehaves better.\nEach node also has a mechanism of maintaining chipcounts for each neighbor to\nmitigate the selfish behavior. A neighbor node earns chips when forwarding a packet\n"
        },
        {
            "page_number": 181,
            "text": "WIRELESS NETWORK SECURITY\n175\non behalf of the node and loses ships when asking the node to forward a packet. If the\nchipcount of the neighbor is below the threshold, packets coming from that neighbor\nwill be denied.\n5.5. Cooperative Intrusion Detection System\nA cluster-based cooperative intrusion detection system, similar to Kachirski and\nGuha’s system [4], has been presented by Huang and Lee [14]. In this approach, an IDS\nis not only able to detect an intrusion, but also to identify the attack type and the attacker,\nwhenever possible, through statistical anomaly detection. Various types of statistics (or\nfeatures), which are proposed in their previous work [15], are evaluated from a sampling\nperiod by capturing the basic view of network topology and routing operations, as well\nas traffic patterns and statistics, in the normal traffic. Hence, attacks could be identified\nif the statistics deviate from the pre-computed ones (anomaly detection).\nStatistics can be categorized into two categories, non traffic-related and traffic-\nrelated. Non traffic-related statistics are calculated based on the mobility and the trace\nlog files, which can be done separately on each node. Some of these statistics are route\nadd count, route removal count, total route change, average route length, etc. Traffic-\nrelated statistics are involved in routing and packet forwarding and can be calculated\nby counting packets going in and out, e.g. the number of packet received, the number\nof packet forwarded, the number of route reply messages, etc. These statistics can be\ncaptured by the node itself or the neighboring nodes who overhear the transmission.\nSeveral identification rules are pre-defined for known attacks by using relationships\nof the mentioned statistics. Once an anomaly is detected, the IDS will perform further\ninvestigation to determine the detailed information of the attack from a set of these\nidentification rules. These rules enhance the system to identify the type of the attack\nand, in some cases, the attacking node. Some notations of statistics are presented as\nfollows. Let M represent the monitoring node and m represent the monitored node.\n#(∗, m): the number of incoming packets on the monitored node m.\n#(∗, [m]): the number of incoming packets of which the monitored node m is\nthe destination.\n#(m, ∗): the number of outgoing packets from the monitored node m.\n#([m], ∗): the number of outgoing packets of which the monitored node m is\nthe source.\n#(m, n): the number of outgoing packets from m of which n is the next hop.\n#([s], M, m): the number of packets that are originated from s and transmitted\nfrom M to m.\n#([s], [d]): the number of packets received on m which is originated from s\nand destined to d.\n"
        },
        {
            "page_number": 182,
            "text": "176\nTIRANUCH ANANTVALEE and JIE WU\n#(∗, m)(TY PE = RREQ): the number of incoming RREQ packets on m.\nThese statistics are computed over a long period L. Let FEATUREL represents\nthe aggregated FEATURE over time L. Some identification rules are defined for\nwell known attacks as follows.\nUnconditional Packet Dropping:This rule uses Forward Percentage (FP) over\na period L to define the attack.\nFPm = packets actually forwarded\npackets to be forwarded\n= #L(m, M) −#L([m], M)\n#L(M, m) −#L(M, [m])\nIf there are packets to be forwarded (denominator is not zero) and FPm = 0,\nthe unconditional packet dropping attack is detected and the attacker is m.\nRandom Packet Dropping: This rule also uses the same FP as unconditional\npacket dropping. However, the threshold ϵF P is defined (ϵF P < 1). If 0 <\nFPm < ϵF P , m is defined as an attacker using random packet dropping.\nSelective Packet Dropping: This rule uses Local Forward Percentage (LFP)\nfor each source s.\nLFP s\nm = packets from source s actually forwarded\npackets from source s to be forwarded\n=\n#L([s], m, M)\n#L([s], M, m) −#L([s], M, [m])\nIf the denominator is not zero and LFP s\nm = 0, the attack is the unconditional\npacket dropping targeted at s. However, if LFP s\nm is less than the threshold\n(ϵLF P < 1), the attack is detected as random packet dropping targeted at s.\nBlackhole: This rule uses Global Forward Percentage (GFP) and it must be\ncomputed on M locally because the rule relies on information available only\non the node. Let N(M) denote M’s 1-hop neighbors.\nGFP s\nm =\npackets to be forwarded\npackets fromN(M)destined to other nodes than itself or anotherN(M)\n=\n#L(∗, M) −#L(∗, [M])\n\u0001\niϵN(M)\n#L(i, M) −\n\u0001\ni,jϵN(M)\n#L(i, [j]) −#L(∗, [M])\nIf the denominator is not zero and GFP = 1, it means that the blackhole attack\nis detected and M is the attacker.\nMalicious Flooding on specific target: This rule uses #L([m], [d]) for every\ndestination d. If it is larger than the threshold the attack is Malicious Flooding.\nHowever, the attacker cannot be determined.\n"
        },
        {
            "page_number": 183,
            "text": "WIRELESS NETWORK SECURITY\n177\nTable 1. Comparison among IDS for Node Cooperation\nTechniques \nWatchdog/ \nPathrater \nCONFIDANT \nCORE \nOCEAN \nCooperative IDS \nArchitecture \nDistributed and cooperative \nStand-alone \nHierarchical \nType of data collection \nReputation \nStatistics \nData distribution \nnegative \nto source node \nnegative \nto friends \npositive \nfrom RREP \nno \nto clusterhead \nself to neighbor \nyes \nyes \nyes \nyes \nyes \nObservation\nneighbor to neighbor \nno \nyes \nno \nyes \nyes \nSelfish – routing \nno \nyes \nyes \nyes \nyes \nSelfish – packet \nforwarding \nyes \nyes \nyes \nyes \nyes \nMalicious – routing \nno \nyes \nno \nno \nyes \nMisbehavior \ndetection \nMalicious – packet \nforwarding \nyes \nyes \nno \nno \nyes \nPunishment \nno \nyes \nyes \nyes \nn/a \nAvoid misbehaving node in route \ndiscovery \nno \nno \nno \nyes \nn/a \nThe authors also presented cluster formation algorithms and ensured that they are\nfair and secure. Each and every node has an equal chance of becoming a clusterhead and\nserves as a clusterhead for an equal service time. In addition, no node can manipulate\nthe clusterhead selection process. Initially, each node forms a clique - a group of\nnodes where every pair of members can communicate via a direct wireless link. Then,\nmembers in the clique perform the selection of a clusterhead. The process of re-election,\nto enforce fairness, and the process of recovery from lost clusterheads are defined as\nwell.\nMonitoring is how data is obtained in order to analyze for possible intrusions,\nhowever it consumes power. Therefore, instead of every node capturing all features\nthemselves, the clusterhead is solely responsible for computing traffic-related statistics.\nThis can be done because the clusterhead overhears incoming and outgoing traffic on\nall members of the cluster as it is one hop away (a clique). As a result, the energy\nconsumption of member nodes is lessened, whereas the detection accuracy is just a\nlittle worse than that of not implementing clusters. Besides, the performance of the\noverall network is noticeably better - decreases in CPU usage and network overhead.\n5.6. Summary of IDS for Detecting Misbehaving Nodes\nAlthough the watchdog is used in all of the above IDS, the authors in [5] have\npointed out that there are several limitations. The watchdog cannot work properly\nin the presence of collisions, which could lead to false accusations. Moreover, when\neach node has different transmission ranges or implements directional antennas, the\nwatchdog could not monitor the neighborhood accurately.\nAll of the above IDS’s presented are common in detecting selfish nodes. However,\nCORE doesn’t detect malicious misbehaviors while the others detect some of them, i.e.,\nunusually frequent route update, modifying header or payload of packets, no report of\nfailed attempts, etc. Table 1 shows the comparison among these IDS.\n"
        },
        {
            "page_number": 184,
            "text": "178\nTIRANUCH ANANTVALEE and JIE WU\n6.\nCONCLUSIONS AND FUTURE DIRECTIONS\nAs the use of mobile ad hoc networks (MANETs) has increased, the security in\nMANETs has also become more important accordingly. Historical events show that\nprevention alone, i.e., cryptography and authentication are not enough; therefore, the\nintrusion detection systems are brought into consideration. Since most of the current\ntechniques were originally designed for wired networks, many researchers are engaged\nin improving old techniques or finding and developing new techniques that are suitable\nfor MANETs.\nWith the nature of mobile ad hoc networks, almost all of the intrusion detection\nsystems (IDSs) are structured to be distributed and have a cooperative architecture. The\nnumber of new attacks is likely to increase quickly and those attacks should be detected\nbefore they can do any harm to the systems or data. Hence, IDS’s in MANETs prefer\nusing anomaly detection to misuse detection [1, 3, 4, 14, 24]. Some techniques are\nproposed to implement on top of the existing protocols [5, 6, 7], others are proposed as\nindependent modules to be added on mobile nodes [1, 3, 4, 14, 16, 24].\nAn intrusion detection system aims to detect attacks on mobile nodes or intrusions\ninto the networks. However, attackers may try to attack the IDS system itself [5].\nAccordingly, the study of the defense to such attacks should be explored as well.\nMany researchers are currently occupied in applying game theory for cooperation\nof nodes in MANETs [20, 21, 22, 23] as nodes in the network represent some charac-\nteristics similar to social behavior of human in a community. That is, a node tries to\nmaximize its benefit by choosing whether to cooperate in the network. There is not\nmuch work done in this area, therefore, it is an interesting topic for future research.\nACKNOWLEDGEMENTS\nThis work was supported in part by NSF grants CCR 0329741, CNS 0422762,\nCNS 0434533, ANI 0073736, EIA 0130806, and a grant from Motorola Inc.\n7.\nREFERENCES\n1. Y. Zhang, W. Lee, and Y. Huang, “Intrusion Detection Techniques for Mobile Wireless Networks,”\nACM/Kluwer Wireless Networks Journal (ACM WINET), Vol. 9, No. 5, September 2003.\n2. A. Mishra, K. Nadkarni, and A. Patcha, “Intrusion Detection in Wireless Ad Hoc Networks,” IEEE\nWireless Communications, Vol. 11, Issue 1, pp. 48-60, February 2004.\n3. P. Albers, O. Camp, J. Percher, B. Jouga, L. Mé, and R. Puttini, “Security in Ad Hoc Networks: a\nGeneral Intrusion Detection Architecture Enhancing Trust Based Approaches,” Proceedings of the 1st\nInternational Workshop on Wireless Information Systems (WIS-2002), pp. 1-12, April 2002.\n4. O. Kachirski and R. Guha, “Effective Intrusion Detection Using Multiple Sensors in Wireless Ad\nHoc Networks,” Proceedings of the 36th Annual Hawaii International Conference on System Sciences\n(HICSS’03), p. 57.1, January 2003.\n"
        },
        {
            "page_number": 185,
            "text": "WIRELESS NETWORK SECURITY\n179\n5. S. Marti, T. J. Giuli, K. Lai, and M. Baker, “Mitigating Routing Misbehavior in Mobile Ad Hoc Net-\nworks,” Proceedings of the 6th Annual International Conference on Mobile Computing and Networking\n(MobiCom’00), pp. 255-265, August 2000.\n6. S. Buchegger and J. Le Boudec, “Performance Analysis of the CONFIDANT Protocol (Cooperation\nOf Nodes - Fairness In Dynamic Ad-hoc NeTworks),” Proceedings of the 3rd ACM International\nSymposium on Mobile Ad Hoc Networking and Computing (MobiHoc’02), pp. 226-336, June 2002.\n7. P. Michiardi and R. Molva, “Core: A Collaborative Reputation mechanism to enforce node coopera-\ntion in Mobile Ad Hoc Networks,” Communication and Multimedia Security Conference (CMS’02),\nSeptember 2002.\n8. D. B. Johnson, and D. A. Maltz, “The Dynamic Source Routing Protocol for Mobile Ad Hoc Networks\n(Internet-Draft),” Mobile Ad-hoc Network (MANET) Working Group, IETF, October 1999.\n9. P. Brutch and C. Ko, “Challenges in Intrusion Detection for Wireless Ad-hoc Networks,” Proceedings\nof 2003 Symposium on Applications and the Internet Workshop, pp. 368-373, January 2003.\n10. M. G. Zapata, “Secure Ad Hoc On-Demand Distance Vector (SAODV) Routing,” ACM Mobile Com-\nputing and Communication Review (MC2R), Vol. 6, No. 3, pp. 106-107, July 2002.\n11. Y. Hu, D. B. Johnson, and A. Perrig, “SEAD: Secure Efficient Distance Vector Routing for Mobile\nWireless Ad Hoc Networks,” Proceedings of the 4th IEEE Workshop on Mobile Computing Systems\nand Applications (WMCSA’02), pp. 3-13, June 2002.\n12. Y. Hu, A. Perrig, and D. B. Johnson, “Ariadne: A secure On-Demand Routing Protocol for Ad hoc Net-\nworks,” Proceedings of the 8th Annual International Conference on Mobile Computing and Networking\n(MobiCom’02), pp. 12-23, September 2002.\n13. A. Perrig, R. Canetti, D. Tygar and D. Song, “The TESLA Broadcast Authentication Protocol,” RSA\nCryptoBytes, 5 (Summer), 2002.\n14. Y. Huang and W. Lee, “A Cooperative Intrusion Detection System for Ad Hoc Networks,” Proceedings\nof the ACM Workshop on Security in Ad Hoc and Sensor Networks (SASN’03), pp. 135-147, October\n2003.\n15. Y. Huang, W. Fan, W. Lee, and P. Yu, “Cross-Feature Analysis for Detecting Ad-Hoc Routing Anom-\nalies,” Proceedings of the 23rd IEEE International Conference on Distributed Computing Systems\n(ICDCS’03), May 2003.\n16. D. Sterne, P. Balasubramanyam, D. Carman, B. Wilson, R. Talpade, C. Ko, R. Balupari, C.-Y. Tseng, T.\nBowen, K. Levitt, and J. Rowe, “A General Cooperative Intrusion Detection Architecture for MANETs,”\nProceedings of the 3rd IEEE International Workshop on Information Assurance (IWIA’05), pp. 57-70,\nMarch 2005.\n17. Y. F. Jou, F. Gong, C. Sargor, X. Wu, S. Wu, H. Chang, and F. Wang, “Design and Implementation of\na Scalable Intrusion Detection System for the Protection of Networks Infrastructure,” Proceedings of\nDARPA Information Survivability Conference and Exposition, Vol. 2, pp. 69-83, January 2000.\n18. E. Y. K. Chan et al., “IDR: An Intrusion Detection Router for Defending against Distributed Denial-of-\nService (DDoS) Attacks,” Proceedings of the 7th International Symposium on Parallel Architectures,\nAlgorithms and Networks (ISPAN’04), pp. 581-586, May 2004.\n19. S. Bansal and M. Baker, “Observation-Based Cooperation Enforcement in Ad hoc Networks,” Research\nReport cs.NI/0307012, Stanford University, 2003.\n20. P. Michiardi and R. Molva, “A Game Theoretical Approach to Evaluate Cooperation Enforcement\nMechanisms in Mobile Ad Hoc Networks,” Modeling and Optimization in Mobile, Ad Hoc and Wireless\nNetworks (WiOpt’03), March 2003.\n21. A.Agah, S. K. Das, K. Basu, and M. Asadi, “Intrusion Detection in Sensor Networks: A Non-\nCooperative Game Approach,” Proceedings of the 3rd IEEE International Symposium on Network\nComputing and Applications (NCA’04), pp. 343-346, 2004.\n"
        },
        {
            "page_number": 186,
            "text": "180\nTIRANUCH ANANTVALEE and JIE WU\n22. R. Mahajan, M. Rodrig, D. Wetherall and J. Zahorjan, “Experiences Applying Game Theory to System\nDesign,” Proceedings of the ACM SIGCOMM Workshop on Practice and Theory of Incentives in\nNetworked Systems (PIN’04), pp. 183-190, September 2004.\n23. S. Zhong, L. Li, Y. G. Liu and Y. Yang, “On Designing Incentive-Compatible Routing and Forwarding\nProtocols in Wireless Ad-hoc Networks: An Integrated Approach Using Game Theoretical and Crypto-\ngraphic Techniques,” Proceedings of the 11th Annual International Conference on Mobile Computing\nand Networking (MobiCom’05), pp. 117-131, 2005.\n24. B. Sun, K. Wu, and U. W. Pooch, “Alert Aggregation in Mobile Ad Hoc Networks,” Proceedings of the\n2003 ACM Workshop on Wireless Security (WiSe’03) in conjuction with the 9th Annual International\nConference on Mobile Computing and Networking (MobiCom’03), pp. 69-78, 2003.\n"
        },
        {
            "page_number": 187,
            "text": "Part II\nSECURITY IN\nMOBILE CELLULAR NETWORKS\n"
        },
        {
            "page_number": 188,
            "text": "8\nINTRUSION DETECTION IN\nCELLULAR MOBILE NETWORKS\nBo Sun\nComputer Science Department\nLamar University\nBeaumont, TX 77710, USA\nE-mail: bsun@cs.lamar.edu\nYang Xiao\nComputer Science Department\nUniversity of Alabama\n101 Houser Hall\nBox 870290\nTuscaloosa, AL 35487-0290 USA\nE-mail: yangxiao@ieee.org\nKui Wu\nComputer Science Department\nUniversity of Victoria\nBC, Canada V8W 3P6\nE-mail: wkui@cs.uvic.ca\nSecurity concerns have attracted a great deal of attentions for both service providers and\nend users in cellular mobile networks. As a second line of defense, Intrusion Detection\nSystems (IDSs) are indispensable for highly secure wireless networks. In this chapter,\nwe first give a brief introduction to wired IDSs and wireless IDSs. Then we address the\nmain challenges in designing IDSs for cellular mobile networks, including the topics of\nfeature selection, detection techniques, and adaptability of IDSs. An anomaly-based IDS\nexploiting mobile users’ location history is introduced to provide insights into the intricacy\nof building a concrete IDS for cellular mobile networks.\n1.\nINTRODUCTION\nThe rapid development of cellular mobile services makes people rely heavily on\ncellular phones in their daily lives for important and sensitive tasks. While providing a\n"
        },
        {
            "page_number": 189,
            "text": "184\nBO SUN et al.\ngreat convenience, these booming new services have brought serious security concerns.\nThe lack of security has become one of the main obstacles in preventing wireless com-\nmunications carriers from providing business such as E-Banking or E-Shopping over\nwireless networks on a large scale basis. Although there are many security mechanisms\nin cellular mobile networks [36], the number of security incidents continues to increase.\nHow to design a highly secure cellular mobile network is still a very challenging issue\ndue to the open radio transmission environment and the physical vulnerability of mobile\ndevices.\nGenerally, two complementary classes of approaches exist to protect the cellu-\nlar mobile networks: prevention-based approaches and detection-based approaches.\nPrevention-based techniques, such as authentication and encryption, can effectively\nreduce attacks by ensuring that users conform to predefined security policies. They\ncan keep most illegitimate users from entering the system. However, security research\nindicates that there are always some weak points in the system that is hard to pre-\ndict, especially for a wireless network, in which open wireless transmission medium\nand low physical security protection of mobile devices pose additional challenges for\nprevention-based approaches. For example, although numerous security measures are\ntaken into account in the design of second-generation and third-generation digital cel-\nlular systems, security flaws have been reported in literature [1] [2]. Security research\nindicates the necessity of multi-layer and multi-level protection because there are al-\nways some weak points in the system that attackers can exploit to break into the system.\nCurrently, tamper-resistant hardware and software are still expensive or unrealistic for\nmobile devices. Therefore, if a device is compromised, all the secrets associated with\nthe device become open to the attackers, rendering all prevention-based techniques\nhelpless and resulting in great damage to service providers. For example, one of the\nbasic threats is the illegitimate use of services, which leads to the serious problem of\nimproper billing and masquerading. To solve these problems, Intrusion Detection Sys-\ntems (IDSs), serving as the second wall of protection, could effectively help identifying\nmalicious activities.\nAlthough IDSs have been widely used in wired networks, not many research ef-\nforts have been dedicated to IDSs in cellular mobile networks. The communications\nparadigm in cellular mobile networks and traditional wired networks are fundamen-\ntally different. This makes attack scenarios in cellular mobile networks more complex.\nMoreover, it is challenging to model the normal and abnormal user behaviors because\nof the potential wide variety of users’ activities. Feature selection, detection tech-\nniques, and adaptability of IDSs in the context of cellular mobile networks are still\nopen research problems.\nIn this chapter, we provide a general introduction to IDSs in cellular mobile net-\nworks. Section 2 presents necessary background knowledge of cellular mobile net-\nworks. Section 3 focuses on the introduction of Intrusion Detection Systems, including\nIDSs for wired networks and IDSs for cellular mobile networks, respectively. Section\n4 addresses one important and challenging topic regarding IDSs - Feature Selection.\nIn Section 5, we discuss the adaptability issue of IDSs. Secion 6 presents the details of\n"
        },
        {
            "page_number": 190,
            "text": "WIRELESS NETWORK SECURITY\n185\nconstructing a mobility-based anomaly detection system for cellular mobile networks.\nIn Section 7, we conclude the chapter.\n2.\nCELLULAR MOBILE NETWORKS\nA mobile wireless network with a cellular infrastructure is illustrated in Figure 1.\nA typical network consists of a wired backbone and a number of Base Stations (BSs).\nEach BS controls a cell, and a group of BSs are managed by a Mobile Switching Center\n(MSC). When a mobile user moves into a different Location Area (LA), a location\nregistration process happens. In cellular mobile networks, the Home Location Register\n(HLR) is a database used for storage and management of subscriptions. Usually, the\nHLR stores the permanent data about subscribers, while the Visitor Location Register\n(VLR) stores temporary information to serve visiting subscribers.\nMSC/\nVLR\nMSC/\nVLR\nHLR\nCell\nBS\nSignaling Network\nFigure 1. An Example of Cellular Mobile Network.\nA mobile station communicates with another mobile station via a BS. To do so, the\nsource mobile station needs to make a request through the BS of its current cell. If the\nrequest is granted by the MSC, a pair of voice channels is assigned. In cellular mobile\nnetworks, location updates often happen when the user traverses the border of an LA.\nWhen the user is inside an LA and is not making a phone call, the Mobile Switching\nCenter, which is responsible for location and paging management, is not updated with\nthe user’s latest location information.\n"
        },
        {
            "page_number": 191,
            "text": "186\nBO SUN et al.\n3.\nINTRUSION DETECTION SYSTEMS\nIntrusions can be defined as any set of actions that compromise the confidentiality,\navailability, and integrity of the system. Intrusion detection is a security technology\nthat attempts to identify individuals who are trying to break into and misuse a system\nwithout authorization and those who have legitimate access to the system but are abusing\ntheir privileges [3], [4]. An Intrusion Detection System (IDS) is a computer system\nthat dynamically monitors the system and users’ actions in order to detect intrusions.\nBecause an information system can suffer from various kinds of security vulnerabilities,\nit is both technically difficult and economically costly to build and maintain a system\nthat is not susceptible to attacks. IDSs, by analyzing the system and user operations\nin search of activity undesirable and suspicious, can effectively monitor and protect\nagainst threats.\nResearch on IDSs began with a report by Anderson [5] followed by Denning¡¯s\nseminal paper [6], which lays the foundation for most of the current intrusion detection\nprototypes. Since then, many research efforts have been devoted to wired IDSs. Nu-\nmerous detection techniques and architectures for host machines and wired networks\nhave been proposed. A good taxonomy of wired IDSs is presented in [4].\nWith the rapid proliferation of wireless networks and mobile computing applica-\ntions, new vulnerabilities that do not exist in wired networks have appeared. Security\nposes a serious challenge in deploying wireless networks in reality. Moreover, the\nvast differences between wired and wireless networks make traditional intrusion detec-\ntion techniques inapplicable. Wireless IDSs, emerging as a new research topic, aim at\ndeveloping new architecture and mechanisms to protect wireless networks.\n3.1. Intrusion Detection for Wired Networks\nFocusing mainly on network traffic and computer audit data, there are two general\napproaches in wired IDSs to detect intrusions: misuse-based intrusion detection (also\nreferred to as knowledge-based detection, or detection by appearance) and anomaly-\nbased intrusion detection (also referred to as behavior-based detection or detection by\nbehavior). They are complementary to each other for intrusion detection.\nMisuse-Based Intrusion Detection Systems\nBased on a database of known attack signatures and system vulnerabilities, misuse-\nbased IDSs try to identify activities matching a signature that is stored in the database.\nAn alarm is triggered whenever a match is found. The main advantage of misuse-based\nIDSs is that the false alarm rate is very low. The triggered alarms are meaningful\nbecause the attack signatures contain the diagnostic information about the cause of the\nalarm. The main disadvantage of misuse-based IDSs is that the attack signatures may\nnot cover all attacks because new attacks are hard to predict. As such, the databases\ncontaining the attack signatures and system vulnerabilities need to be kept up-to-date.\nThis is a tedious task because new attacks and system vulnerabilities are detected on\n"
        },
        {
            "page_number": 192,
            "text": "WIRELESS NETWORK SECURITY\n187\na daily basis. Careful analysis of the vulnerabilities is also time-consuming. Misuse-\nbased IDSs also face the generalization issues because most of the attack knowledge is\nfocused on the different versions of operating systems and applications.\nThere are several approaches in misuse-based detection. They differ in the rep-\nresentation as well as the matching algorithm employed to detect intrusion patterns.\nBelow are the mainly used approaches:\nExpert System: Expert systems provide strategies and mechanisms for process-\ningfactsregardingthestateofagivenenvironment, andderivelogicalinferences\nfrom these facts. Audit events and security policies are mapped to facts that\nare recorded and evaluated by the system. During the process of mapping, a\nsemantic meaning is attached to increase the abstraction level of the audit data.\nThe expert system contains a set of rules that describe the attacks. These rules\nare triggered when certain activities that meet their conditions happen. The\nexecution speed of the expert system shell is usually poor because all of the\naudit data need to import into the shell as facts. Therefore, expert system based\nIDSs only exist in research prototypes, as performance is more important in\ncommercial products.\nEventMonitoringEnablingResponsestoAnomalousLiveDisturbances(EMER-\nALD) [7] is an extension of the Intrusion Detection Expert System (IDES)\n[8],[9] and Next Generation Intrusion Detection System (NIDES) [10] by SRI\nInternational.\nEMERALD uses a rule-based expert system component for\nmisuse-based detection. A forward-chaining rule-based expert system devel-\nopment toolset called the Production Based Expert System Toolset (P-BEST)\n[11] is utilized to develop a modern generic signature-analysis engine. A chain\nof rules is established utilizing P-BEST to form the signature database.\nPattern Recognition: In this approach, known intrusion signatures are encoded\nas patterns (e.g., strings, a sequence of events, etc.) and matched against audit\ndata. An alarm is generated if a match can be found. This method allows a very\nefficient implementation. Therefore, they are commonly used in commercial\ntools, such as RealSecure of Internet Security Systems [12].\nColored Petri Nets: In this method, signatures of known intrusions are mod-\neled as a number of different states, which form Colored Petri Nets (CPNs).\nCompared with other approaches, CPNs have more generalities to represent\nsignatures. This makes it easy to write complex intrusion scenarios. However,\nit is very computationally expensive to manifest misbehaviors in the audit trail.\nIntrusion Detection In Our Time (IDIOT) is the one example that uses CPNs\n[13].\nState Transition Analysis: In this approach, to represent an intrusion scenario,\na sequence of actions is constructed starting from the initial state to the tar-\nget compromised state. State Transition Diagrams identify the steps and the\nrequirements of the penetration. The states that make up the intrusion form a\n"
        },
        {
            "page_number": 193,
            "text": "188\nBO SUN et al.\nsimple chain that has to be traversed from the beginning to the end. It was a\ntechnique proposed by Porras and Kemmerer [14], which was implemented in\nUstat - a real-time intrusion detection system for UNIX [15].\nAnomaly-Based Intrusion Detection Systems\nAnomaly-based IDSs assume that an intrusion can be detected by observing a\ndeviation from normal or expected behaviors of systems or users. Normalcy is defined\nby the previously observed subject behavior, which is usually created during a training\nphase. The normal profile is later compared with the current activity. If a significant\ndeviation is observed, IDSs flag the unusual activity and generate an alarm. The main\nadvantage of anomaly-based IDSs is that they can detect attempts that try to exploit new\nand unforeseen vulnerabilities. They are also less system-dependent. Disadvantages\ninclude that they may have very high false alarm rate and are more difficult to configure\nbecausecomprehensiveknowledgeofexpectedsystembehaviorsisrequired. Inorderto\nbuild the up-to-date normal profiles, they also usually require a periodic online learning\nprocess. Anomaly-based detection techniques are harder to implement, making them\ninappropriate for commercial use.\nSeveral anomaly-based detection techniques exist. They are different in the way\nof representing a normal profile and the method of inferring the difference between the\nnormal profile and the observed activities. Below are the mainly used approaches:\nStatistics: Statistics-based anomaly detection techniques build a statistical pro-\nfile (e.g., statistical distribution) of normal activities from historic data by mea-\nsuring a number of variables over time. Examples of the variables are the\nlogin/logoff times, the time duration of one session, the number of packets\ntransmitted in this session, and so on.\nIn EMERALD [7], the statistical algorithms employ four classes of measures\nto track subject activities: categorical, continuous, intensity, and event distribu-\ntion. The profile is subdivided into short and long-term elements. A short-term\nprofile may characterize recent activities of the system, while a long-term profile\nis slowly adapted to the changes of system activities. Because of the popularity\nof the Internet, many traffic perspectives are used to profile TCP/IP streams\n[7]. For example, all ICMP exchanges can be parsed to analyze ICMP-specific\ntransactions. The application-layer sessions from specific internal hosts to spe-\ncific external hosts can also be analyzed for specific applications.\nNeural Networks: The use of neural networks in IDSs consists of three steps:\nlearning the normal pattern of the system by collecting training data, training\nthe neural networks to identify the subject, and applying the output of the\nneural networks to the observed activity to identify intrusions. Neural networks\nare computationally intensive. Therefore, they are not widely used in IDSs.\nHyperview [16] is an example IDS that utilizes neural networks.\n"
        },
        {
            "page_number": 194,
            "text": "WIRELESS NETWORK SECURITY\n189\nThere are some other anomaly-based detection techniques. Detection techniques\nbased on immunology [17] first capture a large set of event sequences from historic\ndata to construct the normal profile. They then use either negative selection or positive\nselection algorithms to detect the difference of incoming event sequences from event\nsequences in the normal profile [18]. Expert systems can also be used to implement\nanomaly-based techniques [9]. To describe normal behaviors, these expert systems\ncan study the activities of the target system to form a set of rules. Lee et al. proposed\nto use data mining approach to construct intrusion detection models [19]. Anomaly-\nbased detection techniques utilizing Chi-square Test are introduced in [20] and [21].\nThere are also anomaly-based detection techniques that use a first-order or high-order\nMarkov model of event transitions to represent a normal profile [22],[23],[24],[25].\nIn [22], utilizing a Markov Chain model, Jha et al. proposed a general framework to\nconstruct anomaly detectors.\nBesides misuse-based detection and anomaly-based detection, there is a new class\nof detection algorithms: specification-based techniques [27]. They combine the advan-\ntages of both misuse-based detection and anomaly-based detection techniques. These\napproaches are based on manually developed specifications, thus avoiding the high rate\nof false alarms. IDSs detect deviations of observed program behaviors from these spec-\nifications, rather than detect the occurrence of specific attack patterns. Thus, attacks\ncan be detected even though they have not previously been encountered.\n3.2. Intrusion Detection for Cellular Mobile Networks\nMost of the proposed work in the areas of wireless IDSs explores the regularity of\nusers’ behaviors (for example, mobility patterns, calling activities) to construct normal\nprofiles. Regularity is one of the basic assumptions to develop realistic IDSs. For\nexample, in terms of mobility patterns, a mobile user usually travels with a specific\ndestination in mind and tends to follow the shortest path to it. A user’s mobility pattern\nis a reflection of his/her daily routines and most mobile users have favorite routes and\nhabitual movement patterns. In terms of calling activities, most mobile users have\nhis/her regular calling activities. For example, because of the regular working rhythms\nlike daily or weekly business telephone conference, most users demonstrate certain\ncalling patterns. Although an attacker can compromise all the secrets associated with\na mobile device, he/she could not follow the movement pattern of the authentic owner\nand mimic the authentic user’s profile. By establishing an accurate normal profile\nthat can reflect the normal pattern and comparing it with the current observed pattern,\nmisbehaviors can be effectively identified.\nRelatively few research efforts have been devoted to Intrusion Detection for Cel-\nlular Mobile Networks. B¨uschkes et al. [28] applied the Bayes decision rule to user’s\nmobility patterns to increase the security in mobile networks. Through proper behavior\npredictions, they applied anomaly-based detection techniques to profile mobile users.\nSamfat et al. [29] proposed IDAMN (Intrusion Detection Architecture for Mobile Net-\nworks) that included two algorithms to model the behavior of users in terms of both\ntelephony activity and migration patterns. IDAMN can perform intrusion detection in\n"
        },
        {
            "page_number": 195,
            "text": "190\nBO SUN et al.\nthe visited location and within the duration of a typical call. Y. -B Lin [1] presented\nan excellent study to detect the potential fraudulent usage of cloned phones in cellular\nmobile networks. They showed how quickly the fraudulent usage can be detected under\nGSM/UMTS call setup procedures and how to reduce the possibility of fraudulent us-\nage. Exploring mobility patterns of public transportation users, Hall et al. [30] utilized\nan Instance based Learning technique to classify different users’ behaviors. There are\nalso some research efforts dedicated to fraud detection systems in cellular mobile net-\nworks. Hollm´en [31] presented fraud detection techniques in mobile communications\nnetworks by means of user profiling and classification. Call data is used to describe\nbehavioral patterns of mobile users. Neural networks and probabilistic models were\nemployed to learn their usage patterns. Based on these models, abrupt changes from\nestablished usage patterns can be detected.\nIt is worth mentioning that some of the above mentioned schemes require the track-\ning of uses’ locations. This will cause location privacy issues because of the potential\nexposure of users’ whereabouts. Fortunately, there is some work in the literature that\nare aimed to address the privacy issues. For example, He et al. [34] proposed to\nuse blind signature to generate an authorized-anonymous-ID for the server to autho-\nrize the mobile device. Location-based IDSs should be properly integrated with these\nprivacy-enhanced schemes in order to be readily deployed.\n4.\nFEATURE SELECTION\nOne of the most important steps in constructing intrusion detection systems is to\nextract effective features. Features are security related measures that could be used to\nconstruct suitable detection algorithms. Desirable features must be selected to reflect\nthe subject activities. Feature selection plays such a critical role in constructing effective\nfeatures that its importance cannot be overemphasized.\nEach intrusion detection approach is technically suited to identify a subset of secu-\nrity violations to which the system is subject. The selection of security measures should\nbe based on good understanding about the system itself as well as all possible attacks\nthat may influence the system’s normal behaviors. Different attacks may be sensitive\nto different statistical features. Sometimes it requires domain expert knowledge to help\nselecting good features. In the history of IDSs, people have used various features to\nconstruct detection models. They tend to define the normal behavior of a user, a pro-\ngram, or a network element. Since the ground-breaking discovery of S. Forrect [32],\npeople find that the short sequence of system calls of privileged programs is stable in\ncharacterizing system’s behaviors. Therefore, many research efforts have focused on\nconstructing different detection models using the short sequence of system calls since\nthen.\nAlthough there are some theoretical guidelines in optimal feature selection [33],\nit is still challenging to apply them in practice. In [26], Lee et al. utilized data mining\nalgorithms to compute activity patterns from system audit data and extract temporal\nand statistical features from the pattern. They identified intrusion-only patterns from\n"
        },
        {
            "page_number": 196,
            "text": "WIRELESS NETWORK SECURITY\n191\ntraining data (a set of network connection records) and parsed these patterns to define\nfeatures accordingly. Experiments based on test data were also needed to tell whether\nthe selected features can be used to distinguish normal and abnormal activities. This\nprocess was repeated until a satisfactory set of features can be selected.\nToday, features used in most anomaly-based IDSs are still selected empirically. It\nremains an open problem to decide the right set of features to construct IDSs in the\ncontext of cellular mobile networks. Some example features used include call times\nand duration, roaming behavior, location coordinates, the list of traversed cells, and\nso on.\n5.\nADAPTABILITY OF IDSS\nIt is necessary to integrate adaptability into the construction of IDSs. In reality,\nit is highly possible that a single user will demonstrate different mobility behaviors.\nEven if the user demonstrates the same mobility level, a user will have a set of mobil-\nity patterns during weekdays, while demonstrating a different set of mobility patterns\nduring weekends. Therefore, established users’ normal profiles need to be changed\nadaptively in order to reflect users’ activities more accurately. Moreover, in construct-\ning an anomaly-based IDS, a threshold-based scheme is often used. That is, the distance\nbetween observed activities and established normal profiles is compared with a thresh-\nold in order to decide whether the system needs to generate an alarm or not. It is also\nnecessary to adjust the threshold adaptively in order to achieve desirable performance.\nHowever, how to adaptively adjust the normal profile and the threshold of IDSs\nin the context of cellular mobile networks is a very challenging problem. Special\nmechanisms need to integrate with existing detection techniques to achieve adaptability.\nFor example, an individual subject’s activity may change over time. Therefore, it is\nnecessary for the normal profile to be updated in order to reflect the recent activities.\nExponentially Weighted Moving Average (EWMA) techniques [35] provide a suitable\nway to make activities in the recent past weigh more than activities long time ago. In\nthis way, normal profiles can be adjusted accordingly. To adjust the threshold, usually\nan effective metric is needed to reflect the uncertainty of established normal profiles.\nEntropy may be a good choice here. We will see a more detailed example illustrating\nthe integration of adaptability in Section 6.\n6.\nCASE STUDY: AN EXEMPLARY IDS FOR CELLULAR\nMOBILE NETWORKS\n6.1. Introduction\nIt is very difficult to design a once-for-all Intrusion Detection System for cellular\nmobile networks. Instead, an incrementally refined methodology is suitable. In this\nsection, we introduce an exemplary IDS for cellular mobile networks [36],[37] that\nfocuses on the exploitation of users’ mobility patterns. Other important features like\ncalling activities need to be integrated into the system to provide more comprehensive\n"
        },
        {
            "page_number": 197,
            "text": "192\nBO SUN et al.\nprotections. In the sequel, we introduce system assumptions, models (threat model,\nnetwork model, and mobility model), and detailed detection techniques.\n6.2. Assumptions\nFirst, we assume that most mobile users have favorite or regular itineraries. This\nmakes it viable for us to establish each user¡¯s normal profile. This assumption is\nreasonable given that most users have regular daily lives. Studies in [38] conducted\nexperiments over a period of six weeks to study the trajectories that users follow, and\nfound out that users tend to follow regular trajectories more than 70% of time.\nActually, research on intrusion detection has two basic assumptions: 1) subject\nactivities are observable via some system auditing mechanisms, and 2) normal and\nmalicious activities should demonstrate distinct behaviors. Therefore, it is possible\nto reason about the evidence in the data to determine whether the system is currently\nunder attack. If a user has totally random behavior, for example, the movement of a\ntaxi driver, it will be very difficult, if not impossible, to create his normal movement\nprofile. Our mobility-based detection algorithm alone is not suitable for such kind of\nusers. Based on these considerations, our research is not motivated to build a system\nto accurately detect all intrusions. Instead, we aim at providing an optional service to\nend users as well as a useful administration tool to service providers. If the system\nobserves some abnormal behaviors, other channels (e.g., email, phone calls to home)\ncan be used to issue some warnings to the real users. Given the increasing number\nof security related incidents in wireless networks, these kinds of optional services can\nprotect both the service providers and the end users from financial losses.\nSecond, we assume that there is a mobility database for each mobile user that\ndescribes his normal activities. This is a reasonable assumption in cellular mobile\nnetworks because this mobility database could be constructed by location tracking and\nprediction services. This mobility database could be stored together with the mobile\nuser¡¯spersonalinformation, suchasbillinginformation, intheHomeLocationRegister\n(HLR). Note that in realistic networks, the locations of mobile users are actually tracked\nfor the purpose of service provision and smooth handoff, even though the end users\nmay be unaware of such monitoring. We assume that HLR is secure and the movement\ninformation is accurate. Usually, because of its importance, HLR is protected with\nhighly secure measures, and thus it is extremely hard to be compromised. Also, the\nupdate and registration of the location are usually based on the device¡¯s current serving\ncell and the hardware registration such as the serial number of SIM card. Therefore, it\nwill be hard for the attacker to hide or fabricate his location even if he has compromised\nall the secrets of the mobile device. Even if an attacker finds some magical ways to\nfabricate his location, he still has no idea about the normal movement profile of the real\ndevice owner.\nThird, we assume that mobile devices can be compromised and all secrets asso-\nciated with the compromised devices are open to attackers. Under this assumption,\nwe do not need to assume or apply tamper-resistant hardware and software, which\nare still costly and impractical to handheld devices. This assumption justifies our re-\n"
        },
        {
            "page_number": 198,
            "text": "WIRELESS NETWORK SECURITY\n193\nsearch in anomaly detection, since all prevention techniques will be rendered helpless\nonce the mobile device is captured and compromised. Actually, if we could assume\nthe tamper-resistance of hardware/software, the whole security research could become\nmuch easier.\n6.3. Model\nThreat Model\nThe complex wireless mobile network system could incur software errors and\ndesign errors. This could make many attacks possible. One exemplary attack is cell\nphone cloning: a mobile phone card of an authenticate user A is cloned by an attacker\nB, which enables B to use the cloned phone card to make fraudulent telephone calls.\nIf this kind of illegitimate use of service happens, the bills for the calls will go to\nthe legitimate subscriber. Also, the masquerader can fake the International Mobile\nEquipment Identifier (IMEI) and the SIM (Subscriber Identity Module) card in order\nto obtain the service illegally. In the subscription fraud, fraudsters can also subscribe\nthe service using the authentic user’s name and obtain an account without intention to\npay the bill. Our presented IDS can enhance system security to defend against these\nkinds of attacks.\nNetwork Model\nDifferent ways exist to model cellular mobile networks. For example, most previ-\nous work uses structured graph network topology models, such as hexagonal or square\ncell configurations. One disadvantage of this model is that it does not accurately repre-\nsent a cellular network in practice, where the cell shape and size may vary depending\non the antenna radiation pattern and propagation environment. In wireless cellular\nnetworks, each cell usually has a base station to serve it. Therefore, in our system, the\nwireless cellular network is modeled as a generalized graph G = (V,E). The vertex set V\nrepresents all the base stations. If two cells are adjacent to each other, there is an edge\nbetween their two vertices. An example of this model is illustrated in Figure 2.(a) and\nFigure 2.(b). In this example, the vertex set is V = {a, b, c, d, e, f, g, h, i, j}, and the\nedge set is E = {(a, b), (a, c), ...(h, i)}.\nThere may exist other ways to model the networks in order to facilitate the intrusion\ndetectiontasks. Forexample, consideringthefactthatamobileuserusuallydrivesalong\na road, cell-based models may not precisely locate a mobile user or model the trajectory\nof a user because they do not support fine granularity of the road network [39].\nUsually, each user will follow a specific road for daily activities. Most users will\nfollow the speed limit sign when driving. Also, each user has his own habit of traveling\nspeed. Therefore, for a specific path, a user will take roughly the same amount of time\nto travel (if we do not consider the possible traffic jam). In reality, there exist a road\nnetwork and the road network is overlapped with the Location Area, which consists of\nseveral cells. Considering all these factors, a network model as illustrated in Figure 3\ncould be adopted.\n"
        },
        {
            "page_number": 199,
            "text": "194\nBO SUN et al.\nb\na\nc\nd\ne\nf\ng\nh\ni\nj\na\nb\nc\nd\nf\ng\ne\nj\ni\nh\na) An Example of Cellular \nMobile Network with Cells\nb) The Graph Model of the \nCellular Mobile Network\nFigure 2. An Example Cellular Mobile Network and its Graph Model.\na) An Example of Location \nArea in Cellular Mobile \nNetworks\nb) The Graph Model of the \nLocation Area\ncell\nroad\na\nb\nc\nd\na\nb\nc\nd\nFigure 3. An Example Location Area in Cellular Mobile Networks and its Graph Model.\nFigure 3.(a) illustrates the network topology in one LA, which consists of 10 cells.\nIn Figure 3.(a), bold lines represent the road network. a, b, c, and d represent the\nintersection points of the road network and the boundary of the LA. For current mobile\nsystems, location updates happen when the user enters or leaves one LA. This is the one\nof the most common ways to track the cellular mobile phones. This is true whenever a\nuser is making a phone call or not. Considering this, we could adopt the corresponding\nnetwork model as in Figure 3.(b).\nIn Figure 3.(b), each intersection between the road network and the LA is modeled\nas a vertex. In our example, we have four vertices, a, b, c, and d. These vertices form\na fully connected graph, meaning that there is one path between any two vertices. In\nthis way, we can ignore the complex internal road network inside one LA.\n"
        },
        {
            "page_number": 200,
            "text": "WIRELESS NETWORK SECURITY\n195\nIt is possible that in one LA, there are more than one possible path connecting two\nvertices. We assume that in one LA, one user prefers one specific path. This means\nthat in Figure 3.(b), for a specific user, it will take him roughly the same amount of time\nto travel between any two vertices. If one user has variations in his traveling habit, i.e.,\nif he takes two different paths between the same two vertices, we can have two entries\nfor these two vertices in the user¡¯s mobility profile.\nBy integrating the current mechanisms that mobile networks use to track user¡¯s\nlocation information, this network model is more accurate than the model only consid-\nering the cell list traversed by each user. Furthermore, most routes have a speed limit\nand most users have a driving habit. For example, some users want to strictly follow the\nspeed limit, while some others want to drive 10 miles/hour faster. This will cause the\ndifferent time used by different users to traverse a specific route (edge). This network\nmodel is also more realistic because it ignores the potential different routes between\ntwo vertices. Therefore, it is more suitable for intrusion detection systems.\nDifferent network models can be abstracted into different graphs. The vertices of\nthe graph can be treated as the feature to construct different intrusion detection systems.\nIn the following, we only use the cell list traversed by the user as the feature to illustrate\na detailed detection technique. In this way, we denote each cell as a character. A string\ncan be used to denote the cell list traversed by the user.\nMobility Model\nThe random walk model has been widely used in the literature, in which a mobile\nuser will move to any one of the neighboring cells with equal probability after leaving a\ncell. This may not be realistic in practice, since mobile users normally travel with a des-\ntination in mind. Therefore, we adopt a m-th order Markov model. In such a model, the\nmobility of a user can be represented by a sequence of characters, C1, C2, C3, ..., Ci, ...,\nwhere Ci denotes the identity of the cell visited by the mobile. Since the future locations\nof the mobile user are likely to be correlated with its movement history, the sequence of\ncharacters C1, C2, C3, ..., Ci, ... is assumed to be generated by an m-th order Markov\nsource, where the states correspond to the context of the previous m characters. The\nprobability that the user moves to a particular cell depends on the location of the current\ncell and a list of cells recently visited.\n6.4. Mobility-Based Anomaly Detection Systems\nIn this section, we present two mobility-based anomaly detection schemes called\nLZ-based scheme and Markov-based scheme.\nFigure 4 illustrates the LZ-based detection scheme. In the LZ-based detection\nscheme, based on users¡¯ regular itineraries, a mobility trie is constructed from the\naccumulative history of users¡¯ movement patterns. To integrate adaptability , the Ex-\nponential Weighted Moving Average (EWMA) [35] technique is applied to the mobility\ntrie. This EWMA-based mobility trie serves as the normal profile of the user in the\nrecent past, and reflects the stationary part of the user¡¯s regular mobility pattern. Based\n"
        },
        {
            "page_number": 201,
            "text": "196\nBO SUN et al.\ncell string: aabbabcccabaaba\nm-th Markov model\nFeature Extraction based on \nNetwork Models \nMobility Trie Construction\nEWMA\nNormal Profile: \nEWMA_based Mobility Trie\nCompute\nDistance\nGenerate \nAlert or not\nUser Mobility \nActivity\nFigure 4. LZ-based Anomaly Detection Scheme.\non this, we use a blending scheme to calculate the probability of each user¡¯s activity\nin order to decide whether it is normal or not.\nThe second scheme, Markov-based scheme, is based on order-o Markov predictors.\nThat is, given an order o, the probability of being the next cell given the previous o cells\nis constructed. In other words, the probability of the future activity can be calculated.\nBoth the LZ-based and the Markov-based schemes are online predictors, meaning\nthat they examine the history so far, extract the current context, and predict the next cell\nlocation. Once the next location is known, the history is appended with one character\n(standing for one cell), and the predictor updates its history to prepare for the next\nprediction.\nIn the LZ-based scheme, we adopt Lempel-Ziv algorithms [40] [41]. In the rest of\nthe chapter, when we discuss these algorithms, we use the word character. When we\napply them to cellular mobile networks, we use the word cell. These two words have\nthe same meaning in their respective contexts. Similarly, string is used in discussing\nLempel-Ziv algorithms, while cell list is used in cellular mobile networks.\nLZ-based Intrusion Detection\nData compression is a technique that encodes data in order to minimize its repre-\nsentation. Some of the most common lossless compression algorithms used in practice\nare dictionary-based schemes, where a dictionary D = (M, C) is a finite set of phrases\nM and a function C that maps M onto a set of codes. In practice, when no a priori\nknowledge of the source characteristics is available, the problem of data compression\nbecomes considerably complicated. Therefore, we often resort to universal coding\n"
        },
        {
            "page_number": 202,
            "text": "WIRELESS NETWORK SECURITY\n197\nschemes whereby the coding process is interlaced with a learning process for the vary-\ning source characteristics.\nThe family of Lempel-Ziv algorithms belongs to dictionary-based text compres-\nsion and encoding techniques [42]. They are based on a popular incremental parsing\nalgorithm by Ziv and Lempel [40],[41], and have been widely used in data compression.\nSince its invention, many variations have been developed. LZ78 is the most popular\none.\nThe original LZ78 [40] is a word-based data compression algorithm. It parses\nthe input string S of size n in a greedy manner into distinct substrings x1, x2, . . . , xm\nwith the following property: for j > 1, there exists a number i < j, which makes\nxj equal to xi concatenated by c, where c is one character in the alphabet. This is the\nso-called prefix property [42]. In the parsing process, if a phrase is the longest matching\nphrase seen previously concatenated by one character, the phrase, called a new phrase,\nis added to the dictionary. Substring xj is encoded by the value i, using ⌈lg(j−1)⌉bits,\nfollowed by the ASCII encoding of the last character of xj, using ⌈lg α⌉bits, where α\nis the size of the input string’s alphabet. Here the base of the logarithm is 2.\nThe Ziv-Lempel algorithm can be converted from a word-based method to a\ncharacter-based algorithm by building a probabilistic model that feeds probability in-\nformation to an arithmetic coder [43], which encodes a sequence of probability of p\nusing lg( 1\np) = −lg p bits.\nLZ78 is both theoretically optimal and good in practice. When the input text\nis generated by a stationary and ergodic source, LZ78 algorithms enjoy the property\nof being asymptotically optimal as the input size increases. That is, it encodes an\nindefinitely long string in the minimum size dictated by the entropy of the source. Here\nwe omit the detailed proof. Being good in practice means that searching of LZ78 can\nbe implemented efficiently by inserting each phrase in a trie data structure.\nA trie is suitable to store the parsed phrases, and is a multiway tree with any path\nfrom the root to a unique node forming a string. In a trie, only the unique prefix of each\nstring is stored because the suffix can be determined by searching the string. A longest\nmatch is found by following down the tree until no match is found, or the path ends at\na leaf.\nHere is an example of how to parse a string using LZ78 algorithm and construct\na trie. Suppose the alphabet A is (a, b, c), and one possible string S over this al-\nphabet is aababccbababbabb . . .. Each element of the alphabet A could be one pos-\nsible cell the user visits.\nS could be one possible cell list traversed by this user.\nEach substring in the parse is encoded as a pointer followed by an ASCII character.\nBased on the greedy parsing manner, this string will be parsed into phrases as follows:\n(a)(ab)(abc)(c)(b)(aba)(bb)(abb) . . ..\nIn the character-based version of the Ziv-Lempel encoder, a trie is built when the\nprevious substring ends. A trie at the start of the ninth substring is shown Figure 5.(a).\nThe number associated with each node indicates the frequency in terms of number of\ntimes this node has been parsed in the construction of the mobility trie.\nThis trie characterizes the probability model of the string aababccbababba\nbb . . .. There are five previous substrings beginning with an a, two beginning with a b,\n"
        },
        {
            "page_number": 203,
            "text": "198\nBO SUN et al.\nand one beginning with a c. Therefore, the probability of a at the root is 5\n8. Similarly,\nthe probability of b at the root is 2\n8 = 1\n4 and the probability of c at the root is 1\n8. Of the\n5 substrings that begin with an a, 4 begins with b. Therefore, the probability of b from\na is 4\n5.\nProbability Calculation\nThe probability calculation is based on the Prediction by Partial Matching (PPM)\n[44] scheme. Here, we use a context model to predict the next character based on the\nprevious consecutive characters. Specifically, we use a m-th Markov model to model\nthe sequence. That is, we use the consecutive previous m characters to predict the next\ncharacter and calculate its probability. Here m is the order of the Markov model. For\na first-order (m = 1) Markov model, it assumes that the next event only depends on the\nlast event in the past. A high-order (m > 1 order) Markov model assumes that the next\nevent depends on multiple (m) events in the past.\nA trade-off exists here. If the order m is too small, the prediction will be poor in\nthe long run because little audit data will be available to make a decision. However, if\nthe order is too large, most contexts will seldom happen, and initially the probability\nestimation will have to solely rely on the resolve of zero-frequency problems [42].\nBased on these considerations, we take a blending approach, where the predications of\nseveral contexts of different lengths are combined into a single overall probability. It\nuses a number of models with different orders to compute the probabilities respectively,\nassign a weight to each model, and calculate the weighted sum of the probabilities.\nLet’s denote the maximum order as m.\nThe next character, denoted by α, is\npredicted on the basis of previous i characters. For each character α, let pi(α) be\nthe probability assigned to α by the finite-context model of order i. Note that when\ni is zero, the probability of each character is estimated independently of other char-\nacters. If the weight given to the model of order i is wi and the blending weight\nvector is [w0, w1, . . . , wm], the blended probability p(α) is computed as p(α) =\n\u000em\ni=0 wi ∗pi(α), where the sum of weights is normalized to 1. The larger the or-\nder, the larger the weight assigned to it, because context models with larger orders tend\nto be more accurate and should weight more in the current normal profile.\nAnomaly Detection Algorithm\nWe adopt the character-based LZ78 to deal with the anomaly detection problem,\nand a classifier is trained with known “normal” data to distinguish normal behaviors\nfrom anomalous ones.\nIntegration of EWMA into Mobility Trie\nIn anomaly detection, each subject (i.e.,\nuser in this application) has a normal profile. For an individual subject, its activity\nmay change over time. Therefore, it is necessary for the normal profile to be updated\nin order to reflect the recent activities. In our situation, the normal profile of the user\nactivity should be dynamic. Generally, activities in the recent past should weight more\n"
        },
        {
            "page_number": 204,
            "text": "WIRELESS NETWORK SECURITY\n199\nthan activities long time ago. Adaptively modifying the normal profile correspondingly\nis a suitable mechanism.\nBased on the above considerations, we integrate EWMA [35] to the mobility trie.\nThe mobility trie is modified when a new phrase is formed during the string pars-\ning. When a new phrase is inserted, we say an event happens. Note that this event\ncorresponds to a sequence of characters. The insertion of the new phrase needs to\nmodify the existing frequency of the mobility trie. We will call the modified frequency\nEWMA-based frequency hereafter. EWMA-based frequency measures how often the\ncorresponding node appears in the recent past. Note that we do not need to do an extra\ntrie search to modify the frequency. Instead, it is done at the same time with the update\nof the mobility trie to improve efficiency.\nThe EWMA-based frequency of each node in the mobility trie is updated as:\nF(i) = λ ∗1 + (1 −λ) ∗F(i),\n(1)\nwhere node i is one item of the corresponding events;\nF(i) = λ ∗0 + (1 −λ) ∗F(i),\n(2)\nwhere node i is not one item of the corresponding events.\nroot\na, 0.3\na\nroot\na, 0.51\na, ab\nb, 0.3\nroot\na, 0.657\na, ab, abc\nb, 0.51\na: 0.51=0.3*1+(1-0.3)*0.3\nb: Initialized to 0.3\na: Initialized to 0.3\na: 0.657=0.3*1+(1-0.3)*0.51\nb: 0.51=0.3*1+(1-0.3)*0.3\nc: Initialized to 0.3\n(b) When (a) is parsed.\n(c) When (a)(ab) is parsed\n(d) When (a)(ab)(abc) is parsed.\nroot\nc, 1\na, 5\nb, 2\nb, 4\nc, 1\na, 1\nb, 1\n(a) An example mobility trie.\nb, 1\nc, 0.3\nFigure 5. An Example of Mobility trie and an Example of Building Mobility Trie.\nHere F(i) is the EWMA-based frequency value stored in node i after a new phrase\nis inserted. For example, in Figure 5.c, the EWMA-based frequency associated with a\nis 0.51. The EWMA-based frequency associated with b is 0.3. Here λ is a smoothing\nconstant that determines the decay rate. If a node i is not observed for continuous k\n"
        },
        {
            "page_number": 205,
            "text": "200\nBO SUN et al.\nevents (one event happens when a new phrase is inserted), the EWMA-based frequency\nof node i will be decayed to (1 −λ)k. In this way, the EWMA-based frequency of\neach node measures the intensity of this node over the recent past.\nContinuing the example illustrated in Figure 5.(a), we illustrate how to integrate\nEWMA into the construction of the mobility trie. In this example, we let λ be 0.3. When\nthefirstcharacteraisparsed, thecorrespondingmobilitytrieisillustratedinFigure5.(c).\nWhen ab is parsed, the corresponding mobility trie is illustrated in Figure 5.(d). When\nabc is parsed, the corresponding mobility trie is illustrated in Figure 5.(d). As we\ncan see, the EWMA-based frequency value associated with each node is exponentially\nfaded.\nThe Similarity Measure\nEWMA-based mobility trie maintains the stationary part\nof each user’s recent activities. Based on this, we could accurately predict whether the\nfuture activities are normal or not.\nLet the sample space be all the possible cells traversed by a user. Because a user\nhas his favorite routine of activity, this could lead to a small set of sample space. Let\nS = (X1, X2, . . . , Xn) denote the observed activities of the user, where Xi denotes a\ncell number. We want to identify whether or not it is normal based on our constructed\nmobility trie. We use a high-order Markov model to compute its blending transition\nprobabilities.\nGiven an order o of the Markov model, we define the o-th order probability of S\nas:\nPo =\nn−o\n\u0001\ni=1\nP(Xi+o|Xi, Xi+1, . . . , Xi+o−1).\n(3)\nWhen it is order-0 model (o = 0), the probability of S is calculated as Po = P0 =\n\u000en\ni=1 P(Xi).\nTo calculate the probability of the transition (Xi, Xi+1, . . . , Xi+o−1) −→Xi+o\nin equation 3, we need to search (Xi, Xi+1, . . . , Xi+o−1) from the root. Let F(Xi+o)\ndenotetheEWMA-basedfrequencyofnodeXi+o. If(Xi, Xi+1, . . . , Xi+o−1)isfound,\nthe probability P(Xi+o|Xi, Xi+1, . . . , Xi+o−1) is defined as:\nP(Xi+o|Xi, Xi+1, . . . , Xi+o−1) =\nF(Xi+o)\nF(Xx+o−1).\n(4)\nIf (Xi, Xi+1, . . . , Xi+o−1) is not found, its probability is assigned 0.\nTo calculate P(Xi), we compute the sum of the EWMA-based frequency of the\nroot’s children. P(Xi) is then defined as F(Xi)/\u000e F(Xroot’s children).\nIf Xi is not a child of the root, P(Xi) is 0. That is, we only search from the root\nto decide the probability of each Xi.\nTake the trie illustrated in Figure 5.d as an example, P(b) =\n0.3\n0.357+0.3 = 0.4566,\nP(b|a) =\n0.51\n0.657 = 0.7763, P(c|ab) =\n0.3\n0.51 = 0.5882.\n"
        },
        {
            "page_number": 206,
            "text": "WIRELESS NETWORK SECURITY\n201\nSuppose that the blending weight vector is [w0, w1, . . . , wm], where wi is the\nweight value associated with the i-th order Markov model. \u000em\ni=0 wi = 1 and wi ≥\n0, ∀i. The probabilities of string S is defined as P = \u000em\ni=0 wi ∗Pi.\nIntuitively, P increases with the increase of S’s length because more transitions\nwill be considered when S is longer. Therefore, P is not a good metric. We propose to\nuse the following metric as our similarity measure similarity(S) =\nP\nLength(S), where\nLength(S) is the length of string S.\nBased on our definition, the similarity measure could be normalized by the length\nof the string and provides good criteria to evaluate its normalcy. Intuitively, similarity\nindicates how good a mobile user follows its routines.\nFor the input string S, we calculate its similarity(S). When a user follows one of\nits favorite itineraries, because this path is integrated into the mobility trie to construct\nthe normal profile, many of its transitions illustrated in equation 4 at different order\no will be found in the mobility trie, i.e., normal profile. Based on our definition,\nsimilarity(S) will be a relatively large value. However, when the mobile is stolen,\nand the intruder takes an infrequent path, the similarity of this string tends to be a very\nsmall value, because many transitions cannot be found in the mobility trie.\nWeintroduceathreshold, Pthr, whichisadesignparameter. Whensimilarity(S) ≥\nPthr, string S is evaluated as normal, otherwise string S is identified as anomalous.\nBecause our mobility trie records the most frequently used path of a user, it is\nvery sensitive to anomalous paths, even if they are very short strings. This enables\nour detection algorithm to detect the abnormal very quickly - an important quality\nfor reducing potential damage by a malicious user. At the same time, our detection\nalgorithm has a very high detection rate. Also, when a frequently used path is taken,\nour detection algorithm can tolerate slight variations from the path and thus has small\na false positive rate.\nImplementation issues\nIn practice, an important issue is how to store the mobility information in a trie.\nA trie is actually a multiway tree with a path from the root to a unique node for each\nstring represented in the tree. The fastest approach for processing is to create an array\nof pointers for each node in the trie with a pointer for each character of the input\nalphabet. Although this approach is easy for processing, it wastes memory space.\nAnother approach is to use a linked list at each node, with one item for each possible\nbranch. This method uses memory economically, but the processing is intensive. A\ntrie can also be implemented as a single hash table with an entry for each node. For\nfurther details, the reader can consult books on algorithms and data structures.\n6.5. Markov-based Anomaly Detection\nMarkov predictors are a very popular family of predictors. They have been widely\nused and studied in the literature. Let Xt be the cell visited by the user or the state of\nthe user’s activity at time t. The order-o Markov predictor assumes that the location\n"
        },
        {
            "page_number": 207,
            "text": "202\nBO SUN et al.\ncan be predicted from the current context, which is the sequence of the previous o most\nrecent characters in the location history (Xt−o+1, Xt−o, . . . , Xt). Under this Markov\nmodel, the transitions represent the possible cell locations that follow the context.\nA Markov Chain with order-o of only one-step event transitions is a stochastic\nprocess with the following assumptions:\nP(Xt+1\n=\nit+1|Xt = it, Xt−1 = it−1, . . . , X0 = i0)\n=\nP(Xt+1 = it+1|Xt = it, Xt−1 = it−1, . . . ,\nXt−o+1 = it−o+1)\nP(Xt+1\n=\nit+1|Xt = it, Xt−1 = it−1, . . . , Xt−o+1 = it−o+1)\n=\nP(Xt+1 = j|Xt = io, Xt−1 = io−1, . . . , Xt−o+1 = i1)\n≡\np{i1,...,io−1,io}→j.\nIt describes the two important properties of the Markov Chain:\nEquation 5 states that the probability distribution of the user at time t + 1\ndepends on the state at time t, t −1, . . . , t −o + 1, and does not depend on the\nprevious states leading to the states at t, t −1, . . . , t −o + 1.\nEquation 5 states that the state transitions from time t, t −1, . . . , t −o + 1 to\nt + 1 is independent of time.\nIf the system has a finite number of states 1, 2, . . . , s, these probabilities could\nbe represented in a transition probability matrix, where each element in the matrix is\np{i1,...,io−1,io}→j, as illustrated in 5.\n⎡\n⎢⎢⎢⎣\np{1,1,...,1}→1\np{1,1,...,1}→2\n. . .\np{1,1,...,1}→s\np{1,1,...,2}→1\np{1,1,...,2}→2\n. . .\np{1,1,...,2}→s\n...\n...\n...\n...\np{s,s,...,s}→1\np{s,s,...,s}→2\n. . .\np{s,s,...,s}→s\n⎤\n⎥⎥⎥⎦\n(5)\np{i1,...,io−1,io}→j could be learned from the observations of the user’ locations\nin the past. When o ≥1, P(Xt+1 = j|Xt = io, Xt−1 = io−1, . . . , Xt−o+1 =\ni1) = N(Lj)/N(L), where L = {i1, . . . , io−1, io}, N(Lj) denotes the number of\nobservation pairs of L and j. N(L) denotes the number of observations of L.\nWhen o is 0, the formula becomes:\nP(Xt+1 = j) = N(j)\nN\n,\n(6)\nwhere N is the total number of observations (i.e., total number of cells). N(j) is the\nnumber of observations of a.\n"
        },
        {
            "page_number": 208,
            "text": "WIRELESS NETWORK SECURITY\n203\nGiven this estimation, we can calculate the probability of the next location given\nthe previous o locations for a specific user. The larger the probability, the more likely\nit is normal. We can then derive a threshold policy and use it to decide whether the\ncurrent activity is normal or not.\nThat is, given a fixed order value o and an observed activity in terms of a cell list\nSobserved = (X1, X2, . . . , Xn), where each Xi denotes a cell number. For o ≥1, we\nfirst calculate its o-order transition probabilities as Po = \u000en−o\ni=1 P(Xi+o = j|Xi =\ni, Xi+1 = i + 1, . . . , Xi+o−1 = i + o −1) = \u000en−o\ni=1 p{i,i+1,...,i+o−1}→j, where\np{i,i+1,...,i+o−1}→j can be retrieved from the probability transition matrix whose el-\nement is obtained using Equation 5. If the transition does not exist in the transition\nmatrix, we assign P(Xi+o|Xi, Xi+1, ..., Xi+o−1) to 0.\nFor o = 0, its probability could be calculated as Po = \u000en\ni=1 P(Xi = j), where\nP(Xi = j) can be obtained from Equation 6.\nSimilar to LZ-based mechanism, Po increases with the increase of S’s length.\nTherefore, for Markov-based prediction, we also define the following similarity metric:\nsimilarity(S) =\nPo\nLength(S), where Length(S) is the length of string S.\nFor the input string S, we calculate its similarity(S). If most transitions can\nbe found, similarity(S) tends to be large. This indicates that S is more likely to be\nnormal. However, if the mobile is stolen, and an infrequent or new path is taken, the\nsimilarity of the string should be small.\nWhen the mobile is at low mobility, the user usually travels one or two cells during\nthe call. Given a fixed o, it is highly possible that the length of the transition (o +\n1) is larger than the length of the cell. The Markov-based prediction cannot make\na decision under this situation. Therefore, high-order Markov-based prediction will\nbecome useless for low mobility data. We make a random guess when this situation\nhappens. For example, with a probability of 1/2, this cell list is identified as normal\n(abnormal).\nFor Markov-based prediction, we introduce a threshold Pthr markov. If similarity\n(S) ≥Pthr markov, string S is evaluated as normal. Pthr markov should be tuned by\ntaking into consideration both false alarm rate and detection rate.\n6.6. Adaptive Anomaly Detection\nIn this section, we illustrate how to integrate adaptability into LZ-based detec-\ntion schemes. EWMA-based mobility trie itself facilitates the differentiation between\nweekday and weekend routes because when the user changes its mobility patterns, for\nexample, from weekday to weekend routes, the more recent the activities, the more\nweight they should have in the normal profile. The smoothing constant in EWMA\ntechniques plays an important role in determining how much weight the more recent\nactivities should have. Basically the larger the smoothing constant is, the more weight\nthey should have. Therefore, intuitively, the shorter the recent activities last, the larger\nthe smoothing constant should be.\n"
        },
        {
            "page_number": 209,
            "text": "204\nBO SUN et al.\nThe EWMA-based approach only partially addresses the adaptation of normal\nprofiles. In the following, we detail our approach of how to tune the threshold for\ndifferent users and different mobility levels.\nFeedback-based Approach\nOne simple approach to adjust the threshold is to apply the feedback principle. That\nis, based on the output of the detection algorithm (for example, in terms of detection\nrate and false positive rate), the system administrator can adaptively adjust the detection\nthreshold in order to achieve the required performance. If the false positive rate is a more\nimportant metric, for example, when the system has been detected raising too many false\nalarms, the system administrator could lower the detection threshold correspondingly.\nHowever, in this approach, the decrease of the false positive rate is achieved at the risk\nof a decreased detection rate.\nEntropy-based Approach\nWe use Shannon’s entropy measure to identify the uncertainness of the up-to-date\nnormal profile. Based on this, we could adjust the detection threshold correspondingly.\nMetric Selection\nThe first step we need is to identify a metric that can effectively reflect the location\nuncertainty. In our case, it is the EWMA-based mobility trie. Shannon’s entropy\nmeasure [45] is an ideal candidate for quantifying this uncertainty. Our previous work\nshowed that for the non-adaptive mechanism, given a mobility level, the more varied\nthe mobility pattern is, the more dynamic the mobility trie is. This motivates us to use\nentropy as a measure to reflect the dynamic level of the normal profile. The lower the\nuncertainty under the movement pattern, the richer the movement pattern is.\nDefinition 1. Entropy: Suppose X is a dataset, Cx = {Cx[1], Cx[2], . . . , Cx[m]} is\na class set. Each data item of X belongs to a class x ∈Cx[i]. Then the entropy of X\nrelated to this |Cx|-wise classification is defined as H(X) = \u000em\ni=1 −Pi log Pi, where\nPi is the probability of x belonging to class Cx[i].\nEntropy can be interpreted as the number of bits required to encode the classifica-\ntion of a data item. It measures the uncertainty of a collection of data items. The lower\nthe entropy, the more uniform the class distribution. If all data items belong to one\nclass, then its entropy is 0, which means that no bits needs to be transmitted because\nthe receiver knows that there is one class. The more varied the class distribution is,\nthe larger the entropy is. When all of the data items are equally distributed over the m\nclasses, its entropy is log(m) (natural logarithm). In the context of anomaly detection,\nentropy is a measure of the regularity of audit data.\nDefinition 2.\nConditional entropy: Suppose that X and Y are two datasets, and\nCx = {Cx[1], Cx[2], . . . , Cx[m]} and Cy = {Cy[1], Cy[2], . . . , Cy[n]} are two class\nsets. Each data item of X belongs to a class x ∈Cx[i] and each data item of Y belongs\n"
        },
        {
            "page_number": 210,
            "text": "WIRELESS NETWORK SECURITY\n205\nto a class y ∈Cy[i]. Then given Y and Cy, the entropy of X related to Cx is defined\nas H(X|Y ) = \u000em\ni=1\n\u000en\nj=1 Pij log\n1\nPi|j , where Pij is the probability of x ∈Cx[i] and\ny ∈Cy[j], and Pi|j is the probability of x ∈Cx[i] given y ∈Cy[j].\nConditional entropy describes the uncertainness of X given Y , i.e., it indicates\nthe coefficiencies between X and Y . The smaller the conditional entropy is, the more\ncorrelated X and Y are. If X can be determined by Y , H(X|Y ) is 0. In the context of\nanomaly detection, conditional entropy can be used to explore the temporal sequential\ncharacteristics of audit data due to the temporal nature of the system activities.\nCompute the Entropy of a Trie\nWhen we compute the entropy of the EWMA-based mobility trie, we apply a\nweighted scheme at different orders. Specifically, based on the order of different finite\ncontexts of the mobility trie, we calculate conditional entropies respectively and assign\nthem different weights. The larger the order is, the larger the weight should be. The sum\nof these weighted entropies is used as the measurement for adjusting system detection\nthreshold. Let’sconsideramorecomplexstringaaababbbbbaabccbaaaaaacabbbabcacb.\nBy applying LZ78 algorithm [42], we obtain a trie as illustrated in Figure 6.\nroot\nb, 5\na, 7\nc, 3\nb, 3\na, 1\nc, 1\na, 2\nb, 1\na, 3\nc, 1\nb, 2\na, 1\nb, 1\na, 1\nb, 1\nFigure 6. An Example of EWMA-based Mobility Trie.\nThe maximum order m and the corresponding weight wi are design parameters.\nIn this example, let’s assign 2 to m.\nOrder-0 Model\nH(V 1)\n=\n7\n15 log 15\n7 + 5\n15 log 15\n5 + 3\n15 log 15\n3\n=\n1.0438.\n"
        },
        {
            "page_number": 211,
            "text": "206\nBO SUN et al.\nOrder-1 Model\nH(V 2|V 1)\n=\n7\n15[(3\n6 log 6\n3) × 2] + 5\n15[(2\n4 log 4\n2) × 2] + 3\n15[(1\n2 log 2\n1) × 2]\n=\n0.6931.\nOrder-2 Model\nH(V 3|V 1V 2) = 3\n15[(1\n2 log 2\n1) × 4] + 0 = 0.2773.\nWhen the context of a specific length is not found in the trie, we assign its condi-\ntional probability to 0. Note that we treat 0 log 0 as 0.\nGenerally, the larger the order is, the larger the weight assigned to it should be,\nbecause context models with a larger order tend to be more accurate and should weight\nmore in the current normal profile. If we assign 0.1, 0.2, and 0.7 to w1, w2, and w3,\nrespectively, the weighted entropy of the mobility trie in Figure 6 can be calculated as:\nweighted entropy\n=\nw1 × H(V 1) + w2 × H(V 2|V 1) + w3 × H(V 3|V 1V 2)\n=\n0.4371.\nAdaptive Algorithm\nThe algorithm of constructing the adaptive normal profile is illustrated in Figure 7.\nIt summarizes how to use EWMA to adaptively adjust the normal profile and how to\nuse entropy to adaptively adjust the threshold.\n7.\nSUMMARY\nSignificant security concerns exist in wireless networks. Although there are many\nprevention-based protocols in cellular mobile networks, how to design a highly secure\ncellular mobile network is still a very challenging issue due to the open radio transmis-\nsion environment and physical vulnerability of mobile devices. Intrusion detection is\nindispensable to provide an enhanced protection for wireless networks.\nThis chapter presents the current status of major intrusion detection techniques\ndeveloped for wired and wireless networks. We point out corresponding challenges\nthat need to be addressed in the future. In the context of cellular mobile networks, we\nalso present the detailed steps in developing one exemplary intrusion detection system.\nOur presented example mainly exploits users’ location information to identify potential\nfraudsters and masqueraders. Future work may include the integration of users’ calling\nactivities. Because of the potential wide variety of users’ behaviors, it is difficult to\n"
        },
        {
            "page_number": 212,
            "text": "WIRELESS NETWORK SECURITY\n207\nINPUT: Observed user’s mobility activities in terms of a cell list \nOUTPUT: Adaptive normal profile \nInitialize mobility database := null \nLOOP \n \n    Based on the LZ78 algorithm, wait for a sequence s\nIF (The mobility trie of the mobile exists) \n \n \n  IF (A path p corresponding to s is found) \nAdd s to the mobility trie \nUsing EWMA to modify the frequencies of nodes \n \n \n  ELSE\n \n \n    Create new nodes, and initialize their frequencies to λ\nELSE \n \n \n \n1) Create a mobility trie := single sequence s\n \n \n \n2) Initialize the frequencies for every node in sequence  \ns to λ\n \n \nCompute the entropy e1 of the EWMA-based mobility trie \nIF (e1 > e)\n \n \n/* e is the entropy of the previous EWMA-based mobility trie */ \n \n \n \nDecrease the detection threshold by ∆\nELSE \n \n \n \nIncrease the detection threshold by ∆\ne = e1;\nFOREVER \nFigure 7. Adaptive Normal Profile.\naccurately characterize users’ activities. Moreover, considering the randomness of\ncertain users’ behaviors, not all users can be considered as potential candidates for the\nsuccessful applications of anomaly detection techniques.\nIntrusion detection in cellular mobile networks is a challenging problem. Not\nonly will traditional challenges like feature selection continue to exist, but also new\nproblems specific to cellular mobile networks keep appearing. All these deserve the\nfurther attention from the research community.\n"
        },
        {
            "page_number": 213,
            "text": "208\nBO SUN et al.\n8.\nREFERENCES\n1. Y.-B. Lin, M. Chen, and H. Rao, Potential fraudulent usage in Mobile Telecommunications Networks,\nIEEE Transactions on Mobile Computing, Vol.1 No.2 (2002) pp. 123-131.\n2. M. Zhang, and Y. Fang, Security Analysis and Enhancements of 3GPP Authentication and Key Agree-\nment Protocol, IEEE Transactions on Wireless Communication, Vol.4, No.2, (2005) pp. 734-742.\n3. B. Mukherjee, T.L. Heberlein, and K.N. Levitt, Network Intrusion Detection, IEEE Network, Vol.8,\nNo.3, (1994) pp. 26-41.\n4. H. Debar, M. Dacier, and A. Wespi, A Revised Taxonomy for Intrusion-Detection Systems, Annales\ndes Telecommunications, Vol. 55, (2000) pp. 361-378.\n5. J. P. Anderson, Computer Security Threat Monitoring and Surveillance, Technical Report, James P.\nAnderson Co., Fort Washington, PA, (April 1980).\n6. D. E. Denning, An Intrusion-Detection Model, IEEE Transactions on Software Engineering, Vol.13,\nNo.7, (Feb. 1987), pp. 222-232.\n7. P. Porras and A. Valdes, Live Traffic Analysis of TCP/IP Gateways, Proceedings of the 1998 ISOC\nSymposium on Network and Distributed System Security (NDSS’98), San Diego, CA, March 1998.\n8. T. F. Lunt, R. Jagannathan, R. Lee, S. Listgarten, D. L. Edwards, P. G. Neumann, H. S. Javitz, and A.\nValdes, IDES: The Enhanced Prototype ¨C a Real-time Intrusion-Detection Expert System, Technical\nReport SRI-CSL-88-12, SRI International, Menlo Park, CA, Oct. 1988.\n9. H.S. Javitz and A. Valdes, The SRI Statistical Anomaly Detector, Proceedings of 1991 IEEE Symposium\non Research in Security and Privacy, pp. 316-326, May 1991.\n10. R. Jagannathan, T. Lunt, D. Anderson, C. Dodd, F. Gilham, C. Jalali, H. Javitz, P. Neumann, A.\nTamaru, and A. Valdes, System Design Document: Next-Generation Intrusion Detection Expert System\n(NIDES), Technical Report A007/A008/A009/A011/A012/ A014, SRI International, 333, Ravenswood\nAvenue, Menlo Park, CA, March 1993.\n11. U. Lindqvist, and P.A. Porras, Detecting Computer and Network Misuse through the Production-Based\nExpert System Toolset (P-BEST), Proceedings of the 1999 IEEE Symposium on Security and Privacy,\nOakland, CA, pp. 146-161, May 9-12, 1999.\n12. Internet Security Systems, RealSecure Network Protection, Nov. 2003, Available at\nhttp : //www.iss.net/products services/enterprise protection/rsnetwork.\n13. S. Kumar and E. Spafford, A Pattern Matching Model for Misuse Intrusion Detection, Proceedings of\nthe 17th National Computer Security Conference, pp. 11-21, Oct. 1994.\n14. P.A.PorrasandR.Kemmerer, PenetrationStateTransitionAnalysis¨CaRule-BasedIntrusionDetection\nApproach, Proceedings of the 8th Annual Computer Security Application Conference, pp. 220-229, Nov.\n1992.\n15. K. Ilgun, Ustat: A Real-time Intrusion Detection System for Unix, Proceedings of IEEE Symposium\non Research in Security and Privacy, Oakland, CA, pp. 16-28, May, 1993.\n16. H. Debar, M. Becker and D. Siboni, A Neural Network Component for an Intrusion Detection System,\nProceedings of 1992 IEEE Symposium on Research in Security and Privacy, Oakland, CA, pp. 240-250,\nMay, 1992.\n17. S. Forrest, S.A. Hofmeyr, and A. Somayaji, Computer Immunology, Communications of the ACM, vol.\n40, no. 10, pp. 88-96, Oct. 1997.\n18. C. Warrender, S. Forrest, and B. Pearlmutter, Detecting Intrusions Using System Calls: Alternative\nData Models, Proceedings of 1999 IEEE Symposium on Research in Security and Privacy, Oakland,\nCA, pp. 133-145, May 1999.\n"
        },
        {
            "page_number": 214,
            "text": "WIRELESS NETWORK SECURITY\n209\n19. W. Lee, S. J. Stolfo, and K. W. Mok, A Data Mining Framework for Building Intrusion Detection\nModels, Proceedings of the 1999 IEEE Symposium on Security and Privacy, Oakland, CA, pp. 120-\n132, May 1999.\n20. N. Ye, X. Li, Q. Chen, S. M. Emran, and M. Xu, Probabilistic Techniques for Intrusion Detection Based\non Computer Audit Data, IEEE Transactions on Systems, Man, and Cybernetics, vol. 31, no. 4, pp.\n266-274, 2001.\n21. N. Ye, S. M. Emran, Q. Chen, and S. Vilbert, Multivariate Statistical Analysis of Audit Trails for\nHost-based Intrusion Detection, IEEE Transactions on Computers, vol. 51. no. 7, pp. 810-820, 2002.\n22. S. Jha, K. Tan, and R.A. Maxion, Markov Chains, Classifiers, and Intrusion Detection, Proceedings\nof the 14th IEEE Computer Security Foundations Workshop, Cape Breton, Nova Scotia, Canada, pp.\n206-219, 2001.\n23. N. Ye, A Markov Chain Model of Temporal Behavior for Anomaly Detection, Proceedings of 2000\nIEEE Workshop on Information Assurance and Security, United States Military Academy, West Point,\nNY, pp. 171-174, June 6-7, 2000.\n24. N. Ye, T. Ehiabor, and Y. Zhang, First-order versus High-order Stochastic Models for Computer Intru-\nsion Detection, Quality and Reliability Engineering International, vol. 18, no. 3, pp. 243-250, 2002.\n25. M. Nassehi, Anomaly Detection for Markov Models, Technical Report Tech Report RZ 3011(#93057),\nIBM Research Division, Zurich Research Laboratory, March 1998.\n26. W. Lee, and S. J. Stolfo, A framework for constructing features and models for intrusion detection\nsystems, ACM Transactions on Information and System Security (TISSEC), Vol. 3 Issue 4, (2000).\n27. C. Ko, G. Fink, and K. Levitt, Execution Monitoring of Security-Critical Programs in Distributed\nSystems: A Specification-Based Approach, Proceedings of 1997 IEEE Symposium on Security and\nPrivacy, Oakland, CA, pp. 134-144, May 1997.\n28. R. Buschkes, D. Kesdogan, and P. Reichl, How to Increase Security in Mobile Networks by Anomaly\nDetection,¡± Proceedings of the Computer Security Applications Conference, Phoenix, AZ, Dec. 1998,\npp. 3-12.\n29. D. Samfat, and R. Molva, IDAMN: An intrusion detection architecture for mobile networks, IEEE\nJournal on Selected Areas in Communications, Vol.15, No.7, (Sept. 1997), pp. 1373-1380.\n30. J. Hall, M. Barbeau, and E. Kranakis, Anomaly-based Intrusion Detection Using Mobility Profiles of\nPublic Transportation Users, IEEE Wireless and Mobile Computing, Networking and Communications\n(WiMob’2005), (2005), pp. 17-24.\n31. J. Hollm´en, User profiling and classification for fraud detection in mobile communications networks,\nPhD Thesis, Helsinki University of Technology, (2000).\n32. C. Warrender, S. Forrest, and B. Pearlmutter, Detecting Intrusions Using System Calls: Alternative\nData Models, Proceedings of 1999 IEEE Symposium on Research in Security and Privacy, Oakland,\nCA, pp. 133-145, May 1999.\n33. D. Koller, and M. Sahami, Toward Optimal Feature Selection, Proceedings of the 13th International\nConference on Machine Learning, Bari, Italy, July 1996, pp. 284-292.\n34. Q. He, D. Wu, and P. Khosla, Quest for personal control over mobile location privacy, IEEE Commu-\nnications Magazine, Vol. 42, No. 5, May 2004, pp. 130-136.\n35. R.A. Johnson and D.W. Wichern, Applied Multivariate Statistical Analysis, Upper Saddle River, NJ:\nPrentice Hall, 1998.\n36. B. Sun, F. Yu, K. Wu, Y. Xiao, and V.C.M. Leung, Enhancing Security using Mobility-Based Anomaly\nDetection in Cellular Mobile Networks, IEEE Transactions on Vehicular Technology, 2005, in press.\n"
        },
        {
            "page_number": 215,
            "text": "210\nBO SUN et al.\n37. B. Sun, Z. Chen, R. Wang, F. Yu, and V.C.M. Leung, Towards Adaptive Anomaly Detection in Cellular\nMobile Networks, IEEE Consumer Communications and Networking Conference (CCNC’06), Las\nVegas, NV, 2006.\n38. S. Schonfelder, Some notes on space, location and travel behaviour, Swiss Transport Research Confer-\nence, Monte Verita, Ascona, 2001.\n39. H. A. Karimi, X. Liu, A predictive location model for location-based services, Proceedings of the 11th\nACM international symposium on Advances in geographic information systems, New Orleans, LA,\nNov. 2003, pp. 126 - 133.\n40. J. Ziv, and A. Lempel, Compression of individual sequences via variablerate coding, IEEE Transactions\non Information Theory, Vol. 24, Noo. 5, Sept. 1978, pp. 530-536.\n41. J. Ziv and A. Lempel, A Universal Algorithm for Sequential Data Compression, IEEE Transactions on\nInformation Theory, Vol. 23, pp. 337-342, 1977.\n42. T. C. Bell, J. G. Cleary, and I. H. Witten, Text Compression, Prentice-Hall Advanced Reference Series,\nPrentice-Hall, Englewood Cleffs, NJ, 1990.\n43. J. S. Vitter and P. Krishnan, Optimal prefetching via data compression, Journal of ACM, vol. 43, no. 5,\nSept. 1996, pp. 771-793.\n44. J. G. Cleary and I. H. Witten, Data compression using adaptive coding and partial string matching,\nIEEE Transactions on Communications, vol. 32, no. 4, Apr. 1983, pp. 396-402.\n45. T. Cover and J. Thomas, Elements of Information Theory, John Wiley & Sons, 1991.\n"
        },
        {
            "page_number": 216,
            "text": "9\nTHE SPREAD OF EPIDEMICS ON SMARTPHONES\nBo Zheng\nDept. of Computer Science and Technology\nTsinghua University\nBeijing 100084, P. R. China\nE-mail: bzheng@csnet1.cs.tsinghua.edu.cn\nYongqiang Xiong\nMicrosoft Research Asia\n5F, Beijing Sigma Center, No. 49, ZhiChun Road,\nHaidian District, Beijing 100080, P. R. China\nE-mail: yqx@microsoft.com\nQian Zhang\nDept. of Computer Science\nHong Kong University of Science and Technology\nClear Water Bay, Kowloon, Hong Kong\nE-mail: qianzh@cs.ust.hk\nChuang Lin\nDept. of Computer Science and Technology\nTsinghua University\nBeijing 100084, P. R. China\nE-mail: chlin@tsinghua.edu.cn\nThe emergence of epidemics such as worms and viruses on smartphones severely threaten\nthe Internet and telecom networks. Two important features of smartphone, i.e., static short-\ncuts and mobile shortcuts, bring great challenge for traditional epidemic spread model. In\nthis paper, we propose a novel epidemics spread model (ESS) for smartphone which is an\nSIR model based on the analysis of the unique features of smartphones. With this ESS\nmodel, we study the “static shortcuts\" and “mobile shortcuts\" effects brought by smart-\nphones and consider the influence of the epidemic spread rate, network topology, patch-\ning and death rate as well as the initial pre-patch to the propagation of the smartphone\nepidemics. Critical condition of epidemic fast die out is derived from the ESS model,\nand the detailed analysis is given to the individual parameters in the model to study their\n"
        },
        {
            "page_number": 217,
            "text": "212\nBO ZHENG et al.\neffects to the epidemics spread. Extensive simulations in typical network topologies (small-\nworld network, power law graph, and Waxman network) have been performed to verify the\nESS model and demonstrate the effectiveness and accuracy. The guidance to prevent the\nepidemics of smartphones is also given based on our theoretical analysis and the simulations.\n1.\nINTRODUCTION\nAs of 2004 smartphones are an increasingly large part of the mobile phone market.\nAt the same time, epidemics1 begin to appear in the smartphone. In this section, we\nintroduce the features of the smartphone and the attacks on the smartphone. And then\ndescribe the difference between the smartphone epidemics and the PC epidemics.\n1.1. Smartphones\nThe Wikipedia [1] defines the smartphone as the following: A smartphone is\ngenerally considered any handheld device that integrates personal information man-\nagement and mobile phone capabilities in the same device. Often, this includes adding\nphone functions to already capable PDAs or putting “smart\" capabilities, such as PDA\nfunctions, into a mobile phone.\nIn recent years, the global market for smartphones takes on a meteoric rise. Ac-\ncording to analyst house Canalys [2] smartphone shipments increased over 100% from\n2004Q2 to 2005Q2, with over twelve million devices shipped in the latter period. And\naccording to market research from IDC [3], 50 million smartphones will be shipped in\n2005, and more than 110 million smart-phones will be shipped by 2008. In a couple\nyears, it is likely that most phones sold will be considered “smart\", except for disposable\nphones.\nMost Smartphones connect the Internet and telecom networks together. Smart-\nphones tend to unify communications which integrate telecom and Internet services\nonto a single device because it has combined the portability of cell-phones with the\ncomputing and networking power of PCs.\nThe key feature of smartphones is that they has common operating systems (OSes),\nand one can install additional applications to the device.\nThe applications can be\ndevelopedbythemanufacturerofthehandhelddevice, bytheOSvendor, bytheoperator\nor by any other third-party software developer.\nMost common operating systems are Symbian [4, 5] (developed by a group of\nrenowned mobile phone solution providers), Windows CE / Mobile [6, 7] (developed\nby Microsoft), Palm OS [8] (developed by PalmSource), BREW [9] (technically a\nplatform developed by Qualcomm), and Linux (such as Montavista [10]). Although the\ndetailed design and functionality vary among these OS vendors, all share the following\nfeatures [11].\n1 In this chapter, we use the term “epidemic\" to denote the epidemic-like phenomena in the computer and\nsmartphone networks, including worms, viruses and Trojans that can spread from one device to another.\n"
        },
        {
            "page_number": 218,
            "text": "WIRELESS NETWORK SECURITY\n213\nAccess to cellular network with various cellular standards such as GSM /CDMA\nand UMTS.\nAccess to the Internet with various network interfaces such as infrared, Blue-\ntooth, GPRS/CDMA1X, and 802.11; and use standard TCP/IP protocol stack\nto connect to the Internet.\nMulti-tasking for running multiple applications simultaneously.\nData synchronization with desktop PCs.\nOpen APIs for application development.\nWhile common OSes, open APIs, and sophisticated capabilities enable powerful\nservices, they also create common ground and opportunities for security breaches and\nincrease worm or virus spreading potentials. Given the PC-like nature of smart-phones\nand the trend of full-fledged OSes, software vulnerabilities seem inevitable for their\nOSes and applications. Moreover, with the Internet exposure, smartphones become\nideal targets for Internet worms or viruses since smart-phones are always on, and their\nuser population will likely exceed that of PCs, observing from the prevalence of cell\nphone usage today.\n1.2. The Smartphone Attacks\nSmartphones get a rapid growth in 2004, but this rapid growth also draws the at-\ntacker’s attention. In June 2004, Cabir was developed. This worm, capable of spreading\nvia Bluetooth was the first notable piece of malware seen on mobile phones and the\nSymbian OS. It was not released to the public in order to infect phones, however, but\nwas instead sent to security experts as a proof of concept of a “wireless worm\". Up to\nnow, more than 60 attacks [12, 13, 14] (virus, worms, Trojan horse, malware, etc.) are\nfound in the smartphones.\nThe following things make the attacks on the smartphones may have many new\nfeatures compared with attacks on PC. First of all, smartphones connect with many other\nthings, they connect to Internet and telecom network togather, and they often contact\nwith another bluetooth devices and sync with host PC. Moreover, when smartphones\ngot infected, they may infect other Smart-phones/PCs, more seriously, the unauthorized\noutgoing may be phone, SMS, MMS, even the user’s private information, and these are\nnot free! After a extensive survey, we summarize all the attacks we found in [12, 13, 14]\nand some attacks which we thought may appear in the near future. Most attack ways on\nPC appeared on smartphone, we list some special attack way on smartphone in Table 1.\nThe security of smartphone has drawn great attention recently. On the one hand,\nthe open mobile operating system, flexible programmability, and powerful computa-\ntional/network capabilities of smartphones inevitably create opportunities for software\nvulnerabilities. On the other hand, as mentioned before, with the fast growth of the\nsmart-phone customer base, smart-phones have become ideal targets because, with a\n"
        },
        {
            "page_number": 219,
            "text": "214\nBO ZHENG et al.\nTable 1. Attacks which may appear on smartphones\nAttack/Spread Ways\nExplanation\nHot synced\nInfect PC/Smartphone while hot synced\nSPAM\nSpread SPAM via SMS, MMS, or email\nMalformed SMS\nSend a certain malformed SMS and make the victim smart-\nphone shutdown\nLimit some functions\nLimit most functions of the Smart-phone, such as restrict\nthe Smart-phone to only receive phone call, or set random\npassword to media card and make it unaccessible\nOverwrite system ROM\nOverwrite system ROM and make the system crashed\nWorms\nNot only through the internet, smartphones can spread\nworms through MMS, Bluetooth, WiFi, etc.\nSleep deprivation torture attack\nVastly shortened battery life caused by the constant scan-\nning. Because Smart-phone is a resource restraint device,\nenergy is a very important resource to it.\nUnapproved dial (DoS)\nSome applications (usually Trojan or worm) dial a certain\nphone number to make phone-DoS attack\nUnapproved dial (Theft)\nHacker uses the victim’s Smart-phone to make phone call\nthrough some backdoors. They can make phone call paid\nby the victim and receive the voice by a VoIP connection\nto the Smart-phone\nUnapproved SMS/ MMS (DoS/\nSpam/ Worm)\nLike unapproved dial, hacker uses the victim’s Smart-\nphone to send spam SMS/MMS through some backdoors\nor Trojan keeps sending SMS/MMS to some Smart-phone\nto make DoS attacks, or worm sends MMS message in-\ncludes a copy of itself as an attachment.\nSIM Card cloned\nSIM Card is cloned, another person use the cloned SIM\nCard (STK) to make phone call or something else.\nDial/SMS/MMS redirection\nSome malware redirects the dial/SMS/MMS number just\nbefore the user press the send key.\nRemote wiretapping\nHacker wiretaps the Smart-phone through a VoIP connec-\ntion or some Trojan send the phone record via email as an\nattachment\nRemote watcher/ Private informa-\ntion theft\nHackers use the DC/DV in the Smart-phone to watch the\nowner, or steal the private information such as pictures or\nvideos in the Smart-phone\n"
        },
        {
            "page_number": 220,
            "text": "WIRELESS NETWORK SECURITY\n215\nlarge cohort of subverted smart-phones, attackers can cause damage not only to the In-\nternet but also to the telecom infrastructure [15]. Moreover, Smartphones are often used\nto store the private or confidential information, which also attracts crackers launching\nattacks to them. Consequently, attacks to smartphones including worms, email viruses,\nMMS virus, and Trojan horses have emerged recently with growing frequency.\n1.3. The Spread of Epidemic on Smartphones\nIn order to deal better with the smartphone security problem and provide some\nguidance for building security scheme for smartphone in the near future, we need study\nthe propagation behavior of the smartphone worms, viruses, and Trojan horses, and\nidentify which factors will influence the propagation. In the literature, these attacks\ncan be modeled as spread of epidemic through the network. So we also leverage it to\ninvestigate the spread of epidemics in mobile network. However traditional epidemic\nmodels can not be applied to smartphones, because they only consider the static network\ntopology, but the new effects of mobile nodes that bring to the network are not modeled.\nFirstly, the smartphone is movable between networks. This handheld device may carry\nepidemics and spread them to devices encountered while moving in different networks,\nwhich are not physically connected. Secondly, the smartphones often have multiple\nnetworking interfaces, such as GPRS or CDMA for wide area networks, and WiFi\nor Bluetooth for local area networks. So smartphone can connect to both telecom\nnetworks and Internet/enterprise/home networks simultaneously, which would speed\nup the epidemics spread in both networks.\nWe use Figure 1 to illustrate the smartphone’s new effects on spread of epidemics.\nIn this figure, the enterprise LAN is well shielded by firewalls or security gateways,\nso it is difficult for worms, viruses and spams to infect the company’s computers from\noutside. With smartphone, two cases of epidemics can happen, For case A, the smart-\nphone connects to both telecom network and the enterprise WLAN, so it becomes a\nstatic shortcut between the two networks. An MMS worm from the telecom network\nmay compromise the smartphones and then spread to the enterprise network. In case\nB, even though the smartphone has only single interface, if it is compromised outside\nand being carried in the company by an employee, it may infect the computers and\nsmartphones in the LAN, and we call this smartphone creating a mobile shortcut be-\ntween internal corporate LAN and outside networks. When numerous compromised\nsmartphones (with multiple interfaces) move to other places, and infect more and more\nsmartphones and computers, we can conclude the “static shortcuts\" and “mobile short-\ncuts\" may cause the epidemic spread faster than that in normal static network, which\nare not studied in traditional epidemic models in the literature.\nIn this chapter, we propose a novel epidemics spread model on mobile network\ncalled ESS (Epidemic Spread on Smartphones) model. To the best of our knowledge,\nthis is the first work to model smartphones epidemics. The major contributions of our\nwork are as follows.\n"
        },
        {
            "page_number": 221,
            "text": "216\nBO ZHENG et al.\nInternet\nAnother network many\nhops form the LAN\nCase B: A smartphohe moves from\none network to another\nCase A: A smartphohe belongs\nto both networks\nA well shielded\nLAN of a Company\nFigure 1. The mobility of smartphone makes itself belongs to both networks. Smartphones\nmay become the “Mobile shortcut\" between two networks, thus speed up the propagation\nof epidemic.\n1) We model epidemics spread in mobile networks while taking the aforemen-\ntioned two unique characteristics of smartphones into account. When smart-\nphone moves, these mobile and static shortcuts change accordingly, and we\nintegrate this dynamics into the factor of network topology in our model. Based\non the analysis of the smartphone features and spread of epidemics, the ESS\nmodel is a comprehensive SIR (Susceptible-Infected-Removed) model which\nconsiders the influence of the epidemic spread rate, the effect of the network\ntopology, the influence of the nomadism of smartphones. And we also model\nthe patching and death rate as well as the initial pre-patch in our proposal, which\nis also often ignored in previous models2.\n2 Smartphones are computationally powerful and flexibly programmable, thus patches or shields [16]\ntechnology can be applied to smartphones to fix the software vulnerabilities. Pre-patch means some nodes\ncan be patched between time of the software vulnerability release and the time of the corresponding epidemic\n"
        },
        {
            "page_number": 222,
            "text": "WIRELESS NETWORK SECURITY\n217\n2) With this ESS model, we solve the critical conditions problem for the fast die\nout of an epidemic on smartphones. We give the theoretical analysis on the\neffect of different parameters and summarize the importance of each parameter\nin the spread of epidemic in smartphones.\n3) Extensive simulations have been performed to verify the ESS model and the\ncritical condition. We have also performed some experiments to study the\neffect to the spread speed of nomadism and the nodes density. Based on these\ntheoretical analysis and experimental simulations, we give some guidance for\nprevention attacks on smart-phones.\nThe remainder of this chapter is structured as follows. The related work is intro-\nduced in Section 2. Then we present how epidemics spread on smartphones in section 3.\nBased on the study in section 3, the ESS model is proposed, and the effect of topology\non epidemic spread and the critical conditions for the fast die out of an epidemic are\nderived in section 4. Section 5 analyzes the proposed ESS model and the individual\nparameters that affect the propagation behavior of the epidemics. Simulations results in\ndifferent network topologies are found in Section 6. Section 7 summarizes the chapter\nand describes further directions to pursue.\n2.\nRELATED WORK\nMuch work about epidemic or virus spread model has been done in both physics\nfield [17, 20] and computer science field [22, 23, 24] The researchers in physics usually\nmodel the general case. Based on these physical models, the researchers of computer\nscience further study the spread of worm [22, 24], or viruses using the contact list [23]\n(e.g. email). Theyconsidermanyspecificparametersofcomputervirusesandcomputer\nnetworks. The two most well studied classes of epidemic models are SIS and SIR\nmodel [27]. In SIS model, individuals can only exist in two discrete states, namely,\nsusceptible and infected. When an infected individual is cured, it changes back to\nsusceptible one, just like the way of many diseases in the world. While in SIR model,\nindividuals can exist in three discrete states, namely, susceptible, infected and removed.\nWhen an infected individual is cured or dies, it becomes remove one. SIR model is\nsimilar to the condition in the Internet and smartphones, when a device is patched, it\nimmunes to the certain attack.\nWatts and Strogatz presented a simple infectious disease model in [28]. In their\nmodel, the contact infection rate is always 1; the nodes of the infection withdraw the\nsystem after a unit time. The study of its spread time shows that in the regular network,\nsmall world network and random network, the spread time is in direct proportion to the\nshortest path. This explained the function of the shortest path. With this model, the\ninfectious disease will break out in the whole network for any network. So they can\nnot derive the critical outbreak condition, which would be helpful for prevention.\nbased upon such vulnerabilities spread. The patch, shield and pre-patch rate have different impact on the\npropagation of the epidemics\n"
        },
        {
            "page_number": 223,
            "text": "218\nBO ZHENG et al.\nIn [18], Newman studies the percolation and epidemic model in small world model.\nThis proposal mapped the epidemic problem into a percolation problem, and found out\nthe threshold of the break out of the epidemic. If the probability of susceptible nodes\nis greater than the threshold, the epidemic will break out. The study shows that the\nthreshold in small world networks is much smaller than that in regular networks. In [21],\nthe author gives the threshold of arbitrary distribution of vertices degree. However, in\nthese models, there’re no special considerations on the mobile nodes which will affect\nthe propagation as we illustrated.\nIn [19], R. Pastor-Satorras and A. Vespignani study the case on the scale-free\nnetworks and point out that there is no similar threshold exists in infinite scale-free\nnetworks for the SIS or SIR model. In other words, once infectious disease occurs, it\nwill spread out in big scope. Therefore, only curing the inflected nodes is not enough,\nchanging the structure of the network is also needed. The typical method that breaks\nthe network structure is to quarantine or cut down some connections forcedly. The\nmodel for finite scale-free networks is studied in [20, 25, 26]. In such models, they\noften ignore the difference between patch, remove, death and pre-patch which leads to\ninaccuracy of the results.\nIn [22], the authors study the spread of active worm in the Internet. They consider\nthe characteristic of Internet worms such as hit-list, scanning rate, death rate and patch-\ning rate, but they assume the worms randomly scan the Internet to find the victims and\ndo not consider the influence of network topology.\nIn summary, all the previous models treat the network as a static one, focusing on\nthe influence of the distribution of the vertices degree, or network topology. They don’t\nstudy the effect of mobile nodes in the network as we discussed in the section I, which\nmotivate our work described in the following sections.\n3.\nHOW EPIDEMIC SPREAD ON SMARTPHONES\nIn this section, in order to model the propagation of the epidemic, we first describe\nhowtheepidemicspreadonsmartphonesandthenstudywhichparameterswillinfluence\nthe spread speed of the epidemics.\nWhen an epidemic is fired into the mobile Internet connected with many computers\nand movable smartphones and laptops, it attempts to send itself to vulnerable machines\nto infect them. The epidemics may spread from one infected device to its neighbors\nthrough the following ways.\nSome epidemics may disguise as some interesting game or useful software and\nbe published to the Internet waiting for some smartphone users to download\nand play;\nSome spread from PC to smartphones by sending epidemic-contained files\nthrough email as an attachment, or propagating while the smartphones syn-\nchronizing with PC;\nSimilarly, smartphons can infect the PCs using the same way;\n"
        },
        {
            "page_number": 224,
            "text": "WIRELESS NETWORK SECURITY\n219\nBetween smartphones, they can infect each other using Bluetooth, WiFi or other\nwireless connection to scan and spread epidemic to all its neighbors during its\nmovement;\nSmartphones can also send epidemic-contained files through email or MMS as\nan attachment to infect other smartphones in its contact list;\nThe epidemics can also use the neighbors as “hosts\" and infect the devices\nmulti-hops from it. For example, a PC-target epidemic may spread from a PC\nto a smartphone without damage it, and then infect other PCs when they contact\nwith the smartphone;\nMoreover, some epidemics which may infect both smartphone and PC will\nappear in the near future, because both smartphones and PCs use the similar\ncommon operating systems.\nAfter any of the aforementioned propagation successes, a copy of this epidemic is\ntransferred to the new device (smartphone or PC). This newly infected device then tries\nto infect other devices using the same way at a certain probability, which is influenced\nby some factors described in the remaining of this section. Hence, the epidemics can\nspread from PC to smartphone and then infect other PCs, and vice versa. Compromised\nsmartphones may also start attacks to the Internet, and then infect more smartphones.\nThe spread speed of the epidemic is influenced by many factors.\nIt’s mostly\ndetermined by its propagating attempt (or in other word determined by its codes),\ne.g., a worm spread much faster than a download Trojan horse. The topology of the\nnetwork also influence the epidemic spread greatly, as mentioned in the Section II,\nepidemic spread much faster in the scale-free network than in the regular network. The\nmobile nodes also influence the spread speed. If the infected node is a mobile device,\nsuch as a smartphone or laptop, acting as mobile shortcuts, it may move to another\nplace with the user and infect the computers and smartphones in other networks.\nCorrespondingly, there are some prevention approaches which can slow down\nthe spread speed of the epidemic.\nNowadays, the epidemics often come after the\nannouncement of vulnerability [30].\nAfter the announcement of vulnerability and\nrelated patch available, some cautious people pre-patch their smartphone or PC to\nmake their machine immune to this epidemic. The more devices are patched between\nthe announcement of vulnerability and the appearance of associated exploit code, the\nharder the epidemics spread. When the attack is detected, more people will try to slow\nit down or stop it. The patch or shield, which repairs the security vulnerability of the\ndevices, is widely used to defend against the epidemic. When an infected or vulnerable\nnode is patched, it becomes an invulnerable node. During the process of epidemic\nspreading, some nodes might stop functioning properly, crashed, or be shutdown or\nat least made offline by the users; all these make the infected nodes eliminated in the\nnetwork.\n"
        },
        {
            "page_number": 225,
            "text": "220\nBO ZHENG et al.\n4.\nEPIDEMIC SPREAD MODEL ON SMARTPHONES\nIn this section, we describe a comprehensive model of epidemic spread on smart-\nphones which considers the influence of the various factors mentioned in Section III.\nFor convenience we introduce the basic ESS model which considers the death rate,\npatch rate and the nomadism feature of smartphones at first. After that, we enhance\nthe basic ESS model with topology effect and mobile shortcuts effect, and present the\nfinal ESS model.\nTable 2 lists the parameters used in the spread of epidemic on the mobile network.\n4.1. The Basic ESS Model\nThe ESS model is a comprehensive model which combines the influence of the\nepidemic spread rate, patching and death rate, the effect of the network topology, as\nwell as the influence of the nomadism of smartphones.\nIn this chapter we focus on SIR (susceptible-infective-removed) model. In SIR\nmodels, a population of N individuals is divided into three states: susceptible (S),\ninfective (I), and removed (R). In this context “removed\" means individuals who are\neither recovered from the disease and immune to further infection, or dead.\nHowever, the traditional SIR model, such as Kermack- McKendrick model [31],\ndoesn’t consider the uniqueness of epidemics on the computers and smartphones such as\npatch rate. According to the characteristics of the spread of epidemic on computers and\nsmartphones, “removed\" means either vulnerable nodes (includes the infected nodes)\nare patched and immune to further infection, or infected nodes die and eliminated from\nthe network. We use d to denote the death rate and p to denote patch rate.\nIf an infected node moves to other network clusters, it may infect nodes in those\nclusters. The increase of inflected node should plus the nodes that are infected because\nof the movement. We use φ to denote the density of mobile shortcuts, and use m(t) to\ndenote the average move speed (move times in unit time) of mobile nodes at time t.\nLet S(t), I(t) and R(t) denote the proportion of vulnerable nodes, the proportion\nof infected nodes and the proportion of removed nodes at time t (t ≥0) respectively,\nand use S′, I′, R′ to denote the increment of S(t), I(t), R(t) (i.e. S′(t), I′(t), R′(t),\nthe derivative of S(t), I(t), R(t)). The infective nodes contact with randomly chosen\nnodes of all states at an average rate α per unit time. At the beginning of the epidemic\nspread I(0) = I0, and 0 < I0 ≪1 is a very small proportion of the total number of\nvulnerable nodes.\nAssumes that the nodes are fully mixed, meaning that the individuals with whom a\nsusceptible individual has contact are chosen at random from the whole nodes (the effect\nof network topology will be considered in the following subsection), all individuals\nhave approximately the same number of contacts at the same time, and that all contacts\ntransmit the disease with the same probability. In the time t the newly infected nodes\nbecause of normal contact is αS(t)I(t). At the same time, some mobile node move\nto other places and infect some more nodes, the newly infected nodes because of the\nnomadic nodes is αS(t)I(t)φm(t). Because of patch and death (d + p)I(t) infected\nnodes and pS(t) susceptible nodes are removed. Then we get the basic ESS model:\n"
        },
        {
            "page_number": 226,
            "text": "WIRELESS NETWORK SECURITY\n221\nTable 2. Notions Used in The Model\nNotion\nExplanation\nt\nTime\nS, S(t)\nProportion of susceptible nodes\nI, I(t)\nProportion of infected nodes\nR, R(t)\nProportion of removed nodes (patched or dead)\nS′, I′, R′\nThe derivative of S(t), I(t), R(t), i.e. S′(t), I′(t), R′(t)\nS0, I0, R0\nInitial value of S(t), I(t), R(t), i.e. S(0), I(0), R(0)\nSk(t)\nProportion of susceptible nodes in the group of vertices de-\ngree k\nIk(t)\nProportion of infected nodes in the group of vertices degree\nk\nRk(t)\nProportion of removed nodes in the group of vertices degree\nk\nN\nTotal vulnerable nodes\nλ\nInfected probability of each link\nα\nEpidemic spread speed (i.e. contacted rate)\nd\nDeath rate\np\nPatching rate\nt0\nThe average time between the announcement of vulnerability\nand the appearance of associated exploit code\np0\nPatching rate during t0\nφ\nThe density of “mobile shortcut\"\nm, m(t)\nThe average move speed (move times in unit time) of mobile\nnodes\nP(k)\nVertices degree distribution of the whole network\nQ(k)\nVertices degree distribution of the mobile vertices\n"
        },
        {
            "page_number": 227,
            "text": "222\nBO ZHENG et al.\n⎧\n⎪\n⎪\n⎨\n⎪\n⎪\n⎩\nS′ = −αS(t)I(t) −αS(t)I(t)φm(t) −pS(t)\nI′ = αS(t)I(t) + αS(t)I(t)φm(t) −(d + p)I(t)\nR′ = pS(t) + (d + p)I(t)\n0 < S(t), I(t), R(t) < 1, S(t) + I(t) + R(t) = 1, α, φ, d, p > 0.\n(1)\nWhen I′ = αS(t)I(t) + αS(t)I(t)φm(t) −(d + p)I(t) < 0, the epidemic will\ndie out, assume the mobile nodes move at uniform velocity m, and now the sufficient\ncondition of epidemic dies out is α(1+φm)S(t)\nd+p\n< 1, or α(1+φm)(1−I(t)−R(t))\nd+p\n< 1.\nAs mentioned before, some nodes are pre-patched between the announcement of\nvulnerability and the appearance of associated exploit code. We denote this period of\ntime as t0, and the pre-patch rate in this period is p0, and R0 = p0t0. If we want to\nrestrict the spread of epidemic from the beginning, α(1+φm)(1−I0−R0)\nd+p\n< 1 should be\nsatisfied.\nThe epidemiological threshold is now:\nα(1 + φm)(1 −I0 −p0t0)\nd + p\n< 1.\n(2)\nIf the sufficient condition (2) is satisfied, the epidemic will die out and not spread\nall over the network. Usually, at the beginning of virus spread, the infected nodes are\na very small set, i.e. I0 is very small and can be ignored in (2).\n4.2. The Extended ESS Model for Smartphones\nBoth the topology of network and the nomadism of mobile nodes can influence the\nspread of the epidemic. We will enhance the basic ESS model (1) by adding the effect\nof topology and the effect of nomadism in this subsection.\nEffect of Topology on the Spread of Epidemic\nIn our model, we assume a connected network G = (N; E), where N is the number\nof nodes in the network and E is the set of edges. The edges of a node are the set of\nlinks to nodes with whom the node may have contact during the time it is infective,\nsuch as the devices that in the same subnet, in the email or phone call contact list, next\nto the node and can build up a Bluetooth connection, and so forth.\nSo we can vary the number of connections of each node by choosing a particular\ndegree distribution for the network. We use λ to denote the infected probability of each\nlink (assumes that λ is a universal infection rate for each edge connected to an infected\nnode and independent with the vertex degree of the node).\nLet us assume initially that the vertices degree distribution is P(k). For the group\nof vertices that have the same vertex degree k, in time t the proportion of newly infected\nnodes because of normal contact is now: λkSk(t)Θ(t), where Ik(t) and Sk(t) denote\n"
        },
        {
            "page_number": 228,
            "text": "WIRELESS NETWORK SECURITY\n223\nthe probability of infected nodes and the probability of susceptible nodes in the group\nof vertex degree k, and Θ(t) is the probability of a randomly chosen link has an infected\nnode and a susceptible node in each side, since the epidemic only spreads through such\ntype of links. Because the node has higher vertex degree is more possible to have an\ninfected node connect to it, we get:\nΘ(t) = 1\nk\n\u0001\nk\nkP(k)Ik(t)\n\u0001\nk\nP(k)Sk(t),\n(3)\nwhere k is the mean degree of the network, k = \u000e\nk kP(k).\nThe number of infected nodes in the group of vertices degree k been removed in\ntime t because of death and patch is: (d + p)Ik(t)\nEffect of Nomadism on the Spread of Epidemic\nAs mentioned before, when an infected node moves to other network clusters, it\nmay infect other nodes in the clusters. It becomes a mobile shortcut between the two\nnetwork clusters (although it may not connect to the first cluster now, but the epidemic\nhas been taken to the second cluster by it). The increase of inflected node should plus\nthe nodes that are infected because of the movement.\n∆Im,k(t) = λkSk(t)Θm(t),\nΘm(t) =\n1\nk\n\u000e\nk kP(k)Pm(k)Ik(t) \u000e\nk P(k)Sk(t),\n(4)\nwhere ∆Im,k(t) is the incensement of infected nodes in the group of vertex degree\nk in time t because of the movement of the mobile nodes, and Θm(t) is the probability\nof a randomly chosen link connecting an infected mobile node and a susceptible node\nafter movement. Moreover, Pm(k) = φm(t)Q(k)\nP (k)\n, where Q(k) is the distribution of\nthe degree of the movable nodes (The probabilities of mobile nodes are different in\ndifferent group of degree. The leaf nodes of a network which has a small vertex degree\nmay have a higher probability to be a mobile node. But the kernel nodes of the network\nwhich connect a lot of nodes are more likely to be static nodes. There are no such\nprevious researches on the distribution of vertices degree of mobile networks; we use\nsome hypothetic distribution in this chapter).\nNow, we get the ESS model consider both the effect of topology and the effect of\nmobile shortcuts on the spread of epidemic:\nI′\nk = λkSk(t)(Θ(t) + Θm(t)) −(d + p)Ik(t).\n(5)\nRedefine Θ as\nΘ = 1\nk\n\u0001\nk\nkP(k)[1 + Pm(k)]Ik(t)\n\u0001\nk\nP(k)Sk(t).\n(6)\nThen the ESS model (5) becomes\nI′\nk = λkSk(t)Θ(t) −(d + p)Ik(t).\n"
        },
        {
            "page_number": 229,
            "text": "224\nBO ZHENG et al.\nEventually, we get the final ESS model\n⎧\n⎪\n⎪\n⎨\n⎪\n⎪\n⎩\nS′\nk = −λkSk(t)Θ(t) −pSk(t)\nI′\nk = λkSk(t)Θ(t) −(d + p)Ik(t)\nR′\nk = dIk(t) + p(Sk(t) + Ik(t))\n0 < Sk(t), Ik(t), Rk(t) < 1, Sk(t) + Ik(t) + Rk(t) = 1, λ, d, p > 0.\n(7)\n5.\nANALYSIS OF THE ESS MODEL\nIn this section, we give the critical conditions for the fast die out of an epidemic\nfrom our model. We also give the analysis on the effect of different parameters and\nsummarize the importance of each parameter in the spread of epidemic in smartphones.\n5.1. A Critical Condition for Epidemic Fast Die Out\nIn this section, we want to find out the epidemiological threshold from (7).\nBecause Sk(t) + Ik(t) + Rk(t) = 1 applying it to formulation (7), we get the\nfollowing equations:\nI′\nk\n=\nλkSk(t)Θm(t) −(d + p)(1 −Sk(t) −Rk(t)),\n(8)\nI′\nk\n=\nλk(1 −Ik(t) −Rk(t))Θm(t) −(d + p)Ik(t).\n(9)\nTo find out the critical condition, let (8)=0 and (9)=0, then\nSk(t)\n=\n(d + p)(1 −Rk(t))\nd + p + λkΘ\n,\n(10)\nIk(t)\n=\nλk(1 −Rk(t))Θ\nd + p + λkΘ\n.\n(11)\nApply (10) and (11) to (6) we get the following equation:\nΘ\n=\n1\nk\n\u0001\nk\nP(k)[1 + Pm(k)]λk(1 −Rk(t))Θ\nd + p + λkΘ\n\u0001\nk\nP(k)(d + p)(1 −Rk(t))\nd + p + λkΘ\n,\n0 < Θ ≤1.\nWe can find that when the epidemic just appears (t →0) or a very long period\nafter the epidemic begin to spread (t →∞), Θ →0. Using Taylor expansion, when\nΘ →0\nΘ = {\n1\nk(d + p)\n\u0001\nk\nλk2P(k)[1 + Pm(k)][1 −Rk(t)]\n\u0001\nk\nP(k)[1 −Rk(t)]}Θ\n+ AΘ2 + · · ·\n"
        },
        {
            "page_number": 230,
            "text": "WIRELESS NETWORK SECURITY\n225\nThen, the critical condition can be got as follows.\nλ\nk(d + p)\n\u0001\nk\nk2P(k)[1 + Pm(k)][1 −Rk(t)]\n\u0001\nk\nP(k)[1 −Rk(t)] = 1.\n(12)\nWhen the epidemic just appears (t →0), the proportions of removed nodes in every\ngroup are almost equal, we can get λ(1−R0)2\n(d+p)k\n\u000e\nk k2P(k)[1 + Pm(k)] = 1, since we\nhave R0 = p0t0, then\nλ(1 −p0t0)2\n(d + p)k\n\u0001\nk\nk2P(k)[1 + Pm(k)] = 1.\n(13)\nBecause Pm(k) = φm(t)Q(k)\nP (k)\n, assume the mobile nodes move at uniform velocity m,\nthen the critical condition (13) becomes\nλ(1 −p0t0)2\n(d + p)k\n\u0001\nk\nk2[P(k) + φmQ(k)] = 1.\n(14)\nAnd \u000e\nk k2P(k) = k2 = k\n2 + Dev(k), we get\nλ(1 −p0t0)2\n(d + p)k\n[k\n2 + Dev(k) +\n\u0001\nk\nφmQ(k)] = 1.\n(15)\nThis critical condition shows that if two networks have the same mean vertices degree,\nthe network which has larger deviation Dev(k) is more vulnerable to virus spread.\n5.2. Analysis on the Effect of Different Parameters\nAfter getting the critical condition of the spread of epidemic in mobile networks,\nwe would like to analyze the model and find out the influence of each parameter, and\nfind efficient defensive way further.\nIn SIR model, the state of “Removed\" includes the dead nodes and the patched\nnodes. Although both of them are removed from the flow of propagation, they have\ntotally different features. The patched nodes are healthy nodes, but the dead ones mean\nthat the damage already taken. Hence, the prevention of epidemic should not only\nprevent the epidemic from spread but also make the patched ratio as higher as possible.\nFrom the ESS model (7), the increment of the proportion of the patched at time t\nis p(Sk(t) + Ik(t)). It’s hard to give the exact solution of the final patched ratio when\nthe epidemic levels off, in the performance evaluation section, we do experiments to\ngive some numerical results.\nIn the viewpoint of epidemiology, the defense ways of disease can be divided into\nthree kinds: prophylaxis, quarantine and cure. In the following part, we’ll analyze the\n"
        },
        {
            "page_number": 231,
            "text": "226\nBO ZHENG et al.\nparameters in (14) and present relative defense ways. For convenience, the analyses of\nall parameters are listed in Table 3.\nIn the critical condition (14), we can see that it’s affected by network topology,\npatch rate and death rate, as well as pre-patch ratio. In the equation, \u000e\nk k2[P(k) +\nφmQ(k)]/k reflects the influence of topology. We call this factor Topology Factor and\ndenote it as T.\nInside the Topology Factor there are following parameters. The first one is the\ndistribution of the vertices degree k in the whole network (P(k)), which is determined\nby the topology of the whole network. Network topology can be changed to restrict\nthe spread of epidemic. The effective way includes changing the routing table, setup\nfirewall, quarantining infected subnet, or setting black list in the gateways, etc. The\nsecond parameter is the distribution of the “mobile vertices\" (Q(k)). The number of\nmobile device increase rapidly in recent years, it makes the density of mobile device in a\ncertain area increase rapidly too. And with the development of wireless technology, the\nsmartphones will connect more and more mobile devices within a single hop. All these\nwill increase the mean value of the degree distribution of the mobile vertex. It’s hard\nto restrict the spread of epidemic by changing Q(k), unless we limit the access right of\nmobile node to the Internet. The third parameter inside topology factor is the density of\nmobile shortcuts in the whole network. The more nodes are mobile nodes the faster the\nvirus may spread. The mobile nodes increase the mix degree of the whole vertices and\ncause the virus spread faster. Therefore, the frequent movements make the infective\nmobile nodes spread the epidemic widely. According to the trend of smartphone and\ncomputer market, the influence of mobility will become more and more significant.\nIn the equation (14), there are some other parameters affecting the critical condition\nincluding the death rate and patch rate. The higher death rate (d) is, the more hardly\nthe worm spread. But death means that the infected nodes may be crashed or unable\nto access the Internet, which is we unwilling to see (And most worms do not crash the\ncomputer, Witty was the first widely propagated Internet worm to carry a destructive\npayload, it tries to destroy the system after sending 20,000 packets). Patch rate influ-\nences the denominator of the critical condition, the higher the better. We can also see\nthat if we only patch the vertices after the virus appear and take no other prophylactic\ntreatment , the virus will spread out unless (d + p) is greater than λT. It means that\nto avoid the virus spread, (d + p) should at least be the same magnitude as the virus\nspread rate λ, while this would be very difficult when we suffer from fast-spreading\nworms where (d + p) will not catch up with the worm’s spread speed. Moreover, even\nif we have a very high speed patch method to satisfy the condition ((d + p) >= λT),\nthey may still cause congestion in the network just like there are two kind of worm\nspreading in the same period (Thinking about the way using AntiBlaster to remove the\nBlaster).\nAs we can see in the formulation of the critical condition, there is (1 −p0t0)2 in it.\nIt’s not a linear change when we increase this pre-patch ratio (p0t0). Hence, pre-patch\nis very important for preventing the spread of epidemic. However, we can not explicitly\ncontrol the time (t0) between the announcement of vulnerability and the appearance\nof associated exploit code since it is determined by the exploit coding difficulty and\n"
        },
        {
            "page_number": 232,
            "text": "WIRELESS NETWORK SECURITY\n227\nTable 3. The Effect of Different Parameters\nParameters\nDescriptions\nActions needed\nto defense\nepidemics\nEffectiveness a\nλ\nThe spread rate of the epi-\ndemic in each link.\nDecrease\nModerate, but hard to ad-\njustb\nT\nThe total effect of topology,\nincluding the effect of static\nand mobile topology\nDecrease\nModerate, inconvenience to\nadjust,\nquarantine is the\ncommon way\nP(k)\nThe distribution of the ver-\ntices degree k of the whole\nnetwork\nDecrease\nmean\nvalue and stan-\ndard deviation\nModerate. Some topologies\nmay cause the T tending to\ninfinite\nQ(k)\nThe degree distribution of\nthe “mobile vertex\".\nDecrease\nmean\nvalue and stan-\ndard deviation\nModerate. Some topology\nmay cause the T tending to\ninfinite\nφ\nThe percentage of “mobile\nshortcuts\" in the whole net-\nwork.\nDecrease\nModerate.\nHard to manu-\nally control it, and it would\nbecome larger since num-\nber of smartphones will in-\ncrease\nm\nThe move speed (average\nmove times in unit time).\nDecrease\nModerate\nd\nThe death rate. Determined\nby the function way of epi-\ndemic and the action of peo-\nple.\nIncrease\nModerate,\nbut increasing\ndeath rate takes much dam-\nage\np\nPatch rate after the virus ap-\npeared\nIncrease\nModerate, There’re some\napproaches to increase it,\nwhile having difficultiesc.\nt0\nThe time between the an-\nnouncement of vulnerabil-\nity and the appearance of as-\nsociated exploit code.\nIncrease\nGreat, but hard to increase,\nsince it ’s determined by the\ncoding difficulty, the virus\nmaker’s interest, etc.\np0\nPatching rate during t0\nIncrease\nGreat, we have effective\nways to increase p0\na The effectiveness of each parameter to prevent the epidemic from spread out.\nb The spread rate of the epidemic is mostly determined by the exploit code and the capability of victim\ndevices, little can be done to reduce it.\nc Some factors slow down the increase of patch rate after the virus appears. A) Patch needs more tests\nand evaluation before it’s installed. B) Users may not patch their computer timely due to lack of professional\nskills or poor network conditions, but viruses exhaust the system to spread itself very fast. C) The patch size\nis usually larger than the size of virus and there often exists bottleneck at the patch servers, they would also\nslow down the patch speed.\n"
        },
        {
            "page_number": 233,
            "text": "228\nBO ZHENG et al.\nthe virus maker’s interest, etc. However, we can increase the pre-patch speed (p0) and\nmake most of the devices immune the epidemic from the beginning for smartphones,\ne.g, mobile network operators can enforce patch to their subscribers.\nIn summary, as analyzed in Table 3, changing the topology of network can influence\nthespreadofepidemic, butit’shardandbringsinconveniencetothequarantineddevices.\nReducing spread rate λ and increasing death rate d can have moderate effect to slow\ndown the spread of epidemic, but it’s hard to adjust them explicitly and dependant on\nthe user behavior. Patch is a good way, but after the epidemics start propagation, the\npatch rate might not be higher enough to stop the spread. Pre-patch takes great effect\non preventing the spread of epidemic and there’re some effective ways to increase the\npre-patch speed.\n6.\nPERFORMANCE EVALUATION\nWe have performed extensive simulation of epidemic spread to validate the ESS\nmodel and check our analytic results, and to investigate further the behavior of the\nmodels under typical network topologies including small world network, Waxman\nrandom network, and power-law network. We have also conducted some experiments\nto analyze the effect of individual parameters in the proposed ESS model.\nThe simulations are performed using Matlab. In order to compare the simulation\nresults with the analysis results, firstly, we educe the discrete-time ESS model. And\nthen we use Matlab to generate some network topologies and simulate the spread of\nepidemic in these networks. Finally, we compare the simulation results with the analysis\nresults in the same topology or the critical condition we derived from the equation (14).\nThe process of simulations will be described amply in the following subsections.\n6.1. Comparing ESS Model with Simulation Result\nIn order to perform numerical calculations, we transform the continuous-time ESS\nmodel (7) into discrete-time ESS model. Let Sk,t, Ik,t and Rk,t denote the number\nof vulnerable nodes, the number of infected nodes and the number of removed nodes\nin the group of vertex degree k at time tick t(t ≥0) respectively. For the group of\nvertices that have the same vertex degree k, from time tick t to time tick t+1 the newly\ninfected nodes because of normal contact is λkSk,tΘ, where Θ =\n1\nk\n\u000e\nk k[P(k) +\nφmQ(k)]Ik,t\n\u000e\nk P(k)Sk,t. And the number of infected nodes in the group of vertex\ndegree k been removed after time stick t because of death and patch is (d + p)Ik,t.\nThen we get the discrete-time ESS model (For convenience we only list the infected\npart):\nIk,t+1 = Ik,t + λkSk,tΘ −(d + p)Ik,t\n(16)\nWe begin each simulation with a set of randomly chosen infected nodes and a\nset of randomly chosen pre-patched nodes on a given network topology (the number\nof initially-infected nodes and pre-patched nodes does not affect the equilibrium of\nthe propagation). And a set of randomly chosen mobile nodes are also initialized\n"
        },
        {
            "page_number": 234,
            "text": "WIRELESS NETWORK SECURITY\n229\naccording to the “density of mobile shortcut\" φ and the “degree distribution of mobile\nnodes\" Q(k). Simulation proceeds in steps of one time unit. During each step, an\ninfected node attempts to infect each of its neighbors with probability λ, and a mobile\nnode moves to another network cluster with speed (probability) m. In addition, every\nnode is patched with probability p, and every infected node is dead with probability d.\nSmall World Network\nThesmallworldphenomenonisthetheorythateveryoneintheworldcanbereached\nthrough a short chain of social acquaintances. Previous study of many researchers\ndiscovered that many real networks have such small world phenomenon. Watz and\nStrogatz define the following properties of a small world graph[28]:\n1. The clustering coefficient C is much larger than that of a random graph with\nthe same number of vertices and average number of edges per vertex.\n2. The characteristic path length L is almost as small as L for the corresponding\nrandom graph. C is defined as follows: If a vertex v has kv neighbors, then\nat most kv ∗(kv −1) directed edges can exist between them. Let Cv denote\nthe fraction of these allowable edges that actually exist. Then C is the average\nover all v.\nFigure 2 shows the simulation result compared with the analysis result derived\nfrom our model. The results are rather satisfied: our model yields precisely to the\nsimulation result which demonstrates the effectiveness of the ESS model.\nPower-law Network\nPower-law networks [29, 33, 34], including current Internet, are characterized by\nan uneven distribution of connectedness. The nodes in these networks do not have\na random pattern of connections, instead, some nodes act as “very connected\" hubs,\nwhich dramatically influences the way the network operates.\nThe degree distributions of power-law networks have power-law tails, i.e. P(k) ∼\nk−τ, typically 2 < τ ≤3.\nFigure 3 shows the time evolution of epidemic in a power-law network (generated\nusing Inet model [35]). Our model conforms very close to the simulation results.\nWaxman Random Network\nIn the random network, a (fixed) set of nodes is distributed in a plane uniformly\nat random. A link is added between each pair of nodes with a certain probability. The\nWaxman method [36] is an instantiation of this method where the probability of adding\na link is given by:\nP(u, v) = αe−d/βL, where 0 < α, β ≤1, d is the Euclidean distance from node\nu to v, and L is the maximum distance between any two nodes.\n"
        },
        {
            "page_number": 235,
            "text": "230\nBO ZHENG et al.\n0\n500\n1000\n1500\n2000\n0\n500\n1000\n1500\n2000\n2500\n3000\n3500\n4000\nTime tick\nNumber of Infected Nodes\nESS(Susceptible)\nESS(Infected)\nESS(Removed)\nSimulation(Susceptible)\nSimulation(Infected)\nSimulation(Removed)\nFigure 2. Simulation result on small world network with vertices number N = 4000, and\naverage vertices degree is 6.0. λ = 0.0025, m = 0.1, d = 0.001, p = 0.001, φ = 0.02\nand p0t0 = 0.\n0\n500\n1000\n1500\n2000\n0\n500\n1000\n1500\n2000\n2500\n3000\n3500\n4000\nTime tick\nNumber of Infected Nodes\nESS(Susceptible)\nESS(Infected)\nESS(Removed)\nSimulation(Susceptible)\nSimulation(Infected)\nSimulation(Removed)\nFigure 3. Simulations on power-law network with vertices number N = 4000, and average\nvertices degree is 3.3218. λ = 0.0015, m = 0.0005, d = 0.001, p = 0.001, φ = 0.02\nand p0t0 = 0.\n"
        },
        {
            "page_number": 236,
            "text": "WIRELESS NETWORK SECURITY\n231\n0\n500\n1000\n1500\n2000\n0\n500\n1000\n1500\n2000\n2500\n3000\n3500\n4000\nTime tick\nNumber of Infected Nodes\nESS(Susceptible)\nESS(Infected)\nESS(Removed)\nSimulation(Susceptible)\nSimulation(Infected)\nSimulation(Removed)\nFigure 4. Comparison of ESS Model and simulation result on Waxman network with\nvertices number N = 4000, and average vertices degree is 15.5520. λ = 0.002, m =\n0.001, d = 0.001, p = 0.001, φ = 0.02 and p0t0 = 0.\nThe Waxman random network is much like some wireless network, such as Blue-\ntooth and WiFi network, where a mobile device only connects to its neighbors within a\ncertain distance using their wireless connection. Studies in this kind of network topolo-\ngies are helpful to learn the spread of epidemic on smartphones, because nowadays a\nlot of smartphone worms spread through Bluetooth connections.\nFigure 4 shows the simulation result compared with the analysis result derived\nfrom our model. Our model conforms very close to the simulation results.\n6.2. Critical Condition Verification\nIn this subsection, we measure the point of epidemic threshold for the fast die out of\nepidemic for comparison with our analytic results. We generate the network topology\nand simulate the spread of epidemic in these networks with Matlab. The procedure of\nspread of epidemic is simulated as follows. Each infected node scans its neighbors one\nby one, and randomly infects them. Mobile nodes will move to another cluster, and all\nnodes will be patched, or die if infected in some random way. We then calculate at each\nstep the size of infected nodes, susceptible nodes and removed nodes. The position of\nthe percolation threshold can then be estimated from the point at which the derivative of\nthis size with respect to the number of infected nodes takes its maximum value. Since\nthere are N nodes on the network in total and the action of infecting, moving, patching\nand death takes time O(N), such simulation runs in time O(N 2).\n"
        },
        {
            "page_number": 237,
            "text": "232\nBO ZHENG et al.\n0\n200\n400\n600\n800\n1000\n20\n40\n60\n80\n100\n120\n140\nTime tick\nNumber of Infected Nodes\np=0.0016\np=0.0018\np=0.0020\np=0.0022\nFigure 5. Simulations on the small world network. All case for N = 10000 vertices\nnetwork, with the “small world\" short cut density of 0.01. The average vertices degree\nk = 8, and the Topology Factor T = 8.0037, λ = 0.0004, p0t0 = 0, φ = 0.01,\nm = 0.01, and p = 0.0016, 0.0018, 0.002 and 0.0022 (form top to bottom).\nIn order to make the figures clear, we only change one parameter in one group of\nsimulations. According to the simulation setup, we can calculate the critical condition\nof the chosen parameter in advance using equation (14). And then, we choose a set of\nvalue near the critical condition to perform the simulations, and the figure of simulation\nresults can reflect the real critical condition. All the experiments validate the critical\ncondition (14) we presented in section 5.\nSmall World Network\nIn Figure 5 we show the simulation result on a small world network which has\nN = 10000 vertices and the static shortcut density is 0.005. The average vertices\ndegree k = 8, and the Topology Factor T = 8.0037. In these simulations we randomly\nchoose 1% of nodes as mobile nodes i.e. φ = 0.01, and fix λ = 0.0004, p0t0 = 0,\nm = 0.01, d = 0.001, and do the simulation in condition of p = 0.0016, 0.0018,\n0.002 and 0.0022. Following the critical condition (14), when the patch rate p is equal\nto 0.00220148 the critical condition equals to 1. The bottom one is very close to the\nepidemic threshold. As we can see, the number of infected nodes of the epidemic for\np = 0.0022 does not increase from the beginning and then peter out because of patch\nand death. Once we get above the epidemic threshold a large number of cases appear\nand then peter out slowly because of death and patch.\n"
        },
        {
            "page_number": 238,
            "text": "WIRELESS NETWORK SECURITY\n233\n0\n500\n1000\n1500\n2000\n40\n50\n60\n70\n80\n90\n100\n110\n120\nTime tick\nNumber of Infected Nodes\nd=0.003\nd=0.004\nd=0.005\nd=0.006\nFigure 6. Simulation on the power-law network with vertices number N = 10000, av-\nerage vertices degree k=4.1151, and Topology Factor T = 19.8087. λ = 0.0004,φ =\n0.02,m = 0.001, p = 0.003, p0t0 = 0, and d = 0.003, 0.004, 0.005 and 0.006 (form\ntop to bottom).\nPower-law Network\nWe generate a power-law network using Inet model. The network has N = 10000\nnodes, the average vertices degree k = 4.1151, and the Topology Factor T = 19.8087.\nIn these simulations we fix λ = 0.0004, m = 0.001, p = 0.003, φ = 0.02, p0t0 = 0%,\nand then simulate in the condition of death rate d = 0.003, 0.004, 0.005 and 0.006.\nFollowing the critical condition (14), if the death rate d is lower than 0.00049235 the\nepidemic will break out. The last one is below the epidemic threshold and the third\none is very close to the epidemic threshold. As Figure 6 shows, the number of infected\nnodes of the epidemic for d = 0.006 die out fast, and while others increase first then\ndecrease because patch and death.\nWaxman Random Network\nFigure 7 shows the simulation result of the infected nodes number as a function of\ntime on a Waxman random network which has N = 10000 vertices, average vertices\ndegree k = 6.5195, and Topology Factor T = 10.009.\nIn these simulations we\ntake the density of mobile shortcut φ = 0.02, death rate d = 0.0002, patch rate\np = 0.003, mobile nodes move speed m = 0.001, and no pre-patch applied (i.e.\np0t0 = 0). The curves in the figure have (from bottom to top) the epidemic spread rate\nper link λ = 0.0004, 0.0005, 0.0006 and 0.0007, which implies, following (14), that the\n"
        },
        {
            "page_number": 239,
            "text": "234\nBO ZHENG et al.\n0\n200\n400\n600\n800\n1000\n0\n50\n100\n150\n200\n250\nTime tick\nNumber of Infected Nodes\nλ=0.0004\nλ=0.0005\nλ=0.0006\nλ=0.0007\nFigure 7. Simulation on the random Waxman network with vertices number N = 10000,\naverage vertices degree k = 6.5195, and Topology Factor T = 10.009. φ = 0.02,\nm = 0.001, d = 0.002, p = 0.003, p0t0 = 0, and λ = 0.0004, 0.0005, 0.0006 and\n0.0007 (form bottom to top).\nepidemicwillbreakwhiletheepidemicspreadrateperlink λisgreaterthan0.00049955.\nOnly the bottom one is below the epidemic threshold. As we can see, the number of\ninfected nodes of the epidemic for λ = 0.0004 shows that the epidemic die out directly\nwithout getting more nodes infected. And the second curve with λ = 0.0005, which\nis almost equal to the epidemic threshold, maintains its infected nodes number for\na little while and then peter out because death and patch. Once we get above the\nepidemic threshold, the number of infected nodes increases, which indicating the onset\nof epidemic behavior.\n6.3. The Effect of Mobile Shortcut Density and Move Speed\nThe previous researches consider either influence of the distribution of the vertices\ndegree or the density of shortcut. They don’t study the mobility of the nodes in the\nnetwork. But the density of mobile shortcut influence greatly on the spread of epidemic.\nHence, the previous models not accurate enough on the network of smartphones.\nFigure 8 illustrates the simulation result of the spread of epidemic in a small world\nnetwork with the total nodes of N = 1000, the density of static shortcut is 0.01,\naverage vertices degree is 6, and the move speed of mobile nodes m = 0.3. We start\nthe simulation at an initial state of zero mobile nodes in the network and then increase\nthe mobile shortcut density step by step. As we can see, the traditional SIR model (i.e.\nφ = 0, the bottom curve) is inaccurate in mobile condition: the number of infected\n"
        },
        {
            "page_number": 240,
            "text": "WIRELESS NETWORK SECURITY\n235\n0\n50\n100\n150\n200\n0\n50\n100\n150\n200\n250\n300\n350\nTime tick\nNumber of Infected Nodes\nTraditional SIR model \nφ=0.01\nφ=0.05\nFigure 8. The effect of mobile shortcut on a small world network with total nodes number\nN = 1000, the density of static shortcut is 0.01, the average vertices degree k = 6,\nmove speed of mobile nodes is m = 3, and mobile shortcut density φ = 0.01, and 0.05,\ncompared with traditional SIR model.\nnodes increase slower than real condition in mobile network, even a very few (1%)\nmobile shortcuts exist in the network.\nFigure 9 illustrates the simulation result of the same network topology with mobile\nshortcut density φ = 0.05. We speed up the move speed of mobile nodes step by step.\nThe simulations show that with the increase of m from 0 to 0.9, at first, the spread\nspeed of the epidemic increase very fast and then slow down. It’s because that with\nthe movement of mobile nodes, they act as shortcut between deferent network clusters,\nthis greatly decrease the diameter of the network; when the nodes move faster there\nare some mobile shortcuts duplicated, and this makes the increase of epidemic spread\nspeed slow down.\n6.4. The Effect of the Uptrend of Peering Spread of Smartphone\nIn these years, more and more smartphones connect to the Internet. The number of\nsmartphones may even exceed the number of computers in the foreseeable future. As\nmentioned before the epidemics can spread from one smartphone to another locally with\nBluetooth or WiFi connection, what will happen with the rapid growth of smartphone?\nThis uptrend firstly increase the density of mobile shortcut density φ directly. And with\nmore and more smartphone in the world, the density of smartphone in a certain area\nwill increase too. Hence, the uptrend of smartphones will also increase the connectivity\ndegree of most nodes.\n"
        },
        {
            "page_number": 241,
            "text": "236\nBO ZHENG et al.\n0\n50\n100\n150\n0\n0.2\n0.4\n0.6\n0.8\n1\n0\n50\n100\n150\n200\nTime tick\nm\nNumber of Infected Nodes\nFigure 9. The effect of mobile shortcut on a regular network with nodes number N = 1000,\ndegree of each node k = 6, mobile shortcut density φ = 0.05, and move speed of mobile\nnodes m = 0, 0.1, 0.2 . . . 0.9.\nWe simulate this uptrend in the Waxman random networks with fixed maximum\ndistance L and fixed connection probability α, β between any two nodes. We assume\nthat the initial static nodes and mobile nodes are 90% and 10% and the total number\nof the nodes is initialized as 2000. The increase of static nodes and mobile nodes are\n10% and 30% and all these nodes are placed in a 10x10 area. The configuration of each\nexperiment is listed in Table 4.\nIn Figure 10 we show the results of the mean degree and the Topology Factor of\nthe networks as a function of the nodes density on Waxman Random networks. We\ncan see that they are two straight lines in the logarithmic axes. It means that the mean\ndegree and the Topology Factor are power-law functions of the nodes density (and\nnodes density is an exponential function of time).\nFigure 11 shows the simulation results of the relationship of nodes density and\nthe spread of epidemic. The spread speed increase very fast when the nodes density\nincreases. Hence, the threat of epidemic will be more and more serious with the uptrend\nof smartphones in the future.\n6.5. Protection of Epidemic\nThe purpose of the studies of epidemic is to provide some guides to prevent the\nspread of epidemic in the future. As mentioned in Section “Analysis of the ESS Model\",\nin the “Removed\" nodes, only the patched nodes are healthy. Hence, the defense of\nepidemic should not only prevent the epidemic from spread but also make the patched\nratio as higher as possible.\n"
        },
        {
            "page_number": 242,
            "text": "WIRELESS NETWORK SECURITY\n237\n10\n1.4\n10\n1.5\n10\n1.6\n10\n1.7\n10\n1.8\n10\n0\n10\n1\n10\n2\nNodes Density\nMean degree\nTopology Factor\nFigure 10. Mean degree and Topology Factor are power-law functions of nodes density.\n0\n1000\n2000\n3000\n4000\n5000\n20\n40\n60\n80\n0\n0.1\n0.2\n0.3\n0.4\n0.5\nTime tick\nNodes Density\nProportion of Infected Nodes\nFigure 11. The effect of nodes density on Waxman random network with the configurations\nlisted in Table 4.\n"
        },
        {
            "page_number": 243,
            "text": "238\nBO ZHENG et al.\nTable 4. Configuration of Epidemic Spread Parameters on Waxman Random Network\n#\n# Static nodes\n# Mobile nodes\nMean degree k\nTopology Factor\n1\n1980\n260\n5.9375\n9.2820\n2\n2178\n338\n6.5008\n10.0760\n3\n2395\n439\n7.3888\n11.2417\n4\n2635\n571\n8.3054\n12.6225\n5\n2898\n742\n9.5467\n14.2051\n6\n3188\n965\n10.8392\n16.2766\n7\n3507\n1254\n12.3896\n18.3849\n8\n3858\n1631\n14.2911\n21.0538\n9\n4244\n2120\n16.3974\n23.9066\n10\n4668\n2757\n19.3006\n28.1408\nThe configurations of the spread of epidemic on Waxman Random Network with λ = 0.0005,\nd = 0.0003, p = 0.0003 and m = 0.05. And all the Waxman Random Network has α = 0.2, β = 0.05.\nFigure 12 shows the simulation results of healthy nodes at the end of an epidemic\nin a small world network with N = 10000, static shortcut density 0.005, m = 0.01,\nd = 0.005 and no pre-patch. We plot the proportion of final patched nodes as a function\nof patch rate p and epidemic spread rate per link λ. We can find that both increasing p\nand decreasing λ can heighten the final patched ratio, but the effect of increasing p is\nbetter than decreasing λ.\nFigure 13 illustrates the experimental results in the same network topology with\nthe fixed the spread rate per link λ = 0.01, which is the worst case in the pervious\nexperiment. We plot the final patched ratio as a function of patch rate p and pre-\npatched ratio p0t0. We can see that increasing p and pre-patched ratio can both rapidly\nincrease the healthy ratio. Smartphones are different from the Internet; they are well\nmanaged by the operators. Hence, the operator can push the patch using GPRS, MMS,\netc. to the subscriber. This service of pre-patch and patch will be very helpful to prevent\nthe epidemic on the smartphones.\n7.\nCONCLUSION\nIn this chapter, we propose a novel epidemics spread model (ESS) for smartphone\nwhich is based on the analysis of the unique features of smartphones and SIR model.\nWith the ESS model, we study the “static shortcuts\" and “mobile shortcuts\" effects\nbrought by smartphones and consider the influence of the epidemic spread rate, network\ntopology, patching and death rate as well as the initial pre-patch to the propagation of the\nsmartphone epidemics. Critical condition of epidemic fast die out is derived from the\n"
        },
        {
            "page_number": 244,
            "text": "WIRELESS NETWORK SECURITY\n239\n0\n0.002\n0.004\n0.006\n0.008\n0.01\n0\n0.002\n0.004\n0.006\n0.008\n0.01\n0.4\n0.5\n0.6\n0.7\n0.8\n0.9\n1\np\nλ\nProportion of Final Patched Nodes\nFigure 12. Influence to the proportion of patched nodes at the end of epidemic, patch rate\np vs. epidemic spread rate per link λ.\n0\n0.002\n0.004\n0.006\n0.008\n0.01\n0\n0.1\n0.2\n0.3\n0.4\n0.5\n0.4\n0.5\n0.6\n0.7\n0.8\n0.9\n1\np\np0t0\nProportion of Final Infected Nodes\nFigure 13. Influence to the proportion of patched nodes at the end of epidemic, patch rate\np vs. pre-patched ratio p0t0.\n"
        },
        {
            "page_number": 245,
            "text": "240\nBO ZHENG et al.\nESS model, and the detailed analysis is given to the individual parameters in the model.\nWe demonstrate the effectiveness and accuracy of the ESS model using simulations\nwith the typical network topologies.\nFrom the theoretical analysis and experimental simulations, we give some guidance\nto defend attacks on smart-phones. We find that the pre-patch before epidemics spread\nis very important for prevention, and shield is especially useful because of its non-\ninterruptive nature and small size of shield filter. Moreover, some intrusion prevention\nsystem can also be used to help reduce the epidemics spread rate to slow down the\npropagation. This also motivates the future research work on smartphone security.\n8.\nREFERENCES\n1. http://en.wikipedia.org/wiki/Smartphone\n2. http://www.canalys.com/pr/2005/r2005071.htm\n3. http://www.idc.com/getdoc.jsp?containerId=31554\n4. Sander Siezen, Product Manager, Symbian Ltd, Symbian OS Version 9.1 Product description, Revision\n1.1, February 2005\n5. http://www.symbian.com/\n6. Windows CE Home Page on MSDN. http://msdn.microsoft.com/embedded/windowsce/default.aspx\n7. Microsoft Corporation. Windows Mobile-based Smartphones.\nhttp://www.microsoft.com/windowsmobile/smartphone/default.mspx.\n8. http://www.palmsource.com/\n9. http://brew.qualcomm.com/brew/en/about/about brew.html\n10. http://www.mvista.com/\n11. S.J. Vaughan-Nichols. OSs battle in the smart-phone market. IEEE Computer, 36(6), 2003.\n12. http://www.symantec.com/avcenter/\n13. http://www.kaspersky.com/cyberthreats\n14. http://www.trendmicro.com/vinfo/\n15. C. Guo, H. J. Wang, and W. Zhu, Smartphone Attacks and Defenses, in Proc. ACM HotNets 2004.\n16. H. Wang, C. Guo, D. Simon, and A. Zugenmaier, Shield: Vulnerability- driven network filters for\npreventing known vulnerability exploits, In ACM Sigcomm’04, Portland, OR, Aug. 30 - Sep. 3 2004.\n17. C. Moore and M. E. J. Newman, Epidemics and percolation in small-world networks, Physical Review\nE 61, 5678-5682.\n18. C. Moore and M. E. J. Newman, Exact solution of the site and bond percolation on small-world networks,\nPhys. Rev. E, 62(2000), 7059-7064.\n19. R. Pastor-Satorras and A. Vespignani, Epidemic spreading in scale-free networks, Physical Review\nLetters, 86 (2001), 3200-3203.\n20. R. Pastor-Satorras and A. Vespignani, Epidemic dynamics in finite scale-free networks, Physical Review\nE, 65 (2002).\n21. M. E. J. Newman, Spread of epidemic disease on networks, Phys Rev E, 2002, 66, 016128\n"
        },
        {
            "page_number": 246,
            "text": "WIRELESS NETWORK SECURITY\n241\n22. Zesheng Chen, Lixin Gao, and Kevin Kwiat, Modeling the Spread of Active Worms, In Proceedings of\nIEEE INFOCOM 2003, San Francisco, CA, April 2003.\n23. Michele Garetto, Weibo Gong, and Don Towsley, Modeling Malware Spreading Dynamics, In Pro-\nceedings of IEEE INFOCOM 2003, San Francisco, CA, April 2003.\n24. Cliff Changchun Zou, Weibo Gong, and Don Towsley, Code Red Worm Propagation Modeling and\nAnalysis, CCS’02, November 18-22, 2002, Washington, DC, USA.\n25. Y. Wang, D. Chakrabarti, C. Wang and C. Faloutsos, Epidemic spreading in real networks: An eigen-\nvalue viewpoint, Proc. IEEE SRDS, 2003.\n26. A. Ganesh, L. Massouli´e, D. Towsley, The Effect of Network Topology on the Spread of Epidemics,\nIEEE Infocom, 2005.\n27. O. Diekmann and J. A. P. Heesterbeek, Mathematical epidemiology of infectious diseases: model\nbuilding, analysis and interpretation, (JohnWiley & Sons, New York, 2000).\n28. D. J. Watts and S. H. Strogatz, Collective dynamics of ‘small-world’ Networks, Nature, 393(1998),\n440-442.\n29. R´eka Albert and Albert-L´aszl´o Barab´asi, Statistical mechanics of complex networks, Rev. Mod. Phys.\n74(2002), 47-97.\n30. Symantec Internet Security Threat Report, Trends for January 1, 2004 - June 30, 2004, Volume VI,\nPublished September 2004.\n31. W. O. Kermack, and A. G. McKendrick, A Contribution to the Mathematical Theory of Epidemics,\nProc. Roy. Soc. Lond. A 115, 700-721, 1927.\n32. A. Rowstron and P. Druschel, Pastry: Scalable, distributed object location and routing for large-scale\npeer-to-peer systems, Proc. Middleware 2001, Germany, November 2001.\n33. M. Faloutsos, P. Faloutsos and C. Faloutsos, On power-law relationships of the Internet topology, Proc.\nACM Sigcomm, 1999.\n34. A. L. Barab´asi and R. Albert, Emergence of scaling in random networks, Science 286,509-511 (1999).\n35. C. Jin, Q. Chen and S. Jamin, Inet: Internet Topology Generator, Technique Report CSE-TR-433-00,\nUniversity of Michigan, EECS dept. 2000, http://topology.eecs.umich.edu/inet.\n36. B. Waxman, Routing of Multipoint Connections, IEEE J. Select. Areas Commun., SAC-6(9):1617-\n1622, December 1988.\n"
        },
        {
            "page_number": 247,
            "text": "Part III\nSECURITY IN\nWIRELESS LANS\n"
        },
        {
            "page_number": 248,
            "text": "10\nCROSS-DOMAIN MOBILITY-ADAPTIVE\nAUTHENTICATION\nHahnsang Kim\nINRIA, Sophia Antipolis, France\nE-mail: hahnsang.kim@inria.fr\nKang G. Shin\nDepartment of Electrical Engineering and Computer\nScience\nUniversity of Michigan, Ann Arbor, U.S.A.\nE-mail: kgshin@eecs.umich.edu\nWhenmobileuserswithon-goingsessionscrossthedomainboundary, theirre-authentication\naffects significantly the inter-domain handoff latency as each inter-domain handoff requires\nremote contact with the home authentication server across domains, making it difficult to\nemploy existing authentication protocols as they are. This chapter focuses on cross-domain\nauthentication over wireless local area networks (WLANs) that minimizes the need for re-\nmote contact/access. We analyze the security requirements suggested by the IEEE 802.11i\nauthentication standard, and consider additional requirements to help reduce the authentica-\ntion latency without compromising the level of security. We propose an enhanced protocol\ncalled the Mobility-adjusted Authentication Protocol (MAP) that performs mutual authen-\ntication and hierarchical key derivation with minimal handshakes, relying on symmetric\ncryptographic functions. We also introduce security context routers (SCRs) that handle\nsecurity context in conjunction with MAP, eliminating the need for continual remote con-\ntact with the home authentication server. In contrast to Kerberos that favors inter-domain\nauthentication, MAP achieves a 26% reduction of authentication latency without degrading\nthe level of security.\n1.\nINTRODUCTION\nTime-sensitive applications, such as Voice over IP (VoIP) or video streaming, are now\npossible over wireless local area networks (WLANs), such as those based on the IEEE\n802.11 Standard [4], thanks to their high bandwidth. WLAN technologies also allow\n"
        },
        {
            "page_number": 249,
            "text": "246\nHAHNSANG KIM and KANG G. SHIN\nthe mobiles to roam within public/corporate buildings or university campuses. Further-\nmore, we anticipate that mobile users might cross the domain boundary without their\non-going application sessions disrupted. However, VoIP requires a handoff to be com-\npleted in less than 50ms for acceptable Quality-of-Service (QoS) [33], including the\nexecution of the IEEE 802.11i authentication [6] as part of a secure handoff mechanism.\nMinimizing the number of messages to be exchanged is important as cross-domain\nauthentication needs to contact the remote home server. Moreover, the authentication\nlatency increases in proportion to the round-trip time between two points involved in\ninter-domain message exchanges. Optimization of the authentication protocol is of\nutmost importance since an existing redundant combination of authentication and key\nnegotiation functions incurs more rounds of message exchange than necessary.\nWe propose an enhanced protocol for cross-domain authentication, Mobility-ad-\njusted Authentication Protocol (MAP) that relies on far less costly symmetric cryp-\ntography. (1) MAP reduces the cross-domain authentication latency by reducing the\nnumber of message exchanges. MAP requires less message exchanges without degrad-\ning security or the re-authentication mechanism, reducing the authentication latency\nsignificantly. (2) MAP replaces the 4-way handshake of the IEEE 802.11i authenti-\ncation. In coordination with the authenticator within an access point, MAP defines\nhierarchical key derivation and generates consecutive keys during authentication op-\nerations. This leads to optimizing the 802.11i authentication mechanism by removing\nthe need for the 4-way handshake. (3) MAP leverages the concept of security context\nto mostly avoid remote contact. With the mobile moving along, its security context is\ntransferred via security context routers (SCRs) we present in this chapter. An SCR also\nplays a role of an authentication server in a foreign domain; it provides security context\nfor MAP operating as if in the home server. Via a prototype implementation, our eval-\nuation results show that the cross-domain authentication latency of MAP accounts for\n74% and 85% that of Kerberos [17] and Needham-Schroeder symmetric-key protocol\n(NS) [26, 27], respectively. It makes up to 53% improvement in the authentication\nlatency which is proportional to the end-to-end domain distance until the round-trip\ntime counts up to 100ms.\nThe remainder of this chapter is organized as follows. Section 2 gives an overview\nof the 802.11i authentication mechanism, the related cross-domain protocols, design\nrequirements, and prerequisites of BAN logic. Section 3 first describes MAP including\nits architecture and a relevant interaction between SCRs. Subsequently we details\ndefined keys and types of messages, an example of message exchanges for a successful\nauthentication, and the corresponding pseudo code of each module. Section 4 considers\npossible threats and analyzes the security of MAP. Section 5 examines the performance\nvia measurements and simulation. Finally, we discuss related work in Section 6 and\nconclude the chapter in Section 7.\n2.\nOVERVIEWOFAUTHENTICATIONMECHANISMANDREQUIREMENTS\nIn this section, we first introduce the 802.11i authentication scheme and protocols ap-\nplicable to the cross-domain authentication, and then describe the design requirements\nof authentication protocols. Finally, we explore prerequisites to BAN logic.\n"
        },
        {
            "page_number": 250,
            "text": "WIRELESS NETWORK SECURITY\n247\nFigure 1. Message exchanges in the IEEE 802.11 and .11i systems\n2.1. The IEEE 802.11i Authentication\nThe IEEE 802.11i authentication is responsible for mutual authentication and key\nderivation for securing WLANs via the 802.1X and 4-way handshake [6]. Figure 1\nshows a typical scenario of message exchanges in the context of the IEEE 802.11\nand 11i.\nOur focus is on two main steps after (re)association.\nFirst, the 802.1X\nauthentication, where an authentication protocol like TLS [13] operates, is to verify\nthe authenticity of end-to-end principals: the mobile (STA) and the authentication\nserver (AS) via the authenticator (AUTH) (which operates in an AP). In particular,\nthe AUTHs and AS construct an authentication authorization and accounting (AAA)\narchitecture [1]. Successful mutual verification of each identity leads to the derivation\nof a pair-wise master key (PMK). This key is transferred to the AUTH via a secure\ntunnel. Second, the STA and AUTH perform the 4-way handshake, exchanging their\nnonces, so that a pair-wise temporary key (PTK) with which the wireless link will be\nsecured is produced using the PMK as a seed.\nThe performance of the IEEE 802.11i authentication depends on the efficiency of\nthis authentication protocol. Recent efforts on security associations have been limited\nto distribution of keys to access points within a domain [24]. For inter-domain handoffs,\nhowever, the authentication latency is critical to the application QoS.\n2.2. Cross-domain-related Protocols\nThere are two protocols: Kerberos that supports the cross-domain authentication\nand NS that can be effectively extended to do so. We will use the two protocols to\n"
        },
        {
            "page_number": 251,
            "text": "248\nHAHNSANG KIM and KANG G. SHIN\nFigure 2. Message exchanges for a remote access grant in Kerberos. This sequence is\nrepeated each time the mobile is bound to a remote TGS.\ncomparatively evaluate the throughout of our protocol via simulation. The following\nare the descriptions of the message exchanges of each protocol.\nThe Kerberos protocol provides cross-domain operations. By establishing inter-\ndomain keys, the administrators of two domains allow the mobile to receive services\nin a remote domain. It receives a remote ticket granting ticket (TGT) from the ticket\ngranting server (TGS) in the local domain. It then obtains a service granting ticket\n(SGT) from the remote TGS in the other domain by using the issued remote TGT.\nWith the SGT containing a secret key, the mobile and AS can authenticate each other.\nFigure 2 illustrates a sequence of message exchanges for a remote authentication in\nKerberos. The link among TGSs is assumed to be secure; a secret key of each TGS is\nshared to identify itself. In addition to the secure link, the AS has security association\nwith its TGS. The remote TGT issued earlier can be reused to get TGTs in the current\ndomain within a given period of time. However, each time the mobile moves into a\nforeign domain, the mobile needs to get a remote TGT again by contacting its home\nTGS.\nThe NS protocol on which Kerberos is based is not intended to operate over cross-\norganization boundaries. However, it can support cross-domain authentication with\nminor modifications, which we call a modified NS protocol (MNS). At first, the original\nprotocol operates, in principle, as can be seen in Figure 3. The initiator A and its\ncorrespondent B share secret keys AK and BK with the AS, respectively.\nIn the\nbeginning, A obtains two copies of a pair-wise key encrypted with BK and AK by the\nAS, respectively, during their communication. Then, A sends B the BK-encrypted\npair-wise key along with SK-encrypted AN. AN will be returned in the next message in\norder for A to ensure that B with which A is communicating is legitimate. B also adds\n"
        },
        {
            "page_number": 252,
            "text": "WIRELESS NETWORK SECURITY\n249\nFigure 3. An example of message exchanges in symmetry-key-based NS protocol. AK\nand BK are pre-shared between A and AS, and B and AS, respectively. Elements used to\nauthenticate A or AS in messages 1 and 2 are omitted. AN and BN are A’s and B’s nonces,\nrespectively.\nBN to the message encrypted by SK, and verifies the decremented BN that A sends\neventually; those exchanged nonces may be used for key generation for need. We can\nview A as STA, and B and AS as foreign and home ASs, respectively. When the foreign\nAS requires a set of pair-wise keys, the home AS generates and sends a set of multiple\ndifferent keys. Once receiving them, the foreign AS has no need for contacting the\nhome AS, which may lead message exchanges to be reduced into the 3-way handshake.\n2.3. Design Requirements\nThe IEEE 802.11i authentication suggests several requirements that must be pre-\nserved to secure WLANs.\nRequirement 1: The STA and AS must be able to authenticate each other.\nSince the STA establishes a wireless link to the AS via anonymous APs, it\nshould be able to identify the AS, so should the AS.\nRequirement 2: A successful mutual authentication leads to the derivation of\na fresh key for the AS and STA. After the successful mutual authentication,\na 256-bit key (i.e., PMK) is generated by the AS and STA, and is eventually\nused by the STA and AP. This key must not be reused, and becomes obsolete\nwhenever the STA binds with a new AP.\nRequirement 3: Mutual authentication should be strong enough to be protected\nfrom any unauthorized reception. It is uneasy to demonstrate the safety of the\nauthentication protocol, but there are theoretical approaches for this purpose.\nForexample, formalverificationmethodsbasedonmodelcheckingandtheorem\nproving, modal logic, and modular approach are widely used. We will show\n"
        },
        {
            "page_number": 253,
            "text": "250\nHAHNSANG KIM and KANG G. SHIN\na logical proof of MAP using BAN logic in Section 4.1. In addition to these\nrequirements, we present the following recommendations for the authentication\nprotocol design to help achieve fast handoffs in WLANs.\nRecommendation1: Minimizingmessageexchangesduringtheauthentication\nprocess helps improve the performance of cross-domain handoffs. We evaluate\nthe effects of the number of message exchanges.\nRecommendation 2: The use of lightweight cryptographic algorithms helps\nlow-power mobile terminals, like personal data assistants, mitigate the perfor-\nmance overhead of computation-intensive cryptographic algorithms.\nBased on the above requirements, we will design a protocol supporting cross-\ndomain mobility.\n2.4. BAN Logic\nBAN logic [11] is a modal logic developed for authentication protocol analysis. It\npresents the proof that a simple logic could be used to describe the beliefs of trustworthy\ncommunicating parties. It found redundancies or security flaws in authentication pro-\ntocols in the literature [10]. BAN logic reasons that the protocol operates as correctly\nas expected. It is effective to prove the correctness of the authentication mechanisms\nwith logical reasoning.\nWe introduce the several constructs and logical postulates in BAN logic that will be\nused for the proof of MAP. Full details of its rules are given in [11]. First, the following\nare the constructs that we use:\n-\nP believes X: P believes X. In particular, the principal P may act as if X is\ntrue. This construct is essential to the logic.\n-\nP sees X: P sees X. Someone has sent a message containing X to P, who\ncan read and repeat X possibly after doing some decryption.\n-\nP said X: P once said X. The principal P at some time sent a message\nincluding the statement X. It is unknown when the message was sent, but it is\nknown that P believed X then.\n-\nP controls X: the principal P is an authority on X and should be trusted on\nthis matter, e.g., a server is often trusted to generate encryption keys properly.\nThis may be expressed by the assumption that the principals believe that the\nserver has jurisdiction over statements about the generated keys.\n-\nfresh(X): the formula X is fresh, i.e., X has not been sent in a message at any\ntime before the current run of the protocol. This is usually true for nonces that\nis randomly generated for use only once.\n-\nP\nK\n←→Q: P and Q may use the shared key K to communicate. It is never\ndisclosed by any principal except for P and Q.\n"
        },
        {
            "page_number": 254,
            "text": "WIRELESS NETWORK SECURITY\n251\n-\nP\nX\n⇐⇒Q: the formula X is a secret known only to P and Q, and possibly to\nprincipals trusted by them. Only P and Q may use X to prove their identities\nto one another.\n-\n{X}K: this represents the formula X encrypted under the key K.\n-\n< X >Y : this represents X combined with the formula Y ; it is intended that Y\nbe a secret and that its presence prove the identity of whoever utters < X >Y .\nThen, we use the following logical postulates in proof.\nThe message-meaning rules are applied to the interpretation of messages for\nshared keys\nP believes Q K\n←→P, P sees {X}K\nP believes Q said X\nand for shared secrets,\nP believes Q Y\n⇐⇒P, P sees < X >Y\nP believes Q said X\n.\nThe nonce-verification rule represents the check that a message is recent and\nthat the sender still believes in:\nP believes fresh(X), P believes Q said X\nP believes Q believes X\n.\nThe jurisdiction rule states that if P believes that Q has jurisdiction over X\nthen P trusts Q on the truth of X:\nP believes Q controls X, P believes Q believes X\nP believes X\n.\nIn addition, the HMAC (Hash Message Authentication Code) represented as MAC\n(m, K), where m and K denote a message and a pair-wise secret key, respectively,\nis used to verify whether or not the verifiee possesses the same K as the verifier. In\nother words, only if the generated codes are different, the applied Ks are different.\nTherefore, MAC(m, K) is interpreted as a unit of the secret < X >Y .\n3.\nMAP\nIn this section, we describe an authentication architecture that extends the AAA\narchitecture to SCR communications, and design MAP. The description of MAP in-\ncludes the definition of keys and messages, message exchanges, and detailed operations\nin each functional module.\n"
        },
        {
            "page_number": 255,
            "text": "252\nHAHNSANG KIM and KANG G. SHIN\n3.1. Architecture\nAuthentication operations work basically with three entities: STA, AS, and AUTH.\nAn STA represents the end user with a WLAN-interface-equipped device. An AS ver-\nifies the STA’s authenticity and provides each key to secure their wireless link. An\nAUTH relays authentication traffic between the STA and AS. In addition to dealing\nwith these entities, our protocol solves the cross-domain authentication problem by\nintroducing so-called security context routers (SCRs). An SCR is usually placed be-\ntween multiple AUTHs and an AS. The SCR is logically distinct from the AS in terms\nof enforcing authentication policy, although both may reside on the same physical ma-\nchine or the SCR can be integrated into the AS. The SCR functions as follows. After\nreceiving a security context1 issued by MAP on the AS, it can perform re-authentication\non behalf of the AS. The SCRs are distributed in each domain so that they can reduce\nthe authentication latency while the STA roams around the domain. It is assumed that\nin case of the communication of inter-administration domains they have a security as-\nsociation agreement on roaming and are securely connected to one another by sharing\ninter-domain keys. This combination is adaptable to the security architecture of the\nIEEE 802.11i authentication and Wi-Fi Protected Access 2 (WPA2) [2]. The protocol\ndescribing how messages are exchanged between the SCRs is part of our future work.\nIn this chapter, we will give a rough idea of how to exchange messages between SCRs\nshortly.\nFigure 4 depicts the MAP architecture. The MAP server module on the AS, which\nis described in Section 3.6, is an end-point authentication protocol that is assumed to\nsecurely be connected to the AUTHs via the SCR. The AS used in the architecture\nis functionally equivalent to the AAA server. The MAP security context module (SC\nmodule) in the SCR, which is described in Section 3.6, helps the AS communicate\nwith the other MAP-support AS for cross-domain authentication. The AUTH is an\nauthentication client as a pass-through authenticator. It relays authentication traffic\nfrom the STA to the AS, and vice versa. The MAP client module in the STA, which is\ndescribedinSection3.6, isanend-pointauthenticationpartythatrequestsauthentication\nand eventually establishes a secure link with the attached AP.\n3.2. Communication between SCRs\nThe SCRs communicate with each other, based on a peer-to-peer manner. There\nare two ways of transferring security context among the SCRs involved. In case of no\nsecurity context cached in an SCR with which an STA has just associated, the targeted\nSCR fetches security context from the original SCR with which the STA associated\npreviously; reactive transfer introduces the latency of fetching security context. On\nthe other hand, the original SCR may somehow forward the targeted SCR(s) security\ncontext before the mobile is handed off; proactive transfer emphasizes the availability\n1 Its contents vary with individual protocols. MAP is expected to have a set of authentication value pairs,\nidentity (= mobile Id), validity time, time stamp, mean handoff time, counter and other security information.\n"
        },
        {
            "page_number": 256,
            "text": "WIRELESS NETWORK SECURITY\n253\nFigure 4. Authentication architecture\nof the context ahead of time. On the other hand, estimation of the STA’s direction and\nmanagement of security context can be emphasized, which is referred to as predictive\nforwarding of security context. Their combination yields a tradeoff between storage\noverhead and latency performance. Elaboration on such issue is part of our future work.\n3.3. Authentication\nThe MAP’s authentication relies on Message Authentication Code (MAC) algo-\nrithms [18]. The MAC values rely on shared symmetric keys, the management of which\nis uneasy to scale in that two communication parties must somehow exchange the key\nin a secure way, compared to that of asymmetric-key pairs. However, on the other\nhand, signing and verifying public keys are very time-consuming; the MAC values are\npreferred to digital signatures because the MAC computation is two to three orders of\nmagnitude faster. There is a tradeoff between scalability and CPU usage; we chose\ncost efficiency since it matches our design goal.\n3.4. Defined Keys\nWe define three types of keys for different purposes: primary key (PK), domain\nkey (DK) and temporary key (TK). PK is a long-term symmetric key which may be\nperiodically updated and deployed, e.g., online subscription to a service provider or\noff-line set-up with a purchased card. PK is assumed to have guaranteed protection\nagainst disclosure for a sufficiently long period of time. DK is a quasi-primary key in\na (sub)domain, which is derived from PK and the previous DK. The STA generates a\n"
        },
        {
            "page_number": 257,
            "text": "254\nHAHNSANG KIM and KANG G. SHIN\nFigure 5. Defined keys hierarchy and boundary. An SCR controls a DK derived from the\nPK. A subnet uses a DK+ hashed from the DK to generate TK that will be used for each\nassociation.\nnew DK as it changes a domain; an old DK must be revoked. In addition, DK+, an n-\ntime-hashed DK, is defined for use in a subnet within a domain—if no concept of such\nsubnet is applicable DK+ is generated from each DK; it plays a role of loose coupling\nof DK and TK. TK that is derived from DK+ is a link key affiliated with securing a\nwireless link established between the STA and AP. TK binds with the addresses of two\ninvolved physical devices. Therefore, in case of re-associating or changing a binding\naddress, TK is also changed. Figure 5 shows a hierarchical derivation and boundary of\nthe defined keys. An association is made of each TK; the disclosure of any TK has no\neffect on other (re)associations. DK+ also provides a key-disclosure barrier for relation\nbetween TK and DK. AS affects only the generation of DKs.\n3.5. Defined Messages\nWe define six types of messages exchanged, with the server, SC, authenticator, and\nclient modules interacting with each other during initial and re- authentication in MAP.\nThe first four messages are used during the initial authentication and the last two are\nused during re-authentication.\nAuth-req message, sent by the client module in the STA, triggers a negotiation\non authentication and key agreement from scratch.\nAuth-chal message, sent by the server module, as a return message, is used for\nthe purpose of challenging the STA, with an encrypted code used for verifying\nthe AS’s authenticity to the STA.\n"
        },
        {
            "page_number": 258,
            "text": "WIRELESS NETWORK SECURITY\n255\nChal-res message, sent by the client module, as a response message, contains\na nonce-response encrypted code so that the AS verifies the STA’s authenticity.\nAuth-res message, sent by the server module, is a reply to a challenge-response\nmessage.\nReauth-req message is sent by the client module in the authenticated STA. The\nSCR captures this message and verifies if the authentication code is legitimate.\nReauth-res message is a reply to the reauthentication-request message including\nthe authentication result.\n3.6. Message Exchanges\nThe following is an example of exchanging messages in case of a successful au-\nthentication. Only authentication-related information is highlighted in the messages.\nM1.STA →AS: Auth-req(STAId, SNi)\nThe STA sends the AS an authentication request containing its identity (Id) and a\nfresh nonce. On receipt of the message, the AS fetches credential corresponding\nto the Id and extracts its key from it.\nM2.AS →STA: Auth-chal(MACP K[STAId, SNi, ASNj, ‘authch’])\nThe AS uses the STA’s nonce to compute a MAC, which protects from a reply\nattack.\nM3.STA →AS: Chal-res(MACP K[STAId, SNi+1, ASNj, ‘authres’])\nIf the received MAC is matched with the one that the STA generates, the AS\nis authenticated to the STA. Subsequently, the STA responds to the challenge.\nOtherwise, the message is silently ignored.\nM4.AS →STA: Auth-res(ENCDK+[SNi+1, ASNj+1, AUNk])\nIf the STA is successfully authenticated as well, the AS adds a secret value, i.e.,\nASNj+1 to a response message. During transfer of the message, the authentica-\ntor inserts a newly generated nonce AUNk that is used to compute a temporary\nkey (TK). Meanwhile, the AS computes and sends a set of authentication value\npairs (AVPs) to the SCR.\nWhen the STA re-associates with another AP in (another) subnet, the following\nmessages are exchanged.\nM5.STA →SCR: Reauth-req(MACP K[STAId, ASNj+1, DKi, ‘reauth’])\nThe STA computes a MAC, using the secret value obtained in the previous\nround of authentication. The SCR can verify if the STA holds the same nonce.\nM6.SCR →STA: Reauth-res(ENCDK+[SNi+2, ASNj+2, AUNk+1])\nIf the STA is authenticated successfully, the SCR adds another nonce for the\nnext challenge in the message. The STA can authenticate the SCR by verifying\nif the nonce is identical of the one that it sent previously.\n"
        },
        {
            "page_number": 259,
            "text": "256\nHAHNSANG KIM and KANG G. SHIN\nIn the subsequent section, we describe in details how MACs and hierarchical keys\nare computed and used in each module.\nMAP Server Module\nThe server module handles two types of incoming messages (i.e., auth-req and\nchal-res) that are related only to authentication from scratch. The following is the\ndescription of the pseudo code of the server module.\nvar: sn1..n, cn1..n := 0; %Server and client nonce queues are initialized.\nfor all i: auth-req of Idi in buffer do\nsni=refresh(sni); %A fresh nonce is generated.\nsend auth-chal: sni | MACP Ki(Idi, cn′, sni, “authch”);\ncni=cn′; %Client nonce from the message is buffered.\nend for\nfor all i: chal-res of Idi in buffer do\nDKi,j−1 = PRF(PKi, cni, sni); %cni is obtained from auth −req.\nif MACP Ki(Idi, cn′, sni, “authres”) && MICDKi,j−1 verified\n%cn′ is obtained from chal-res.\nsni=refresh(sni);\nDK+=Hαi(DKi,j−1)\nsend auth-res: sni | cn′ | DK+; %DK+ is transferred to the authenticator.\nmake SCi:\nfor e = 1..n do\nMACP Ki(Idi, sni, DKi,j−1, “reauth”);\nDKi,j=PRF(PKi, DKi,j−1, sni);\nAVPe:(Idi, sni, MAC, DKi,j) ∈\u0002\n1..e−1 AVP;\nsni=refresh(sni);\nend for\nend if\nend for\nA MAC, including client nonce cn′ from the received message and server nonce\nsni, is computed and sent to the STA of Idi. The MAC allows the STA to verify the\nAS’s authenticity. DK is computed by calculating an n-bit key generating pseudo ran-\ndom function (PRF)—in most cases n=128 is sufficient—with PK and the previously-\nexchanged nonces. An MIC provides a means of verifying authenticity once the associ-\natedMACisverifiedsuccessfully. Ahasheddomainkey, DK+, isgeneratedbyapplying\nα times a cryptographic one-way function H, equivalently Hα(x) = Hα−1H(x) and\nH0(x) = x. The α value is a sync-one shared between the STA and the AS/SCR.\nDK+ allows DK to be hidden from the authenticators. After the message exchanges,\nthe server module creates the STA’s security context that is composed primarily of the\nset of AVPs. It is then transferred to the corresponding SCR. The AVPs enable the SCR\nto conduct the re-authentication and re-keying process on behalf of the AS.\n"
        },
        {
            "page_number": 260,
            "text": "WIRELESS NETWORK SECURITY\n257\nMAP SC Module\nThe SC module handles an incoming message (i.e., reauth-req) and an outgoing\nmessage (i.e., reauth-res) which are related to re-authentication. In particular, this\nmodule can be implemented, combined with the server module. The following is the\ndescription of the pseudo code of the SC module.\nfor all i: reauth-req of Idi in buffer do\nAVPl=(Idi, sni, MAC, DKi,j) ←\u0002\nl..n AVP; %Select one of AVPs.\nif MAC && MICDKi,j verified do %The integrity of the message is checked.\nsend reauth-res: cn′|sni|Hαi(DKi,j); %DK+ is derived by α-time hashing.\nend if\nend for\nThe SC module first retrieves one of AVPs from the security context corresponding\nto Idi and then verifies MAC̸=MAC′ or MICDKi,j(reauth-req)̸=MIC′. If they are\nmatched correctly, it computes DK+ and sends the authenticator it along with the\nexchanged and retrieved nonces. If DK is not allowed to be reused, the AVP is dethroned\nwhen it is notified somehow that the STA of Idi de-associates with the current AP. If\nno more AVP exists, the re-authentication request is forwarded to the AS which will,\nin turn, handle the request from scratch. Note that the SC module does not possess any\nPK.\nAuthenticator\nA primary role of this module is to relay incoming messages. It also computes an\nTK with which the STA and AP establish a secure link after a successful authentication.\nvar: an; %This is an authenticator nonce.\nif auth-req | auth-chal | chal-res | reauth-req received\nrelay it;\nend if\nif auth-res | reauth-res received\nif success in authentication %This is determined by AS/SCR.\nan=refresh(an); % A new an is used to generate TK.\nsend auth-res: ENCDK+[sn′ | an | cn′];\n%DK+ and cn′ are obtained from the message.\nTK=PRF(DK+, AddrST A | AddrAP , an | cn′);\nend if\nend if\nAuthenticator is beyond access to DK; DK+ received from the AS/SCR is used to\ncompute TK by calculating a PRF—the key-size varies with cryptographic protocols\nto be used for securing a wireless link, yet it is either 256 or 512 bits. TK binds with\nmedia access control addresses of the STA and AP; de-association revokes TK and a\nnew TK must be recomputed.\n"
        },
        {
            "page_number": 261,
            "text": "258\nHAHNSANG KIM and KANG G. SHIN\nMAP Client Module\nThe client module incurs an authentication request message (e.g., auth-req or\nreauth-req) when the STA (re)associates with an AP. It also handles incoming mes-\nsages (i.e., auth-chal, auth-res, and reauth-res) and outgoing messages (i.e., chal-res).\nvar: secret := 0, cn; %cn is a client nonce.\nif (re)associated\ncn=Refresh(cn);\nif !secret %In case of authentication from scratch\nsend auth-req: Id | cn;\nelse %In case of the previous successful authentication\nDKi=PRF(PK, DKi−1, secret);\nsend reauth-req: Id|MACP K(Id, secret, DKi−1, “reauth”)|cn|MICDKi;\nend if\nend if\nif auth-chal received\nif MACP K(Id, cn, sn′, “authch”) verified %It authenticates AS.\nDKi−1=PRF(PK, cn, sn′); %sn′ is obtained from auth-chal\ncn=Refresh(cn);\nsend chal-res: Id|cn|MACP K(Id, cn, sn′, “authres”)|MICDKi−1;\nend if\nend if\nif auth-res | reauth-res received\nDK+=Hα(DKi−1 or DKi);\nDECDK+[ENC[sn′ | an′ | cn′]];\nif cn==cn′ %It authenticates AS.\nsecret=sn′; %sn′ is stored as secret\nTK = PRF(DK+, AddrST A | AddrAP , an′ | cn);\n%cn is obtained from the previous message.\nend if\nend if\nIt retains the secret value provided by the AS after completion of the previous suc-\ncessful authentication. Confidentiality of the secret is guaranteed since it is transferred\nin ciphertext. The secret determines whether the authentication process is conducted\nfrom scratch. The α value is matched to that of the AS/SCR.\n4.\nSECURITY CONSIDERATIONS\nIn this section, first, using BAN logic, we show the logical proof that MAP performs\nits authentication mechanism correctly as it is expected, and then examine security\nthreats to our protocol.\n"
        },
        {
            "page_number": 262,
            "text": "WIRELESS NETWORK SECURITY\n259\n4.1. Protocol Analysis\nThe analysis procedure works as follows. First, we translate the original protocol\ninto the idealized one and then make assumptions about the initial state. Finally, we\nmake logical formulas as assertions and apply the logical postulates to the assumptions\nand assertions to arrive at conclusions.\nTranslation; we extract the encrypted forms of messages from MAP communications\nas follows:\nM1.B →A: < Na, Nb >P K\nM2.A →B: < Nb, Na >P K\nM3.B →A: {Nb′, Na′}DK\nM4.A →B: < Nb′, ADK\n←→B >P K\nM5.B →A: {Nb′′, Na′′}K\nab′\nWe have STA and SCR, referred to as A and B, respectively—the functionality of AS\nand AUTH is integrated into SCR for simplicity; DK+ is identical of DK. We also omit\ncommunication in clear-text. There is a slight difference by representing (Na ⊕Nb)\nas (Na, Nb), which is acceptable since this means that Na and Nb were uttered at the\nsame time and their XOR-ed value is straightforwardly obtained.\nFor authentication, each party verifies the MAC which requires the nonces gener-\nated by itself and the other. That is, the correct MAC can only be generated with the\nfresh nonces from the two. Thus, authentication between A and B might be deemed\ncomplete if each of the two believes that the other has recently sent the nonce, and\nproving sound mutual authentication is sufficiently satisfied by deriving the facts:\nA believes B believes Na and B believes A believes Nb\nfor initial authentication and\nA believes B believesNa′′ and B believes A believes Nb′\nfor re-authentication.\nMaking assumptions; we then write the following assumptions:\n(1)A believes A P K\n⇐⇒B, (2) B believes A P K\n⇐⇒B,\n(3)A believes A DK\n←→B, (4) B believes A DK\n←→B,\n(5)A believes A DK′\n←→B, (6) B believes A DK′\n←→B,\n(7)A believes fresh(Na), (8) B believes fresh(Nb),\n(9)A believes fresh(Na′), (10) B believes fresh(Nb′),\n(11)A believes fresh(Na′′), (12) B believes fresh(Nb′′),\n(13)A believes fresh(Nb′), (14) A believes fresh(Nb′′),\n(15)A believes B controls Nb′,\n"
        },
        {
            "page_number": 263,
            "text": "260\nHAHNSANG KIM and KANG G. SHIN\n(16)A believes B controls Nb′′.\nAssumptions (1) and (2) are made from the fact that A and B initially share a secret,\nPK. Assumptions (3), (4), (5) and (6) are derived from the fact that only A and B can\ngenerate a shared key only if the sound authentication is achieved. Assumptions (7)\nto (12) state that A and B believe that the nonces generated by themselves are fresh;\nfreshness of nonces holds by verification of MAC and MIC associated with the nonces.\nThe nonces, Nb′ and Nb′′, also play a role of secrets since they are transferred with\nproper encryption. Thus, A can believe that B has generated the nonces that was not\nused in the past,which leads to Assumptions (13) and (14), and also (15) and (16),\nindicating that A trusts B to generate the secret.\nReasoning; we analyze the idealized version of MAP by applying the logical postulates\npresented in Section 2.4 to the assumptions.\nA receives Message M1. The annotation rule yields that A sees < Na, Nb >P K\nholdsafterward. Withthehypothesisof(1), themessage-meaningruleforsharedsecrets\napplies and yields A believes B said (Na, Nb). Breaking conjunctions produces A\nbelieves B said Na. With the hypothesis of (7), we apply the nonce-verification rule\nand yield A believes B believes Na. On the other hand, B receives Message M2\nand the following result is obtained in the same way as that of Message M1, via the\nmessage-meaningandnonce-verificationruleswithhypotheses(2)and(8), respectively,\nB sees < Nb, Na >P K and B believes A believes Nb. This concludes the analysis\nof Message M2. The analysis of Messages M1 and M2 confirms that MAP performs\nmutual authentication successfully.\nA receives Message M3 and the annotation rule yields that A sees {Nb′, Na′}DK\nholds thereafter. The message-meaning rule for shared keys with the hypothesis of\n(3) via breaking conjunctions yields: A believes B said Nb′, and A believes B said\nNa′. Taking the former, with hypotheses (13) and (15), the nonce-verification and\njurisdiction rules apply and yield A believes B believes Nb′, and A believes Nb′,\nrespectively. Taking the latter, the nonce-verification rule with hypothesis (9) yields A\nbelieves B believes Na′. This concludes the analysis of Message M3. This message\nmay appear redundant since authentication is completed from Message M1, but it is\nessential not because it is for authentication, but because it is for transmission of a\nsecret, nonce Nb′.\nB receives Message M4 and the annotation rule yields that B sees < Nb′, A and\nDK\n←→B >P K holds thereafter. By applying the message-meaning rule for the secrets\nwith (2) via breaking conjunctions, we obtain: B believes A said (Nb′, ADK\n←→B), and\nB believes A said Nb′. The nonce-verification rule with hypothesis (10) yields that B\nbelieves A believes Nb′. On the other hand, A receives Message M5 and the annotation\nrule yields that B sees {Nb′′, Na′′}DK′ holds thereafter. By applying the message-\nmeaning rule for the shared keys with hypothesis (6) via breaking conjunctions, we\nobtain A believes B said Nb′′ and A believes B said Na′′. Taking the former, the\nnonce-verification and jurisdiction rules with (14) and (16) yield A believes B believes\nNb′′, and A believes Nb′′, respectively. Taking the latter, nonce-verification with (11)\n"
        },
        {
            "page_number": 264,
            "text": "WIRELESS NETWORK SECURITY\n261\nyields that A believes B believes Na′′. The analysis of Messages M4 and M5 confirms\nthat MAP also achieves mutual re-authentication.\n4.2. Possible Attacks\nKey recovery attack: This relies on finding the key K itself from a number of message–\nMAC pairs. Ideally, any attack allowing key recovery requires about 2k operations\nwhere k is the length of K. The adversary tries all possible keys with a small number\nof message–MAC pairs available. Choosing a sufficiently long key is a simple way\nto thwart a key search. Another possible attack is to choose an arbitrary fraudulent\nmessage and append a randomly-chosen MAC value. Ideally, the probability that this\nMAC value is correct is equal to 1/2m, where m is the number of bits in the MAC value.\nRepeated trials can increase the corresponding expected value, but a good implemen-\ntation will be alert to repeated MAC verification errors.\nForgery attack: This attack relies on prediction of MACK(x) for a message x with-\nout initial knowledge of K.\nFor an input pair (x, x′) with MACK(x)=g(H) and\nMACK(x′)=g(H′), where g denotes the output transformation and H is a chaining\nvariable, a collision occurs if MACK(x) = MACK(x′). Its feasibility depends on an\nn-bit chaining variable and the MAC result. Given g that is a permutation, a collision\ncan be found using an expected number of\n√\n2 · 2n/2 known text-MAC pairs of at least\ntwo divided blocks each. A simple way to counter this attack is to ensure that each se-\nquence number at the beginning of every message is used only once within the lifetime\nof the key.\nImpersonating attack: Note that the AUTH, SCR and AS maintain a security associ-\nation with each other. Therefore, neither of them can be used to impersonate the other.\nInstead, this attack occurs between the STA and AUTH, which causes an authentication\nfailure or misconduct of the principals. Oracle-based impersonating attacks are that\nthe attacker exploits one of principals as an oracle to obtain cryptographic messages in\na session since it has no knowledge of K. The attacker applies the obtained messages\nto the other principal party in another session. For example, it runs a session with an\nAUTH to obtain a MAC value, impersonating a legitimate STA. It runs another session\nwith an STA and exploits the MAC value on the STA, impersonating the legitimate\nAUTH. This attack can be countered by exchanging nonce with each other and using a\nsequence counter.\n5.\nPERFORMANCE EVALUATION\nWe evaluate the efficiency of MAP via experimentation and simulation, contrasting it\nwith other protocols. We first describe simulation methodology and model and then\nanalyze the MAP’s performance benefits via the simulation results and in comparison\nwith other protocols. Finally, we discuss the storage overhead caused by security-\ncontext transfer.\n"
        },
        {
            "page_number": 265,
            "text": "262\nHAHNSANG KIM and KANG G. SHIN\n5.1. Simulation Methodology\nThe probe phase, discovering the next AP in WLAN handoffs, takes a large latency\n(ranging from 50ms to 350ms), depending on different vendors [22]. Even if the recent\neffort in [32] to reduce the latency by 84%, the large variance is an obstacle to highlight\nthe effectiveness of our protocol on a real testbed. We therefore use Matlab-based\nsimulation, relying on experimental data. We assume that network traffic is stable with\nsmall variations, e.g., the latency of establishing a (re)association with an AP including\nthe probe phase is 30ms with 3% jitter, and the round-trip time (RTT) between two\ncommunicating servers across a domain is about 20ms with 4% jitter. In addition, the\nRTT between the AP and SCR/AS is less than 3ms. We use these values throughout the\nsimulation. In cryptographic computations, we conducted an experiment using three\nmachines: Linux v.2.4.19 iPAQ 206MHz ARM processor with 64 megabyte memory\n(iPAQ), Linux v.2.4.2 Laptop Mobile Pentium 366MHz processor with 128 megabyte\nmemory (MP2) and Linux v.2.4.23 Desktop Intel Xeon 3Ghz bi-processor with 2GB\nmemory (Xeon). We compiled crypto libraries [12] in gcc v.3.3 with an option of\nLevel-1 optimization.\nTable 1. Throughput of hash/symmetric and asymmetric algorithms (in Megabit per sec-\nond)\nAlg.\\Pow.\niPAQ\nMP2\nXeon\nSHA-1\n15.8 Mbps\n18 Mbps\n104.9 Mbps\nSHA-256\n3.4 Mbps\n9 Mbps\n64.0 Mbps\nSHA-512\n0.2 Mbps\n4.3 Mbps\n24.8 Mbps\nMD5\n15.8 Mbps\n41 Mbps\n290.9 Mbps\nAES-128\n2.7 Mbps\n10 Mbps\n80 Mbps\nRSA enc.\n15.1 Kbps\n138.9 Kbps\n625 Kbps\nRSA dec.\n0.9 Kbps\n4.6 Kbps\n21.6 Kbps\nRSA sig.\n0.9 Kbps\n4.4 Kbps\n21.2 Kbps\nRSA ver.\n15.1 Kbps\n138.9 Kbps\n625 Kbps\nTable 1 shows the computation throughput of symmetric-key and public-key al-\ngorithms, respectively. With these measurement data, we numerically calculate the\ntime to perform each authentication protocol while ignoring the overhead of running\napplications for simplicity.\n"
        },
        {
            "page_number": 266,
            "text": "WIRELESS NETWORK SECURITY\n263\nFigure 6. The simulation model for inter-domain handoffs. A circle and hexagon indicate\nan AP’s radio coverage range and a domain, respectively. Each SCR controls its domain\nand is securely connected with its neighbor. An STA initially associates with a.2 and move\naround in its local domain (from a.5 to a.6 via a.4). It crosses Domain b and finally associates\nwith c.4 in Domain c.\n5.2. The Simulation Model\nFigure 6 shows the simulation model we used. Each AS constructs a domain\nconsisting of an SCR and several APs. The SCR and AS may reside on the same\nmachine as mentioned before.\nHandoff Pattern\nThe handoff pattern for STAs is basically random; the STAs cross the boundary\nafter hopping a random number of times. Random pattern is sufficient to evaluate the\noverall efficiency performance. Nevertheless, to notice the comparative effectiveness\nof our protocol, we additionally set a regular handoff pattern; after association in the\nhome domain, STAs hop three times and then cross a domain boundary. In a visited\ndomain, every five hops they traverse the domain.\nSCR Configuration\nWhether or not the “visitor\" can use storage resources in a domain affects the\nperformance of its handoffs. There can be three system configurations according to the\n"
        },
        {
            "page_number": 267,
            "text": "264\nHAHNSANG KIM and KANG G. SHIN\nstorage availability in the SCR of the visited domain. First, if only relaying security\ncontext is allowed, the authentication process takes place in the AS/SCR of the home\ndomain. The SCR in the visited domain serves as a relay agent. Second, if caching\nsecurity context is allowed, the foreign SCR serves as a proxy authentication server.\nIn this case, security context is transferred and stored in the visited domain, which\nenables avoiding contact with the home server. Third, if pre-caching security context\nis allowed somehow, i.e., security context is transferred to the foreign SCR before the\nSTA arrives, then the latency of fetching security context from the home server/SCR\ncan be eliminated. We will evaluate the caching effect via simulation.\n5.3. The Simulation Results\nMAP performs an optimized re-authentication procedure based on the security\ncontext generated after the initial authentication. It allows one to (1) consolidate the\nre-authentication procedure (with two-message exchanges, the mutual authentication\nis completed) and (2) avoiding contact with the home server from the visited domain.\nFigure 7 clearly shows that from re-authentication, the authentication latency dramati-\ncally drops by up to 45% thanks to (1). As a regular handoff pattern, after three hops\nin the local domain (the first handoff corresponds to the initial authentication in the\nfigure), the STA crosses the domain boundary at every 5 handoffs, which triggers the\nforeign SCR to request the security context from the home server. As a result, the la-\ntency increases in proportion to the RTT between the end-to-end points of two domains.\nEven if the STA roams in the foreign domain, it shows the same latency performance\nas in the home domain thanks to (2). In this case, the SCR in the foreign domain sup-\nports caching security context. After the 15-th handoff in the figure, the cross-domain\nauthentication encounters the case of relaying security context in the SCR of the visited\ndomain, which triggers the authentication procedure to be performed in contact with\nthe home server for each hop in the visited domain.\nFigure 8 shows the results with a random handoff pattern, illustrating the cumu-\nlative distributions of the authentication latency for three cases supporting SCR. The\nfigure shows the effect of pre-caching and caching security context to achieve more im-\nprovements in time efficiency than just relaying security context which is characteristic\nof the legacy protocols that are unable to generate security context. For example, more\nthan 70% and 80% of authentication processes in the cases of caching and pre-caching\nsecurity context, respectively, take less than 36ms.\nWe evaluated the increase in storage availability via the number of authentication\nrequests with a random handoff pattern. Figure 9 shows that the higher inter-domain\nhandoff frequency the home SCR has, the higher its storage availability. The x-axis is\nthe ratio of authentication request queries in inter-domain handoffs to the total number\nof queries, and the y-axis is the ratio of the network traffic in the foreign SCRs. Let\nAQr denote the foreign server’s overhead and AQl denote the home server’s overhead.\nThen, the ratio of the gain in storage availability with MAP to the overall overhead is\nexpressed as, 1−AQl/(AQl+AQr) which grows as the frequency of the inter-domain\nhandoffs increases.\n"
        },
        {
            "page_number": 268,
            "text": "WIRELESS NETWORK SECURITY\n265\n0\n5\n10\n15\n20\n25\n30\n30\n35\n40\n45\n50\n55\n60\n# of handoff\nSimulated authentication latency (ms)\nintra−domain\nhandoff\ninter−domain\nhandoff\nFigure 7. Authentication latency variations in different configurations of foreign servers\n30\n35\n40\n45\n50\n55\n60\n0\n0.2\n0.4\n0.6\n0.8\n1\nAuthentication latency (ms)\nCumulative Distribution (%)\nPre−caching SC\nCaching SC\nRelaying SC\nFigure 8. Cumulative distributions of authentication latency under each different configu-\nration. SC stands for security context.\nAs shown in Figure 10 that plots the results with a random handoff pattern, the\nperformance in authentication efficiency (caching allowed) improves up to 53% over\na legacy method (relaying allowed) until the end-to-end domain distance continues\n"
        },
        {
            "page_number": 269,
            "text": "266\nHAHNSANG KIM and KANG G. SHIN\n0.1\n0.2\n0.3\n0.4\n0.5\n0.6\n0.05\n0.1\n0.15\n0.2\n0.25\n0.3\n0.35\n0.4\n0.45\nY = 0.016+ 0.76X\nMSE = 0.015\nDegree of cross−domain authentication occurrence ratio\nStorage availability ratio\nSimulation\nRegression\nFigure 9. System storage availability affected by the inter-domain handoff authentication\noccurrence ratio. MSE is Mean Squared Error of the above regression function.\nto increase up to RTT=100ms. In case of security context pre-cached in the visiting\ndomain, MAP makes a 10% additional improvement with RTT=100ms. Therefore, the\neffectiveness of MAP increases dramatically as the distance gets larger.\n5.4. Comparison with Other Protocols\nFigure 11 shows the cumulative CPU usage (represented in millisecond) crypto-\ngraphic primitives of required in ten consecutive times of authentication in symmetry-\nkey-based protocols including MAP, MNS, and Kerberos, and public-key-based pro-\ntocols including PNS and TLS. We chose one-way hash functions (i.e., MD5 [30],\nSHA [3]) and block ciphers (i.e., AES [5]) for symmetric-key protocols, and RSA [16]\n1024-bit modulus for the public-key protocols. The symmetric-key protocols are shown\nto be two orders of magnitude faster thanks to the inherent advantage over modulo op-\nerations. MAP is faster than the MNS and Kerberos protocols, respectively, by 12.6%\nand 21.5% CPU usage gains. This is a considerable impact on the performance gain in\nview of millions of runs for authentication in a single server.\nRegarding the number of message exchanges, MAP achieves the cross-domain\nauthentication only with 2-way handshake, the cost of which is minimal, compared to\nMNS and Kerberos requiring 3-way and 4-way handshakes, respectively. This con-\ntributes to the further enhancement of latency performance. Figure 12 shows the com-\nparison of authentication latency of MAP with that of the MNS and Kerberos protocols\nwhile mobiles are hopping with a regular pattern. MAP outperforms the others in both\ninter- and intra-domain roaming. It accounts for 74% of cross-domain authentication\n"
        },
        {
            "page_number": 270,
            "text": "WIRELESS NETWORK SECURITY\n267\n10\n−3\n10\n−2\n10\n−1\n0\n0.2\n0.4\n0.6\n0.8\n1\nRelative end−to−end domain distance (ms/max(L))\nAuthentication latency degree\nRelaying SC\nCaching SC\nPre−caching SC\nFigure 10. End-to-end domain distance vs. authentication latency. The distance is scaled\ndown at a rate of the maximum authentication latency (max(L)). SC stands for security\ncontext.\n10\n0\n10\n1\n10\n2\n10\n3\n10\n4\nStacks of Authentication Latency (ms)\nCryptographic protocols\nMAP\nMNS\nKerberos\nPNS\nTLS\nFigure 11. CPU utilization. Ten consecutive times of authentication. MNS and PNS stand\nfor modified symmetry-key-based and public-key-based Needham Schroeder protocols,\nrespectively. TLS is Transport Layer Security protocol.\nlatency of Kerberos and 85% of that of MNS. It reduces the intra-domain authentication\nlatency by 5% for Kerberos and 7% for MNS.\n"
        },
        {
            "page_number": 271,
            "text": "268\nHAHNSANG KIM and KANG G. SHIN\n0\n5\n10\n15\n20\n25\n30\n30\n40\n50\n60\n70\n80\n# of handoffs\nSimulated authentication latency (ms)\nMAP\nMNS\nKerberos\nFigure 12. Latency comparison of MAP with MNS and Kerberos. Fetching security context\nwhile the mobile crosses the boundary increases the authentication latency.\n5.5. Storage Overhead\nSecurity context is transferred and stored in a foreign server (SCR) for cross-\ndomain authentication. It consists mainly of a set of AVPs each of which is composed\nof nonce (128 bits), MAC (128 bits), DK (128 or 256 bits) and Identity (about 320 bits).\nIn addition, a value (of 40 bits) may be reserved for security context validity and other\ninformation. The security context can be of 64·n+45 bytes where n is the number of\nAVPs. Approximately, given a 1 kilobyte security context per STA, manipulating one\nmillion STAs requires 1 gigabyte storage capacity, which is usually a small overhead\nto the server system.\n6.\nRELATED WORK\nThere have been several studies on how to achieve fast handoffs and enhance the per-\nformance of authentication mechanisms, including WLAN protocols.\nMichra et al. [24, 23] presented a keys distributing method by means of proactive\ncontext caching. The idea of proactive caching is for an AP to broadcast its cached\ncontext to its neighbor APs in advance by using neighbor graphs and IAPP. However,\nthis method is limited to intra-domain handoffs since APs are required to be functionally\nidentical. Pack et al. [29] presented a pre-authentication method that skips the 802.1X\nauthentication phase by distributing the key to a certain number of selected APs and\ncomputing the likelihood based on the analysis of past network behavior. Bargh et\nal. [8] presented the applicability of the pre-authentication method for inter-domain\n"
        },
        {
            "page_number": 272,
            "text": "WIRELESS NETWORK SECURITY\n269\nhandoffs. However, a pre-authentication method creates a higher risk of compromising\nsecurity.\nWong et al. [35] proposed a hybrid protocol based on a certificate containing a\nsymmetric key signed with a public key which is suitable for wireless communications.\nAn asymmetric method for wireless communications presented in [15] uses Diffie-\nHellman key exchange combined with Schnorr signatures. In addition, there are several\nlegacy authentication protocols [36, 37, 31, 11, 28, 13] for the general purpose in the\nliterature.\nThere are several approaches to analyzing the security of authentication protocols.\nOne is the formal methods that model and verify the protocol using specification lan-\nguages and verification tools [21]. It consists of model checking and theorem-proving\nmethods. Application examples [19, 25, 14, 20] demonstrated the feasibility of for-\nmally verifying the authentication protocols with general-purpose verification tools.\nAlso proposed in [9, 34, 7] are modular approaches aiming to establish a sound formal-\nization and a security analysis for the authentication problem.\n7.\nCONCLUSIONS\nThe cross-domain authentication requires retrieval of security context from the server\nof the previously-visited or home domain. Contacting a remote server may increase the\nauthentication latency significantly. the longer the end-to-end distance, the larger the\nlatency reduction. The longer the end-to-end distance, the larger the latency reduction.\nIf security context is allowed to be pre-cached/transferred before the mobile arrives,\nthe latency can be reduced significantly. In this chapter we designed and evaluated\na mobility-adjusted authentication protocol, MAP, by leveraging symmetric-key cryp-\ntography for cross-domain authentication and key generation. MAP can be configured\nto make tradeoffs between performance and storage usage. MAP introduces three con-\ncepts to the cross-domain authentication: (1) a re-authentication mechanism based on\na 2-way handshake; (2) the temporary-key generation of the IEEE 802.11i authentica-\ntion; and (3) security context eliminating the need to contact a remote server. MAP\nperforms best in cases of long end-to-end domain distances and high cross-domain\nauthentication traffic.\n8.\nREFERENCES\n1. Authentication Authorization and Accounting IETF WG.\n2. Wi-fialliance. http://www.wi-fi.org/.\n3. Secure Hash Standard. In Federal Information Processing Standards Publication 180-1. NIST, Apr.\n1995.\n4. Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications:\nSpecification for Robust Security. In ANSI/IEEE Std 802.11: 1999(E). ISO/IEC 8802-11, 1999.\n5. Advanced Encryption Standard (AES). In Federal Information Processing Standards Publication 197.\nNIST, Nov. 2001.\n"
        },
        {
            "page_number": 273,
            "text": "270\nHAHNSANG KIM and KANG G. SHIN\n6. Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications:\nSpecification for Robust Security. In IEEE Std 802.11i/D3.1. ISO/IEC 8802-11, 2003.\n7. William Aiello, Steven M. Bellovin, Matt Blaze, Ran Canettiand John Ioannidis, Angelos D. Keromytis,\nand Omer Reingold. Efficient, DoS-Resistant, Secure Key Exchange for Internet Protocols. In Conf.\non Computer and Comm. Security. ACM Press, 2002.\n8. M. S. Bargh et al. Fast Authentication Methods for Handovers between IEEE 802.11 Wireless LANs.\nIn Int. Workshop on Wireless Mobile App. and Services on WLAN Hotspots (WMASH), pages 51–60.\nACM, Oct. 2004.\n9. Mihir Bellare, Ran Canetti, and Hugo Krawczyk. A Modular Approach to the Design and Analysis\nof Authentication and Key Exchange Protocols. In 30th Symposium on Theory of Computing, pages\n419–428. ACM Press, 1998.\n10. Michael Burrows, Martin Abadi, and Roger Needham. A Logic of Authentication. Technical Report 39,\nDigital Equipment Corporation, Palo Alto Calif., February 1989.\n11. Michael Burrows, Martin Abadi, and Roger Needham. A Logic of Authentication. ACM Trans on\nComput. Systems, 8(1):18–36, 1990.\n12. Wei Dai. Crypto++, http://www.eskimo.com/∼weidai/ cryptlib.html.\n13. Tim Dierks, Alan O. Freier, Martin Abadi, Ran Canetti, Taher Elgamal, Anil R. Gangolli, Kipp E.B.\nHickman, and Hugo Krawczyk. Transport Layer Security protocol version 1.0. RFC 2246, Jan. 1999.\n14. James Heather and Steve Schneider. Towards automatic verification of authentication protocols on an\nunbounded network. In 13th Computer Security Foundations Workshop, page 132. IEEE Computer\nSociety, 2000.\n15. Markus Jakobsson and David Pointcheval. Mutual Authentication for Low-Power Mobile Devices. In\nFinancial Cryptography 2001, Grand Cayman Island, British West Indies, Feb. 2001.\n16. Burt Kaliski. PKCS #1 RSA Encryption Version 1.5. RFC 2313, Mar. 1998.\n17. John Kohl and B. Clifford Neuman. The kerberos network authentication service (v5). RFC 1510, Sep.\n1993.\n18. Hugo Krawczyk, Mihir Bellare, and Ran Canetti. HMAC: Keyed-Hashing for Message Authentication.\nRFC 2104, Feb. 1997.\n19. Gavin Lowe. Breaking and fixing the Needham-Schroeder public-key protocol using FDR. In Tools\nand Algorithms for the Construction and Analysis of Systems (TACAS), volume 1055, pages 147–166.\nLNCS, 1996.\n20. Paolo Maggi and Riccardo Sisto. Using spin to verify security properties of cryptographic protocols.\nIn 9th SPIN Workshop on Model Checking of Software, volume 2318, pages 187–204. LNCS, 2001.\n21. Catherine A. Meadows. Formal Verification of Cryptographic Protocols: A Survey. In ASIACRYPT:\nInt. Conf. on the Theory and Application of Cryptology, volume 917, pages 135–150. LNCS, Dec. 1994.\n22. Arunesh Mishra, Minho Shin, and William Arbaugh. An Empirical Analysis of the IEEE 802.11 MAC\nLayer Handoff Process. SIGCOMM: CCR, 33(2):93–102, 2003.\n23. Arunesh Mishra, Minho Shin, and William Arbaugh. Context Caching using Neighbor Graphs for Fast\nHandoffs in a Wireless Network. In INFOCOM, Hong Kong, Mar. 2004. IEEE.\n24. Arunesh Mishra, Minho Shin, and William Arbaugh. Pro-active Key Distribution using Neighbor\nGraphs. Wireless Comm. Magazine, 11 Issue 1:26–36, Feb. 2004.\n25. J. C. Mitchell, M. Mitchell, and U. Stern. Automated analysis of cryptographic protocols using murφ.\nIn Symposium on Security and Privacy, pages 141–153. IEEE Computer Society, 1997.\n26. Roger M. Needham and Michael D. Schroeder. Using encryption for authentication in large networks\nof computers. Comm. of the ACM, 21(12):993–999, 1978.\n"
        },
        {
            "page_number": 274,
            "text": "WIRELESS NETWORK SECURITY\n271\n27. Roger M. Needham and Michael D. Schroeder. Authentication Revisited. SIGOPS: OSR, 21(1):7,\n1987.\n28. Dave Otway and Owen Rees. Efficient and timely mutual authentication. SIGOPS: OSR, 21(1):8–10,\n1987.\n29. Sangheon Pack and Yanghee Choi. Pre-Authenticated Fast Handoff in a Public Wireless LAN based\non IEEE 802.1x Model. In IFIP TC6 Personal Wireless Comm. 2002, Singapore, Oct. 2002.\n30. Ronald L. Rivest. The MD5 Message-Digest Algorithm. RFC 1321, Apr. 1992.\n31. M. Satyanarayanan. Integrating security in a large distributed system. Trans. on Comp. Systems,\n7(3):247–280, 1989.\n32. Minho Shin, Arunesh Mishra, and William Arbaugh. Improving the Latency of 802.11 hand-offs using\nNeighbor Graphs. In Mobisys, Boston, Jun. 2004. ACM.\n33. R. Shirdokar, J. Kabara, and P. Krishnamurthy. A QoS-based Indoor Wireless Data Network Design\nfor VoIP. In Vehicular Technology Conf. (VTC’01), volume 4. IEEE, Oct. 2001.\n34. Duncan S. Wong and Agnes H. Chan. Efficient and Mutually Authenticated Key Exchange for Low\nPower Computing Devices. In ASIACRYPT: Int. Conf. on the Theory and Application of Cryptology\nand Information Security, volume 2248, pages 272–289. LNCS, 2001.\n35. Duncan S. Wong and Agnes H. Chan. Mutual Authentication and Key Exchange for Low Power\nWireless Communications. In MILCOM 2001, pages 39–43, USA, Oct. 2001. IEEE Press.\n36. Thomas Y. C. Woo and Simon S. Lam. A Lesson on Authentication Protocol Design. SIGOPS: OSR,\n28(3):24–37, 1994.\n37. Yuqing Zhang, Chunling Wang, Jianping Wu, and Xing Li. Using SMV for cryptographic protocol\nanalysis: a case study. SIGOPS: OSR, 35(2):43–50, Apr. 2001.\n"
        },
        {
            "page_number": 275,
            "text": "11\nAAA ARCHITECTURE AND AUTHENTICATION\nFOR WIRELESS LAN ROAMING\nMinghui Shi, Humphrey Rutagemwa, Xuemin (Sherman) Shen\nand Jon W. Mark\nDepartment of Electrical and Computer Engineering\nUniversity of Waterloo\nWaterloo, Ontario, N2L 3G1, Canada\nE-mail: {mshi, humphrey, xshen, jwmark}@bbcr.uwaterloo.ca\nYixin Jiang, Chuang Lin\nDepartment of Computer Science and Technology\nTsinghua University\nBeijing, 100084, P.R. China\nE-mail: {yxjiang, clin}@csnet1.cs.tsinghua.edu.cn\nA wireless LAN service integration architecture based on current wireless LAN hotspots\nis proposed to make migrating to new service cost effective. The AAA (Authentication,\nAuthorization and Accounting) based mobile terminal registration signaling process is dis-\ncussed. An application layer end-to-end authentication and key negotiation protocol is\nproposed to overcome the open air connection problem existing in wireless LAN deploy-\nment. The protocol provides a general solution for Internet applications running on a mobile\nstation undervariousauthenticationscenariosandkeeps the communications private to other\nwireless LAN users and foreign networks. A functional demonstration of the protocol is\nalso given. The research results should contribute to rapid deployment of wireless LANs\nhotspot service.\n1.\nINTRODUCTION\nIEEE 802.11b/g wireless LAN products have become a de facto standard com-\nponent in mobile devices. An increasing number of wireless Internet access services\nhave been appearing in places such as airports, cafes, and bookstores. Annual industry\nrevenues already exceed US$1 billion, and is expected to pass US$4 billion by 2007 [1].\n"
        },
        {
            "page_number": 276,
            "text": "274\nMINGHUI SHI et al.\nFigure 1. Wireless service integration architecture\nIn addition, mobile devices with both cellular network and wireless LAN acess are be-\ncoming widely available. Demand for integrating multiple mobile computing services\ninto a single entity is preeminent.\nFigure 1 shows a global architecture of the public wireless Internet. Mobile IP\nis used throughout the architecture to support user roaming. The home network has\na network prefix matching to that of the home address of the mobile terminal. When\nthe mobile terminal moves from one network to another, such as roaming between\nforeign networks, it performs registration and updates its registration information with\nitshomeagent(HA),eitherdirectlyorindirectly. ThehotspotsaremostlybasedonIEEE\n802.11b/g WLAN standard. The layout of the hotspots can be adjacent or distributed in\nthe cellular networks. When the services of WLAN and cellular network are integrated,\nthemobileuserscanroambetweentheheterogeneousnetworksbyseamlesslyswitching\nbetween the associated network access interfaces. Two types of handoff may occur:\nhorizontal handoff and vertical handoff. Horizontal handoff supports user roaming\nbetween WLANs or cellular networks, and vertical handoff handles roaming between\nWLAN and the cellular networks. Since most WLAN hotspots do not overlap, vertical\nhandoff is most common seen when the mobile terminal enters and leaves the WLAN\nhotspot. Thischapterfocusesonproposingahigh-performancesecurecellularnetwork-\nintegratable wireless LAN service framework.\nMany wireless Internet service providers (WISPs), such as T-mobile, provide pub-\nlic wireless LAN Internet access at hotspots using a network access server (NAS). The\nNAS allows only legitimate customers to use the service and provides intra-domain\nroaming because the hotspots from one WISP share the same customer base. However,\nit lacks an architecture to provide inter-domain roaming and Mobile IP support. A user\ncannot access hotspots of service provider B with his/her account from A, even though\nA and B would like to have a roaming agreement.\n"
        },
        {
            "page_number": 277,
            "text": "WIRELESS NETWORK SECURITY\n275\nIn interworking implementation, handoff delay should also be considered. Mobile\nIP handoff delay can be divided into two parts: movement detection and signaling for\nregistration. Several proposed approaches are actually only effective on registration\nsignaling delays. A micro-mobility approach [2] divides a network in a hierarchical\nmanner, and location management is handled locally when the mobile station moves\nwithin a smaller area at the lower hierarchy level. For simultaneous binding in [3],\nmultiple care-of address bindings for the mobile station are maintained and packets\ndestined for the mobile station are transmitted to all care-of addresses to reduce packet\nloss during handoff. However, it cannot be used in an IEEE802.11 network, because\ncurrent wireless LAN cards can only access one access point or channel at a time.\nOn the other hand, Wired Equivalency Protocol (WEP) is a security protocol for\nWLAN defined in the 802.11b standard and is designed to provide the same level of\nsecurity as that of a wired LAN. However it has several problems in both transmission\nprivacy and deployment. Various studies show that WEP is vulnerable to several attacks\n[4, 5], especially in a heavily loaded wireless network. WEP uses a single key shared\nbetween the access point and clients. Malicious clients are able to tap into the commu-\nnication traffic of other clients who are associated with the same access point. Most\nhotspots do not use data encryption due to this technical limitation. Authentication can\nbe used to negotiate a shared session key to further encrypt data traffic in the session\n[6, 7, 8, 9]. Although there are many authentication protocols published, they do not\ngenerally support Internet applications for wireless mobile devices. For example, the\nauthentication protocols proposed in [19, 20, 21] allow a mobile station to communicate\nwith another one directly, but there is no solution for a mobile station to communicate\nwith a fixed Internet server, which is found in FTP (File Transfer Protocol) applications.\nProtected transmission based on Secure Shell (SSH) and/or Secure Sockets Layer\n(SSL) has been suggested to secure wireless transmission. SSH requires a previously\ngenerated public/private key pair, so it may never be applied to authentication between\nparties who have not contacted with each other before. SSL is not suitable for extension\nto mobile wireless Internet either, because the operation of SSL relies on certification\nverification by certificate authority (CA) servers. It is not practical for CA servers\nto store the certificate of every mobile station because the number of mobile stations\nis too large (for the same reaso, client authentication in SSL is optional). The home\nnetwork would not like to register every mobile station to CA servers either. In the case\nof a wireless LAN hotspot, the service access is controlled by medium access control\n(MAC) addresses of the mobile stations. Usually there is no key negotiation during the\nnetwork authentication and Mobile IP registration phases.\nThe objective of this chapter is to propose a secure wireless LAN service integration\narchitecture and necessary signaling process design. It is divided into three categories:\nIEEE802.11 service integration functionality: The architecture should be able\nto integrate into cellular networks. Since third-generation (3G) and beyond cellular\nnetworks use or very likely will continue to use AAA structure and protocol to control\nnetwork access and manage user accounts, the IEEE802.11 roaming architecture and\nsignaling processes should work with cellular networks.\n"
        },
        {
            "page_number": 278,
            "text": "276\nMINGHUI SHI et al.\nWireless network security: The security issues include network access control,\nuser account management, and transmission privacy. The first two items can be taken\ncare of by using the AAA structure. For the third item, a wireless LAN hotspot has no\ngeneral solution to guarantee data transmission privacy due to the poor design of WEP.\nService quality: It mainly refers to handoff speed and packet loss rate. Naive\nhandoff acceleration solutions do not apply to an IEEE802.11 network interface cards,\nbecause they can only talk to one another, so the solutions cannot guarantee no packet\nloss.\nThe remainder of the chapter is organized as follows. We first overview AAA\nmechanism and propose the AAA based infrastructure of wireless LAN roaming. We\nthen present a security mechanism for wireless LAN transmission and related demon-\nstration results, followed by a summary of the work.\n2.\nAAA OVERVIEW\nAAA is an architectural framework for configuring a set of three independent\nsecurity functions in a consistent manner. AAA provides a modular way of performing\nthe following services:\n1. Authentication is the way a user is identified prior to being allowed access to\nthe network and network services. AAA authentication can be configured by\ndefining a named list of authentication methods, and then applying that list to\nvarious interfaces on a per-user or per-service basis.\n2. Authorization is the process of determining whether the actions, such as ac-\ncessing a resource is permitted for the corresponding identity. Authorization\nworks by assembling a set of attributes that describe what the user is autho-\nrized to perform. These attributes are compared to the information contained\nin a database for a given user and the result is returned to AAA to determine\nthe user’s actual capabilities and restrictions. Remote security servers, such\nas RADIUS[13, 14, 15] and TACACS+[16, 17] , authorize users for specific\nrights by associating attribute-value pairs, which define those rights with the\nappropriate user.\n3. Accounting tracks the services that users are accessing as well as the amount\nof network resources they are consuming. When accounting is activated, the\nnetwork access server reports user activity to the RADIUS or TACACS+ secu-\nrity server in the form of accounting records. This data can then be analyzed\nfor network management, client billing, and/or auditing.\nAccess control is the way to control who will be allowed to access to the network server\nand what services they are allowed to use. AAA network security services provide the\nprimary framework to set up access control on the router or access server. The three\nsecurity functions are used together, for example networks or services, to control which\nusers are allowed access, what functions they are allowed to use and how much resource\n"
        },
        {
            "page_number": 279,
            "text": "WIRELESS NETWORK SECURITY\n277\nIdentity Authenticaton\nUser / Subject\nAuthorization Database\nAuditing Database\nAccess \nControl\nResource / Objects\nFigure 2. The relations between access control and AAA\nthey have used. Access Control protects system resources against unauthorized access.\nThe use of system resources is regulated according to a security policy and is permitted\nby only authorized entities (users, programs, processes, or other systems) according to\nthat policy [10]. In this chapter, AAA is adopted to excise service admission for users\nin WLAN roaming.\nAAA deals with access control of systems and environments based on policies set\nby the administrators and users of the systems. The access policy may be implied in\nboth the authentication, which can be restricted by the time of day, number of sessions,\ncalling number, etc., and the attribute-values authorized [2]. Access control provides\nthe limited access according to the authorization policies between a subject and objects\nwhen a subject access the related resources (objects). Objects mainly include passive\nentities (file, storage area) while subjects mainly contain active entities (processes,\nusers). Subjects obtain information by accessing objects. Fig. 2 shows the basic\nrelations between access control and AAA services.\nWhen a user, or the subject,\nneeds to access the resources, the authentication is first preformed to verify its identity.\nAccording to the access policy corresponding to the user’s identity, which is stored in\nauthorization database, access control allows the user to access the defined resource.\nThe activity of the subject during the session is logged in the auditing database.\nAAA scheme provides the following benefits: (1) increased flexibility and control\nof access configuration; (2) scalability; (3) standardized authentication methods, such\nas RADIUS, TACACS+, and Kerberos [18]; (4) multiple backup systems. In many\ncircumstances, AAA uses protocols such as RADIUS, TACACS+, or Kerberos to ad-\nminister its security functions. If the router or the access server is acting as a network\naccess server, the communication is established between the network access server and\n"
        },
        {
            "page_number": 280,
            "text": "278\nMINGHUI SHI et al.\nFigure 3. The proposed network structure for IEEE802.11 service integration\nthe RADIUS, TACACS+, or Kerberos security server through AAA. In wireless LAN\nenvironment, there is a strong application requirement for Mobile IP AAA [12].\n3.\nIEEE 802.11 WIRELESS LAN ROAMING\nFigure 3 shows the infrastructure of the mobile networks under consideration.\nThe Internet offers much larger bandwidth and lower transmission error rates than\nwireless links. The home network is considered as a private network, which only\nallows its users access. The foreign networks are the real WISPs. After completion of\na registration process, the mobile station and the corresponding foreign network will\nshare a key for further encrypted communications. Fixed nodes represent common\nweb sites. Authentication is required for accessing some of those sites. The cellular\nnetworks and base stations are 3G based. Access points, which form a hotspot, are\nthe attachment points that allow mobile stations to access the network through wireless\nconnection. A mobile station, as a member of its home network, is allowed to access the\nresources in the home network whenever it is within or outside the home network. CA\nservers are special servers that issue and verify certificates of fixed nodes or networks\nupon request so that they have proofs to identify themselves. CA servers are organized\nin a tree topology and working in a distributed way, so that it is not necessary to connect\nall Internet servers to one CA server. Mobile stations do not contact CA severs directly\nbecause of the large population size. CA shares independent secret key with the servers\nwhich it is connected with.\nThe proposed IEEE802.11 roaming structure is based on an AAA broker with\na Remote Authentication Dial-In User Service (RADIUS) server proxy.\nRADIUS\nis popular and easier to use for integrating hotspot service into AAA based cellular\nnetworks. The broker model is suitable for large-scale and commercial implementation\n"
        },
        {
            "page_number": 281,
            "text": "WIRELESS NETWORK SECURITY\n279\nNAS / AAAF\nAAAH\nFigure 4. Network structure of AAA brokers\nbecause a RADIUS server can simply have one simple security association or a pre-\nsetup shared secret with the RADIUS proxy. RADIUS proxies forward authentication\nand accounting requests from different domains to their destination.\n3.1. Radius Proxy\nRADIUS servers of multiple ISPs can be interconnected via a series of forward-\ning servers. The RADIUS server retrieves the remote servers domain from the users\nrequest that includes the network access identifier (NAI) [22, 23, 24] in the form of\nidentifier@domain name, which identifies a users name and the domain where the user\ncomes from. Then it forwards the request to the remote server identified by the domain.\nThe remote server also replies through the forwarding server.\nA group of forwarding servers with secured communication tunnels between each\nother are used as AAA brokers (AAABs). Figure 4 illustrates the network structure of\nAAABs. A mobile user whose NAI is alice@homedomain.org moves from its home\ndomain to another domain (e.g., foreigndomain.org). The NAS located in the foreign\ndomain authenticates the mobile user, and forwards this request via RADIUS protocol\nto the foreign AAA (AAAF). According to NAI, the AAAF forwards the request to the\nhome AAA (AAAH) through the AAABs.\nWhen the number of domains increases, it is no longer feasible to connect all\nthe AAA servers to one AAAB network. The AAABs will be grouped according to\ngeographical distribution of the network domains. In this way the complexity of each\nAAA broker can be reduced. The performance of an AAAB cluster is evaluated by the\nnumber of hops to forward the AAA request from the originator to the destination.\n"
        },
        {
            "page_number": 282,
            "text": "280\nMINGHUI SHI et al.\nInternet\nHome Domain\nForeign Domain\nAccess Point\nMIP registration\nWireless LAN \nGateway\nClient\nAAAF\nAAAH\nHA\nNAS/FA\nAAABs\nMIP FA Adv w/ \nChallenge\nFigure 5. Wireless LAN roaming\n3.2. IEEE 802.11 Horizontal Roaming\nThe IEEE802.11 horizontal roaming architecture is shown in Figure 5. The hotspot\nis connected to the Internet through a gateway. Each network domain is interconnected\nby AAABs. In order to provide IP mobility, the functionality of a foreign agent (FA)\nis placed into the NAS. The FA located in the NAS periodically sends advertisements\nwith challenge packets, and all mobile stations register via the FA. The challenge is\na piece of data used to verify if the user device has knowledge of the secret (e.g., a\npassword) without sending it explicitly via a communication link. The architecture is\nable to process two horizontal roaming scenarios:\nThe current IEEE802.11 device connects to the network via the NAS: The\nnetwork can provide IP mobility, however the roaming is not seamless. When a mobile\nuser requests Mobile IP services by sending Mobile IP Registration, the NAS blocks\nthe Registration until the mobile user has been authenticated via the AAA architecture.\nThe NAS prompt the user to enter his or her credential, such as the username and the\npassword for authentication. Once the mobile user is authenticated successfully, the\nnormal Mobile IP registration will retained.\nSeamless roaming: Authentication is completely done by the home agent (HA).\nThemobilestationisrequiredtosupportMobileIPChallenge/Responseextensionswith\na Mobile-AAA authentication extension so that the user credential can be processed by\nthe program automatically.\nIn the following, we focus on developing efficient signaling process for the two\nroaming scenarios. The design shares as many common signaling messages as possible.\nIn order to have further integration with 3G cellular networks, the signaling process\nshould also be able to share with the AAA signaling process for 3G networks. Based\non the architecture in Figure 3, Figure 6 illustrates the internal design of an NAS/FA.\nIt has two modes: one for compatibility of current wireless LAN deployment, and the\nother for seamless wireless LAN roaming.\nIn the compatible mode, when a mobile station registers, it may use its home\naddress or the mobile station NAI to identify itself in its Registration Request. A\nmobile station associates itself with an access point and starts sending IP packets, such\nas Mobile IP requests, to its HA via an FA that relays the Registration Request. After\nthe HA authenticates the request and sends a reply via the same FA, the HA and FA\n"
        },
        {
            "page_number": 283,
            "text": "WIRELESS NETWORK SECURITY\n281\nWait for user \ncredentials\nStart authentication\nAuthenticated?\nLoad authentication \nweb page\nForward MIP Reg. \nReq. / Challenge to \nHA\nNo\nYes\nAccept?\nNo\nYes\nNetwork \naccess granted\nNetwork access \ndenied\nForward Access \nRequest to HA\nCompatible or \nseamless type?\nCompatible, \nMIP Reg. Req.\nSeamless, MIP Reg.\n Req. + Challenge\nMessage received\nFigure 6. The flow diagram of the NAS/FA\nboth update their bindings. Sometimes an FA forces all its serving mobile stations to\nregister through it. If a mobile station does not send the user credentials, including\nNAI and password, along with the Mobile IP request, the user will be redirected to a\nlogin page. By extracting the domain portion of NAI, the authentication request will be\nforwarded to the AAA server of the WISP. After successful authentication, the Mobile\nIP request is forwarded to the HA of the mobile station and the Mobile IP registration\nis completed.\nIn seamless roaming mode, a mobile station associates with the access point and\nresponds to a Mobile IP FA Advertisement packet with a Challenge, and sends the\nMobile IP Registration Request with the NAI and Challenge. The user credentials are\nincluded in the Mobile IP request. When the FA receives the reply from the mobile\nstation, it realizes that the mobile station can do seamless roaming. It encapsulates the\nrequest in the AMR and forwards it to the AAAH and HA. After the HA processes\nthis request, it sends a HAA containing Mobile IP Registration Reply. The AAAH and\nAAAL forward the encapsulated Mobile IP Registration Request to the FA in the AMA\npackets. The FA then sends a Mobile IP Registration Reply.\nComparing signaling processes of the two methods, it can be seen that the processes\nare designed to be quite similar, such as the signaling messages and the signal path.\n"
        },
        {
            "page_number": 284,
            "text": "282\nMINGHUI SHI et al.\nSo some components in the network do not need to differentiate the message type for\neach mode. Only one signaling processing mechanism needs to be designed. The FAs\nown local clients still can access the hotspot as they can use AAAF to authenticate\nthemselves.\n3.3. Mobile IP Handoff Performance Improvement\nIn order to roam between a wireless LAN and a cellular network, the mobile station\nshould be equipped with corresponding network access interfaces. The data packets\nfrom the corresponding server are routed to the mobile station through its HA. When the\nmobile station roams to the foreign network, the two network access cards are assigned\na temporary care-of address by the FA.\nThe switching of the two interfaces can be considered a care-of address change in\nMobile IP. When the mobile station decides to switch the interface, it informs the HA\nby updating its current care-of address to the IP address of the other network access\ninterface. The HA redirects the data flow to the new IP address. This method ensures\nthat the process of network access interface switch over is dealt with using the switching\nprocess in Mobile IP.\nFor typical data applications such as web surfing, it is not necessary to use a\nreal-time seamless handoff as for cellular telephony. A gap of a few seconds while\na connection is being rerouted should be fine with the applications. However, with\nthe growth of real-time Internet applications, like voice or streaming video, Mobile IP\nhandoff latency and packet loss performance have become more and more critical. In\norder to provide high-quality applications in a wireless LAN environment, the key issue\nis to support efficient and seamless network handoff. When a mobile station moves\nfrom the coverage of one access point to another, it re-associates with a new AP. This is\ncalled a layer 2 handoff. On the other hand, a Mobile IP handoff (layer 3) is the process\nthat takes place when changing FAs. The latency generated by both layer 2 and Mobile\nIP handoffs should be reduced. In order to reduce the latency of Mobile IP handoff in a\nwireless LAN, link layer update frames and movement notification packets can be used\nto assist Mobile IP handoff. A MAC bridge or data tunnel is established between the\nnew FA and old FA servers to improve the latency of Mobile IP handoff in the wireless\nLAN environment. The pre-registration and authentication data can be sent to the\nmobile station before it moves, and/or the data packets that arrive at the old FA during\nmovement can be sent to the mobile station via the new FA. Additional flow control\nshould be taken in the handover period, because the connection speed of the old and new\naccess point/base station could be quite different if the mobile station performs handoff\nbetween IEEE 802.11 and cellular networks. If the data source is not informed in a\ntimely way, data may block the channel if the device is moving from high speed to low\nspeed connection, or the user cannot get better quality of service otherwise. Therefore,\neffective congestion control is very important, especially for media streaming service\nthat uses the protocol without an inherent congestion mechanism. Measures should\nalso be taken to ensure that the pre-authentication data transfer between the two FAs is\nprivate and unaltered. So the two FAs authenticate each other via a CA server using the\n"
        },
        {
            "page_number": 285,
            "text": "WIRELESS NETWORK SECURITY\n283\nHome agent\n in N’s home network\nHome agent\n in M’s home network\nFixed Internet nodes\n(Restricted)\nCA Servers\nFixed Internet nodes\n(public)\nMobile station N\nMobile station M\nAuth. required, \nno shared key\nAuth. required, \nshared key\nNo auth, required\nshared key\nNo auth, requred\nno shared key\nFigure 7. Authentication for Internet applications\nprotocol proposed in the next section, and a temporary session key can be negotiated\nto encrypt the pre-authentication data.\n4.\nWIRELESS TRANSMISSION PRIVACY\nAlthough the architecture proposed earlier prevents an unauthorized user from\nusing the service, the wireless transmission is still kept open. Using built-in WEP\nencryption cannot guarantee data transmission privacy in a public hotspot, since a\nWEP key is unique for each access point and there is no privacy among the mobile\nstations associated with an access point. A separate authentication and key negotiation\nmechanism is required to keep wireless transmission private.\nThis section presents a protocol that operates at the application layer to avoid\nany hardware or low-level protocol modification, and the authentication messages are\ncarried in the payload of data packets used in Mobile IP networks. User location updates\nare transparent to the protocol since user mobility is handled at the network layer. It\nis an end-to-end solution, so it secures not only the wireless data link in hotspots, but\nalso the entire data path. The FA just forwards the authentication message between the\nmobile station and its home network, and vice versa.\n4.1. Analysis of Authentication for Current Internet Applications\nFigure 7 shows various types of authentication with different security requirements,\nwhich may occur in applications running on a mobile station. For example, mobile user\n"
        },
        {
            "page_number": 286,
            "text": "284\nMINGHUI SHI et al.\nM and its home network shares a secret key. Its home network may only be accessed by\nM. On the other hand, a fixed public Internet site may be visited without authentication.\nFor clarity, these situations are sorted into three categories:\nAuthenticating parties share a secret key: authentication between a mobile\nstation and its HA. The secret key can be stored in either the mobile station or its\nSubscriber Identity Module (SIM) card.\nAuthenticating parties do not share a secret key: authentication between two\nmobile stations or between a mobile station and a fixed Internet server, and so on.\nSince the two parties have no common secret key to share, more public key algorithm\ncomputations are involved.\nVisit the Internet public resource: Since the resource is open to the public, no\nauthentication is needed.\nThus, the parties authenticated with mobile stations are divided into two categories:\nhome network and any authentication parties other than the home network. This design\nsimplifies the protocol and the implementation on the mobile station. In cases other\nthan authentication between a mobile station and its home network, the home network\nperforms the major authentication job and then passes the authentication result to the\nmobile station.\n4.2. Characteristics of Proposed Authentication and Key Negotiation\nProtocol\nIn Figure 7, fixed networks are identified by the information issued by the CA\nserver. Identity verification is carried out using the public key encryption and digital\nsignature algorithms. Since CA servers are responsible for large amounts of certificate\nissuing, the task for CA servers in the proposed protocol is simple, no more than looking\nup the database and sending the necessary information, such as the public key message,\nto the corresponding receiver. A mobile station never contacts the CA server in the\nprotocol, since it is not practical for a CA server to record the certificate information\nof all mobile stations because of their enormous population. The certificate of each\nmobile station is stored in its home network. Thus, each home network server can be\nconsidered as a CA server of its mobile stations.\nA CA server works as a bridge connecting the domain servers, such as HAs and\nfixed servers. A fixed server can be considered as an HA without clients. The proposed\nprotocol puts the corresponding daemon programs in each node. It is designed with\nthe following considerations to compensate for salient features or limitations in both\nhardware and transmission environments:\nThe protocol should be intelligent.\nThe design should enable the protocol\nto adapt to various application scenarios. The adaptation should be mainly\nimplemented in wired servers.\nThe number of different types of message for mobile stations should be limited\ncompared to the home network such that the design simplify the implementation\nin the mobile station.\n"
        },
        {
            "page_number": 287,
            "text": "WIRELESS NETWORK SECURITY\n285\nIt is desirable to move much of the computation to the corresponding HAs\nwhich have more computation power, high speed, and reliable wired network\nconnections. At the mobile station, intensive computations are limited. Only\ncritical data such as secrets and their hash values are encrypted using a public\nkey algorithm. The public key encryption and digital signature algorithms are\nnot used simultaneously in one message.\nThe length of messages will be collected for protocol latency evaluation. Ac-\ncording to the network structure, the major presence of latency should be in\nthe wireless part, especially when the client is connected to a cellular network.\nThe design goal is that the time taken to transmit all messages in the slowest\nconnection method be less than 3 s.\n4.3. A Wireless Transmission Privacy Protocol\nThe wireless transmission privacy protocol1 serves as an authentication service\nprovider to other wireless Internet applications. Before an Internet application begins\nto send data, the mechanism does the authentication first and negotiates a shared key of\nwhich the foreign network has no knowledge. At the sender side, all the upcoming data\ngenerated by the Internet application with security requirements are encrypted by this\nshared key. The encrypted or wrapped data are then sent to other data processing blocks.\nFor example, they can be further encrypted by the key acquired by the registration\nprocess. At the receiver side, the process is reversed. Thus, a foreign network cannot\nget plain text even if it holds a key generated during the registration process, and the\nwireless transmission part is also secured.\nThere are a few authentication scenarios. We assume that mobile station 1 (MS1)\nwants to establish a connection with mobile station 2 (MS2) via wireless Internet. MS1\nand MS2 belong to different home networks and have no shared key. This is the most\ncomplicated scenario and other scenarios are considered its subsets. The mechanism\nworks this way and is shown in Figure 8. The numbers in the figure represent the\nsequence of steps.\n1. MS1 finds MS2’s home address and and sets IPauth.desk = IPMS2. MS1\ncreates a nonce N with the corresponding hash value Hash(N). The nonce is\nused to verify the identity of MS2. N and Hash(N) are encrypted with HA’s\npublic key pubHA1. MS1 sends the authentication request\nIPMS1 : EkeyHA1−MS1 {IDMS1, IPauth.dest, E[pubHA1, < N, Hash(N) >]}\nto HA1. The whole message is encrypted by the shared secret key of MS1 and\nHA1 keyHA1−MS1.\n1 Earlier version of the protocol has been published in Minghui Shi, Xuemin (Sherman) Shen, and\nJon W. Mark, \"IEEE802.11 roaming and authentication in wireless LAN/cellular mobile networks,\" IEEE\nWireless Communications Special Issue on Mobility and Resource Management, vol. 11, issue 4, Aug. 2004,\npp. 66-75\n"
        },
        {
            "page_number": 288,
            "text": "286\nMINGHUI SHI et al.\nCA\nHA1\nMS1\nHA2\nMS2\n1\n2\n3\n3\n5\n6\n7\n8\n9 10\n8\n10\n4\n7\nFigure 8. Authentication and key negotiation protocol between two mobile stations be-\nlonging to different home networks\n2. HA1 decrypts the message from MS1using keyHA1−MS1 and privHA1and\ngets IDMS1, N, Hash(N) and IPdest.\nHA1 realizes that MS1 intends to\nauthenticate with a third party. HA1 is able to find MS2’s HA, HA2, from\nthe IP of MS2. In order to discover if HA2 is legal, HA1 contacts CA for\nidentification information of HA2, such as the public key of HA2. HA1sends\nIPHA1 : E{IDHA1, IPHA2}\nto CA.\n3. CA decrypts the message from HA1 by keyCA−HA1 and gets IDHA1 and\nIPHA2. CA verifies IDHA1. CA searches its database, and finds the public\nkeys of both HA1 and HA2 and the device ID of HA2 IDHA2. CA does not\nneed to check the information requester strictly since the information CA sends\nis for public use. What CA needs to do is to ensure the authority and accuracy of\nthe information it sends. CA finds pubHA1, pubHA2 and IDHA2. CA attaches\nthe digital signature of the message and transmits HA1’s public key and device\nID\nIPCA : EkeyCA−HA2 {IDCA, IDHA1, IPHA1, pubHA1} : sigCA\nto HA2 and HA2’s\nIPCA : EkeyCA−HA1 {IDCA, IDHA2, IPHA2, pubHA2} : sigCA\nto HA1.\n4. HA1 decrypts the message from CA by keyCA−HA1, and gets IDCA, HA2’s\nIP IPHA2, public key pubHA2and device ID IDHA2. HA1 verifies the validity\n"
        },
        {
            "page_number": 289,
            "text": "WIRELESS NETWORK SECURITY\n287\nof the message sent by CA by its digital signature. Any changes to the message\nafter it was sent can be detected. HA1 verifies if DCA matches keyCA−HA1. If\nall validation passes, HA1 stores the pubHA2 and IDHA2 pair. HA1 generates\nthe temporary session key keytemp. HA1 set IPauth.orig to IPMS1, IPauth.dest\nto IPMS2, and forwards the authentication request and temporary session key\nIPHA1 : EpubHA2 [keytemp, Hash(keytemp)] : Ekeytemp{\nIPMS1, IPMS2, IDMS1, pubMS1} : sigHA1\nto HA2. The key is encrypted by HA2’s public key. So far, there are two\nmessages in step 3 and 4 sent to HA2.\n5. HA2 will buffer the latter if the latter comes before the former. By receiving\nmessage in step 3, HA2 can get HA1’s device ID IDHA1, IP IPHA1, and public\nkey pubHA1. HA2 then verifies the validity of the message in Step 4 by the\nattached digital signature of HA1 and decrypt the first part of the message using\nits private key privHA2 to get keytemp. Then HA2 can further decrypt the\nsecond part of the message to get the IPauth.orig, IPauth.dest, and information\nof authentication originator, such as MS1’s public key and device ID in this\ncase. Since HA2 knows that MS1 wants to authenticate with MS2, it initiates\nthe authentication with MS2, because it is not secure to send the identification\ninformation before HA2 verifies MS2. HA2 temporally stores keytemp and the\ninformation of MS1. HA2 send authentication request\nIPHA2 : EkeyHa2−MS2 {IDHA2, IPMS1}\nto MS2.\n6. Similar to step 1, MS2 starts authentication with HA2 . MS2 decrypts the\nmessage from HA2using keyHA2−MS2 and gets IDHA2 and IPMS1. Because\nthe IP address received is different from its home network’s address prefix, MS2\nknows that a third party wants to authenticate with it. MS2 creates a nonce N\nand its hash value pair Hash(N). N and Hash(N) are encrypted by HA2’s\npublic key. MS2 sends the message\nIPMS2 : EkeyHA2−MS2 {IDMS2, IPauth.dest, EpubHA2 [N, Hash(N)]}\nto HA2.\n7. HA2 decrypts the message MS2 using keyHA2−MS2 and gets IDHA2. HA2\ndecrypts N, Hash(N) and IPdesk using its private key privHA2. HA2 verifies\nthe identity of MS2, N and Hash(N). HA2 sends MS1’s identify information\nand the session key keytemp\nIPHA2 : EKeyHA2−MS2 {IDHA2, Hash(N), IDMS1,\npubMS1, EpubMS2 [keytemp, Hash(keytemp)]}\n"
        },
        {
            "page_number": 290,
            "text": "288\nMINGHUI SHI et al.\nto MS2. HA2 informs HA1 that MS2 has accepted the authentication request\nand tell HA1the identity information of MS2. also HA2 sends\nIPHA2 : Ekeytemp{Hash(keytemp), IDMS2, pubMS2} : sigHA2\nto HA1. HA2’s work in the protocol ends here.\n8. MS2 receives the message from HA2 and decrypts it by keyHA2−MS2. MS2\nverifies Hash(N) and IDHA2.\nIf they are valid, MS2 uses its private key\nprivMS2 to get ketemp and verifies it using Hash(N). MS2 get the identity\ninformation of MS1. MS2 sends a confirmation\nIPMS2 : Ekeytemp{Hash(Keytemp)} : sigMS2\nto HA2. MS2 generates a new nonce newN, and computes its Hash value\nHash(newN). MS2transmits the acknowledgement to MS1\nIPMS2 : Ekeytemp{newN, Hash(newN)} : sigMS2\nby using a new nonce NewN and its hash value encrypted by the session key.\n9. HA1 receives the message in step 7 from HA2. HA1 verifies the validity of the\ndigital signature signed by HA2. HA1 decrypts the first part of the message\nsuing keytemp and verify Hash(keytemp). HA1 gets the identity information\nof MS2. HA1sends MS2’s identity information and the session key\nIPHA1 : E{IDHA1, Hash(N), IDMS2, pubMS2, EpubMS1\nto MS1. HA1’s work in the protocol ends here.\n10. MS1 receives the message in step 8 from MS2 and the message in step 9 from\nHA1. Since the decryption of the former message depends on the information\ncontained in the latter message, the former will be buffered until the latter is\nreceived. MS1 get the identity of MS2 IDMS2and pubMS2. MS1then is able to\nverify and decrypt the message from MS2. If the signature is valid, and newN\nand Hash(newN) match, MS1 sends acknowledgement\nIPMS1 : Ekeytemp{Hash(keytemp)} : sigMS1\nto HA1 and replies\nIPMS1 : Ekeytemp{Hash(newN)} : sigMS1\nto MS2 by sending the hash value of the new nonce.\n"
        },
        {
            "page_number": 291,
            "text": "WIRELESS NETWORK SECURITY\n289\n(a) a mobile station to a fixed Internet server\n(b) a fixed Internet server to a mobile station\n(c) a mobile station to another within the same home\nnetwork\n(d) a mobile station to a home agent and a home\nagent to a mobile station\nFigure 9. Protocol variation in other authentication scenarios\n11. MS2 receives the message from MS1 and verifies the digital signature of MS1\nand compares Hash(newN) with the original. If they are both correct, the\nauthentication process is now complete.\nThe protocol is also adaptive to other scenarios. For example, if a mobile station\nwants to authenticate with a fixed server, we consider HA2 and MS2 as one unit, and\nsteps 5, 6, 7, and 8 are not necessary. The extended scenarios are shown in Figure 9.\n4.4. Security Analysis\nSecurity analysis presented here includes data privacy, a built-in feature for dealing\nwith certain security compromises. Device-related information is divided into two\ncategories. Device ID and public key belong to normal sensitive data, which means they\nwill not do harm to the system even if they are leaked. Shared secret key and private key\nbelong to permanent critical information that must not be compromised. The nonce\nand session key generated in the protocol belong to short-term critical information\nthat can affect the ongoing session in which an attacker can discover communication\ncontents. However, if the permanent critical information is still good, short-term critical\ninformation is safe because it is encrypted by the permanent critical information.\nIn the authentication protocol the exchanged message, except for digital signature\nin the authentication process, are all encrypted by a shared secret key between the\n"
        },
        {
            "page_number": 292,
            "text": "290\nMINGHUI SHI et al.\nnonce\nkeytemp\npubMS1\npubMS2\nidMS1\nidMS2\npubHA2\nidHA2\nHA1\nHA2\nMS2\nMS1\nprivMS1\nprivMS2\nprivHA2\nprivHA1\nkeyHA2MS2\noutside\nipMS1\nipHA1\nipHA2\nipMS2\npubHA1\nidHA1\nkeyHA1MS1\nFigure 10. Secret and identification information control\nHA and the client, or the session key using a symmetric encryption algorithm. More\nimportant information like the session key is further encrypted by a public key algorithm\nand capsulated by symmetric encryption. The authentication message is not different\nfrom the normal payload of a TCP (Transmission Control Protocol) packet, and so the\nforeign network can route it to the destination. The foreign network and other intruders\nare not able to discover the information inside because they do not have the shared key;\nand they must have both corresponding shared key and matched private key to get the\nsession key.\nFigure 10 shows the distribution of sensitive data after authentication is completed.\nThe circles in the figure denote the knowledge of the information Normal sensitive\ninformation is spread to trusted sites and devices only.\nThe protocol ensures that\nno sensitive information is released before the information receiver is identified. No\npermanent critical information is sent in any form during the authentication. Temporary\ncritical data are spread to trusted sites only.\nThe protocol should be designed to resist certain security compromises. In our\nauthentication protocol, illegal possession of someone’s device ID, home IP address,\nand public key will do virtually no harm to the system. The home network always uses\nthe corresponding shared secret key to process messages according to the carried home\n"
        },
        {
            "page_number": 293,
            "text": "WIRELESS NETWORK SECURITY\n291\nIP address. It is the shared secret key and private key that build up the real or final\nauthentication process.\nComparing the shared secret key and private key, a shared secret key is more likely\nto be compromised because at least two parties, the home network and mobile station,\nhave a copy of the key. Only one copy exists in the mobile station for the private key\ncase. A private key is never given out because it is not necessary due to the nature of\npublic key algorithms. In our protocol, for example, if the shared secret key is leaked,\nthe intruder can get only the device ID, IP address, and public key, which belong\nto normal sensitive information and do virtually no harm to the system, because the\ndevice ID is used for quick identification, and the public key itself is originally open\nto the public. But it could be harmful if this key were also used on other occasions\nsuch as mobile station registration, since the registration process should be done very\nquickly to avoid the connection being dropped, so there is no time to execute additional\ntime-consuming public key algorithms.\nIf a shared secret key is leaked and an intruder tries to use it without a proper\nprivate key, the system can detect the compromise of the shared secret, because in our\nauthentication protocol, each entity involved is required to return a hash value that can\nonly be achieved by its private key or attach a digital signature to the message. Once the\nsystem detects this flaw, it indicates that the common secret key is leaked and the user\nshould be warned immediately. The system cannot detect the private key flaw though,\nbecause without a proper shared secret key, the system cannot look into the message. So\nonly when the shared secret and corresponding private keys are broken simultaneously\ncan the intruder access the network illegally. The system is able to detect a security\ncompromise of the shared secret key, but not of the private key. Fortunately, the private\nkey is unlikely to be leaked due to the nature of public key algorithms.\nThe protocol is also designed to resist replay attack. Every authentication session\nbetween an HA and a mobile station, or two mobile stations or two HAs is completed\nby using fresh nonces and fresh session key, so replay attack has no effect on it. Since\nthe information exchanged between an HA and a CA server represents facts on the\nclients identifications and public keys, simply replaying this message does no harm to\nthe system unless the intruder can change the payload and the corresponding digital\nsignature, which is very hard unless the intruder can get the CAs private signature.\nIn order to totally duplicate component (mobile station, home network server, CA\nserver), the malicious user at least needs a proper home IP address, device ID, shared\nsecret key, and private key to satisfy the authentication protocol completely, or it will\nbe rejected at the corresponding step where the item is checked.\n5.\nAUTHENTICATION AND KEY NEGOTIATION DEMONSTRATION\nFigure 11 shows the demonstration program for all the considered scenarios.\nThe demonstration shows the authentication progress, the way in which the proto-\ncol self-adapts to each case, and message lengths sent and received by each node.\nThe demonstration uses RSA as the public key algorithm, DES as the symmetric\nalgorithm, and MD5 as one-way hash functions.\nSince the proposed protocol is\n"
        },
        {
            "page_number": 294,
            "text": "292\nMINGHUI SHI et al.\nFigure 11. Demonstration of the authentication and key negotiation protocol\ncryptographic-algorithm-independent, other stronger or lighter algorithms can be used\nto accommodate specific application requirement. The demonstration also shows that\nthe total amount of data for the mobile node is less than 2 kbytes. If the slowest network\nconnection speed is 14.4 kb/s in the cellular network with overhead of the transmission\nconsidered, the data transmission can be finished in less than 3 s.\n6.\nSUMMARY\nIn this chapter, a network architecture and a set of signaling mechanisms are pro-\nposed to support current available wireless LAN hotspot roaming. The proposed ar-\nchitecture offers a smooth transition of wireless LAN hotspots from non-roaming-\nsupported to seamless-roaming-supported, and therefore previous investment can be\nprotected.\nMeanwhile, wireless transmission security is carefully considered.\nAn\napplication layer authentication and key negotiation protocol is developed to keep end-\nto-end transmission secure. The results can enable wireless LAN roaming, enhance\nwireless communications, and speed up the deployment of public wireless LAN appli-\ncations.\n"
        },
        {
            "page_number": 295,
            "text": "WIRELESS NETWORK SECURITY\n293\nACKNOWLEDGEMENT\nThis work has been supported by a Natural Science and Engineering Council\n(NSERC) Postgraduate Scholarship and a research grant from Bell University Labora-\ntories (BUL), Canada. We would like thank Matthew Cruickshank for his help in the\ndemonstration programming.\n7.\nREFERENCES\n1. C. S. Loredo and S. W. deGrimaldo, “Wireless LANs: Global Trends in the Workplace and Public\nDomain,” The Strategies Group, 2002.\n2. A. T. Campbell et al., “IP Micro-Mobility Protocols,” ACM SIGMOBILE Mobile Comp. and Commun.\nRev., vol. 4, Oct. 2001, pp. 45-54.\n3. C. Perkins, “IP Mobility Support,” RFC 2002, Oct. 1996.\n4. J. R. Walker, “Unsafe at Any Key Size: An Analysis of the WEP Encapsulation,” IEEE doc. 802.11-\n00/362, Oct. 2000.\n5. W. A. Arbaugh, “An Inductive Chosen Plaintext Attack Against WEP/WEP2,” IEEE doc. 802.11-\n01/230, May 2001.\n6. J. Zhang and J. W. Mark, “A Secured Registration Protocol for Mobile IP,” Master’s Thesis, Apr. 1999.\n7. Sufatrio and K. Y. Lam, “Mobile IP Registration Protocol: A Security Attack and New Secure Minimal\nPublic-Key Based Authentication,” I-SPAN 99, Fremantle, Australia, 1999, pp. 364-369.\n8. J. S. Stach, E. K. Park, and Z. Su, “An Enhanced Authentication Protocol for Personal Communication\nSystems,” IEEE Wksp. App.-Specific Software Eng. Tech., Dallas, TX, 1999, pp. 128-132.\n9. H. Lin and L. Harn, “Authentication Protocols with Nonrepudiation Services in Personal Communica-\ntion Systems,” IEEE Commun. Lett., vol. 3, 1999, pp. 236-238.\n10. R. Shirey, “Internet Security Glossary,” IETF RFC 2828, 2000.\n11. A. Westerinen, J. Schnizlein, J. Strassner, M. Scherling, B. Quinn, J. Perry, S. Herzog, A.N. Huynh,\nM. Carlson, and S. Waldbusser, “Terminology for policy-based management,” IETF RFC 3198, 2001.\n12. IETF AAA Working Group, “Mobile IP AAA Requirements,” IETF RFC2977, October 2000.\n13. C. Rigney, “Remote Authentication Dial In User Service (RADIUS),” IETF RFC2138, 1997.\n14. C. Rigney, “RADIUS Accounting,” IETF RFC2139, 1997.\n15. Network Working Group, “RADIUS Accounting,” IETF RFC 2866, June 2000.\n16. C. Finseth, “An Access Control Protocol, Sometimes Called TACACS,” IETF RFC1492, 1993.\n17. D. Carrel and L. Grant, “The TACACS+ Protocol: Version 1.78,” IETF Internet Draft, <draft-grant-\ntacacs-02.txt>, 1997.\n18. J. Kohl and C. Neuman, “The Kerberos Network Authentication Service (V5),” IETF RFC1510, 1993.\n19. U. Carlsen, “Optimal Privacy and Authentication on a Portable Communication System,” Op. Sys. Rev.,\nvol. 28, 1994, pp. 16-23.\n20. C. Park et al., “On Key Distribution and Authentication in Mobile Radio Networks,” Proc. Advances\nin Cryptology-Eurocrypt 93, Szombathely, Hungary, 1993, pp. 461-465.\n21. M. Tatebayashi and D. B. Newman, “Key Distribution Protocol for Digital Mobile Communication\nSystems,” Proc. Advances in Cryptology-Crypto 89, Houthalen, Belgium, 1989, pp. 324-333.\n"
        },
        {
            "page_number": 296,
            "text": "294\nMINGHUI SHI et al.\n22. B. Aboda and M. Beadles, “The Network Access Identifier,” IETF RFC 2486, Jan. 1999.\n23. P. Calhoun and C. Perkins, “Mobile IP Network Access Identifier Extension for IPv4,” IETF RFC 2794,\nMar. 2000.\n24. E. Shim and R. D. Gitlin, \"Reliable and Scalable Mobile IP Regional Registration,\" IETF Internet Draft,\n<draft-shimmobileip-reliable-reg-00.txt>, Apr. 2001\n"
        },
        {
            "page_number": 297,
            "text": "12\nAN EXPERIMENTAL STUDY ON\nSECURITY PROTOCOLS IN WLANS\nAvesh Kumar Agarwal\nDepartment of Computer Science\nNorth Carolina State University\nE-mail: akagarwa@unity.ncsu.edu\nWenye Wang\nDepartment of Electrical and Computer Engineering\nNorth Carolina State University\nE-mail: wwang@eos.ncsu.edu\nWireless Local Area Networks (WLANs) are vulnerable to malicious attacks due to their\nopen shared medium. Consequently, provisioning enhanced security with strong crypto-\ngraphic features and low performance overhead becomes exceedingly necessary to actualize\nreal-time services in WLANs. In order to exploit full advantage of existing security proto-\ncols at various layers, we study the cross-layer interactions of security protocols in WLANs\nunder different network scenarios. In particular, we present a detailed experimental study\non the integration of commonly used security protocols such as WEP, 802.1x and EAP,\nIPsec and RADIUS. First, we classify individual and hybrid policies, and then, define se-\ncurity index and cost functions to analyze security strength and overhead, quantitatively,\nof each policy. By setting-up an experimental testbed, we measure performance cost of\nvarious policies in terms of authentication time, cryptographic cost and throughput using\nTCP/UDP traffic streams. Our results demonstrate that in general, the stronger the secu-\nrity, the more signaling and delay overhead, whereas, the overhead does not necessarily\nincrease monotonically with the security strength. Therefore, it is suggested to provide\nsubstantial security at a reasonable cost of overhead with respect to mobile scenarios and\ntraffic streams. Also, we notice that authentication time will be a more significant factor\ncontributing towards QoS degradation than cryptographic cost, which is critical to real-time\nservice in wireless networks.\n1.\nINTRODUCTION\nWireless Local Area Networks (WLANs) have become prevalent for providing\nubiquitous internet access for mobile users. However, security is of utmost concern\nin WLANs, because interception and eavesdropping of data in transit by malicious users\n"
        },
        {
            "page_number": 298,
            "text": "296\nAVESH KUMAR AGARWAL and WENYE WANG\nbecome easier due to shared and broadcast air medium [1], [2]. Therefore, several\nsecurity protocols such as wired equivalent privacy (WEP), 802.1x port access control\nwith extensible authentication protocol (EAP) support are proposed to address security\nissues [3], [4], [5], [6], [7]. Moreover, due to strong security provided by IP security\n(IPsec) in wired networks, it is considered a good option for establishing virtual private\nnetworks (VPNs) [8] in wireless network also. However, existing studies demonstrate\nvarious types of malicious attacks on these protocols [3], [9], [10]. Consequently,\nresearchers have studied these security protocols individually with respect to crypto-\ngraphic properties to enhance security in the network [11], [12], [13]. In this study,\nwe explore cross-layer interactions of existing security protocols by integrating the\nprotocols at several layers to enhance security in WLANs.\nMoreover, security protocols incur performance overhead due to the configuration\nand messages at different layers in the network. Meanwhile, many real-time wireless\napplications have shown an increasing demand for better quality of service (QoS) in\nreal networks [14]. Therefore, it becomes mandatory to determine the performance\nimpact of the security protocols in real-time networks for better QoS. Existing studies\nin the past have focused mainly on improving cryptographic perspective of security\nprotocols, whereas lacking detailed quantification of performance overhead associated\nwith the protocols [11], [12], [13]. In this study, we provide comprehensive real-time\nmeasurements of performance overhead associated with security protocols at various\nlayers in WLANs.\nMeasurementsareveryimportanttodeterminetherealisticviewoftheperformance\noverhead associated with the security mechanisms. Therefore, to gain fundamental\nunderstandingofperformanceimpactduetosecurityprotocols, experimentalstudiesare\ncarried out in the past in various network environments [8], [15], [16], [17]. However,\nthese studies have explored security protocols as a stand-alone mode. Moreover, these\nstudies perform experiments in few network scenarios providing less detailed real-time\nresults. In this work, we study the cross-layer integration of security protocols in\nvarious non-roaming and roaming network scenarios to gain a deeper understanding\nof the associated performance overhead. Measurements provided in this study are\nexplained to show how integration of quality of service (QoS) and security service\naffects system performance.\nTo achieve above goals, we have setup a real-time experimental testbed. The test-\nbed is a miniature of existing wireless networks, which ensures that our experimental\nresults can be mapped to large-scale wireless networks. The testbed consists of two\nsubnets for configuring various network scenarios. We install open source versions\nof commonly used security protocols such as 802.1x, EAP, IPsec and RADIUS in the\ntestbed. Security protocols are classified into individual and hybrid security policies to\nstudy cross-layer interactions. Moreover, we define security index and cost functions\nto analyze security strength and overhead associated with each security policy, respec-\ntively. Authentication time, cryptographic cost and throughput are the performance\nmetrics evaluated under TCP and UDP traffic streams. Moreover, we discuss various\nattacks on security policies and demonstrate that hybrid security policies (cross-layer\nintegration of security protocols) are less vulnerable than individual security policies.\n"
        },
        {
            "page_number": 299,
            "text": "WIRELESS NETWORK SECURITY\n297\nOur observations demonstrate that there is always a tradeoff between security and\nperformance associated with a security policy, depending upon the network scenario\nand traffic types. In general, we observe that security policy with higher strength is\nnot always the best option for all scenarios. We find that the cross-layer integration\nof security protocols may provide the strongest protection, but with more overhead\ntogether. Our results demonstrate that in general, the stronger the security, the more\nsignaling and delay overhead, whereas, the overhead does not necessarily increase\nmonotonically with the security strength. Moreover, we observe that IPsec policies\nprovide the best tradeoff between security and performance for authentication time;\n802.1x-EAP-TLS policy is the best suitable option for low cryptographic cost and better\nsecurity strength in many scenarios. In addition, experimental results for throughput\nreveal that authentication time will be a more significant factor contributing towards\nQoS degradation in the network than cryptographic cost.\nThe rest of the chapter is organized is as follows. Section 2 discusses the back-\nground of existing security protocols for WLANs. Details of the testbed, network\nscenarios and classification of security policies are provided in Section 3. Security\nindex and cost functions to analyze the security strength and performance overhead as-\nsociated with security policies are presented in Section 4. Section 5 explains procedure\nto carry out experiments. Detailed performance evaluation of experimental results is\nprovided in Section 6. Section 7 discusses vulnerabilities associated with individual\nsecurity policies with respect to malicious attacks and countermeasures in cross-layer\nintegration of security policies. Finally, Section 8 concludes the chapter.\n2.\nBACKGROUND\nTo address securityissues, manyprotocolsaredeveloped, whichoperateatdifferent\nnetwork layers. Wireless Equivalent Privacy (WEP), 802.1x with Extensible Authen-\ntication Protocol (EAP), Remote access dial in user service (RADIUS) and IP security\n(IPsec) are some of the protocols used in wireless networks. We focus on studying these\nsecurity protocols because they operate at different network layers, which will help us\nto analyze the overhead introduced by security services across network layers. These\nprotocols are widely adopted in the wireless networks providing a very close analysis,\nwhich will be useful for the real-time networks. Brief description of these protocols is\nas follows:\nMAC Layer Protocols: WEP is the very first protocol to be considered for wireless\nnetworks. WEP has been identified to be susceptible to many type of attacks [3]. To\novercome WEP weaknesses, IEEE 802.1x standard is designed to provide stronger\nsecurity [4], [5], [6], [7]. 802.1x works at MAC layer and provides port-based access\ncontrol for wireless nodes. In addition, 802.1x exploits the use of EAP (MD5,TLS),\nwhich is used as transport mechanism [4]. Besides considering MAC layer security\nprotocols, we also evaluate network layer and application layer security protocols such\nas IPsec and RADIUS in the experimental testbed.\n"
        },
        {
            "page_number": 300,
            "text": "298\nAVESH KUMAR AGARWAL and WENYE WANG\nHigher Layer Protocols: IPsec is a network layer protocol, originally designed for\nwired network, which is now being considered for wireless networks due to its strong\nauthentication and cryptographic methods. Further, TSL is a transport layer protocol\nand successor to Socket Security Layer (SSL), which is the most widely deployed\nsecurity protocol on the Internet. At application layer, we consider RADIUS protocol,\nwhich is based on client-server architecture.\nExisting security protocols have some drawbacks and are prone to several attacks.\nFor example, according to previous studies, WEP and 802.1x are susceptible to many\ntypes of attacks [3], [9] and [10]. In addition, there are other studies which explain\nthe security aspects of WLANs providing overview of various security protocols such\nas [2]. To overcome these problems, researchers have come up with many solutions\nto improve the security aspects of these protocols in recent years. For example, re-\ncently a new authentication protocol is proposed for wireless networks in [19]. In\naddition, other works have proposed solutions to improve security for mobile wireless\nnetworks [11], [12] and [13]. Moreover, there are other studies, which focus on perfor-\nmance aspects of security protocols. For example, a performance analysis of different\nprotocols of IPsec is provided in [15]. Similarly, IPsec performance is also analyzed as\nvirtual private networks (VPN) in [8]. In addition, a proposal is provided to implement\nwireless gateway for WLAN based on IPsec protocol in [16]. But, we observe that most\nof the research is focused on security aspects with little thoughts given to performance\nimpact of security protocols on system performance. Therefore, we conduct compre-\nhensive experimental analysis to uncover performance issues associated with security\nprotocols in mobile wireless LANs.\nOur study focuses on the impact of security protocols on different user’s mobility\nscenarios in combination with IP mobility in WLAN roaming. Moreover, our analysis\nhas considered a wide range of security protocols at different layers such as 802.1x,\nWEP, SSL other than just IPsec. Unlike previous studies, we focus on the quality of\nservice (QoS) aspects of the network determining impact on QoS when security services\nare enabled in the wireless networks. To our knowledge, this is the first experimental\nstudy on this issue, which analyzes security protocols in various mobility scenarios.\n3.\nIMPLEMENTATION SETUP\nIn this section, we provide details of the experimental testbed including hardware\nequipments and software configurations. Figure 1 shows an example of testbed archi-\ntecture in which two subnets are illustrated. Although, we show only two subnets; with\ndifferent combinations in hardware and software, virtually we create a heterogeneous\nenvironment that captures mobile scenarios of wireless local area networks.\n3.1. Hardware Configuration\nMobile IP is used to support mobility and routing in our testbed. A mobile node\n(MN) is defined as the wireless node, which is able to change its point of attachment\n[20]. Different mobile nodes used in our testbed consist of iPAQ (Intel StrongARM\n"
        },
        {
            "page_number": 301,
            "text": "WIRELESS NETWORK SECURITY\n299\nCisco AP1\nRouter\nNetwork Switch\nWEP, 802.1x\nWEP,802.1x\nIP\n802.1x\nApplication\nApplication\nMAC\nPhysical\nPhysical\nPhysical\nPhysical\nRADIUS\nRADIUS\nTCP/ UDP\nTCP/UDP/SSL/TSL\nTCP/UDP\nTCP/UDP/SSL/TSL\nIP/IPSEC/MIP\nIP/IPSEC/MIP\nIP/IPSEC/MIP\nHome Agent\nHome Agent\nSubnet A\n(A−HA)\nA−Host\nA 1\nA2\nB1\nB2\n(B−HA)\nB−Host\nSubnet B\nCisco AP2\nFigure 1. Wireless LAN Testbed.\n206 MHZ), Sharp Zaurus (Intel XScale 400 MHz) and Dell Laptop (Celeron Processor,\n2.4GHZ). Home agents (HA), A-HA and B-HA, are the gateways in a mobile node’s\nhome network (HN) where the mobile node registers its permanent IP address. In our\ntestbed, home agents (HA) are gateways for Subnets A and B and are Dell PC with\nPentium IV 2.6 GHZ. Foreign agents (FA) are the gateways in a foreign network (FN)\nwhere a mobile node obtains a new IP address to access to the network. Home agents\nhave the functionalities of foreign agents as well in our testbed. They are connected\nto Cisco Access Points (Cisco Aironet 1200 series) to provide wireless connectivity.\nIn addition, the home agents have functions of IPsec gateways and RADIUS server\nfor IPsec and 802.1x, respectively. An IPsec tunnel is setup between home agents to\nprovide security over the wired segment in our testbed. Hosts A-Host and B-Host act\nas wired correspondent nodes in Subnets A and B and are Dell PC with Pentium IV 2.6\nGHZ. Cisco Catalyst 1900 series is used as a network switch to provide connectivity\nbetween two subnets via the router. We have used Netgear MA 311 and Lucent Orinoco\nGold wireless cards in all mobile devices.\n3.2. Software Configuration\nAll systems use Redhat Linux 9.0 kernel 2.4.20. We have installed open-source\nsoftware components for various protocols in the testbed as follows:\n"
        },
        {
            "page_number": 302,
            "text": "300\nAVESH KUMAR AGARWAL and WENYE WANG\nFreeSwan open source is installed on home agents and mobile nodes for IPsec\nfunctionality [21].\nXsupplicant, which provides 802.1x client functionality, has been installed\non mobile nodes [22].\nRADIUS server functionality has been provided by FreeRadius and has been\ninstalled on home agents [23].\nOpenSSL open source software is installed on home agents [24].\nTo introduce user mobility in our network, Mobile IP implementation from\nDynamic is installed on mobile nodes and home agents [25].\nEthereal packet analyzer is used for packet capturing.\nIperf and ttcp are used for generating TCP/UDP traffic streams.\n3.3. Network Scenarios\nNetwork scenarios are classified into non-roaming (N) and roaming (R) based on\nuser’s current location, i.e., whether a user is in its home domain or foreign domain,\nrespectively. To make the description of scenarios clear, we assume that subnet A is the\nhome domain for mobile nodes A1 and A2; and subnet B is the home domain for mobile\nnodes B1 and B2. All scenarios are demonstrated in Figure 2. Non-roaming scenarios,\nrepresented as N, are defined as the scenarios when both communicating mobile users\nare in their home domain. Following are the details of various non-roaming scenario\nconfigured in the testbed.\nB2\nA1\nScenario : N1\nA−HA\nB−HA\nB1\nInternet\nA2\nSubnet A\nSubnet B\nSubnet A\nSubnet B\nA2\nInternet\nB2\nA1\nA−HA\nB−HA\nB1\nScenario : N2\nSubnet A\nSubnet B\nA2\nInternet\nB2\nA1\nA−HA\nB−HA\nB1\nSubnet A\nSubnet B\nA2\nInternet\nA1\nA−HA\nB−HA\nScenario : R1\nA2\nSubnet A\nSubnet B\nInternet\nB2\nA1\nA−HA\nB−HA\nB1\nB1\nScenario : R2\nB1\nScenario : N3\n(a)\n: Roaming\n:  Communication\n(b)\n(c)\n(d)\n(e)\nFigure 2. Non-Roaming and Roaming Scenarios.\n"
        },
        {
            "page_number": 303,
            "text": "WIRELESS NETWORK SECURITY\n301\nScenario N1: It deals with the situation when both mobile nodes are in the same\nsubnet, which is their home domain also. For example, when communication\noccurs between A1 and A2 as shown in Figure 2(a).\nScenario N2: Mobile nodes communicate with their home agent that is acting\nas an application server providing services to mobile clients in the network.\nHere, a part of the communication path is wired, which is not the case in\nscenario N1. As shown in Figure 2(b), this scenario occurs when home agent\nA-HA communicates with A1 or A2.\nScenario N3: It is to capture the impact of security services when partici-\npating mobile nodes are in different domains. For example, when A1 or A2\ncommunicates with B1 or B2 as shown in Figure 2(c).\nWhen at least one of two communicating mobile users is in a foreign domain, we\nrefer it as roaming scenario, represented as R. The following roaming scenarios are\nconfigured in our experimental testbed.\nScenario R1: This scenario specifies when one end node, which is in a foreign\ndomain, is communicating with the other node, which is in its home domain,\nbut two nodes are in different domains. It aims to analyze the effect of security\nservices on data streams when one node is roaming. As shown in Figure 2(d),\nthis scenario occurs when node A2 roams to subnet B and communicates with\nA1.\nScenario R2: This scenario occurs when both nodes are in the same domain\nbut one node is roaming. Therefore, current network is the foreign domain for\none node, whereas it is the home domain for other node. It helps us in analyzing\nperformance impact on data streams when roaming node is communicating with\na non-roaming node in the same domain. For example, when node B1 roams\nto subnet A and communicates with A1 or A2 as shown in Figure 2(e).\n3.4. Security Policies\nSecurity policies are designed to demonstrate potential security services provided\nby the integration of security protocols at different layers. Each security protocol uses\nkey management protocols, various authentication, and cryptographic mechanisms.\nTherefore, a variety of security policies are configured in our experiments by combining\nvarious mechanisms of security protocols. Let P = {P1, P2, . . . , P12} represent the set\nof individual and hybrid security policies configured in the network. Next, we explain\nthese security policies and their significance in detail.\nIndividual Security Policies\nWhen a policy involves mechanisms in a single security protocol, it is called an\nindividual security policy. \"No security\" means that there are no security services\n"
        },
        {
            "page_number": 304,
            "text": "302\nAVESH KUMAR AGARWAL and WENYE WANG\nenabled in the network. \"No Security\" policy helps us in comparing the overhead asso-\nciated with others in terms of end-to-end response time, throughput and performance\noverhead. In the following paragraphs we discuss security policies for each security\nprotocol.\nWEP Policies: WEP supports 40-bit and 128-bit encryption keys. Although,\nwe carried out experiments on WEP with 40-bit and 128-bit keys, we found\nthat results in both cases showed little variations. Therefore, in this paper,\nwe present WEP only with 128-bit due to longer key size. Although WEP\nhas been shown vulnerable to many attacks [3], we study WEP in this paper\nfor two reasons. First, WEP is still being used in many real-time networks\nfor dynamic session keys along with other security protocols such as EAP-\nTLS with 802.1x framework [5]. Second, comparing WEP’s performance with\nother security protocols provides a complete study of the performance impact of\nexisting security protocols for WLANs. P2 is the only individual WEP policy\nconfigured in the testbed.\nIPsec Policies: IPsec protocol supports a large set of cryptographic and au-\nthentication algorithms, and provides strong security. Since we use Freeswan\n[21] for IPsec functionality, our analysis is restricted to the security services\nprovided by Freeswan open source implementation. Freeswan includes 3DES\nas an encryption mechanism, and MD5 and SHA as authentication algorithms.\nSince IPsec tunnel mode is considered better by providing stronger security\nservices than IPsec transport mode, we analyze only IPsec tunnel mode in our\nsetup. P3 is the only individual IPSEC policy configured in the testbed.\n802.1x Policies: In case of 802.1x, we use RADIUS as a backend server main-\ntaining users’ secret credentials. EAP is used as a transport mechanism that\ninvolves MD5 and TLS modes. Since FreeRadius open source also supports\nMD5 and TLS, we analyze 802.1x with EAP in both TLS and MD5 modes.\nAlthough EAP-MD5 is not considered a very strong authentication mechanism\nfor WLANs [26], it can provide better security when configured with other\nsecurity protocols. Therefore, we believe that inclusion of EAP-MD5 makes\nour study complete. Moreover, as a discussing performance aspect of various\nsecurity protocols is the main objective of this paper, inclusion of EAP-MD5\nenables us to provide comprehensive performance measurements of the existing\nsecurity protocols in WLANs. Policies P5 and P6 are two 802.1x individual\npolicies configured in the testbed.\nHybrid Security Policies\nWhen security policies involve mechanisms belonging to multiple security proto-\ncols at different network layers, they are called hybrid security policies. Such policies\nare required, if visiting clients have security support at more than one network layer.\nTherefore, the network can fulfill the needs of a large number of clients. Also, security\n"
        },
        {
            "page_number": 305,
            "text": "WIRELESS NETWORK SECURITY\n303\nfunctionalities required by the network may not be fulfilled by one security protocol,\nleading to the need for configuration of more than one security protocol in the network.\nOur study incorporates security services provided by WEP, IPsec and 802.1x in\ndifferent ways. Initially we focus on the combination of IPsec and WEP. We first\nanalyze the overhead associated with IPsec (3DES, MD5 and SHA) and WEP (40 or\n128 bits), but here we present results only for IPsec (3DES, SHA) and WEP (128 bits)\ndue to page limit. Then we perform experiments with 802.1x and WEP to capture\ncombined effects of all security services at MAC layer and transport layer. Finally, we\nunite different security services of 802.1x, WEP and IPsec together for analysis. P4,\nP7, P8, P9, P10, P11 and P12 are hybrid security policies configured in our testbed.\nIntegration of different security protocols helps us answer a vital question, i.e., whether\nit is beneficial to combine security mechanisms at different network layers at the cost of\nadding extra overhead. A subset of security policies and associated features are shown\nin TABLE 1.\nTable 1. Features of Security Policies.\nPolicy No.\nSecurity Policies\n“A”\n“C”\n“D”\n“N”\n“M”\nP1\nNo Security\nP2\nWEP-128 bit key\nY\nY\nP3\nIPsec-3DES-SHA\nY\nY\nY\nY\nY\nP4\nIPsec-3DES-SHA-WEP-128\nY\nY\nY\nY\nY\nP5\n8021x-EAP-MD5\nY\nP6\n8021x-EAP-TLS\nY\nY\nY\nP7\n8021X-EAP-MD5-WEP-128\nY\nY\nP8\n8021X-EAP-TLS-WEP-128\nY\nY\nY\nY\nP9\nP7 +IPsec-3DES-MD5\nY\nY\nY\nY\nY\nP10\nP8 + IPsec-3DES-MD5\nY\nY\nY\nY\nY\nP11\nP7 + IPsec-3DES-SHA\nY\nY\nY\nY\nY\nP12\nP8 + IPsec-3DES-SHA\nY\nY\nY\nY\nY\nIn the above table, “A” denotes authentication; “C” denotes confidentiality; “D”\ndenotes data integrity; “N” denotes non-repudiation; and “M” denotes mutual authen-\ntication.\n"
        },
        {
            "page_number": 306,
            "text": "304\nAVESH KUMAR AGARWAL and WENYE WANG\n4.\nSECURITY INDEX AND COST ANALYSIS\nIn this section, we present a simple model to analyze the security strength of various\npolicies. Then, we develop cost functions to evaluate the associated security features of\neach policy. Our model is based on security features such as authentication, encryption,\ndata integrity, non-repudiation, access control and mutual authentication required by a\nsecurity policy.\n4.1. Security Index (SI)\nIn this work, we aim to represent the security services that can be configured\nin security protocols. Our goals regarding quantifications of security of a system are\nmanifolds. Generally, therearerequirementstoquantifythesecurityevenbeforesystem\nis deployed so that an appropriate security policy can be chosen. Therefore, it is not\npossible to observe the system behavior in advance for security quantification. In\naddition, the approach for quantifications should be simple and practically feasible\nwith regards to processing time and implementation, so that it can be implemented\neven in resource constrained environments. Moreover, quantification should have fine\ngranularitytohavecleardistinctionamongthestrengthsofsecuritypolicies. Asexisting\nstudies lack one or more goals desired by us, we define security quantification method\nbased on linear sum of weights assigned to various mechanisms in a security policy.\nEvery security policy provides some security features such as authentication and\nconfidentiality in our experimental study. However, it is difficult to quantify the security\nstrength delivered to a system or a network by a security policy based on its features.\nThis is due to the fact that it is almost impossible to predict that when a system or a\nnetwork can be compromised in the future during the configuration of a security policy.\nGenerally, it is not easy to be fair in comparing two policies with different features.\nFor example, assume that a security policy Pα consists of 2 features which are very\nstrong, and another security policy Pβ has of 4 features which are relatively weak. If\nwe compare two policies with respect to the 2 features of Pα, then we can conclude\nthat Pα provides stronger security than Pβ. However, if we compare Pα and Pβ with\nrespect to the 2 features not in Pα but in Pβ, we find that Pβ is better than Pα. The\njustification of which security policy is better than the other depends upon network\nrequirements, policies installed, and features activated in a network. We define few\nterms to make discussion clear while defining security index and assigning weights to\nsecurity features.\nSecurity Feature: Security services, such as authentication, mutual authentica-\ntion, confidentiality, data integrity and non-repudiations, are defined as security\nfeatures.\nSecurity Mechanism: Various protocols, such as EAP-MD5, EAP-TLS, IPsec,\nWEPandsoon, whichconsistsofdifferentalgorithmsandprotocols, aredefined\nas security mechanisms. A security mechanism can provide more than one\n"
        },
        {
            "page_number": 307,
            "text": "WIRELESS NETWORK SECURITY\n305\nsecurity features. For instance, EAP-MD5 can provide authentication and data\nintegrity security features.\nWe define security index to quantify and understand the strength of security policies\nby using weights associated their security feature. By defining security index, we aim\nto achieve the following goals.\nAs it is generally intuitive that security policy with more number of features is\nconsidered stronger than a security policy with less number of features. There-\nfore, our definition of security index should follow this intuition.\nIn addition, if two policies have the same number of security features associated\nwith them, the policy with strong security features is considered stronger than\nthe policy with weak security features. Therefore, our weight assignments to\ndifferent security mechanisms ensure that policy with strong security features is\nassigned a higher value of security index than other policies with weak security\nfeatures.\nWeights assigned to different security mechanisms and the resultant security\nindices of security policies signify whether a security policy is stronger or not\nfrom other policies. Security indices do not imply the absolute security strength\nof a policy, which is hard to quantify.\nAlthough, our definition of security index can be applied in different other sce-\nnarios not explored here, the weight assignment to various security mechanisms\nis unique to this study. If security protocols with different security mechanisms\nare considered, then weights assignments may require the modifications to var-\nious weights to accommodate different security mechanisms.\nNow, we define security index as described below.\nWe consider five security\nfeatures, authentication, mutualauthentication, confidentiality, dataintegrity\nand non repudiation, for assigning weights to different security mechanisms. Let\nwi\nA\nbe the weight of a mechanism i on authentication.\nwi\nC\nbe the weight of a mechanism i on confidentiality.\nwi\nT\nbe the weight of mechanism i on data integrity.\nwi\nR\nbe the weight of a mechanism i on non-repudiation.\nwi\nM\nbe the weight of a mechanism i on mutual authentication.\nAssume a security policy Pα consists of n security mechanisms. Then, security\nindex security policy Pα of is a metric which is defined as\nSI(Pα) =\nn\n\u0001\ni=1\nwi\nAIA + wi\nCIC + wi\nT IT + wi\nRIR + wi\nMIM\n(1)\n"
        },
        {
            "page_number": 308,
            "text": "306\nAVESH KUMAR AGARWAL and WENYE WANG\nIn the above expression, I(·) is an indicator function, which equals to 1 if that\nparticular security feature is provided by the mechanism i, otherwise zero. Next, we\nexplain weight assignments in detail.\nThe purpose of weight assignment to each security mechanism is to quantify\nstrengths of different security mechanisms with respect to various security features.\nWeight assignment to these security mechanisms is based on several criteria such as\nthe key length and use of digital certificates used in a particular mechanism, which are\nexplained in detail below.\nSince WEP-128, 802.1x-EAP-MD5, IPsec and 802.1x-EAP-TLS provide au-\nthentication feature in the testbed, four different weights are assigned to each\nof them. Weights assigned to each of these mechanisms are based on their\nrelative strengths. WEP-128 is assigned the lowest weight of 1 due to its weak\ncryptographic algorithm [3]. 802.1x-EAP-TLS is assigned the highest weight\nof 4 due to its use of digital certificate for signing private keys. IPsec is as-\nsigned weight of 3 which is lower than the weight assigned to 802.1x-EAP-TLS,\nbecause IPsec uses public key cryptography without certificates unlike 802.1x-\nEAP-TLS. Although digital certificate can be used with IPsec as well, but we\nuse IPsec without certificate due to some practical problems in configuring\nthem together in the testbed. On the other hand, 802.1x-EAP-MD5 is assigned\nweight of 2 which is lower than those of IPsec and 802.1x-EAP-TLS, because\nit uses weak unencrypted user-password mechanism [27].\nIn case of mutual authentication, IPsec and 802.1x-EAP-TLS mechanisms are\nconsidered, and are assigned weights of 1 and 2, respectively. The reason for\nassigning a higher weight to 802.1x-EAP-TLS than IPsec is the same as we\ndescribed for the authentication feature.\nIn addition, WEP-128 and 3DES offer confidentiality feature for various secu-\nrity policies in the testbed. 3DES encryption mechanism is allocated higher\nweight of 2 than the weight of 1 assigned to WEP-128, because 3DES provides\ncomplex and more secure cryptographic algorithm than WEP-128.\nIPsec with SHA/MD5 and 802.1x-EAP-MD5 provide the data integrity security\nfeature. Since SHA uses longer keys than MD5 [28], IPsec with SHA is\nassigned a higher weight of 2 than those of IPsec with MD5 and 802.1x-EAP-\nMD5. IPsec with MD5 and 802.1x-EAP-MD5 are assigned the same weight\nof 1, because both of them use MD5 algorithm.\nWe have considered IPsec and 802.1x-EAP-TLS as security mechanisms with\nrespect to non-repudiation. 802.1x-EAP-TLS policy is assigned a higher weight\nof 2 than the weight 1 of IPsec.\nIn general, we notice that TLS has been assigned a higher weight than MD5,\nbecause it makes use of digital certificates which provide stronger authentication and\n"
        },
        {
            "page_number": 309,
            "text": "WIRELESS NETWORK SECURITY\n307\naccess control mechanism than MD5 [27].\nNote that the weight assigned to each\nprotocol signifies only its relative strength corresponding to other protocols. These\nweights do not imply any quantification of absolute security strength associated to a\nsecurity protocol. For instance, if two mechanisms providing authentication feature\nare assigned weights of 4 and 1, respectively, it does not mean that the mechanism\nwith weight 4 is four times stronger than the mechanism with weight 1 with respect to\nauthentication feature. It only signifies that the mechanism with weight 4 is stronger\nthan the mechanism with weight 1 with respect to authentication feature. The weights\nassigned to each protocol are shown in Table 2.\nTable 2. Weights Associated with Security Protocols.\nSecurity Feature\nSecurity Mechanism\nWeight\nAuthentication\nWEP-128 (Shared)\n1\n(wA)\n802.1x-EAP-MD5\n2\nIPsec\n3\n802.1x-EAP-TLS\n4\nMutual (wM)\nIPsec\n1\nAuthentication\n802.1x-EAP-TLS\n2\nConfidentiality\nWEP-128\n1\n(wC)\n3DES\n2\nData Integrity\nMD5 (IPsec/802.1x-EAP)\n1\n(wT )\nSHA (IPsec)\n2\nNon-repudiation\nIPsec (ESP)\n1\n(wR)\n802.1x-EAP-TLS\n2\nNext let us look at an example of how SI is obtained. We notice from Tables 1 and\n2 that P12 (802.1x-EAP-TLS-WEP-128-IPsec-3DES-SHA) consists of 3 mechanisms:\nIPsec-3DES-WEP, WEP-128 and 802.1x-EAP-TLS. These 3 mechanisms consist of 10\nfeatures: 5 by IPsec-3DES-WEP, 2 by WEP-128, and 3 by 802.1x-EAP-TLS. Let i, j\nand k represent IPsec-3DES-WEP, WEP-128 and 802.1x-EAP-TLS, respectively. By\nusing Table 2, weights of features provided by IPsec-3DES-SHA are wi\nA = 3, wi\nM = 1,\nwi\nC = 2, wi\nT = 2, and wi\nR = 1. The corresponding weights of features in WEP-128\nare wj\nA = 1, wj\nM = 0, wj\nC = 1, wj\nT = 0, and wj\nR = 0. With 802.1x-EAP-TLS, we\nobtain the weight as wk\nA = 4, wk\nM = 2, wk\nC = 0, wk\nT = 0, and wk\nR = 2. Although\n802.1x-EAP-TLS can provide confidentiality and data integrity, but in our testbed, it is\nused in access control for authentication in wireless network without its confidentiality\n"
        },
        {
            "page_number": 310,
            "text": "308\nAVESH KUMAR AGARWAL and WENYE WANG\nand data integrity features. Therefore, we do not take into account the confidentiality\nand data integrity features in 802.1x-EAP-TLS. According to (1), P12 has an index of\nSI(P12)\n=\nwi\nA + wi\nM + wi\nC + wi\nT + wi\nR\n+wj\nA + wj\nM + wj\nC + wj\nT + wj\nR\n+wk\nA + wk\nM + wk\nC + wk\nT + wk\nR.\n(2)\nBy substituting the weights of various features, the value SI for policy P12 is\n3+1+2+2+1+4+2+2+1+1 = 19. For comparative study, we normalize the SI\nvalues of other policies based on the highest value of 19 of security policy P12. Table 3\nlists actual and normalized SI (NSI) values of security policies in the increasing order.\nTable 3. Security Index.\nPolicy\nP1\nP2\nP5\nP7\nP6\nP3\nP8\nP4\nP9\nP11\nP10\nP12\nSI\n0\n2\n3\n5\n8\n9\n10\n11\n12\n13\n18\n19\nNSI\n0\n10.5\n15.8\n26.3\n42.1\n47.4\n52.6\n57.9\n63.2\n68.4\n94.7\n100\n4.2. Cost Analysis\nNow we analyze performance cost associated with various security policies in\nterms of authentication time, cryptographic cost and throughput. Metrics, authentica-\ntion time and cryptographic cost, are associated with authentication phase and encryp-\ntion/decryption process of a security policy, respectively. On the other hand, throughput\nhelps us in quantifying QoS degradation in a network.\nAuthentication Time\nAuthentication time is defined as the total time consumed in an authentication\nphase of a security policy. We consider authentication time, represented as (CA), as the\ncost associated with authentication phase of a security policy. It is due to the fact that\ntime involved in an authentication phase is one of the important factors contributing\ntowards performance impact in a network. Here, we describe steps to calculate the\nauthentication time (CA) as follows:\nAssume that security policy Pα is configured in the network. Let the total time\ninvolved in transmitting, receiving and processing kth packet by Pα during its\nauthentication phase be denoted as Tk(Pα).\nAssume that n packets are exchanged during authentication phase. Then, au-\nthentication time can be represented as,\n"
        },
        {
            "page_number": 311,
            "text": "WIRELESS NETWORK SECURITY\n309\nCA(Pα) =\nn\n\u0001\nk=1\nTk(Pα).\n(3)\nCryptographic Cost\nCryptographic cost represents the performance overhead associated with a security\npolicy. Since we compute the cost of authentication phase of a security policy separately\nin terms of authentication time, cryptographic cost involves overhead due to other\nsecurity features, such as encryption/ decryption, data integrity and so on. Below we\ndescribe the procedure for calculating the cryptographic cost of a security policy.\nLet Pα denote the case that there is no security policy configured in the network\nand Pα denote security policy when there is some security service configured in the\nnetwork where α = {2, 3, . . . , 12}. Let ts(k, Pα) denote the time required to process\nkth packet by a sender s during the configuration of security policy Pα in the testbed.\nThe time duration, ts(k, Pα), usually involves adding extra header by security policy,\nencryption of packet and so on. Let tr(k, Pα) denote the time required to process kth\npacket by a receiver r during the configuration of security policy Pα in the testbed. The\ntime duration, tr(k, Pα), usually involves removing extra header of security policy,\ndecryption of packet and so on. Let tsr(k, Pα) denote the time taken by the kth packet\nin traversing the network between the sender and the receiver during security policy Pα.\nTherefore, the total time involved in processing the kth packet, denoted by T(k, Pα),\nbetween the sender and the receiver during policy Pα is the sum of three time periods\ndefined above, and is given by,\nT(k, Pα) = ts(k, Pα) + tr(k, Pα) + tsr(k, Pα).\n(4)\nAssume that n packets are transmitted between the sender and the receiver, then the\ntotal time required for processing n packets during security policy Pα is the sum of\ntime involved in processing all n packets, that is,\nn\n\u0001\nk=1\nT(k, Pα) =\nn\n\u0001\nk=1\n[ts(k, Pα) + tr(k, Pα) + tsr(k, Pα)].\n(5)\nConsider that the number of bits in each packets may be different, for example, the size\nof kth packet is bk bits. Then the total number of bits in n packets, denoted as Bn, is,\nBn =\nn\n\u0001\nk=1\nbk.\n(6)\nNow we compute bit rate associated with various security policies to measure the\nassociatedcryptographiccostwitheachpolicy. LetRB(Pα)denotethebitrate(bits/sec)\nthat can be experienced during security policy Pα. Using (5) and (6), bit rate for security\npolicy Pα can be obtained as:\n"
        },
        {
            "page_number": 312,
            "text": "310\nAVESH KUMAR AGARWAL and WENYE WANG\nRB(Pα) =\nBn\n\u000en\nk=1(ts(k, Pα) + tr(k, Pα) + tsr(k, Pα).\n(7)\nAssume that CC(Pα) denotes the cryptographic cost associated with security policy\nPα. In this work, we evaluate the cryptographic cost as the difference between the bits\nrates for security policies (Pα) and (P1). Then, CC(Pα) can be calculated as follows:\nCC(Pα) =\nBn\n\u000en\nk=1(ts(k, P1) + tr(k, P1) + tsr(k, P1))\n−\nBn\n\u000en\nk=1(ts(k, Pα) + tr(k, Pα) + tsr(k, Pα)).\n(8)\nThroughput (bits/second) (η)\nIt is defined as the data transferred per unit time between participating nodes during\nthe configuration of a security policy in the network. According to this definition, we\nobserve that if the data is represented in bits, then the throughput associated with a\nsecurity policy is same as the bit rate associated with a security policy that we computed\npreviously during the calculation of cryptographic cost. Therefore, throughput (η)\nduring security policy Pα can be represented as follows:\nη(Pα) =\nBn\n\u000en\nk=1(ts(k, Pα) + tr(k, Pα) + tsr(k, Pα).\n(9)\n5.\nDATA ACQUISITION\nFor each security service configured in our testbed, experimental data are collected\nin two phases. The first phase collects measurements during the initial negotiation of\nprotocols. The second phase focuses on generating streams, and then collecting data\nsuch as throughput and response time for different security policies. In addition, the\ntransmission rate for each wireless card has been set to 11Mbps.\nIn the First phase, we concentrate on taking data that is related to initial negotia-\ntions, which take place during the handshake stage of any protocol. We use Ethereal\nnetwork packet analyzer to capture the packets exchanged in handshake phase. Using\ntimestamp option provided in every packet, we record the time difference between the\nfirst and last packet of the negotiation phase. Since in our analysis, we interpret initial\nnegotiation phase as the authentication phase, data obtained in this manner is used to\ncompute and compare authentication time for different security services.\nThe Second phase in our study includes generating different traffic streams in the\nnetwork between two participating nodes. We use ttcp and Iperf traffic generators,\nbecause they can generate TCP and UDP traffic. Moreover, these utilities provide\ndifferent types of statistics such as end-to-end delay, throughput, packet loss, and so\non. Also, we can verify whether measurements provided by one tool are in consistent\nwith experimental data provided by other tools.\n"
        },
        {
            "page_number": 313,
            "text": "WIRELESS NETWORK SECURITY\n311\nInitially, we generate TCP and UDP streams with different data sizes. But after\nanalyzing experimental data obtained, we observed that, for smaller size data, differ-\nences in measurements of security services are not visible, so they are not helpful in\nthe analysis. Then, we focus our measurement on larger data size such as 16MB from\nwhich we can notice significant differences in the measurements. The data obtained in\nthis phase is used to investigate and compare network throughput and protocol over-\nhead for different security services configured in the testbed. Moreover, we repeat\nexperiments more than 15 times to obtain accurate measurements. The average values\nof these measurements are further used in our analysis and comparison.\n6.\nEXPERIMENTAL RESULTS\nIn this section, we discuss experimental results obtained for afore-mentioned se-\ncurity policies in various mobility scenarios. We provide experimental data for authen-\ntication time, cryptographic cost and throughput defined in Section 4. Experimental\nresults presented in this section are particular to the open source software used in the\nnetwork. As other existing implementations of security protocols may demonstrate var-\nied performances depending upon the software design, coding methods and language\nused; actual quantitative results may vary slightly as compared to the results presented\nin this paper. Moreover, to draw useful conclusions, we categorize the security policies\nin three groups having low, middle and high security strength as described below.\nLow Security Group: P2, P5, P6 and P7 belong to this group and are the\nsecurity policies with SI values below 45%.\nMiddle Security Group: Security policies with SI values between 45% and\n70%, such as P3, P4, P8, P9 and P11, are in the middle security group.\nHigh Security Group: Security policies, P10 and P12, have SI values ap-\nproaching to 100% and belong to the high security group.\n6.1. Authentication Time\nAuthentication time is associated with the initial phase of a security policy as\ndefined in Section 4. During this period, a mobile node provides its credentials to the\nauthentication server, such as home agent or foreign agent in the testbed, to access a\nnetwork. Messages exchanged during the initial phase of a security policy vary with the\nsecurity mechanisms involved in the policy. Moreover, authentication time for various\npolicies is obtained for non-roaming and roaming mobility scenarios, respectively.\nTable 5 shows authentication time for individual protocols, whereas Table 4 shows\nauthentication time (CA in sec) for IPsec and 802.1x policies. Since WEP does not\ninvolve exchange of control messages, there is no authentication time involved with it.\nSinceMobileIPisusedforenablingmobilityinthetestbed, authenticationtime(CA)for\nIPsec and 802.1x involves Mobile IP authentication time as well. In addition, Figures\n"
        },
        {
            "page_number": 314,
            "text": "312\nAVESH KUMAR AGARWAL and WENYE WANG\n3 and 4 demonstrate the authentication versus SI. Note that SI values are demonstrated\nin an increasing order in the figures.\nTable 4. Authentication Time.\nPolicy\nNon-Roaming\nRoaming\nIPsec (sec)\n1.405\n2.837\n802.1x-EAP (MD5)\n0.427\n2.176\nwithout IPsec (sec)\n802.1x-EAP (MD5)\n1.722\n3.471\nwith IPsec (sec)\n802.1x-EAP (TLS)\n1.822\n4.966\nwithout IPsec (sec)\n802.1x-EAP (TLS)\n3.117\n6.281\nwith IPsec (sec)\nTable 5. Individual Authentication Time.\nProtocol\nTime (Sec)\nMobile IP(HA)\n0.11\nMobile IP(FA)\n1.432\nIPsec\n1.295\n802.1x-EAP-MD5\n0.317\n802.1x-EAP-TLS\n1.712\nWe observe from Figures 3 and 4 that 802.1x-EAP-TLS policies cause the longest\nauthentication time among all policies. This is due to the fact that 802.1x-EAP-TLS\nuses digital certificate for mutual authentication, which involves exchange of several\ncontrol packets. We find that a total of 17 control packets are exchanged during the\ninitial phase of 802.1x-EAP-TLS, which is much higher than 8 and 9 control packets\nexchanged in 802.1x-EAP-MD5 and IPsec authentication phases, respectively. More-\nover, IPsec policies generate longer authentication time than 802.1x-EAP-MD5 (with-\nout IPsec) policies because of IPsec tunnel establishment. In addition, we can see\nthat the security policies create longer authentication time in roaming scenarios than\nnon-roaming scenarios due to the reauthentication in a foreign network. Besides these\n"
        },
        {
            "page_number": 315,
            "text": "WIRELESS NETWORK SECURITY\n313\n5\n7\n6\n3\n8\n4\n9\n11 10 12\n15.8\n26.3\n42.1\n47.4\n52.6\n57.9\n63.2\n68.4\n94.7100\n0\n0.5\n1\n1.5\n2\n2.5\n3\nSecurity Policies\n SI\nAuthentication Time (Sec)\nFigure 3. Non-Roaming Scenarios: Authentication Time vs. SI.\n5\n7\n6\n3\n8\n4\n9\n11 10 12\n15.8\n26.3\n42.1\n47.4\n52.6\n57.9\n63.2\n68.4\n94.7100\n0\n1\n2\n3\n4\n5\n6\nSecurity Policy\nSI\nAuthentication Time (Sec)\nFigure 4. Roaming Scenarios: Authentication Time vs. SI.\ngeneral observations, we notice that authentication time does not increase proportion-\nally with respect to the SI of security policies. For example, we recognize that the P3\n(IPsec) induces lower authentication time than the P6 (802.1x-EAP-TLS) in all scenar-\nios although it has higher SI value than the P6. Although P10 and P12 cause longer\nauthentication time than other policies but these policies consist of highest SI values\ndue to more than one levels of security mechanisms involved.\n"
        },
        {
            "page_number": 316,
            "text": "314\nAVESH KUMAR AGARWAL and WENYE WANG\nWe observe that policy P12 (with the highest SI value) consists of longest authen-\ntication and incurs around 7 and 3 times longer authentication time than P5 which has\nthe shortest authentication time in non-roaming and roaming scenarios, respectively.\nThis observation suggests that variations in authentication time values are less in roam-\ning scenarios than in non-roaming scenarios. In other words that even policies with\nlower SI values induce higher authentication time in roaming scenarios. The reason\nfor this phenomenon is that registration to foreign agent takes a very long time (1.432\nsec). Further, authentication time of P6 is approximately 3 and 1.3 times longer than\nauthentication time of P7 in non-roaming and roaming scenarios, respectively, which\nis due to the much higher SI value of P6 than P7. Similar behavior can be observed\nbetween P10 and P11. However, we find that although SI value of P4 is higher than\nthat of P8, authentication time of P4 is less than that of P8 in both scenarios. There-\nfore, it can be concluded that authentication time for security policies does not increase\nmonotonically with SI values.\nFurther, P12 with highest SI value incurs around two times longer authentication\ntime than the security policies in the middle group, for instance IPsec policies, in both\nroaming and non-roaming scenarios. Moreover, we discover that security policies,\nwhich are in the middle security group, for example P3, P8, P4 and P9, do not exhibit\nmuch variations in authentication time, and IPsec policies, P3 and P4, induce the lowest\nauthentication time (1.4 sec in non-roaming and 2.8 sec in roaming) among them. In\naddition, authentication time of a security policy in roaming scenarios is about twice of\nits authentication time in non-roaming scenarios. An exception is 802.1x-EAP-MD5\n(without IPsec) policy which exhibits 5 times longer authentication time due to large\ndifference between its authentication time (0.432 sec) and registration time to a foreign\nagent by mobile node (1.432 sec).\nBased on these observations, we conclude that policies in the middle security\ngroup provide the best tradeoff between security and performance overhead, and IPsec\npolicies, P3 and P4, are the best among them. On the other side, P12 (802.1x-EAP-TLS\nwith IPsec) is the best suitable for the network carrying very sensitive data.\n6.2. Cryptographic Cost\nNow, we discuss cryptographic cost associated with security policies in roaming\nand non-roaming scenarios. By analyzing cryptographic cost, we capture encryption\nand decryption time associated with security policies during the data transmission.\nIn addition, we have normalized experimental data for comparing results in various\nscenarios. Tables 6 and 7 list cryptographic costs in non-roaming and roaming scenarios\nfor low, middle, and high security, respectively. Values presented in italics in Tables\n6 and 7, represent the best-recommended security policies in each security group in a\nparticular scenario. On the other hand, values presented in bold face indicate the overall\nrecommended security policy for a particular network scenario.\nWe notice from Tables 6 and 7 that cryptographic costs associated with policies P4,\nP9, P11, P10 and P12 in non-roaming scenarios are very close to each other, showing\nlittle variations. This is due to the fact that these policies use the same IPsec and WEP\n"
        },
        {
            "page_number": 317,
            "text": "WIRELESS NETWORK SECURITY\n315\nTable 6. TCP Cryptographic Cost (Kbits/sec): Low Security.\nScenarios\nNo Security\nLow\nSecurity\nP1\nP2\nP5\nP7\nP6\nN1\n0\n71.10\n11.88\n75.88\n11.47\nN2\n0\n70.90\n2.09\n101.88\n3.92\nN3\n0\n108.78\n7.29\n105.11\n4.33\nR1\n0\n90.43\n1.97\n104.81\n6.15\nR2\n0\n208.04\n25.60\n232.66\n1.79\nTable 7. TCP Cryptographic Cost (Kbits/sec): Middle and High Security.\nScenarios\nMiddle\nSecurity\nHigh\nSecurity\nP3\nP8\nP4\nP9\nP11\nP10\nP12\nN1\n264.90\n77.21\n302.80\n286.89\n313.94\n291.48\n301.77\nN2\n273.45\n57.15\n311.70\n347.33\n298.78\n299.25\n296.13\nN3\n304.59\n118.71\n331.68\n343.84\n378.10\n382.87\n343.52\nR1\n209.54\n92.19\n216.27\n246.49\n251.56\n259.97\n260.81\nR2\n318.32\n230.78\n367.53\n393.13\n391.12\n381.70\n395.29\nmechanisms which are the dominating factors contributing towards their cryptographic\ncosts. Generally, the policies P4, P9, P11, P10 and P12 exhibit 16% higher crypto-\ngraphic costs than P3, and around 3.5 times higher than that of P2, P7 and P8. The\nreason is that policies P4, P9, P11, P10 and P12 have more than one levels of encryption\nand decryption mechanisms associated with them. Further, we observe that P5 and P6\nexhibit negligible cryptographic costs, which is due to the fact that these policies do\nnot consist of any encryption/decryption mechanisms associated with them. Although,\ntheoretically, cryptographic costs of policies P5 and P6 should be zero, but the small\nvalues obtained are due to some external factors in real-time environments. A closer\nlook at the table reveals that cryptographic cost increases corresponding to SI values.\n"
        },
        {
            "page_number": 318,
            "text": "316\nAVESH KUMAR AGARWAL and WENYE WANG\nHowever, we see that P8 is the policy with a higher SI value but with lower crypto-\ngraphic cost. Specifically, P8 exhibits almost half of cryptographic cost of policies P4,\nP9, P11, P10 and P12, and almost similar to policies P2 and P7.\nWe also notice the similar behavior for UDP traffic in various non-roaming sce-\nnarios in our experiments. However, cryptographic costs of security policies for UDP\ntraffic are less than that of TCP traffic. It is due to the fact that TCP requires acknowl-\nedgment for each packet, leading to the transmission of more number of packets through\nthe networks than UDP. So TCP results in higher encryption and decryption processing\noverhead, leading to increased cryptographic cost. Therefore, the observations suggest\nthat P8 (802.1x-EAP-TLS with WEP) provides the best tradeoff for both TCP and\nUDP types of traffic streams in non-roaming scenarios. However, we recognize that in\nnon-roaming scenarios during UDP traffic, difference between cryptographic costs of\npolicies P9, P10, P11, P12 (with high SI values), and P8 is relatively less. Therefore,\nP12 is a good choice with little extra overhead in these scenarios due to its very strong\nsecurity features.\nComparing the cryptographic costs in roaming scenarios, we find that crypto-\ngraphic cost of P12 (with the highest SI value) is about two times higher than those of\npolicies P2, P7 and P8, and 25% higher than that of P3 for TCP traffic in R1 scenario.\nWhereas, P12 exhibits almost twice the cryptographic cost of P2, P7 and P8, and 24%\nhigher than P3 in R2 scenario during TCP. On the other side, P12 demonstrates about\n4 times higher overhead than P2, P7 and P8, and almost twice of P3 for UDP traffic\nin R1 scenario. In addition, P12 shows almost twice the overhead of P2, P7 and P8,\nand 40% higher than P3 during UDP traffic in R2 scenario. Moreover, we observe\nthat P9, P10 and P11 show cryptographic cost very close to P12 with little variations.\nCryptographic costs for P5 and P6 are negligible in almost all scenarios due to the same\nreason cited previously. Therefore, we notice that P8 provides the best tradeoff in all\nroaming scenarios due to the low overhead associated with it. However, we observe that\nvariations between cryptographic costs of P12 and P8 are small. Therefore, it suggests\nthat P12 is also an alternative in roaming scenarios.\n6.3. Throughput\nTo understand the impact of security policies on the network performance, Tables 8\nand 9 enumerate throughputexhibitedbysecuritypoliciesindifferentnetworkscenarios\nfor TCP traffic. We follow the similar methodology for representing values in these\ntables as in the tables for cryptographic cost. The only difference is that Tables 8 and 9\nillustrate only the overall recommended policy in each scenario. This is due to the fact\nthat variations in throughput across security policies are not very significant, therefore,\nwe prefer to represent the overall recommended security policy in each scenario rather\nthan illustrating the most suitable security policy in each security group.\nWe observe that the highest variations in throughput for various security policies\nduring TCP traffic are up to 12%, 13.5%, 14.6% in N1, N2 and N3 scenarios, respec-\ntively. Whereas, variations in throughput for UDP traffic are up to 6%, 5.6%, 8% in\nN1, N2 and N3 scenarios, respectively. On the other side, roaming scenarios, R1 and\n"
        },
        {
            "page_number": 319,
            "text": "WIRELESS NETWORK SECURITY\n317\nTable 8. TCP Throughput (Kbits/sec): Low Security.\nScenarios\nNo Security\nLow\nSecurity\nP1\nP2\nP5\nP7\nP6\nN1 (1.0e+03)\n2.90\n2.83\n2.89\n2.83\n2.89\nN2 (1.0e+03)\n5.64\n5.51\n5.64\n5.45\n5.64\nN3 (1.0e+03)\n2.97\n2.86\n2.96\n2.86\n2.97\nR1 (1.0e+03)\n2.83\n2.74\n2.83\n2.73\n2.83\nR2 (1.0e+03)\n2.86\n2.65\n2.83\n2.62\n2.86\nTable 9. TCP Cryptographic Cost (Kbits/sec): Middle and High Security.\nScenarios\nMiddle\nSecurity\nHigh\nSecurity\nP3\nP8\nP4\nP9\nP11\nP10\nP12\nN1 (1.0e+03)\n2.64\n2.83\n2.60\n2.62\n2.59\n2.61\n2.60\nN2 (1.0e+03)\n5.11\n5.53\n5.04\n4.97\n5.06\n5.06\n5.07\nN3 (1.0e+03)\n2.67\n2.85\n2.64\n2.63\n2.59\n2.59\n2.63\nR1 (1.0e+03)\n2.62\n2.74\n2.62\n2.59\n2.58\n2.57\n2.57\nR2 (1.0e+03)\n2.54\n2.63\n2.49\n2.46\n2.47\n2.48\n2.46\nR2, exhibit variations up to 10% and 16% for TCP traffic, respectively. In addition,\nfrom our experiments we find that R1 and R2 demonstrate variations up to 12% and 7%\nfor UDP traffic, respectively. We notice that variation in throughput values are around\n10% in most of the scenario and in some scenarios even lower than 10% Therefore, it\nsuggests that the effect of security policies over throughput during data transmission\nis not very significant. This is based on the fact that we have not taken into account\nthe cost of authentication time for calculating throughput, because throughput for a\ndata stream is calculated using the total time involved in transmission of the entire data\nafter authentication phase is over. Therefore, variations in throughput values presented\nin this paper are only because of cryptographic costs. Another reason to segregate\n"
        },
        {
            "page_number": 320,
            "text": "318\nAVESH KUMAR AGARWAL and WENYE WANG\nauthentication phase from throughput phase is to measure the authentication cost inde-\npendently, which would be helpful in comparing authentication cost and cryptographic\ncost to uncover some useful facts below.\nIn addition, we believe that, in the future, as hardware becomes faster, crypto-\ngraphic cost (i.e., time involved in encryption/decryption process) will be reduced\nfurther. Moreover, based on our previous observations from Figure 4, authentication\ntime in roaming scenarios is very high, and it may affect mobile applications signifi-\ncantly as user’s mobility increases. As we observe that variations in throughput across\nsecurity policies are almost similar, we speculate that QoS degradation in a network\nmay be more significant due to the authentication cost than the cryptographic cost in\nthe future.\n7.\nSECURITY ANALYSIS\nIn this section, we discuss security issues associated with security policies against\nmalicious attacks. In addition, we demonstrate the advantages of the cross-layer inte-\ngration of security protocols in providing enhanced security.\n7.1. Authentication Forging\nIn the low security group, policy P2 (WEP) offers open or shared key authentica-\ntion mechanisms. However, P2 (WEP) is known to be highly vulnerable, and therefore,\nprone to authentication forging by malicious users. It is because of the small key space\nand the reuse of the same initialization vector (IV) [3] involved in WEP. Therefore, WEP\nis not advisable to be used in stand-alone mode in WLANs. Similarly, P5 (802.1x-EAP-\nMD5) and P7 (802.1x-EAP-MD5-WEP-128) in the low security group are vulnerable\nto authentication forging due to the inclusion of MD5 in which passwords are trans-\nmitted in clear-text form [26]. Only policy P6 (802.1x-EAP-TLS) in the low security\ngroup is immune to authentication forging due to digital certificate used in its authenti-\ncation process. Policies in the middle and high security groups either include IPsec or\nEAP-TLS mechanisms which use public key cryptography and digital certificate, and\ntherefore, are not vulnerable to authentication forging. Here, we observe that integrated\npolicies in the middle and high security groups are able to overcome the vulnerability\nexhibited by the policies in the low security group.\n7.2. Man In The Middle Attack (MITM)\nPolicies in the low security group are vulnerable to the man in the middle attack\n(MITM). For example, policies P5 (802.1x-EAP-MD5) and P7 (802.1x-EAP-MD5-\nWEP-128) transmit response in plain-text form, are vulnerable to MITM attack [26].\nSince it is easy to decipher WEP security keys [3], it is possible to perform MITM\nattack with the policy P2 as well. On the other hand, security policies with IPsec and\nTLS protocols are not vulnerable to MITM attack. Although, IPsec employs public key\ncryptography, which is vulnerable to MITM attack, ISAKEMP key agreement protocol\n"
        },
        {
            "page_number": 321,
            "text": "WIRELESS NETWORK SECURITY\n319\nused in IPsec prevents the attack. Whereas, TLS protocol use digital certificate and is\nnot vulnerable to MITM attacks.\n7.3. User Access Control (or Unauthorized Participation)\nBesides previous attacks, such as authentication forging and MITM, we observe\nthat user access control is not strong enough in the policies in the low security group.\nMoreover, policies P3 and P4 in the middle security group are not appropriate for the\naccess control at the user level as well. It is due to the fact that IPsec mechanism used\nin policies P3 and P4 works at network layer 3, and provides system authentication\nbut not user level authentication [21]. For example, if a system is authenticated using\nIPsec, any user, authorized or unauthorized, with access to the system can use network\nresources. Therefore, for user authentication, P3 and P4 must be employed with higher\nlayer security protocols. For instance, policies P9 and P11 provide strong security\nat layer 3 due to IPsec, and enable user access control by using 802.1x-EAP-MD5.\nHowever, user access control with policies P9 and P11 is not secure because MD5\ntransmits passwords in clear-text form [26]. Policies P8, P10 and P12 use EAP-TLS\nmechanism, and provide strong access control due to the use of digital certificates, and\ndynamic keys which are refreshed periodically during a session as well [27]. Here\nagain, we observe the advantages of integrating security protocols at different layers.\n7.4. Fabrication of Messages\nFabrication of message is possible in the network configured with policies in the\nlow security group, such as P2, P5 and P7. It is due to the fact that it is easier to\ndecipher security keys used in these policies. However, policies in the middle and\nhigh security groups use IPsec or EAP-TLS with public key cryptography and digital\ncertificate mechanisms, respectively. Therefore, modification of messages in the transit\nis not possible over these policies.\n7.5. Denial of Service (DOS) and Traffic Analysis\nAll policies are vulnerable to denial of service (DOS) and traffic analysis attacks\n[29], [30], [31]. The reason is that these attacks are hard to prevent even with very\nstrong cryptographic features.\nIn addition, WEP is susceptible to dictionary attacks, statistical cryptanalysis,\nknown plaintext, and partial known plaintext attacks. Moreover, due to security is-\nsues with 802.1x framework, policies P5 to P12 suffer from packet spoofing attacks\n[4]. In general, policies P3 and P4 provide enhanced security due to the strong crypto-\ngraphic and key management protocols used in IPsec. Moreover, policies P10 and P12\nare very strong because they use IPsec and EAP-TLS together with WEP [21], [27].\nConsequently, P10 and P12 provide strong authentication, confidentiality, mutual au-\nthentication, non-repudiation and data integrity features.\nIn summary, we observe that when security policies are used in stand-alone mode,\nthey are prone to many attacks. But when we configure protocols at various layers\n"
        },
        {
            "page_number": 322,
            "text": "320\nAVESH KUMAR AGARWAL and WENYE WANG\ntogether, the attacks associated with a weak protocol can be prevented more effectively\nby the strong protocols at other layer. Therefore, the policies in the middle and the\nhigh security groups provide enhanced security and prevent vulnerabilities related to a\nsingle protocol. Therefore, the cross-layer integration of security protocols seems an\nadvantageous choice in providing better security solutions for many wireless applica-\ntions.\n8.\nCONCLUSIONS\nIn this work, we addressed the issue of performance overhead and security strength\nassociated with security protocols in WLANs. Specifically, we studied the cross-layer\nintegration of various security protocols with respect to authentication time, crypto-\ngraphic cost and throughput in different network scenarios with TCP and UDP data\ntraffic. Moreover, we performed a comprehensive study to obtain the experimental\nresults of performance metrics associated with security policies. We found that IPsec\npolicies, P3 and P4, provide the best tradeoff between security and performance re-\ngarding authentication time. Moreover, we observed that P8 (802.1x-EAP-TLS) is the\nmost suitable option for low cryptographic cost and better security strength in many\nscenarios. However, we also found that P12 (802.1x-EAP-TLS with IPsec) is also a\nsuitable option with little extra overhead during some network scenarios.\nWe noticed that there is always a tradeoff between security and performance as-\nsociated with a security policy depending upon the network scenario and traffic types.\nIt is seen that security policy with higher strength may not always be the best option\nfor all scenarios with respect to the tradeoff between security strength and performance\noverhead. We found that cross-layer integration of security protocols at many layers,\nfor example, policies P9, P10, P11, P12 provide very strong security but with higher\noverhead. Therefore, we suggest that these policies are most suitable for the networks\ncarrying very sensitive data. Moreover, we noticed that the variations in throughput\nunder different scenarios are not very large, and concluded that authentication time may\nbe a more significant factor contributing towards QoS degradation than cryptographic\ncost as hardware becomes faster in the future.\nIn summary, our results recommended the appropriate security policy for each\nscenario. In addition, we provided the quantification of performance overhead. We\nbelieve that combination of these results can lay a very strong foundation for designing\nnew security protocols or improving the existing ones. Moreover, the real-time nature\nof our results can enable network designers make intelligent decision regarding the\nimplementation of security policies in a network. Also, performance measurements for\nauthentication cost, cryptographic cost and throughput are helpful in deciding about\nwhich security feature should be enabled in a particular scenario while keeping overhead\nlow. Therefore, our study provide first-hand valuable results, which will be very useful\nto the design of network protocols for security and flexible quality of service in future\nmobile wireless networks.\n"
        },
        {
            "page_number": 323,
            "text": "WIRELESS NETWORK SECURITY\n321\n9.\nREFERENCES\n1. T. Karygiannis and L. Owens, “Wireless Network Security 802.11, Bluetooth and Handheld Devices,”\nNational Institute of Technology, Special Publication, pp. 800–848, November 2002.\n2. Y. Zahur and T. A. Yang, “Wireless LAN Security and Laboratory Designs,” Journal of Computing\nSciences in Colleges, vol. 19, pp. 44–60, January 2004.\n3. N. Borisov, I. Goldberg, and D. Wagner, “Intercepting Mobile Communications: The Insecurity of\n802.11,” in Proceedings of the Seventh Annual International Conference on Mobile Computing And\nNetworking, July 2001.\n4. “IEEE 802.1X,” http://www.ieee802.org/1/pages/802.1X-rev.html, 2004.\n5. “IEEE 802 Standards,” http://standards.ieee.org/getieee802.\n6. A. Hecker and A. H. Laboid, “Pre-authenticated Signaling in Wireless LANs using 802.1X Access\nControl,” in Proceedings of IEEE Global Telecommunications Conference, GLOBECOM ’04, vol. 4,\npp. 2180 – 2184, November-December 2004.\n7. A. Hecker and A. H. Laboid, “A New EAP-based Signal Protocol for IEEE 802.11 Wireless LANs,” in\nProceedings of IEEE 60th Vehicular Technology Conference, VTC-Fall, 2004 , vol. 5, pp. 3214 – 3218,\nSeptember 2004.\n8. W. Qu and S. Srinivas, “IPSEC-Based Secure Wireless Virtual Private Networks,” in Proceedings of\nIEEE MILCOM, pp. 1107–1112, October 2002.\n9. D. B. Faria and D. R. Cheriton, “DoS and Authentication in Wireless Public Access Networks,” in\nProceedings of ACM Workshop on Wireless Security (WiSe), pp. 47–56, September 2002.\n10. W. A. Arbaugh, N. Shankar, J. Wang, and K. Zhang, “Your 802.11 Network Has No Clothes,” IEEE\nWireless Communications Magazine, December 2002.\n11. S. Kasera, S. Mizikovsky, G. S. Sundaram, and T. Y. Woo, “On Securely Enabling Intermediary-Based\nServices and Performance Enhancements for Wireless Mobile Users,” in Proceedings of ACM Workshop\non Wireless security (WiSe), pp. 61–68, September 2003.\n12. J. Kong, S. Das, E. Tsai, and M. Gerla, “ESCORT: A Decentralized and Localized Access Control\nSystem for Mobile Wireless Access to Secured Domains,” in Proceedings of ACM workshop on Wireless\nsecurity (WiSe), pp. 61–68, September 2003.\n13. Y. Matsunaga, A. Merino, T. Suzuki, and R. H. Katz, “Secure Authentication System for Public WLAN\nRoaming,” in Proceedings of The 1st ACM international workshop on Wireless mobile applications and\nservices on WLAN hotspots, pp. 113–121, 2003.\n14. M. Li, H. Zhu, S. Sathyamurthy, I. Chlamtac, and B. Prabhakaran, “End-to-End Framework for QoS\nGuarantee in Heterogeneous Wired-cum-Wireless Networks,” in Proceedings of The First International\nConference on Quality of Service in Heterogeneous Wired/Wireless Networks, 2004, pp. 140 – 147, Oct\n2004.\n15. O. Elkeelany, M. M. Matalgah, K. Sheikh, M. Thaker, G. Chaudhary, D. Medhi, and J. Qaddour,\n“Perfomance Analysis Of IPSEC Protocol: Encryption and Authentication,” in Proceedings of IEEE\nCommunication Conference (ICC), pp. 1164–1168, May 2002.\n16. A. Godber and P. Dasgupta, “Secure Wireless Gateway,” in Proceedings of ACM Workshop on Wireless\nSecurity (WiSe), pp. 41–46, September 2002.\n17. G. Hadjichristofi, N. D. IV, and S. Midkiff, “IPSec Overhead in Wireline and Wireless Networks for\nWeb and Email Applications,” in Proceedings of IEEE International Performance, Computing, and\nCommunications Conference, 2003, pp. 543 – 547, April 2003.\n18. K. Wang and S. Tripathi, “Mobile-End Transport Protocol: An Alternative to TCP/IP Over Wireless\nLinks,” in Proceedings of IEEE INFOCOM, pp. 1046–1053, April 1998.\n"
        },
        {
            "page_number": 324,
            "text": "322\nAVESH KUMAR AGARWAL and WENYE WANG\n19. M. D. Corner and B. D. Noble, “Zero-Interaction Authentication,” in Proceedings of IEEE/ACM MO-\nBICOM, pp. 1–11, September 2002.\n20. C. Perkins, “IP Mobility Support,” http://www.ietf.org/rfc/rfc2002.txt, October 1996.\n21. “IPSEC,” http://www.freeswan.org.\n22. “802.1x Supplicant,” http://www.open1x.org.\n23. “RADIUS,” http://www.freeradius.org.\n24. “OpenSSL,” http://www.openssl.org.\n25. “Mobile IPv4,” http://dynamics.sourceforge.net.\n26. I.-G. Kim and J.-Y. Choi, “Formal Verification of PAP and EAP-MD5 Protocols in Wireless Networks:\nFDR Model Checking,” in Proceedings of The 18th International Conference on Advanced Information\nNetworking and Applications, vol. 2, pp. 264 – 269, March 2004.\n27. B. Aboba and D. Simon, “PPP EAP TLS Authentication Protocol,” RFC 2716, October 1999.\n28. A. Satoh and T. Inoue, “ASIC-Hardware-Focused Comparison for Hash Functions MD5, RIPEMD-\n160, and SHS,” in Proceedings of International Conference on Information Technology: Coding and\nComputing, ITCC 2005, vol. 1, pp. 532 – 537, April 2005.\n29. C. B. McCubbin, A. A. Selcuk, and D. Sidhu, “Initialization Vector Attacks on the IPsec Protocol\nSuite,” in Proceedings of EEE 9th International Workshops on Enabling Technologies: Infrastructure\nfor Collaborative Enterprises, (WET ICE 2000), pp. 171–175, June 2000.\n30. A. Mian and A. Masood, “Arcanum: A Secure and Efficient Key Exchange Protocol for the Internet,”\nin Proceedings of International Conference on Information Technology: Coding and Computing, ITCC\n2004, vol. 1, pp. 17 – 21, April 2004.\n31. J. Rejeb, M. Vohra, and T. Le, “IKE-based Secure Wireless and Mobile Networks,” in Proceedings of\nThe IEEE 6th Circuits and Systems Symposium on Emerging Technologies: Frontiers of Mobile and\nWireless Communication, 2004, vol. 2, pp. 567 – 570, June 2004.\n"
        },
        {
            "page_number": 325,
            "text": "13\nSECURITY ISSUES IN\nWIRELESS SENSOR NETWORKS USED IN\nCLINICAL INFORMATION SYSTEMS\nJelena Miˇsi´c and Vojislav B. Miˇsi´c\nDepartment of Computer Science\nUniversity of Manitoba\nWinnipeg, Manitoba R3T 2N2, Canada\nE-mail: {jmisic, vmisic}@cs.umanitoba.ca\nHigh quality healthcare is an important aspect of the modern society. In this chapter we\naddress the security and networking architecture of a healthcare information system com-\nprised of patients’ personal sensor networks, department/room networks, hospital network,\nand medical databases. Areas such as diagnosis, surgery, intensive care and treatment, and\npatient monitoring would greatly benefit from light untethered devices which can be unob-\ntrusively mounted on patient’s body in order to monitor and report health-relevant variables\nto the interconnection device mounted on the patient’s bed. Interconnection device should\nalso have larger range wireless interface which should communicate to the access point in\nthe patient’s room, operation room or to the access points within the healthcare institution.\nThe results of measurements will then be stored in central medical database with appropri-\nate provisions for protecting the patient privacy as well as the integrity of personal health\nrecords. We review confidentiality and integrity polices for clinical information systems\nand discuss the feasible enforcement mechanisms over the wireless hop. We also compare\ncandidate technologies IEEE 802.15.1 and IEEE 802.15.4 from the aspect of resilience of\nMAC and PHY layers to jamming and denial-of-service attacks.\n1.\nINTRODUCTION\nHealthcare is an important area for deployment of wireless sensor and personal\narea networks. The IEEE 1073 Medical Device Communications standards organiza-\ntion is currently in the process of developing the specifications for wireless interface\ncommunications. The main objective for this effort is to develop universal and inter-\noperable devices for medical equipment which are transparent to the user and easily\nre-configurable. The group has recognized that developing new wireless technologies\n"
        },
        {
            "page_number": 326,
            "text": "326\nJELENA MI ˇSI ´C and VOJISLAV B. MI ˇSI ´C\nis not an option and is looking instead in deployment of existing wireless technologies\nbelonging to IEEE 802 family in the healthcare applications.\nThere are many research issues related to sensor and Wireless Personal Area\n(WPAN) networks in healthcare. First, there are different healthcare applications which\nmonitor vital signs, electrocardiogram signals (ECG), electroencephalogram signals\n(EEG), as well as signals from other electronic or electro-mechanical devices that may\nbe used in healthcare (dialysis, infusion, . . . ). All these applications require some min-\nimum event detection reliability, i.e., the minimum number of data bits per second, as\nthe result of sampling and digitizing analog variables related to patient health or proper\nfunctioning of electronic and electro-mechanical devices. Therefore, it is important\nto pair the medical application with the low rate WPAN technology from the aspect\nof sufficient bandwidth as well as from the aspect of supported security mechanisms.\nAs the bandwidth requirements of different variables and devices vary, it is probably\nunrealistic to assume that any given WPAN technology can cope with requirements of\ndifferent medical applications. Then, among the candidates for one application it is\nnecessary to address several issues:\n1. We need to define the security policies to be utilized for management and use\nof patient medical records within the clinical information system. The policies\nshould aim to protect the confidentiality and integrity of data from its very entry\ninto the system at the patient WPAN.\n2. To this end, it is necessary to develop appropriate security and network architec-\nture for sensor networks, WLANs, and WPANs that might be deployed within\nthe clinical information system. That architecture (or architectures) will serve\nas the foundation upon which individual health applications can monitor the\nhealth of individual mobile patients without harming their health or life habits.\n3. We need to consider secure location management whenever the patient changes\nlocation within the hospital, either because the patient walks around, or his/her\nbed is moved from one room to another.\n4. We need to look at the security issue of denial of service at the physical and\nMAC layers (jamming) which can cut the flow of patient’s data to the monitoring\nstation. This problem is related to the interference issues since every mobile\npatient or patient’s bed presents independent WPAN(s). They might interfere\namong themselves, and with WLAN running in the room or WPAN carried by\nthe doctor/nurse.\n5. We need to provide secure interconnections among different WPANs among\nthemselves and with WLAN. The efficiency of interconnecting devices will\ndetermine the scalability of our secure healthcare network design.\n6. The packet delay issue, which is related to the Medium Access Control (MAC)\nprotocol used in particular technology. It is also necessary to look at the packet\nsize for given technology since all measured health variables, which are analog\n"
        },
        {
            "page_number": 327,
            "text": "WIRELESS NETWORK SECURITY\n327\nand hence have to undergo analog-to-digital conversion as well as encryption,\nproduce a stream of bits with a constant rate. In such cases, packetization delay\nbecomes an issue.\nWe begin the chapter by reviewing clinical information security policies. Then,\nwe propose networking and security architecture of clinical information system which\nincludes patient sensor networks, wireless local area networks which belong to the\ndepartments, and the central medical database where results of patient examinations are\nheld. Enforcement of policy rules using cryptographic mechanisms over networking\ninfrastructure is discussed, followed by a discussion of the classification of medical\napplications and pairing with WPAN technologies. We also compare some candidate\ntechnologies for wireless sensor networks from the aspects of MAC and physical layer\nsecurity and sensing reliability. A brief summary concludes the chapter.\n2.\nSECURITY POLICY FOR HEALTHCARE SENSOR NETWORKS\nAS PART OF CLINICAL INFORMATION SYSTEMS\nSensor networks in medical applications are the edge component of the clinical\ninformation system. The wireless data flows with health variables measurements are\npart of ‘personal health information’ and must be protected from the aspect of integrity\nand patient privacy before they can be stored in the patient medical record. Actually,\nhealth sensing information forms the most important part of the medical record. The\nsecurity policies for medical records have been extensively studied; they have to be\ncarefully designed in order to (a) limit the number of actors, clinical physicians and\nothers, that can access the patient record, and (b) control the operations over the record\n[2, 4]. The policies are typically expressed as a number of security rules, including the\nfollowing:\n1. Each medical record has an associated access control list which names the\nindividuals and groups that may read, update, and append the information to\nthe record. The system must restrict the access to those identified on the access\ncontrol list.\n2. One of the clinicians on the access control list (called the responsible clinician)\nmust have right to add other clinicians to the access control list.\n3. The responsible clinician must notify the patient of the names on the access\ncontrol list whenever the patient medical record is accessed.\n4. Each time the record is accessed, the name of the clinician, the date and time,\nand the manner of access have to be recorded.\nThe purpose of previous four access rules is to control the confidentiality of the\nmedical record. Patient must consent to the treatment and he/she must have the access\nto his/her record at any time. Moreover, the patient must be informed whenever any\n"
        },
        {
            "page_number": 328,
            "text": "328\nJELENA MI ˇSI ´C and VOJISLAV B. MI ˇSI ´C\nclinician accesses the record. In all previous cases, if the patient is incapacitated to\nmake the decisions for him- or herself, the authority rests with the legal guardian or\nanother person with the appropriate power of attorney.\nIntegrity of the patient’s medical record is protected by the following set of rules:\nCreation Whenanewmedicalrecordiscreated, thecliniciancreatingtherecordshould\nhave access to it, as should the patient. If the medical record is created due to\nthe referral from another referring clinician, he/she should also be authorized\nto access the record.\nDeletion Clinical information cannot be deleted from the medical record until the\npredefined time period has passed.\nConfinement Information from one medical record may be appended to a different\nmedical record if and only if the access control list of the second record is a\nsubset of the access control list of the first.\nAggregation Aggregation of patient data must be prevented.\nEnforcement Any computer system that handles medical records must have a subsys-\ntem that enforces previous rules.\nThe need for wireless sensor networks which are integrated in the medical infor-\nmation system presents a big challenge to the implementation of aforementioned rules.\nUnfortunately, previous access principles can’t be implemented in the network envi-\nronment through simple access lists. Instead, we will need to use some cryptographic\ntechniques which we will discuss in the next section.\n3.\nSECURITY ARCHITECTURE OF THE WIRELESS LAYER OF THE\nMEDICAL INFORMATION SYSTEM\nLet us consider the medical information system infrastructure including the wire-\nless sensor networks, as shown in Fig. 1. Important parts of the architecture are the\npatient security processor (PSP) which is attached to bed and the wireless access point\nin the patient room. The PSP is module that implements networking as well as security-\nrelated functions.\nFrom the networking aspect, PSP is the coordinator of sensing nodes which belong\nto the patient’s Personal Area Network (PAN) and participates in the Medium Access\nControl function of the nodes. For example, for IEEE 802.15.1 technology (Bluetooth)\nPSPwillbeexecutedonthepiconet’smasterandforIEEE802.15.4PSPwillbeexecuted\non the cluster’s coordinator.\nFrom the security aspect, it generates the symmetric encryption key by which all\ndata packets with sensed health information are encrypted. It distributes the symmetric\nkey to the sensing nodes by encrypting it with public key which is common for all\nsensing nodes. Sensing nodes are pre-configured with the private key by which they\n"
        },
        {
            "page_number": 329,
            "text": "WIRELESS NETWORK SECURITY\n329\npatient security\nprocessor and\nPAN coordinator\n(PSS)\nroom\naccess point\nward network (wired)\nward room\npatient sensor\nnode\nFigure 1. Security architecture of wireless part of medical information systems.\ncan decrypt the symmetric key. Sensing nodes will send packets with encrypted payload\nand completely authenticated to the patient security processor which forwards them,\npossibly aggregated, to the room access point.\nThe patient room access point is further connected to the central medical record\ndatabase through a suitable wired network. The access point forwards encrypted and\nauthenticated packets to the central database. Data packets which carry measurements\nof personal health variables must be authenticated and encrypted in the way which\nwe discuss below. From the networking point of view access point is interconnection\ndevice which interconnects Personal Area Network technology (IEEE 802.15.1 or IEEE\n802.15.4) with the hospital network which might be implemented using wireless LAN\nand mesh technologies.\nMedical personnel might carry their own PAN nodes and communicate directly to\nmedical health database through the patient’s room access point.\nSecurity of medical applications over sensor networks has to be protected at every\nnetworking layer. At the physical and MAC layer there exists possibility of denial of\nservice attack by generating to much interference or by generating unnecessary traffic.\nTherefore, MACs should be evaluated from this perspective also. Payload of packets\nwith sensed data should be encrypted when needed. Also, in some situations, patient’s\nlocation should be hidden as well. Given the hierarchical application architecture, there\nshould exist layered security architecture with different keys and possibly different\nencryption algorithms at WPAN, room and hospital level. The encryption standards\nused at particular level should match importance and vulnerability of the data. For\nexample the traffic at the WPAN level has to be encrypted at the MAC level but the\ntraffic between access points should be protected by IPSec.\nHowever, security measures will affect the delay and throughput of sensed health\ndata and this impact has to be carefully evaluated. Initial work on performance eval-\nuation of IPSec is presented in [9] but much more needs to be done for multi-tier\ncommunication architecture built over WPANs. We plan to develop multi-layer secu-\n"
        },
        {
            "page_number": 330,
            "text": "330\nJELENA MI ˇSI ´C and VOJISLAV B. MI ˇSI ´C\nrity architecture which will match confidentiality and integrity of the sensed data and\nevaluate the performance of overall application architecture.\n4.\nENFORCEMENT OF PRIVACY AND INTEGRITY RULES\nIn order to protect from the attacks from the outside world, all hospital equipment\nand personnel must possess the secret ‘hospital/department/room’ key KH. This key is\nused to sign and authenticate network packets generated by the equipment and personnel\nbelonging to specific medical department. Authentication is achieved by calculating the\nhash function over the packet with measurement data, hospital/department/room key\nand timestamp Ts with time of packet generation. For hash function, we adopt Secure\nHash Algorithm (SHA) [27]. Let us denote i-th packet containing measurements of\nsome health variable as Pi, its Medium Access Control header as Hi and its payload\nas Di. Packet authentication code for packet i(PACi) can then be calculated as\nPACi = H(KH, Ts,i, Hi||Di).\n4.1. Patient privacy\nAforementioned access policy rules require that only patient and clinicians have\naccess to patient’s medical record and that patient must be informed of any access to\nhis/her record. Therefore, this small group must have dedicated secret session key\nKp (p comes from patient who is the principal of the group), but no one from this\ngroup must have the capability to derive the key without the participation of other\nmembers. Particularly, participation of the patient is necessary in all accesses. This\nkey will be used as encryption key of an symmetric encryption system such as 3-DES\n(Data Encryption Standard) or AES (Advanced Encryption Standard). Operations of\nencryption and decryption with patient’s key will be denoted as EKp() and DKp()\nrespectively. Encryption using public key cryptography takes long time and generates\nhigh packet payload which is a problem for existing candidate technologies for wireless\nsensor networking.\nProcess of generating patient’s key requires attention. If the patient is unable to\nparticipate in the decisions regarding his/her healthcare, then his part of the key gener-\nation must be done either by proxy person or by central hospital authority. Clinicians\nwho are supposed to participate in the key generation are responsible (principal) clin-\nician and referring clinician. Therefore we assume that minimum three entities must\nparticipate in the generation of patient’s key.\nOne approach which we adopted for patient’s key generation is the concept of\nsecret sharing with threshold. Secret is divided into n parts called shadows and in order\nto recover it, m shadows are needed. This idea was first independently proposed in\n[28] and [5]. It was further elaborated in [3, 18] and nice overview of the work in this\narea is given in [29].\nPriority among the users can be modeled by giving important user more shadows.\nFor example, for emergency cases central hospital authority together with responsible\n"
        },
        {
            "page_number": 331,
            "text": "WIRELESS NETWORK SECURITY\n331\n(principal) clinician should be able to reconstruct the patient’s key. The basic math-\nematical idea behind the key generation among m entities is to create the system of\nm equations with m variables by using the polynomial with random coefficients. For\nexample for m = 3 we start from the polynomial:\nF(x) = (ax2 + bx + Kp) mod p\nwhere p is public random prime number, a, b < p are secret random numbers and Kp\nis patient’s symmetric key. Assume that each participant j in key generation has some\nnumerical representation of his/her identity IDj. Then the shadows become\nF(IDpt)\n=\n(aID2\npt + bIDpt + Kp) mod p\npatient’s shadow\nF(IDpc)\n=\n(aID2\npc + bIDpc + Kp) mod p\nprincipal clinician’s shadow\nF(IDrc)\n=\n(aID2\nrc + bIDrc + Kp) mod p\nreferring clinician’s shadow\nF(IDca)\n=\n(aID2\nca + bIDca + Kp) mod p\ncentral authority’s shadow\nTo generate the patient’s key and start the measurement of health variables, three\nshadows are needed and must be presented to PSP. For the start of measurement, pa-\ntient’s shadow, principal clinician’s shadow and central authority’s shadow are suffi-\ncient. Three shadows are also needed in order to decrypt the medical record from the\nmedical database which is also encrypted with Kp. In this case, central authority should\nbe excluded and key should be recovered from patient’s shadow, principal clinician’s\nshadow and referring clinician’s shadow. In this case, patient will be always notified\nwhen his/her record is accessed and he/she will be sure that record is not changed.\nShadows should be changed frequently.\n4.2. Timestampingthesensedrecordsasresultsofpatient’sexamination\nThe fourth access rule calls for recording of all accesses for the purpose of auditing.\nAuditing requires that accesses are recorded together with the date, time and name of\neach person who accessed the record. This problem can be solved by linking current\nrecord of access (timestamp, list of persons involved) with previous records as proposed\nin [19, 20, 21, 29, 4]. It is also facilitated by the fact the central medical database can\nbe associated with the trusted timestamping server. Server builds a tree of hashes of\ntimestamping requests received for given time period (second, minute). Server further\nsends to the medical database signed hashes from the leaf generated by the opening\nof patient’s record till the root of the tree. Assume that information about patient’s i\nrecord is n-th leaf in the tree counting from the root and it has format:\nRIDpt,n = Tn, Ln, Kp, IDpc, IDrc.\nwhere Ln denotes the record lifetime. Let us denote Hn = H(RIDpt,n). Let us also\nassume that timestamping server has public/private key pair Kt and that encryption\n"
        },
        {
            "page_number": 332,
            "text": "332\nJELENA MI ˇSI ´C and VOJISLAV B. MI ˇSI ´C\nwith public and private key is denoted as VKt and SKt respectively. Then timestamping\nserver will associate information about access to the patient’s record with:\nSKt(H(H0, H1, H2...Hn)).\nwhere H0 represents the hash of the information at the root of the tree and Hi are hashes\nof the access information along the path to the root of the tree.\nTimestamping is also related to the deletion principle which states how long pa-\ntient’s medical record must be kept before deletion. The lifetime of the patient’s exam-\nination record Ln which is entered into medical database must be also protected using\nthe timestamping service. Patient’s record lifetime can be determined staring from the\nmoment when record is generated. If the particular record is missing but its hash exists\nin the timestamping tree, the integrity of the patient’s record is corrupted.\n4.3. Enforcement of the confinement principle\nPatient must be informed when clinician non-familiar to his/her medical record\naccesses the record. On the other hand responsible clinician must be able to add other\nclinicians to the access list. In that case the number of secret shadows has to change\n(increase) and central clinical authority has to increase the number random parameters\nin the equation which determines secret shadows. For example, if second clinician has\nto be added to the access list, the system of secret shadow equations becomes:\nF(IDpt) = (aID3\npt + bID2\npt + cIDpt + Kp) mod p patient’s shadow\nF(IDpc) = (aID3\npc + bID2\npc + cIDpc + Kp) mod p principal clinician’s shadow\nF(IDsc) = (aID3\nsc + bID2\nsc + cIDsc + Kp) mod p second clinician’s shadow\nF(IDrc) = (aID3\nrc + bID2\nrc + bIDrc + Kp) mod p referring clinician’s shadow\nF(IDca) = (aID3\nca + bID2\nca + bIDca + Kp) mod p central authority’s shadow\nIn this case four out of five shadows are needed to generate or access the patient’s\nexamination record so this presents (4,5)-threshold scheme.\n4.4. Enforcement of the aggregation principle\nAggregation of patients’ records must be prevented in the case the principal/second\nclinician becomes corrupted. This is mostly prevented by sharing the secret encryption\nkey through the shadows. Another helpful thing would be to encrypt the database\nrecords [11, 10]. The index filed can be the hash of last name of the patient concatenated\nwith his/her ID number. Data fields must be encrypted by the secret key assembled\nfrom m secret shadows. In this way, the list of the patients is hidden as well as their\nmedical records.\n"
        },
        {
            "page_number": 333,
            "text": "WIRELESS NETWORK SECURITY\n333\n2400MHz\n2483.5MHz\n802.11\n11 channels\n@ 22MHz\n802.15.1\n79 channels\n@1MHz\n802.15.4\n16 channels\n@2MHz\n2412+5k (k=0..10)\n2402+k (k=0..78)\n2405+5k (k=0..15)\nFigure 2. Spectrum usage for various WPAN technologies running in ISM\n5.\nIMPACT OF THE WIRELESS PAN TECHNOLOGIES\nWe plan to evaluate current WPAN standards namely, IEEE 802.15.1, and 802.15.4\nand their interworking among themselves and with IEEE 802.11b WLANs as major\ncandidates for implementations of healthcare sensor networks. We agree with [7] that\nthe success of wireless sensor networks as a technology rests on the success of the\nstandardization efforts to unify the market and avoiding the proliferation of proprietary,\nincompatibleprotocolsthat, although, perhapsoptimalintheirindividualmarketniches,\nwill limit the size of overall wireless sensor market.\n5.1. Classification of healthcare applications and pairing with WPAN\ntechnologies\nWe will analyze a number of healthcare applications from the aspects of bandwidth\nand delay. For example electrical signals from the heart are sampled at the rate of 500\nsamples per second and each sample is digitized to 8 bits giving data flow of 4000bps.\nFurthermore, samples must be taken from several points on the body. Each flow can not\nbedelayedmorethanfewhundredsofmillisecondsandflowsmustbesynchronized. We\nwill look at the following issues which are of foremost importance for sensor networks\nand which follow from the requirement for controlled event detection reliability at the\nnetwork sink and use them as criteria to match the technology with the application.\n1. How much is physical layer immune to the interference errors? We note that\nall candidate technologies run in Industrial Scientific and Medical (ISM) band\nbetween 2400 and 2483.5MHz. They use different modulations at the physical\nlayer, for example, the networks compliant with the 802.15.4 and 802.11 use\nDirect Sequence Spread Spectrum (DSSS), while those compliant with the\n"
        },
        {
            "page_number": 334,
            "text": "334\nJELENA MI ˇSI ´C and VOJISLAV B. MI ˇSI ´C\n802.15.1 (Bluetooth) use Frequency Hopping Spread Spectrum.\nTherefore\ndynamic channel allocation algorithms and interference mitigation techniques\nwill be needed to avoid excessive interference at the physical layer. Some work\non interference mitigation between 802.151 and 802.11b is reported in [14] but\nmuch more is needed for the interworking with 802.15.4. The channel layout\nfor all the technologies under consideration is given in Fig. 2.\n2. The MACs for candidate technologies can be classified as TDMA with polling\nand CSMA-CA. Node access delay has to be evaluated for both MAC classes\nunder varying number of nodes and packet rate from node. Is acknowledged\ntransfer necessary for achieving desired event reliability and which packet spac-\ning it induces? How much of the buffering is reasonable to have at the source\nnodes? For specific MAC, maximum effective bandwidth left to the application\nhas to be evaluated and paired with the delay.\n5.2. Design and evaluation of interconnection devices\nThere will be a need to interconnect different WPAN and to interconnect\nWPAN/WLANnetworksinordertoregulatethescaleofpower, distance, andbandwidth-\nrelated issues. The example of location of interconnecting devices (bridges) is given in\nFig. 1.\nThese devices have to be designed in the scope of MAC, channel and buffering\nissues and their performance has to be evaluated. The operation of interconnection\ndevice is very important for the overall network design since it affects the end-to-\nend delay and the scalability of the overall design. Some work in this area exists for\ninterconnection of Bluetooth piconets. The work in [32, 31] requires computation of\nnon-overlapping rendezvous points for bridges while the work in [25] allows bridge to\nvisit the piconet at will trading delay for scalability. However, little is known about\ninterconnections between IEEE 802.15.4 with other WPANs and WLANs.\n5.3. Reliable event detection\nThe rate at which data is propagated from source nodes at patient’s body to mon-\nitoring devices at the patient’s bed and monitoring room (sink) must be high enough\nto obtain the desired event detection reliability R, which is commonly defined as the\nnumber of data packets required per second for reliable event detection at the sink [1].\nAt the same time, sensor nodes operate on battery power which means that energy\nefficiency must be maintained.\nReliable event detection using minimal energy resources requires simultaneous\nachievement of several sub-goals. First, packet loss along the path from source to\nthe sink has to be minimized; at the Physical (PHY) layer, packets can be lost due to\nnoise and interference, while at the Medium Access Control layer (MAC) layer, losses\nmay be incurred by collisions. Second, packet waiting has to be minimized, including\nqueueing delays experienced in various devices along the data path towards the sink,\nbut also delays due to potential congestion in the network. Congestion control has to be\n"
        },
        {
            "page_number": 335,
            "text": "WIRELESS NETWORK SECURITY\n335\naddressed as a cross-layer problem and solved at the MAC level since excessive active\nnodes have to be turned off [23]. Finally, packet propagation should take place along\nthe shortest paths, while avoiding congested nodes and paths; this is the responsibility\nof the network layer.\nGiven that the protocol stack on sensor nodes—which operate on battery power\nand have limited computational capabilities—has to be as simple as possible, we con-\nclude that simultaneous minimization of packet losses and improvement in efficiency\n(with the goal of maximizing the lifetime of the network) necessitate that some of the\naforementioned functions of different layers are performed together. In other words,\ncross-layer optimization of network protocol operation is needed; the feasibility of this\noptimization is determined by the communication technology used to implement the\nnetwork. This problem has classically been treated only as a graph-theoretical problem\nwhere only connectivity has been addressed [30, 15] or addresses the collision based\nMAC [8] although it does not allow the active node to get into the sleep state and achieve\nload balancing among the nodes. Second group of proposals [17, 12] tries to regulate\nthe event sensing reliability but without looking at MAC and PHY properties at all. The\ncongestion problem is particularly important in networks which use a collision-based\nMAC protocol such as CSMA-CA, e.g., in 802.15.4 [16]. The decrease in through-\nput due to congestion may lead the coordinator to the erroneous conclusion that the\nnumber of active nodes is too low. Therefore it is important to look at the cross-layer\nimplementation of power/congestion control in wireless sensor networks.\n5.4. Handling patient’s mobility\nThe patient wearing wireless sensors will either walk within the hospital or he will\nlie in his bed while the bed is moved to another room. Therefore, sensed data will have\nto be sent to new access points and experience new level of interference and congestion.\nThe handover procedure can be handled at the MAC layer or at the networks layer. The\nhandover between 802.11 MACs is analyzed in [22]. However there is an open issue\nabout how the handover between 802.15.4 and 802.11 or between 802.15.1 and 802.11\nhas to be executed, and how much of data flow interruption will occur. We plan to\ndesign and analyze secure MAC layer handover procedures between involved WPAN\nand WLAN technologies.\nBesides MAC layer handover, it can happen that network layer handover is also\nneeded if the IP subnets covering access points have changed. Once when MAC layer\nhandover is finished, the mobile node (bridge on the patient’s bed) has to discover the\nnetwork layer information on the link, i.e. the new care-of-address router and network\nprefix. Foreign routers periodically advertise this in Router Advertisement using mobile\nIPv6. When mobile node learns the new care-of-address it registers this address with\nits home agent. We have to model and evaluate the acceptability of latencies and packet\nlosses during secure network layer handover.\n"
        },
        {
            "page_number": 336,
            "text": "336\nJELENA MI ˇSI ´C and VOJISLAV B. MI ˇSI ´C\n6.\nCOMPARISON BETWEEN TWO TECHNOLOGIES REGARDING\nTHE DEPLOYMENT IN SENSOR NETWORKS\nAfter individual descriptions of IEEE 802.15.1 and IEEE 802.15.4 we will give\ndirect comparison of their properties against the criterion of feasibility of their deploy-\nment in sensor networks.\n6.1. How much is physical layer immune to the noise errors\nBoth standards, 802.15.1 and 802.15.4 with 250kbps rate, operate in 2450MHz\nband known as Industrial, Scientific and Medical – ISM. This band is already hosting\nwireless LAN/PAN standards such as 802.11b and 802.15.1 (Bluetooth) and a lot of\ninterference is expected. It is also worth mentioning that Bluetooth packets can be 1, 3\nor 5 slots long which results in payload sizes of 17, 121, 224 bytes for DM 1, 3 and 5\npacket types respectively with Forward Error Correction (FEC) or in payload sizes of\n27, 183 and 339 bytes for DH-type , 1, 3 and 5 packet types without FEC. On the other\nhand, 802.15.4 does not have FEC and allows maximum packet size of 127 bytes. This\npacket size includes all headers from physical and MAC layer which minimum size is\n15 bytes giving the actual maximum payload size of 112 bytes. Therefore, it makes\nsense to compare these two technologies only in the case of payload size of 27 bytes\n(DH1).\nAsmentioned, BluetoothusesFHSSandisveryresilienttointerference. According\nto the exhaustive simulation results reported in [33] when 10 fully loaded piconets each\nwith 7 slaves are placed in the room with dimensions 10m x 20m, (and interfere with\neach other) packet error rate for DH1 packets was 0.03. When the same experiment\nwas repeated with 100 co-located piconets, packet error rate was 0.3.\nIEEE 802.15.4 standard in the 2450 MHz range (ISM band) uses 16-ary quasi-\northogonal modulation technique. Four data bits represent one modulation symbol and\nthat symbol is further encoded into 32 bit chip sequence. There are 16 nearly-orthogonal\nPseudo-Noise chip sequences. Each chip sequence is modulated onto the carrier using\noffset quadrature phase shift keying (O-QPSK). Since the chip rate is 2Mcps and raw\ndata rate is 250kbps the maximum supported ratio of bit energy to the noise power\nspectral density of Eb\nN0 = 8. According to the properties of QPSK, the Bit Error Rate\nis determined using known expression given for example in [13]. Therefore, without\nthe interference, we should expect BER slightly less than 10−4. This is confirmed in\nthe section 6.1.6 of the standard where Packet Error Rate (PER) of 1% is expected on\npackets which have 20 bytes including MAC and physical level headers. However, in\nthe presence of interference in the ISM band, it is more realistic to expect BER around\n10−3 and Packet Error Rate more than 28% for packets with 27 bytes of payload and\n15 bytes of headers. (Packet Error Rate can be calculated as PER = 1−(1−BER)X\nwhere X is packet length including MAC and physical layer header expressed in bits).\nAlthough, Z¨urbes’ experiment can not be directly translated into BER, 10 co-\nlocated piconets present interference probably much larger than what the physical layer\nof 802.15.4 can handle.\n"
        },
        {
            "page_number": 337,
            "text": "WIRELESS NETWORK SECURITY\n337\n6.2. The access delay\nBluetooth has a polling based MAC protocol and its access delay depends on the\norder in which master polls the slaves and on the amount of packets which are exchanged\nbetween master and slave in one visit. Mathematically speaking, packet service time\ndirectly depends on the piconet cycle time i.e. the time needed for the master to visit\neach slave. It has been shown [26] that under low traffic exhaustive scheduling (where\nmaster exchanges packets with slave as long as one of them has packets in the queue)\noffers the lowest access delay compared with other limited round-robin polices where\nmaster can exchange at most M packets per one polling cycle. However, under high\nloads exhaustive scheduling is not the best one compared to limited polices and fairness\nissue raises since one station can keep the master busy for a long period of time. Under\nlimited policies every station has equal amount of bandwidth and piconet cycle time is\nlimited. Therefore, if one or more slaves have excessive traffic their packets will suffer\nfrom the large delay, but the other slaves with lower traffic will not.\n6.3. Can wireless sensor network reach the regime when delays are\nunacceptable?\nBluetooth piconet can reach such regime only if duration of piconet cycle becomes\nextremely long and this can happen only under exhaustive scheduling of slaves. This\ncan represent also a security problem, since one malicious node can bring the whole\npiconet down.\nIEEE 802.15.4 network can reach this saturation regime if the number of nodes\nand packet arrival rates exceeds certain limits. For example, for packet size of 30 bytes\n(including PHY and MAC headers) saturation is reached with 30 nodes each having\npacket arrival rate of 3 packets per second (total of 45 bytes per second). Under packet\nsize of 90 bytes, saturation is reached with 15 nodes with packet arrival rate of 3 packets\nper second. Saturation also can represent a security problem since a couple of malicious\nnodes can quickly bring the network down as shown in [24].\n6.4. How much of the buffering is reasonable to have at the source\nnodes?\nAssuming thattheentiremeasurementcanfitinasinglepacket, transmittingseveral\npackets from the node buffer means that some slightly older information is sent. Also,\nthe inter-packet time will be less than in the case where each node sends one packet\nonly. Therefore, exhaustive scheduling of active sensor’s periods with large buffers\nincreases spatial and temporal correlation of sensed data. This fact is important in\nthe applications where controlled reliability means controlled inter-packet spacing or\nin applications with security concerns where mal-functioning node with exhaustive\nscheduling can inject large amount of bogus data into the network. Therefore, buffer\nsizes at the nodes should not exceed several packet sizes.\n"
        },
        {
            "page_number": 338,
            "text": "338\nJELENA MI ˇSI ´C and VOJISLAV B. MI ˇSI ´C\n6.5. Whatistheeffectivebandwidthlefttotheapplication, i.e. whatisthemaximum\npossible event detection reliability for particular MAC?\nThe main concern in sensing applications of Bluetooth is that the major part of\nthe traffic is targeted toward the network coordinator, i.e., in the uplink direction.\nBecause of the Bluetooth polling mechanism, the downlink packet slots—which are\nstill necessary—will be empty. This wastes the bandwidth and limits the maximum\nthroughput of the network to something of the order of 723kbps out of 1Mbps in\nBluetooth version 1.2 with DH5 packets. The recent Enhanced Data Rate option in\nBluetooth version 2.0 allows for maximum data rates over 2Mbps [6], however the\nsame conceptual problem remains.\nOn the other hand, the 802.15.4 standard allows for maximum raw data rate of\n250kbps, i.e. about one-quarter of that obtainable under Bluetooth version 1.2. Due to\nthe random backoff countdown procedure and the need to listen to the medium before\nattempting transmission, the traffic intensity of one node affects the activities of the\nothers. Under large traffic volume (which may be expected when the network has\nmany nodes), there will be many collisions and many deferred transmissions. This\nusually results in severe congestion and all nodes experience large delays. In severe\ncases when saturation occurs, the network throughput drops to few percent of the raw\ndata rate. Since the backoff window can not exceed value of 31 and the packet size\nis limited to 127 bytes, an 802.15.4 network can easily reach saturation regime. Our\nresults show that the highest throughput of around 25% of the theoretical maximum\noccurs for packet size of around 90 bytes (including PHY and MAC headers), with five\nactive stations in the network (we did not check for networks with smaller number of\nstations), and under the superframe size of 48 backoff periods. This puts a limit of\neffective data rate of 62.5 kbps per cluster, or around 12.5 kbps per node. However,\nwith fifteen active nodes the total throughput drops to only about 18%, and this drop\ncontinues in proportion with the increase of the cluster size.\n7.\nSUMMARY\nIn this chapter we have addressed security and networking architecture of the clin-\nical information systems with emphasis on the wireless hop. Wireless hop includes\nsensor networks and possibly wireless local area or mesh networks. We have reviewed\nconfidentiality and integrity polices for clinical information systems and proposed the\npolicy enforcement mechanisms which cover the wireless hop. We have also com-\npared two candidate technologies: IEEE 802.15.1 (also known as Bluetooth) and IEEE\n802.15.4, from the aspect of resilience of MAC and physical layers to the jamming and\ndenial-of-service attacks.\n8.\nREFERENCES\n1. O. B. Akan and I. F. Akyildiz. ESRT: Event-to-Sink Reliable Transport in Wireless Sensor Networks.\nIn IEEE/ACM Transaction on Networking (to appear), October 2005.\n2. R. Anderson. A security policy model for clinical information systems. In Proc. of the 1996 IEEE\n"
        },
        {
            "page_number": 339,
            "text": "WIRELESS NETWORK SECURITY\n339\nSymposium on Security and Privacy, pages 34–48, 1996.\n3. C. Asmuth and J. Bloom. A modular approach to key safeguarding. it, 29(2):208–210, 1979.\n4. M. Bishop. Computer Security – Art and Science. Pearson Education, Inc., Boston, MA 02116, 1st\nedition, 2003.\n5. G. R. Blakely. Safeguarding cryptographic keys. In Proceedings of the National Computer Conference,\nAmerican Federation of Information Processing Societies, volume 48, pages 313–317, 1979.\n6. Bluetooth SIG. Draft Specification of the Bluetooth System. Version 2.0, Nov. 2004.\n7. E. H. Callaway, Jr. Wireless Sensor Networks, Architecture and Protocols. Auerbach Publications,\nBoca Raton, FL, 2004.\n8. A. Cerpa and D. Estrin. Adaptive self-configuring sensor network topologies. In Proceedings Twenty-\nFirst Annual Joint Conference of the IEEE Computer and Communications Societies IEEE INFOCOM\n2002, volume 3, pages 1278–1287, New York, NY, June 2002.\n9. O. Elkeelany, M. M. Matalgah, K. P. Sheikh, M. Thaker, G. Choudry, D. Medhi, and J. Qaddour.\nPerformance analysis of IPSec protocol: Encryption and authentication.\nIn Proceedings of IEEE\nInternational Conference on Communications ICC 2002, pages 1164–1168, 2002.\n10. J. Feigenbaum, M. Liberman, and E. Grosse. Cryptographic protection of membership lists. Newsletter\nof the International Association of Cryptologic Research, 9:16–20, 1992.\n11. J. Feigenbaum, M. Liberman, and R. N. Wright. Cryptographic protection of databases and software.\nDistributed Computing and Cryptography, J. Feigenbaum and M. Merritt eds., pages 161–172, 1991.\n12. J. Frolik. QoS control for random access wireless sensor networks. In Proc. WCNC 2004, Atlanta, GA,\nMar. 2004.\n13. V. K. Garg, K. Smolik, and J. E. Wilkes. Applications of CDMA in Wireless/Personal Communications.\nPrentice Hall, Upper Saddle River, NJ, 1998.\n14. N. Golmie. Bluetooth dynamic scheduling and interference mitigation. ACM/Kluwer Journal on Special\nTopics in Mobile Networking and Applications (MONET), 9(1):21–31, 2004.\n15. H. Gupta, S. Das, and Q. Gu. Connected sensor cover: self organization of sensor networks for efficient\nquery execution. In Proceedings 2003 ACM International Symposium on Mobile ad hoc networking &\ncomputing, volume 1, pages 189–200, Annapolis, MD, June 2003.\n16. Standard for part 15.4: Wireless MAC and PHY specifications for low rate WPAN. IEEE Std 802.15.4,\nIEEE, New York, NY, Oct. 2003.\n17. R. Iyer and L. Kleinrock. QoS control for sensor networks. In Proc. ICC’03, volume 1, pages 517–521,\nAnchorage, AK, May 2003.\n18. E. D. Karnin, J. W. Greene, and M. E. Hellman. On sharing secret systems. it, 29(2):35–41, 1983.\n19. R. Merkle. Method of providing digital signatures, Jan. 1982.\n20. R. Merkle. A digital signature based on a conventional encryption function. In Proceedings of the\nAdvances in Cryptology - CRYPTO’87, pages 369–378, 1988.\n21. R. Merkle. A certified digital signature. In Proceedings of the Advances in Cryptology - CRYPTO ’88,\npages 218–238, 1990.\n22. A. Mishra, M. Shin, and W. Arbaugh. An empirical analysis of the IEEE 802.11 MAC layer handoff\nprocess. SIGCOMM Comput. Commun., 33(3):93–102, 2003.\n23. J. Miˇsi´c, G. R. Reddy, and V. B. Miˇsi´c. Activity Scheduling based on cross layer information in\nBluetooth sensor networks. Computer Communications, to appear, 2006.\n24. V. B. Miˇsi´c, J. Fung, and J. Miˇsi´c. Mac layer security of 802.15.4-compliant networks. In Proc.\nWSNS’05, held in conjunction with IEEE MASS05 2005, Washington, DC, Dec. 2005.\n"
        },
        {
            "page_number": 340,
            "text": "340\nJELENA MI ˇSI ´C and VOJISLAV B. MI ˇSI ´C\n25. V. B. Miˇsi´c, J. Miˇsi´c, and K. L. Chan. Walk-in scheduling in Bluetooth scatternets. Cluster Computing,\n8(2/3):197–210, 2005.\n26. Miˇsi´c, J. and Miˇsi´c, V. B. Performance Modeling and Analysis of Bluetooth Networks: Network\nFormation, Polling, Scheduling, and Traffic Control. Boca Raton, FL: CRC Press, July 2005.\n27. National Institute of Standards. Digital Signature Standard. US Department of Commerce, 1994.\n28. A. Samir. How to share a secret. IEEE Computer, 22(11):612–613, 1979.\n29. B. Schneier. Applied Cryptography. John Wiley & Sons, Inc., New York, N.Y., 2nd edition, 1996.\n30. C. Schurgers, V. Tsiatis, S. Ganeriwal, and M. Srivastava. Topology management for sensor networks:\nexploiting latency and density. In Proceedings 2002 ACM International Symposium on Mobile ad hoc\nnetworking & computing, volume 1, pages 135–145, Lausanne, Switzerland, June 2002.\n31. G. Tan and J. Guttag. A locally coordinated scatternet scheduling algorithm. In Proceedings of the 26th\nAnnual Conference on Local Computer Networks LCN 2002, pages 293–303, Tampa, FL, Nov. 2002.\n32. W. Zhang and G. Cao. A flexible scatternet-wide scheduling algorithm for Bluetooth networks. In\nProc. 21st IEEE International Performance, Computing, and Communications Conference IPCCC\n2002, Phoenix, AZ, Apr. 2002.\n33. S. Z¨urbes. Considerations on link and system throughput of Bluetooth networks. In Proceedings of the\n11th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications PIMRC\n2000, volume 2, pages 1315–1319, London, UK, Sept. 2000.\n"
        },
        {
            "page_number": 341,
            "text": "14\nKEY MANAGEMENT SCHEMES\nIN SENSOR NETWORKS\nVenkata Krishna Rayi\nDepartment of Computer Science\nThe University of Memphis\nMemphis, TN 38152 USA\nYang Xiao\nComputer Science Department\nUniversity of Alabama\n101 Houser Hall\nBox 870290\nTuscaloosa, AL 35487-0290 USA\nE-mail: yangxiao@ieee.org\nBo Sun\nDepartment of Computer Science\nLamar University\nBeaumont, TX 77710 USA\nE-mail: bsun@cs.lamar.edu\nXiaojiang (James) Du\nDepartment of Computer Science\nNorth Dakota State University\nFargo, ND 58105 USA\nE-mail: Xiaojiang.Du@ndsu.edu\nFei Hu\nComputer Engineering Department\nRochester Institute of Technology\nRochester, NY 14623 USA\nE-mail: fxheec@rit.edu\n"
        },
        {
            "page_number": 342,
            "text": "342\nVENKATA KRISHNA RAYI et al.\nIn the near future, sensor networks are going to be a part of everyday life. Traffic moni-\ntoring, military tracking, building safety, pollution monitoring, wildlife monitoring, patient\nsecurity are some of the applications in sensor networks. Sensor networks vary in size and\ncan consist of 10 to 1,000,000 sensor nodes. They can be deployed in a wide variety of\nareas, including hostile environments, demanding secure measures for data transfer. Sensor\nnodes used to form these networks are resource-constrained, which makes these types of\nsecurity applications a challenging problem. A basic technique to protect data is encryption;\nbut, due to resource constraints, achieving necessary key agreement for encryption is not\neasy. Many key establishment techniques have been designed to address this challenge, but\nwhich scheme is the most effective is still debatable. Our work aimed to generate a brief\nknowledge about different key management schemes and their effectiveness. We noticed\nthat no key distribution technique is ideal to all the scenarios where sensor networks are\nused; therefore the techniques employed must depend upon the requirements and resources\nof each individual sensor network.\n1.\nINTRODUCTION\nDistributed Sensor Networks (DSNs) are going to be widely used in the near fu-\nture due to their breadth of applications by military, exploration teams, researchers, and\nso on. It is not possible to use general wireless techniques for DSNs since they are\nresource-constrained and security measures are required. Distribution techniques that\nare applicable employ assorted key management methods, such as public key cryptog-\nraphy, and require numerous communication and computation capabilities. Therefore,\nit is important to examine the different requirements, constraints and evaluation metrics\nof sensor networks as well as single network-wide key scheme, which is the simplest of\nkey management techniques, before discussing the various germane key management\ntechniques.\n1.1. Requirements\nSensor networks must arrange several types of data packets, including packets of\nrouting protocols and packets of key management protocols. The key establishment\ntechnique employed in a given sensor network should meet several requirements to\nbe efficient. These requirements may include supporting in-network processing and\nfacilitating self-organization of data, among others. However, the key establishment\ntechnique for an secure application must minimally incorporate authenticity, confiden-\ntiality, integrity, scalability, and flexibility.\nAuthenticity: The key establishment technique should guarantee that the com-\nmunication nodes in the network have a way for verifying the authenticity of\nthe other nodes involved in a communication, i.e., the receiver node should\nrecognize the assigned ID of the sender node.\nConfidentiality: The key establishment technique should protect the disclosure\nof data from unauthorized parties. An adversary may try to attack a sensor\nnetwork by acquiring the secret keys to obtain data. A better key technique\ncontrols the compromised nodes to keep data from being further revealed.\n"
        },
        {
            "page_number": 343,
            "text": "WIRELESS NETWORK SECURITY\n343\nIntegrity: Integrity means no data falsification during transmissions. Here in\nterms of key establishment techniques, the meanings are explained as follows.\nOnly the nodes in the network should have access to the keys and only an\nassignedbasestationshouldprivilegetochangethekeys. Thiswouldeffectively\nprevent unauthorized nodes from obtaining knowledge about the keys used and\npreclude updates from external sources.\nScalability: Efficiency demands that sensor networks utilize a scalable key\nestablishment technique to allow for the variations in size typical of such a\nnetwork. Key establishment techniques employed should provide high-security\nfeatures for small networks, but also maintain these characteristics when applied\nto larger ones.\nFlexibility: Key establishment techniques should be able to function well in\nany kind of environments and support dynamic deployment of nodes, i.e., a key\nestablishment technique should be useful in multiple applications and allow for\nadding nodes at any time.\n1.2. Constraints\nOne of the challenges in developing sensor networks is to provide high-security\nfeatures with limited resources. Sensor networks cannot be costly made as there is\nalways a great chance that they will be deployed in hostile environments and captured\nfor key information or simply destroyed by an adversary, which, in turn, can cause huge\nlosses. Part of these cost limitation constraints includes an inability to make sensor\nnetworks totally tamper-proof. Other sensor node constraints that must be kept in mind\nwhile developing a key establishment technique include battery life, transmission range,\nbandwidth, memory, and prior deployment knowledge.\nBattery Life: Sensor nodes have a limited battery life, which can make using\nasymmetric key techniques, like public key cryptography, impractical as they\nuse much more energy for their integral complex mathematical calculations.\nThis constraint is mitigated by making use of more efficient symmetric tech-\nniques that involve fewer computational procedures and require less energy to\nfunction.\nTransmission Range: Limited energy supply also restricts transmission range.\nSensor nodes can only transmit messages up to specified short distances since\nincreasing the range may lead to power drain. Techniques like in-network\nprocessing can help to achieve better performance by aggregating and trans-\nmitting only processed information by only a few nodes, thereby saving the\ndissipated energy.\nBandwidth: It is not efficient to transfer large blocks of data with the limited\nbandwidth capacity of typical sensor nodes, such as the transmitter of the UC\nBerkeley Mica platform that only has a bandwidth of 10Kbps. To compensate,\n"
        },
        {
            "page_number": 344,
            "text": "344\nVENKATA KRISHNA RAYI et al.\nkey establishment techniques should only allow small chunks of data to be\ntransferred at a time.\nMemory: Memory availability of sensor nodes is usually 6-8Kbps, half of\nwhich is occupied by a typical sensor network operating system, like TinyOS.\nKey establishment techniques must use the remaining limited storage space\nefficiently by storing keys in memory, buffering stored messages, etc.\nPrior Deployment Knowledge: As the nodes in sensor networks are deployed\nrandomly and dynamically, it is not possible to maintain knowledge of every\nplacement. A key establishment technique should not, therefore, be aware of\nwhere nodes are deployed when initializing keys in the network.\n1.3. Evaluation Metrics\nA key establishment technique is not judged solely based upon its ability to provide\nsecrecy of transferred messages, but must also meet certain other criteria for efficiency\nin light of vulnerability to adversaries, including the three Rs of sensor networks: resis-\ntance, revocation, and resilience. Though scalability may be considered an evaluation\nmetric, it is not discussed here since we have included it in DSN requirements.\nResistance: An adversary might attack the network by compromising a few\nnodes in the network and then replicating those nodes back into the network.\nUsing this attack the adversary can populate the whole network with his repli-\ncated nodes and thereby gain control of the entire network. A key establishment\ntechnique must resist node replication to guard against such attacks.\nRevocation: If a sensor network become invaded by an adversary, the key\nestablishment technique should provide an efficient way to revoke compromised\nnodes, a lightweight method that does not use much of the network’s already\nlimited capacity for communication.\nResilience: If a node within a sensor network is captured, the key establish-\nment technique should ensure that secret information about other nodes is not\nrevealed. A scheme’s resilience is calculated using the total number of nodes\ncompromised and the total fraction of communications compromised in the\nnetwork. Resilience also means conveniently making new inserted sensors to\njoin secure communications.\n1.4. Single Network-Wide Key\nUsing a single network-wide key is by far the simplest key establishment technique.\nIn the initialization phase of this technique, a single key is preloaded into all the nodes of\nthenetwork. Afterdeployment, everynodeinthenetworkcanusethiskeytoencryptand\ndecrypt messages. Some of the advantages offered by this technique include minimal\nstorage requirements and avoidance of complex protocols. Only a single key is to be\n"
        },
        {
            "page_number": 345,
            "text": "WIRELESS NETWORK SECURITY\n345\nstored in the nodes’ memory and once deployed in the network, there is no need for a\nnode to perform key discovery or key exchange since all the nodes in communication\nrange can transfer messages using the key which they already share.\nThough a single network-wide key may seem advantageous, the main drawback is\nthat compromise of a single node causes the compromise of the entire network through\nthe shared key. This scheme counters several constraints with less computation and\nreduced memory use, but it fails in providing the basic requirements of a sensor network\nby making it easy for an adversary trying to attack.\n1.5. Organization of of this chapter\nThe key establishment technique employed in a given sensor network should take\ninto consideration all the requirements, constraints, and evaluation metrics discussed.\nIn our work we have assessed different types of key establishment techniques, each\nranging in efficiency by providing various necessary characteristics. In Section 3 of\nthis chapter we explain the Basic Scheme for key establishment in sensor networks\n[1]; Section 4 discusses three more-efficient schemes, two of which are extensions of\nthe Basic Scheme, the Q-Composite Scheme [2] and the Multipath Key Reinforcement\nScheme, and one of which is the Random Pairwise Scheme; Section 5 describes the\nPolynomial-Based Key Predistribution and two efficient instances of that scheme [3];\nSection 6 details SPINS and its two building blocks (SNEP and µTESLA) [5]; Section 7\nelaborates on LEAP and its implementation [6]; Section 8 clarifies a key management\nscheme using the deployment knowledge [7]; Section 9 concludes this essay with a\nbrief review of all the schemes evaluated.\n2.\nBASIC SCHEME\nIn this section we specify the Random Key Predistribution Scheme proposed by\nEschenauer and Gligor [1], which we refer to as the Basic Scheme. First we will\ndescribe the structure and features of the Basic Scheme and then how it may be eval-\nuated using two of the three Rs of efficient sensor networks, revocation and resilience\n(through rekeying). We will then analyze the pros and cons of the Basic Scheme for\nkey establishment in a sensor network.\n2.1. Key Distribution\nIn the Basic Scheme, key distribution is divided into three stages: key predistri-\nbution, shared-key discovery, and path-key establishment. In the key predistribution\nstage, a large key pool of |S| keys and their identifiers are generated. From this key\npool, K keys are randomly drawn and pre-distributed into each node’s key ring, in-\ncluding the identifiers of all those keys. At the point that each node has K keys and the\nidentifiers of those keys, trusted nodes in the network are selected as controller nodes,\nand all the key identifiers of a key ring and the associated sensor identifier on controller\nnodes are saved. Following this, the i-th contoller node is loaded for each node with\n"
        },
        {
            "page_number": 346,
            "text": "346\nVENKATA KRISHNA RAYI et al.\nthe key that is shared with that node. This key predistribution process ensures that,\nthough the size of the network is large, only a few keys need to be stored in each node’s\nmemory, thereby saving storage space. These few keys are enough to ensure that two\nnodes share a common key, based on a selected probability.\nAfter the key predistribution stage, we move to the shared-key discovery stage.\nOnce the nodes are initialized with keys, they are deployed in the respective places\nwhere they are needed, such as hospitals, war fields, etc. After deployment, each node\ntries to discover its neighbors with which it shares common keys. There are many ways\nfor finding out whether two nodes share common keys or not. The simplest way is to\nmake the nodes broadcast their identifier’s list to other nodes. If a node finds out that it\nshares a common key with a particular node, it can use this key as the communication\nlink. This approach does not give the adversary any new attack opportunities and only\nleaves room for launching a traffic analysis attack in the absence of key identifiers.\nMore secure alternate methods exist for finding out the common keys shared between\ntwo nodes though. For example, for every key on a key ring, each node could broadcast\na list α, EKi(α), i = 1, ..., k where α is a challenge. The decryption of EKi(α) with\nthe proper key by a recipient would reveal the challenge α and establish a shared key\nwith the broadcasting node [1].\nA link exists between two nodes only if they share a key, but the path key establish-\nment stage facilitates provision of the link between two nodes when they do not share a\ncommon key. Let us suppose that node u wants to communicate with node v, but they\ndo not share a common key between them. Node u can send a message to node y saying\nthat it wants to communicate with v; this message is then encrypted using the common\nkey shared between u and y and, if node y has a key in common with v, it can generate\na pairwise key Kuv for nodes u and v, thereby acting like a key distribution center or\na mediator between the communication of nodes u and v. As all the communications\nare encrypted using their respective shared keys, there will not be a security breach in\nthis process. After the shared-key discovery stage is finished there will be a number\nof keys left in each sensor’s key ring that are unused and can be put to work by each\nsensor node for path key establishment.\nA compromised sensor node can cause a lot of damage to a network and therefore,\nrevocation of a compromised node is very important in any key distribution scheme. In\nthe Basic Scheme, node revocation is conducted by the controller node. When a node is\nrevoked, all the keys in that particular node key ring have to be deleted from the network.\nLet us assume that the controller node has knowledge about a compromised node in\nthe network and broadcasts a message to all the nodes in the network, the message\nwill include a list of the key identifiers of the compromised node’s key ring. To sign\nthe list of key identifiers, the controller node uses a signature Ke and then encrypts\nits message with Kci, which is the key that the controller node shares with the nodes\nduring the key predistribution stage. Once each node receives the message, it decrypts\nthe message using the key they already share with the controller node. When the\nsignature is verified, the nodes search their key rings for the list of identifiers provided\nin the message and, if there is any match, they are deleted from the key ring. After the\nmatching keys are completely deleted from all the nodes, there may be links missing\n"
        },
        {
            "page_number": 347,
            "text": "WIRELESS NETWORK SECURITY\n347\nbetween different ones and they then have to reconfigure themselves starting from the\nshared key discovery stage so that new links can be formed between them. As only few\nkeys are removed from the network, the revocation process only affects a part of it and\ndoes not include much communication overhead.\nThe keys used in a sensor network must be rekeyed to lessen the chance that\nan adversary may access all of the network keys when a few nodes and their keys are\ncaptured. Rekeying effectively increases a network’s resilience without incurring much\ncommunication and computation overhead.\n2.2. Analysis of the Basic Scheme\nLet us assume that the probability of a common key existing between two nodes in\nthe network is p, and the size of the network is n. The degree of a node d is derivable\nusing both p and n since the degree of any node is simply the average number of edges\nconnecting that node with other nodes in its neighborhood; therefore, d = p × (n −1).\nFirst we have to find the value of d such that a DSN of n nodes is connected with a\ngiven probability Pc. We then must calculate the key ring size k and the size of the key\npool | S |.\nAccording to Random Graph Theory, a random graph G(n, p) is a graph consisting\nof n nodes and p representing the probability of establishing a link between two nodes.\nErdos and Renyi [12] showed that there exists a probability state p, which moves from\nstate zero to state one for large random graphs. The function that defines p is called the\nthreshold function of a property. If we are given a desired probability (Pc) for graph\nconnectivity, then p is given as\nPc = lim\nn→∞Pr[G(n, p)is connected] = ee−c\n(1)\np = ln(n)\nn\n+ c\nn\n(2)\nwhere c is a real constant.\nThen, to calculate the key ring size k and the size of the key pool | S |, we need\nto first note that wireless constraints limit the number of nodes in a range to be smaller\nthan n, represented by the value n′. Now the probability of sharing a key between two\nneighbor nodes varies to p′ = d/(n′ −1), for a given d value. Also, p′ can be denoted\nas the difference between the total probability and the probability that two nodes do not\nshare a common key; i.e., p′ = 1 −Pr [two nodes do not share any key] and, thus,\np′ = 1 −\n(1 −k\np)2(p−k+ 1\n2 )\n(1 −2k\np )(p−2k+ 1\n2 )\n(3)\nwhere |S| is the size of the key pool and k is the key ring size.\nEschenauer and Gligor have shown that for a pool size S = 10, 000 keys, only 75\nkeys need to be stored in a node’s memory to have the probability that they share a key\nin their key rings be p = 0.5. If the pool size is ten times larger, i.e., S = 100, 000,\n"
        },
        {
            "page_number": 348,
            "text": "348\nVENKATA KRISHNA RAYI et al.\nthen the number of keys required is still only 250. Thus, the Basic Scheme is a key\nmanagement technique that is scalable, flexible and can also be used for large DSNs.\nTrade-offs in the Basic Scheme can be made between sensor memory and connectivity\nbut, it does not provide the node-to-node authentication property that ascertains the\nidentity of a node with which another node is communicating. This property is very\nuseful when revoking misbehaving nodes from the network and also helps in resisting\nthe node replication attack.\nMany key management schemes are proposed as extensions of the Basic Scheme\nto make it even more secure and reliable.\n3.\nEFFICIENT KEY ESTABLISHMENT TECHNIQUES\nChan, Perrig, and Song [2] have introduced three scheme variations that more\nefficiently perform key establishment and meet more requirements, constraints, and\nevaluation metrics of a DSN. These schemes include two expansions of the Basic\nScheme (Q-CompositeRandomKeyPredistributionandMultipathKeyReinforcement)\nand a variation of the commonly known Pairwise Scheme (Random Pairwise Key) that\nare efficient in a particular DSN environment. Each comes with a different kind of\ntrade-off and is not, therefore, widely applicable. The Q-Composite Scheme achieves\nsecurity under small scale attacks while being vulnerable under large scale attacks and is\nuseful in DSNs where large attacks are easily detected. The Multipath Reinforcement\nScheme offers good security with additional DSN communication overhead for use\nwhere security is more of a concern than bandwidth or power drain. Compared to the\nQ-Composite and the Multipath, the Random Pairwise Scheme offers the best security\nfeatures in its perfect resilience to node capture with the only drawback being limited\nscalability.\n3.1. Q-Composite Random Key Predistribution Scheme\nIn the Basic Scheme, two nodes share a unique key for establishing a communica-\ntion link. A given network’s resilience to node capture can be improved by increasing\nthe number of common keys that are needed for link establishment. The Q-Composite\nRandom Key Predistribution Scheme does just this by requiring that two nodes have at\nleast q common keys to set up a link [2]. As the amount of key overlap between two\nnodes is increased, it becomes harder for an adversary to break their communication\nlink. At the same time, to maintain the probability that two nodes establish a link with\nq common keys, it is necessary to reduce the size of the key pool | S |, which poses a\npossible security breach in the network as the adversary now has to compromise only\na few nodes to gain a large part of S. So the challenge of the Q-Composite Scheme is\nto choose an optimal value for q while ensuring that security is not sacrificed.\nIn the key predistribution stage of both the Basic and Q-Composite Schemes, k\nrandom keys are picked from S and initialized in each node’s key ring. In the shared-\nkey discovery phase, each node has to find the common keys which it shares with other\nnodes by either making all the nodes broadcast their key identifiers or by selecting a\n"
        },
        {
            "page_number": 349,
            "text": "WIRELESS NETWORK SECURITY\n349\nslower and more secure method of posing puzzles such as the Merkle Puzzle [18]. For\nthis puzzle method, each node issues m client puzzles to each neighboring node and\nany node that comes up with the correct solution to the puzzle is identified as sharing\nthe associated key. After this, the two schemes differ in that the Q-Composite Scheme\nrequires each node identify neighboring nodes with which they share at least q common\nkeys while the Basic Scheme only requires one shared key. This restriction in the Q-\nComposite Scheme allows the number of keys shared to be more than q but not less,\nrepresented by the value q. At this stage in the process, nodes will fail to establish\na link if the number of keys shared is less than q; otherwise, they will form a new\ncommunication link using the hash of all the q keys, i.e., K = hash(k1||k2||...||kq).\nS, the size of the key pool, is the critical parameter that must be calculated for the\nQ-Composite Scheme to be efficient. If S is large, then the probability is decreased\nthat two nodes share a common key and therefore can communicate. However, if S is\ndecreased, an adversary’s job may be easier as he can gather most of the keys in the key\npool by capturing only a few nodes. Thus, S must be chosen such that the probability\nof any two nodes sharing at least q keys is ≥p.\nChan, Perrig, and Song’s [2] method to calculate S is\np(i) =\n\f\n|S|\ni\n\r \f\n|S| −i\n2(m −i)\n\r \f\n2(m −i)\nm −i\n\r\n\f\n|S|\nm\n\r2\n(4)\nwhere p(i) is the probability that any two nodes have exactly i, which is the number of\nkeys in common; and m is the key ring capacity for a given node. There are\n\f\n|S|\ni\n\r\nways to pick i and | S | −i is the remaining keys in the key pool after i is picked. There\nare\n\f\n|S|\nm\n\r\ndifferent ways to pick m and\n\f\n|S|\nm\n\r2\ntotal number of ways for both\nnodes to pick m.\nAlso, to assign the remaining keys 2(m −i) distinct keys are picked from the key\npool for each node and the number of ways to do this is\n\f\n|s| −i\n2(m −i)\n\r\n. There are\n2(m −i) ways to partition the keys equally between the two nodes.\nLet Pc be the probability of any two nodes sharing sufficient keys to form a secure\nconnection. Therefore, Pc = 1−(the probability that the two nodes share insufficient\nkeys to form a connection) or\nPc = 1 −(p (0) + p (1) + .... + p (q −1))\n(5)\nNow the largest | S | such that Pc ≥p is chosen.\nThe evaluation of the Q-Composite Scheme can be done by verifying its resilience\nagainst node capture. Even though this scheme does not provide resistance against\nnode replication or a means for node-to-node authentication since the keys from the\nkey pool are used more than once, it does improve resilience against node capture when\n"
        },
        {
            "page_number": 350,
            "text": "350\nVENKATA KRISHNA RAYI et al.\nan adversary has successfully captured some other nodes in the DSN. As the same keys\nare used repeatedly in a network, a situation may arise in which two nodes effectively\nhave their communications exposed due to the compromise of another two nodes that\nshare the same key(s). Chan, Perrig, and Song [2] have calculated the probability that\na secure link that is made between two uncompromised nodes will be compromised as\nm\n\u0001\ni−q\n\f\n1 −\n\f\n1 −m\n|S|\n\rx\ri p(i)\np\n(6)\nwhere x is the number of nodes captured, i is the number of keys in common, m is the\nnumber of keys in the key ring of a node, and p is the probability of setting up a secure\nlink.\nThe Q-Composite Scheme offers greater resilience compared to the Basic Scheme\nwhen a small number of nodes have been captured in the network. The amount of\ncommunications that are compromised in a given DSN with the Q-Composite Scheme\napplied is 4.74 percent when there are 50 compromised nodes, while the same DSN\nwith the Basic Scheme applied will have 9.52 percent of communications compromised.\nThough the Q-Composite Scheme performs badly when more nodes are captured in a\nDSN, this may prove a reasonable concession as adversaries are more likely to commit\na less expensive smaller attack and preventing smaller attacks can push an adversary to\nlaunch a larger attack, which is far easier to detect with vast node failure.\nStill, random key pre-distribution schemes like Q-Composite and Basic cannot\nbe securely used for large networks because they use keys more than once, which\nresults in the compromise of a larger fraction of communications when just a few nodes\nare compromised. Since random key pre-distribution schemes are not scalable, the\nmaximum network that can be supported should be measured using the Limited Global\nPayoff Requirement, which states that given a DSN is secure, an adversary should not\nlearn anything about the communications of nodes in the network other than those\nof captured nodes. Let fm be the maximum compromise threshold past where the\nadversary gains an unacceptable high confidence of guessing the sensor readings of\nthe entire network. If xm is the number of nodes compromised, then the total fraction\nof secure links compromised after the key setup phase due to this xm nodes being\ncompromised is f(xm), if the total fraction of secure links compromised reaches the\nthreshold value with the xm nodes being compromised ,i.e., fm = f(xm), then Chan,\nPerrig, and Song [2] have calculated the maximum allowable size of the network to be\nn ≤2xm\n\f\n1 + 1\nfm\n\r\n(7)\nFor example, when p = 0.33, fm = 0.1, and m = 200, the maximum supportable\nnetwork size for a Q-Composite Scheme (q = 2) is 1, 415 nodes. Compared to the\n1, 159 node maximum of the Basic Scheme, the advantage is obvious. Thus, the Q-\nComposite Scheme is more efficient than the Basic Scheme in providing more resilience\nto node capture and significantly increases the maximum allowable size of a DSN.\nSince both schemes fail to provide node-to-node authentication or resistance against\n"
        },
        {
            "page_number": 351,
            "text": "WIRELESS NETWORK SECURITY\n351\nnode replication, it is important to review other schemes that work more efficiently in\nDSNs requiring such security measures.\n3.2. Multipath Key Reinforcement Scheme\nThe idea of using a multipath to reinforce links in a random key establishment\nscheme was first explored by Anderson and Perrig [10]. Chan, Perrig, and Song [2]\nfurther developed the Multipath Key Reinforcement Scheme for establishing a link\nbetween two nodes of a given DSN that is stronger than that in the Basic Scheme.\nThe links formed between nodes after the key discovery phase in the Basic Scheme\nare not totally secure due to the random selection of keys from the key pool allowing\nnodes in a DSN to share some of the same keys and, thereby, possibly threaten multiple\nnodes when only one is compromised. To solve this problem, the communication key\nbetween nodes must be updated when one is compromised once a secure link is formed.\nThis should not be done via the already established link, as an adversary might decrypt\nthe communication to obtain the new key, but should be coordinated using multiple\nindependent paths for greater security.\nIf node A needs an updated communication key with node B, all possible disjointed\npaths to B must be used. Assume that there are h such disjointed paths from node A to\nnode B. Then A generates h random values (g1, g2, ...., gh), each equal to the size of an\nencryption key, and sends one down each available disjointed path to B. When node B\nhas received all h random values, it computes the new encryption key at the same time\nas node A does forming a new, secure communication link, or\nk′ = k ⊕g1 ⊕g2 ⊕.... ⊕gh\n(8)\nwhere k is the original key.\nWith the new link in place, the only way an adversary can decrypt the communi-\ncations is to compromise all the nodes involved in the formation of the key. The higher\nh is, the more paths and nodes involved and the greater the security of the new link.\nThis increase in DSN communications causes excessive overhead in finding multiple\ndisjointed paths between two nodes. Also, as the size of a path increases, it may grow\nso long as it leaves a chance for an adversary to eavesdrop, which makes the whole path\ninsecure. A 2-hop approach to the Multipath Key Reinforcement Scheme considers\nonly 2-link paths to minimize the overhead of path length by using disjointed paths that\nare only one intermediate node away from the two original nodes (A and B).\nTo take such an approach, first the number of common neighbors between the\noriginal nodes must be calculated in a planar deployment of sensors. In [2], Chan,\nPerrig, and Song calculated the overlap of communication radius between two nodes\nin a network to be 0.5865πr2 in a planar deployment where r is the communication\nrange of sensors; therefore, the expected number of common neighbors with whom both\nnodes share a secure link is 0.5865p2n′, where p is the probability of sharing sufficient\nkeys to communicate and n′ is the number of neighbors in each node. Expressed\nas 0.5865d2/n = k, both nodes share a secure link with an expected k neighbors,\ne.g., if d = 20 and n′ = 60, then k = 3.91. To assess the efficiency of the 2-hop\n"
        },
        {
            "page_number": 352,
            "text": "352\nVENKATA KRISHNA RAYI et al.\napproach, the new probability for compromising the link between two nodes needs to\nbe derived. If an adversary’s basic probability of compromising the link is b, then the\nprobability of compromising at least one hop on any given 2-hop path is the probability\nof compromising hop 1 in the path plus the probability of compromising hop 2 in the\npath minus the probability of compromising both hops in the path or 2b−b2 [2]; hence,\nthe final probability of breaking the link will be b′ = b(2b −b2)k.\nIf b = 0.1 and the number of neighbors (k) is 3, then the chance of eavesdropping\nafter reinforcement improves to 6.86 × 10−4, that is about 1 in 1, 458. Chan, Perrig,\nand Song [2] calculated the total additional communication overhead incurred to be\nat least 2 × 0.5865p2n′ times more in the 2-hop approach compared to the normal\nsetup. If, for instance, p = 0.33 and n′ = 60, additional overhead can be at least 7.66\ntimes. Given these results, the minimal network overhead of finding the neighbors that\nshare a common key becomes a reasonable trade-off when using a 2-hop Multipath Key\nReinforcement Scheme to increase the security of DSNs, though this scheme remains\nconstrained by certain vital factors, including the deployment density characteristics of\nthe DSN.\n3.3. Random Pairwise Key Scheme\nNode-to-node authentication not only helps to reduce overhead since sensor nodes\ninstead of a base station take actions when a node is compromised in the DSN, but\nalso entails that each node use a unique identity, which helps nodes to identify exactly\nwhich ones are compromised. We have seen that though the Basic Scheme is somewhat\nefficient, it does not provide node-to-node authentication. The Q-Composite extension\nand 2-hop Multipath approach do not provide node-to-node authentication too. The\nPairwise Key Establishment Scheme, however, is one of the most efficient key estab-\nlishment schemes in DSNs because it does offer many additional features compared\nto other schemes, including node-to-node authentication and resilience against node\nreplication.\nFor a DSN of n nodes in the Pairwise Scheme, the key predistribution is done by\nassigning each node a unique pairwise key with all the other nodes in the network, or\nn −1 pairwise keys, which are retained in each node’s memory so that each node can\ncommunicate with all the nodes in its communication range. With each node sharing\na unique key with every other node in the network, this scheme offers node-to-node\nauthentication. Each node can verify the identity of the node it is communicating with.\nThis scheme also offers increased resilience to network capture as a compromised node\ndoes not reveal information about other nodes that are not directly communicating with\nthe captured node. Through increased resilience, the scheme minimizes the chance for\nnode replication. The drawback with the Pairwise Scheme is the additional overhead\nneeded for each node to establish n−1 unique keys with all the other nodes in the DSN\nand maintain those keys in its memory. Utilizing such a scheme makes a DSN size-\nprohibitive since, as the number of nodes in the network increases, so do the number\nkeys that must be stored in each node’s memory. If there is a network of 10, 000\nnodes, then each node must store 9, 999 keys in their memory. Since sensor nodes are\n"
        },
        {
            "page_number": 353,
            "text": "WIRELESS NETWORK SECURITY\n353\nresource-constrained, this significant overhead limits the scheme’s applicability, but it\ncan be effectively used for smaller networks.\nChan, Perrig, and Song [2] developed the Random Pairwise Scheme as an extension\nof the Pairwise Scheme to help overcome this drawback. They stated that not all n −1\nkeys are required to be stored in a node’s key ring. As we have already seen with the\nBasic Scheme, not all nodes must be connected as long as node connections meet some\ndesired probability Pc, which dictates that only np keys are needed to be stored in a\ngiven node’s key ring (n being the number of nodes in the network and p being the\nprobability that two nodes can communicate securely). Given this, if k is the number\nof keys in a node’s key ring, the maximum allowable network size can be determined\nwith n = k/p for the Random Pairwise Scheme.\nIn the initialization stage of this scheme, n unique node identifiers are created,\neach paired with m other randomly selected node identities. A pairwise key is then\ngenerated for each such pair. Both the generated pairwise key and the identity of the\nother node that shares the key are stored in each node’s memory. Additional identifiers\ncan be generated to allow for scalability of the DSN, i.e., the number of nodes may\noriginally be fewer than n created to allow for adding nodes. In the key discovery\nphase, each node broadcasts its identity to the other nodes in the network. For example,\nif node A wants to communicate with other nodes in the network, it broadcasts its\nidentity to other nodes in the network; if the neighboring nodes share a pairwise key\nwith node A, they perform a cryptographic handshake with A, thereby forming a secure\ncommunication link. This process of broadcasting can also be extended beyond the\ncommunication range of a node by making the intermediate nodes rebroadcast the node\nidentity to a certain number of hops, which in turn helps in increasing the maximum\nallowable size of the DSN. This process of range extension must be done cautiously as\nit leaves a vulnerable opening for an adversary to perform a ‘denial of service attack’.\nThe attack involves the adversary’s introducing foreign nodes into the DSN to generate\nrandom node identities that flood the DSN with rebroadcasted identities, making the\nwhole scheme slow and inefficient. This type of attack can be avoided by restricting\nthe number of hops for range extension.\nRevoking compromised nodes from a DSN helps avert various attacks such as\ndenial of service, implanting clones, dropping legitimate reports, etc. Revocation of\nsensor nodes through the base station can be a slow process due to the high latency in\ncommunications with the sensor nodes. To overcome this difficulty, Chan, Perrig, and\nSong [2] also developed a distributed node revocation method for the Random Pairwise\nScheme. Assume that the scheme can detect compromised nodes. If node A finds a\ncertain node B to be compromised then it casts a public vote against B. If a threshold\nof t such votes have been cast against node B by other nodes in the DSN, node A will\ndisconnect all its communication with B. This process continues until all the nodes in\nthe DSN break their links with B, thereby deleting B from the network. All the nodes\nthat vote against B are called the ‘voting members’ of B and, as B shares exactly k\npairwise keys with other nodes, there will k voting members of B.\nThis voting method of node revocation must have certain important properties to\nfunction properly: make the broadcasted public votes without replay value, disallow a\n"
        },
        {
            "page_number": 354,
            "text": "354\nVENKATA KRISHNA RAYI et al.\nvoting member from forging another vote, provide a means for each voting member to\nverify the validity of the votes that are being broadcasted, etc. In this voting method,\neach of the k voting members of B is initialized with a random key Ki, and should\nknow the hash values of the remaining k −1 voting members. To revoke node B node\nfrom the network, node A broadcasts its Ki key. All other voting nodes verify the key\nby calculating the hash value of the key. Once verified, the key is replaced with a flag\nsignifying the vote has already been used.\nGiven this process, the nodes in the DSN must store an additional k −1 hash\nvalues, a voting key and the pairwise keys, which drastically increases the overhead\nin a sensor node’s memory. Chan, Perrig, and Song [2] proposed using a Merkle Tree\n[11] to authenticate the k hash values to reduce overhead by requiring verification and\nstoring of only one hash value, which reduces the memory size required on the node,\nbut also increases the size of the voting information to O(log(k)) as each node must\nstill recall which ones have already been received from public vote to remove possible\nreplay.\nOther precautionary measures with the Random Pairwise Scheme taking the public\nvoting approach include the critical issue of choosing the threshold or t value. If t is\nhigh, there may not be enough neighboring nodes to revoke a node that has been\ncompromised; however, if t is low, then a group of compromised nodes may cause the\nrevocation of many legitimate nodes. For instance, a DSN of 1, 000 to 10, 000 nodes\nshould have a t value from 1 to 5. Also, since this approach requires a node have at\nleast t neighbors in its communication range to be revoked, an adversary can attack the\nDSN by selectively disrupting a given node such that only t −1 legitimate nodes are\nable to communicate with it so that it cannot be revoked.\nBeyond problems with t value, in the public voting approach each and every vote\nthat is cast is being transmitted to all the nodes in the network, which may lead to a\n‘denial of service attack’. To solve this, only the voting members should be required to\nrebroadcast votes between each other while the remaining nodes are forced to ignore the\ncommunication thereby decreasing the degree of vulnerability to falsely rebroadcasted\nidentities. Additionally, the node that first receives the correctly verified vote rebroad-\ncasts it only a fixed number times to increase the probability of successful transmission\nto neighboring voting members.\nIn similar action, an adversary that tries to compromise a fixed number of nodes\ncan compromise a significant portion of a DSN when public voting is used to perform\ndistributed node revocation since each node can potentially cast a vote against k others.\nTo prevent this problem, only nodes that establish direct communication are given the\nability to revoke a compromised node by distributing masked revocation keys to voting\nmembers in a non-working form. Each node would then complete the key discovery\nphase by sharing the secret key only with other nodes with whom they already share a\npairwise key connection.\n"
        },
        {
            "page_number": 355,
            "text": "WIRELESS NETWORK SECURITY\n355\n4.\nPOLYNOMIAL POOL-BASED KEY PREDISTRIBUTION\nEvery key distribution scheme previously discussed has one or more trade-offs or\ncompromises to be considered and what they fundamentally lack is greater probability\nof key establishment despite part of the DSN being compromised. To this end, Liu and\nNing [3] proposed the Polynomial Pool-Based Key Predistribution Scheme that offers\nseveral efficient features the other schemes lack, including:\nany two sensors can definitely establish a pairwise key when there are no com-\npromised sensors;\neven with some nodes compromised, the others in the DSN can still establish\npairwise keys;\na node can find the common keys to determine whether or not it can establish\na pairwise key and thereby help reduce communication overhead.\nIn the initialization stage of the Polynomial Pool-Based Scheme, the setup server\nrandomly generates a bivariate t-degree polynomial f(x, y) over a finite field Fq, where\nf(x, y) =\nt\n\u0001\ni,j=0\naijxiyj\n(9)\nThe value of q is a prime number which can accommodate a cryptographic key. The\nequation f(x, y) has the property f(x, y) = f(y, x). The setup server then generates a\npolynomial share of the equation for every node in the sensor network; e.g., node i in\nthe network receives an f(i, y) share and node j receives an f(j, y) share. If both nodes\ni and j want to establish a common key f(i, j) between them, then node i can compute\nthe common key by computing f(i, y) at node j and then node j can compute f(j, y)\nat node i for the common key f(i, j). This methodology is secure and reveals nothing\nabout the communication between other nodes until t nodes have been compromised,\nmaking it t collusion resistant where the t value depends upon the memory available in\nthe sensors.\nEach node in this scheme must store a t-degree polynomial which occupies (t +\n1)log(q) storage space. Increasing the size of the DSN increases the chance of compro-\nmising more than t nodes, but modifications based on the Basic Scheme can earn good\nresults. For this, instead of using a single t-degree polynomial, a pool of polynomials\nis used. During the initialization phase, randomly selected polynomials are deployed\ninto each node’s memory. When there is only one polynomial remaining in the pool,\nthe scheme falls back to the Polynomial Pool-Based Key Distribution; but, if all of the\npolynomials are 0-degree, then distribution resembles the Basic Scheme.\nIn the key predistribution stage of the Polynomial Scheme, the setup server gen-\nerates a set of bivariate t-degree polynomials over a field Fq. Each polynomial is then\nassigned with a particular ID for the server. A subset of these polynomial shares are\nthen picked up by the server and placed in each of the DSN’s nodes. While polynomial\n"
        },
        {
            "page_number": 356,
            "text": "356\nVENKATA KRISHNA RAYI et al.\nplacement is the main issue of this stage, in the key discovery stage each sensor node\nfinds a node with which it shares the same bivariate polynomial and both nodes estab-\nlish a common key. The complex issue is to find whether two nodes share the same\npolynomial or not, for which there are two techniques: predistribution and real-time\ndiscovery.\nIn the predistribution approach, the knowledge of the nodes with which each node\nwill share a polynomial is pre-loaded. This is a basic method in which each node carries\nthe node IDs of those with which they share a polynomial. The concessions of this\nmethod are that it does not offer the flexibility of adding new nodes into a DSN and\nit leaves the network vulnerable to attack. Since information is predistributed in this\napproach, an adversary may attack a node and gain access to the stored data, which\nwould help in targeting certain nodes in the DSN. Conversely, nodes must uncover\nwith which others they share a polynomial after deployment when applying the real-\ntime discovery method. This discovery can be done by broadcasting the IDs of the\npolynomials that nodes share or by challenging the nodes with puzzles that are only\nsolvable if the nodes share part of the bivariate t-degree polynomial. Though this\nhandles problems faced with the predistribution approach, real-time discovery increases\nthe communication overhead of the DSN, which makes weighing these factors critical\nin choosing a method.\nAfter key discovery, if two nodes do not find a common polynomial share, they\nmust communicate through a path key. If node P wants to communicate with node\nQ and the two nodes do not have a common polynomial share, node P must find a\npath through which it can communicate with node Q and either node can then send a\nrequest to establish a pairwise key for communication. The problem with this stage is\nthat intermediate nodes should be able to communicate with both nodes and, similar to\nthe previous stage, there are two customary techniques for finding intermediate nodes:\npredistribution and real-time discovery.\nIn the predistribution approach, the setup server preloads each node with informa-\ntion such that, if a node is given an ID, each node can find a path to it. In this stage, the\npredistribution method suffers from the same problems as faced in key predistribution-\nno scalability and vulnerability to attack. With real-time discovery, nodes try to find\na communication path on-the-fly. A source node sends a message to adjoin interme-\ndiate nodes that it wants to establish a pairwise key with the destination node and, as\nthe source node already shares a common key with the intermediate nodes, there is\nno security threat in this communication. If an intermediate node of the source node\nshares a common key with the destination node, then a communication path has been\ndiscovered between the source and destination nodes through which they may discover\na common key. Again, the concession of real-time discovery is additional overhead of\ncommunication.\n4.1. Random Subset Key Predistribution [3]\nA Polynomial Pool-Based Scheme using a random subset key assignment is an\nextension of the Basic Scheme [1], in which random keys are selected from a large key\n"
        },
        {
            "page_number": 357,
            "text": "WIRELESS NETWORK SECURITY\n357\npool and then assigned to each node in a DSN. In the Random Subset Scheme, random\npolynomials are selected from a polynomial pool and assigned to each node in a DSN\nto avoid the Basic Scheme’s vulnerability in possibly using a key in more than one\nnode. With Random Subset, the pairwise keys generated by each node are unique and\nbased upon the each node’s ID. If no more than t shares of the same polynomial have\nbeen disclosed, it is very difficult to attack the communication between two nodes.\nThe Random Subset works similarly to the Polynomial Pool-Based in the three\nstages of key establishment. In the key predistribution stage, the setup server generates\na set F of s-bivariate t-degree polynomials and then initializes each node with a subset\nof s′ polynomials from F. In the key discovery stage, each node attempts to determine\nthe nodes with which they share a common key by employing the real-time discovery\ntechnique as information in not preloaded in the nodes prior deployment. In the path key\nestablishment phase, a source node sends a message to its intermediate nodes seeking\nto establish a connection with a destination node and, if an intermediate node shares a\ncommon key with both the source and destination nodes, then a communication path\nis formed between the two. Generally, the communication range of the source node is\nlimited to lessen vulnerability.\nThe probability of two sensors sharing the same bivariate polynomial is the same\nas the probability of the two sharing a common key as described in the Basic Scheme\ndiscussion,\np = 1 −\ns′ −1\n\u001d\ni = 0\ns −s′ −i\ns −i\n(10)\nThis can also be applied to calculate the probability that any two sensors can\nestablishacommonpairwisekeyusingboththekeydiscoveryandpathdiscoverystages.\nIf there are d neighbors to a node and any one of them can act as the intermediate node,\nthe probability that one of them share a common key with the source and destination\nwill be p2; therefore, the probability that the sensor nodes establish a pairwise key in\neither the key discovery or the path key establishment stage will be\nPs = 1 −(1 −p)(1 −p2)d\n(11)\nIf p = 0.3 and d = 30, then Ps = 0.959. Assuming that an attacker has compro-\nmised Nc sensors in the network, where Nc > t, the scheme is known to be secure until\nthe adversary compromises fewer than t sensors. With a pool of F polynomials, the\nprobability that a polynomial is used i times and its probability of being compromised\nwhen more than t nodes are compromised must be calculated to determine security\nefficiency. Given this, the probability that a polynomial is chosen for a sensor node\nwill be s′/s and the probability that this polynomial is chosen exactly i times among\nthe Nc compromised nodes is\np(i) =\nNc!\n(Nc −i)|i|\n\fs′\ns\n\ri \f\n1 −si\ns\n\rNc−i\n(12)\n"
        },
        {
            "page_number": 358,
            "text": "358\nVENKATA KRISHNA RAYI et al.\nThus, the probability of a polynomial being compromised is\nPc = 1 −\n\u0001t\ni=0 p(i)\n(13)\nThough this makes the DSN more secure, it is still vulnerable to attack because if\nan adversary somehow knows the distribution of polynomial shares, specific nodes can\nbe targeted for attack to compromise communications. It is enough for an adversary to\ncompromise t+1 particular nodes to compromise a polynomial and, in effect, the DSN.\nThis problem is addressed by restricting the use of polynomial shares to a maximum\nof t + 1 times in the network so that an attacker must now compromise all the t + 1\nnodes to compromise a polynomial. Though efficient, use of random subsets like this\ndecreases the potential size of a DSN. The maximum number of nodes in a network\nwhen random subsets are implemented is (t + 1)s/s′; however, using this scheme is\nunnecessary as it is relatively difficult for an adversary to compromise t + 1 selected\nnodes.\nPolynomial Pool-Based Key Predistribution using random subsets offers greater\nsecurity and flexibility when compared to other schemes until a certain number (60\npercent) of compromised nodes has been reached at which point any scheme would\nprove ineffective. Compared to the Random Pairwise Scheme, which offers perfect\nresilience to node capture as no key in the network is used twice, the Polynomial Pool-\nBasedoffersthesameresilienceifapolynomialshareisusednomorethanttimes. Also,\nthe Polynomial Pool-Based Scheme offers certain advantages over Random Pairwise in\nthat sensors can be added dynamically without consulting the already deployed sensors\nwhile dynamically deploying nodes in Random Pairwise demands that the server has\npredesignated unassigned space for additional nodes, which may never be deployed.\nBecause of this, the Random Pairwise Scheme can only offer limited scalability, while\nthe more attractive Polynomial Pool-Based Scheme allows for undetermined network\ngrowth.\n4.2. Grid-Based Key Predistribution\nA Polynomial Pool-Based Scheme [3] using a grid-based key assignment offers\nall the attractive properties of the Polynomial Pool-Based key predistribution and guar-\nantees that two sensors can establish a pairwise key when there are no compromised\nnodes and the nodes can communicate with each other. Even if some nodes are cap-\ntured, there will still be a great chance for key establishment between uncompromised\nnodes using this approach, which also reduces DSN communication overhead. With\ngrid-based key predistribution, a sensor node can determine whether it can establish a\npairwise key with another node or not, and can say which polynomial should be used\nfor key establishment.\nIf a DSN consists of N sensor nodes, the approach involves constructing an m×m\ngrid with a set of 2m polynomials, calculated as {f c\ni (x, y), f r\ni (x, y)}, i = 0, ..., m−1,\nwhere the value of m is the square root of N; each row i in the grid is associated with\na polynomial f r\ni (x, y) and each column of the grid is associated with a polynomial\n"
        },
        {
            "page_number": 359,
            "text": "WIRELESS NETWORK SECURITY\n359\nshare f c\ni (x, y). The setup server distributes an intersection in the grid to each node,\nand then distributes the polynomial shares of that particular column and row to the\nnode to provide each node with the information required for key discovery and path\nkey establishment. Although the Grid-Based Scheme can be extended to n-dimension,\nLiu and Ning [3] considered only a 2-dimension with many polynomials, so that is the\nexample discussed here.\nIn the first stage of key establishment, the setup server generates 2m t-degree\nbivariate polynomials over a finite field Fq and assigns each node to an unoccupied\nintersection in the grid for deployment in the DSN. If the intersection is < i, j >, then\nthe node ID is < i, j >. The server provides each node with its ID and the row and\ncolumn polynomial shares of that grid intersection. To facilitate path discovery, all\nnodes are densely placed in a rectangular area in the grid.\nIn the second stage, polynomial share discovery, if node i wants to establish a\npairwise key with node j, it checks for common rows or columns with j, i.e., ci = cj\nor ri = rj. The pairwise key can be established using the polynomial shares of a\nrow or column that matches. Should none match, then nodes i and j must find an\nalternate path to each other in the path key establishment stage. To do so, node i finds\nan intermediate node through which it can establish a pairwise key with node j. Even\nif some intermediate nodes are compromised, node i can still find a path to node j since\nthere are many connecting paths in the grid between the two nodes; but, as the number\nof compromised nodes increases, so does the length of the path.\nIn this case the nodes remember the graph composing the grid; however, with large\nnetworks, it is not feasible for a node to remember the entire graph or run an algorithm\nfor finding the path between the nodes. Discovering key paths using two intermediate\nnodes limits the demands on the nodes for this scheme to function in large DSNs also.\nIf, for instance, there are two sensor nodes attempting to establish a path key between\nthem, the source node S determines the set of N nodes with which it can communicate,\nand then selects some nodes randomly from that set. S also generates a random number\nr and a counter c. Node S sends each node U of the subset N a message containing the\nIDs of S and D, the counter value c and Kc in an encrypted form.\nHere the value of Kc = F(r, c) where F is a pseudo random function. Encryption\nof the message is done using the pairwise key that S shares with the intermediate node\nU. After receiving the message from S, node U checks for the authentication of the\nmessage and if the message is authenticated, U attempts to find a non-compromised\nnode V. It then sends to V the message sent by S in the encrypted form using the pairwise\nkey which it shares with V. If V receives the message and discovers that it can establish\na pairwise key with D, it sends the message to D in encrypted form using the shared\nkey. Once the destination node D receives the message, it knows that node S wants\nto establish a pairwise key with it and then sends S the counter value c and the new\ncommunication key Ks,d = Kc.\nWith grid-based key predistribution, an adversary may try to attack the connection\nbetween two nodes by either compromising the pairwise key or by preventing the two\nnodes from establishing a shared key. If an adversary wishes to attack the entire DSN,\nthe foe may attempt to lower the probability of establishing a pairwise key between\n"
        },
        {
            "page_number": 360,
            "text": "360\nVENKATA KRISHNA RAYI et al.\nnodes. This may be done through attacking a pair of nodes and finding their common\nkey without actually compromising the nodes by compromising the polynomial which\nthe two nodes share. To discover the polynomial share, the adversary must compromise\nat least t + 1 nodes as stated previously. The DSN may avert such an attack even when\nthe adversary successfully finds out the polynomial which two nodes share by the nodes’\nestablishing a common key through path key establishment.\nFrom the scheme we can see that there are still m −1 nodes that can help nodes\nU and V establish a common key. An adversary must compromise at least one node in\neach pair to arrest path key establishment; thus, the adversary must compromise t + 1\nnodes to learn the pairwise key and t+m sensor nodes to prevent two from establishing\na pairwise key via intermediates. An adversary can also compromise the polynomial\nshares in a pool by knowing the subset assignment mechanism. Supposing that the\nadversary has compromised some l polynomials from the pool, there are about ml\nsensors with at least one polynomial share disclosed. The attacker has compromised\nabout (t+1)l sensor nodes, but only affects the common keys in ml sensors, including\nthose of the compromised nodes. The adversary may also attack the sensors randomly to\ndisrupt path establishment and thereby make key establishment an expensive process.\nIf Pc nodes have been compromised, then the probability that exactly k polynomial\nshares on a particular polynomial are disclosed is\nP(k) =\nm!\nk!(m −k)!pk\nc(1 −pc)m−k\n(14)\nThe probability that one particular polynomial is compromised would be calculated\nas\nPc = 1 −\n\u0001t\ni=0 p(i)\n(15)\nThis grid-based approach to the Polynomial Pool-Based Scheme has reasonable\noverhead when compared to other schemes. Each node must store 2 bivariate t-degree\npolynomials and IDs of the compromised nodes with which it can establish a pairwise\nkey; therefore, the total overhead for each node is, at most, 2(t + 1)log(q) + 2(t + 1)l\nbits. The DSN overhead is almost null when there is direct key establishment between\nnodes. There is slight communication overhead when two nodes must find a common\nkey through path key establishment and this overhead increases with each additional\nnode compromised. The approach offers many attractive properties other schemes\ndo not, including nice resilience to node capture until a certain percentage of nodes\nare compromised (60 percent). The Basic Scheme and Q-Composite Scheme offer\nthis same resilience, but the grid-based approach offers less overhead on both DSN\ncommunication and computations. Compared to Random Pairwise Scheme, the grid-\nbased method offers the same degree of security when the same number of sensors and\nstorage overhead are considered. More than any other scheme, the grid-based approach\noffers greater probability of key establishment when there are no compromised nodes\nas well as greater probability of key establishment with some nodes compromised.\n"
        },
        {
            "page_number": 361,
            "text": "WIRELESS NETWORK SECURITY\n361\nFinally, there will be a greater chance for nodes to establish a pairwise key with others\nwithout communication overhead as the sensors are deployed in a grid-like structure.\nThus, the Polynomial-Based Key Predistribution and the two instantiations of the\nscheme, polynomial pool-based and grid-based, provide some attractive and efficient\nproperties for key establishment in DSNs. In the near future, these schemes might\nbe extended as part of the research performed in sensor networks. For example, the\ngrid-based approach may be extended into n-dimension or hypercube-based. Also,\nresearch must be done for DSNs with Polynomial-Based Key Predistribution scheme\nand mobile properties.\n5.\nSPINS: SECURITY PROTOCOLS FOR SENSOR NETWORKS\nPerrig, Szewczyk, Tygar, Wen and Culler[5] at UC Berkeley presented a suite of\nsecurity protocols optimized for sensor networks that they called ‘SPINS’. The suite is\nbuiltupontwosecurebuildingblocks, eachperformingindividualrequiredwork: SNEP\nand µTESLA. SNEP offers data confidentiality, authentication, integrity, and freshness,\nwhile µTESLA offers broadcast data authentication. The µTESLA protocol, used on\nregular networks, is modified as a SPINS for use in resource-constrained DSNs. SPINS\nincorporates TinyOS (operating system) in each node, all of which communicate with a\nbase station. Most DSN communications pass through the base station and involve three\ncommunication types: node-to-base station, base station-to-node, and base station-to-\nall nodes.\nThe main goal of SPINS protocol is to design a key establishment technique based\non SNEP and µTESLA to prevent an adversary from spreading to other nodes in the\nnetwork through a compromised node. Each node in this scheme shares a secret key\nwith the base station that is initialized before deployment. The following are some of\nthe representations in this scheme used to illustrate how this works:\nA and B are two communicating nodes in the network;\nNa is generated by node A;\nXab is the master key shared between nodes A and B;\nKab and Kba are the encryption keys shared between A and B, which are derived\nfrom the master key Xab;\nK′\nab and K′\nba are the secret MAC keys shared between A and B, which are\nderived from the master key Xab;\n{M}Kab denotes the encryption of message M with key Kab;\nMAC(K′\nab, M) denotes the computation of MAC for message M with MAC\nkey K′\nab.\n"
        },
        {
            "page_number": 362,
            "text": "362\nVENKATA KRISHNA RAYI et al.\n5.1. SNEP: Data confidentiality/authentication/freshness\nA combination of two schemes forms SNEP including a counter for semantic se-\ncurity and a bootstrapping scheme. Using this combination, SNEP is able to offer a\nnumber of advantages and only adds 8 bytes per message by reducing the communi-\ncation overhead of the network. It uses a counter, like many other protocols, to offer\nauthentication and freshness, but does so using means that also provide semantic se-\ncurity. Two counters are shared between nodes attempting to communicate with each\nother for which some of the source node’s cryptographic techniques send the shared\ncounters with a message to the destination node. General encryption can be used as a\nsimple form of confidentiality, but is not sufficient to protect messages; whereas, seman-\ntic security offers far greater security by making it harder for an adversary to derive the\noriginal data even after obtaining one or more encrypted messages. In DSNs, sending\nmessages with a counter can cause overhead; but, the energy can be saved by sharing\nthe counter between both nodes and incrementing it each time the destination node\nreceives a message. As with other schemes, for better security the same keys should\nnot be used again and again. In SNEP, independent keys are used for encryption and\nMAC operations. The secret key shared between source node A and destination node\nB is used for deriving the encryption and MAC keys for each direction. The encrypted\ndata has the form E = D(K, C) where D is the data, K is the encryption key and C\nis the counter. The MAC is M = (K′, C||E).\nIn SNEP then, the total message that node A sends to B is: A →B : D(Kab, Ca),\nMAC(K′\nab, Ca||D(Kab, Ca)).\nThe semantic security property is satisfied as each time the message is encrypted,\nthe counter value is incremented to a different value; thus, though the same message is\nencrypted, an adversary would not be able to decode the message.\nWith SNEP, an adversary does have a chance of performing a DOS attack by\nconstantly sending the requests for counter synchronization, but this can be prevented\neither by sending the counter value with each encrypted message or by attaching a\nshort MAC to the message that does not depend on the counter. Data authentication\nis done using the MAC. The counter value in the message prevents an adversary from\nreplaying old messages, which would cause confusion and overhead in a DSN. As\nthe counter value is kept at both ends of communication and the ID is not transferred\nwith every message, communication overhead is negligible. The counter scheme also\nallows achieving weak freshness. If the counter value is verified correctly, it reveals the\nsequence of the messages, but only guarantees the sequence of messages, not that the\nreply from node B is caused by the message from A. To achieve strong freshness that\nincludes delay estimation, a nonce must be included with messages. To achieve strong\nfreshness, node A sends a nonce Na along with a reply message to B, which resends the\nnoncewithareplymessage. Thisprocesscanbeoptimizedbyimplicitlyusingthenonce\nin the MAC computation; therefore, the entire SNEP protocol with strong freshness is:\nA →B : Na, Ra and B →A : {Rb}(Kba,Cb), MAC(K′\nba, Na||Cb||{Rb}(Kba,Cb))\nIf the MAC correctly verifies, node A will know that the reply from B is a reply\nto its message. In this method, it is assumed that both communicating parties know\n"
        },
        {
            "page_number": 363,
            "text": "WIRELESS NETWORK SECURITY\n363\nthe counter value so that it need not be sent with every message; though, in reality,\nmessages might get lost or tampered and cause inconsistencies in the counter value.\nProtocols needed to synchronize the counter value include bootstrapping the counter\nvalue in the following manner: A →B : Ca with B →A : Cb, MAC(K′\nba, Ca||Cb)\nand A →B : MAC(K′\nab, Ca||Cb)\nThe counter value need not be encrypted since the protocol needs strong freshness\nfor which both communicating parties use the counter as nonce. Also, MAC need\nnot include the names A and B as the keys they use, Kab, state which nodes are\nparticipating in the communication. If node A realizes that the counter Cb of node B\nis not synchronized, it may request the counter of B with a message including Na for\nstrong freshness, or A →B : Na and B →A : Cb, MAC(K′\nba, Na||Cb).\n5.2. µTESLA: Authenticated broadcasts\nAuthenticating broadcasted data is a critical issue in DSNs, but previous solutions\nto this problem suffer from too much communication and computation overhead, and\ntherefore, are not so useful in resource-constrained DSNs. TESLA, one of these solu-\ntions, provides an inefficient scheme for broadcasting data with authentication by using\nthe digital signatures technique, which adds 24 bytes of overhead to each message that\nare typically only allotted a pocket size of 30 bytes. Thus, using TESLA can cause\nalmost all of the packet size to be occupied for the code only. Also, TESLA discloses\nthe key with every message packet it sends and receives, which can use a great deal of\na DSN’s energy. Finally, TESLA authenticates keys using a one-way key chain, which\nis not possible to be stored in each sensor node. Perrig et al. modified TESLA for\nauthenticating broadcasted data in a way that involves no significant overhead. Called\nµTESLA, this innovative method reduces energy needed by authenticating data using\nasymmetric mechanisms. Also unlike TESLA, which discloses the key every time a\npacket is sent or received, µTESLA does so only once in an epoch. The only limit with\nµTESLA is that it restricts the number of authenticated senders as it is expensive to\nstore the one-way key chain in a sensor node.\nµTESLA is able to provide the asymmetric cryptographic type of authenticated\nbroadcast through delayed disclosure of symmetric keys. For broadcasting authenti-\ncated information between the base station and nodes of a DSN, µTESLA requires that\nthe base station and nodes are loosely time synchronized and that each node knows an\nupper bound on the maximum synchronization error. When the base station wants to\nsends a packet to all the nodes in a given network, it computes a MAC on the packet\nbeforehand. Since all the nodes in the network are sure that only a base station can\ncompute the MAC, the MAC key is not disclosed at this point in time so they will not\nbe vulnerable to attacks from an adversary. The packets sent to the nodes are stored in\ntheir buffers until the base station discloses the corresponding keys. Once disclosed,\nthe keys can be authenticated by the nodes’ using the one-way function F. If a key is\ncorrect, a node can use it to authenticate the packet stored in its buffer.\nEach MAC key is a sequence of keys generated by the function F. The sender\nchooses the last key Kn of the chain randomly and then generates the one-way key\n"
        },
        {
            "page_number": 364,
            "text": "364\nVENKATA KRISHNA RAYI et al.\nchain by repeatedly applying F. Supposing the base station has sent packets P1 and\nP2 in the time interval t1, P3 and P4 in t2, P5 in t3, and P6 in t4, the nodes receiving\nthe packets cannot verify their authentication immediately, so the nodes store them\nin buffers. Packets sent in a particular time interval are authenticated using the key\nthat corresponds to that time interval. Let the difference of the time interval be two\nin this case, the receiver node is loosely time synchronized with the base station and\nknows key K0. Assuming that all the messages sending the key information about\npackets P1 −P5 are lost and only the message that carries the key information about\npacket P6 arrives, the receiver node can still authenticate the keys of the other packets\nby deriving the key information supplied for P6. Thus, though some of the packets\nmay have been lost, the nodes can still authenticate them using the keys received. To\ndo this, µTESLA has multiple phases that perform a particular job each, including\nSender Setup, Broadcasting Authenticated Packets, Bootstrapping New Receivers, and\nAuthenticating Broadcast Packets.\nSender Setup: In this phase, the sender wanting to broadcast messages in the\nDSN generates a one-way key chain, randomly selects the last key Kn, and\ngenerates the other values by applying the one-way function F on the chain for\ngenerating a length n. As F is a one-way function that any node can compute,\nthe keys are generated forward but not backward; i.e., given Kj+1, K0, ..., Kj\ncan be computed, not Kj+2.\nBroadcasting Authenticated Packets: The sender of the packet uses the partic-\nular key for the corresponding time interval. For example, in the time interval\nI the sender uses the key K1 and in the time interval I + 1, the node uses the\nkey K2. The packets sent in the particular time interval are authorized using\nthe corresponding key. In the time interval (I + X), the sender reveals the key\nK1.\nBootstrapping New Receivers: The keys in a one-way key chain are self-\nauthenticating. If the receiver has one key in the key chain it can efficiently\nauthenticate the other keys in the chain. If the receiver has value Kj in the key\nchain, it can easily authenticate Kj+1. Also, the sender and the receiver are re-\nquired to be loosely time synchronized with the receiver having the knowledge\nof the sender’s time disclosure schedule. Authenticating the key chain and hav-\ning loose time synchronization establishes strong freshness and point-to-point\nauthentication. A receiver R sends a nonce NR in the request message to the\nsender S, which replies to the message containing the following components:\nTS →the current time of the sender, Ki →a key in the one way key chain,\nTI →starting time, Tint →duration of time interval, and δ →disclosure delay.\nThe secret key shared between the node and the base station is used as the key\nfor the MAC.\nAuthenticating Broadcast Packets: An adversary sometimes knows the key\nused in the time interval I and may also have knowledge about the one-way\n"
        },
        {
            "page_number": 365,
            "text": "WIRELESS NETWORK SECURITY\n365\nkey chain, so the receiver should ensure that the packet received is from an\nauthenticated sender and not from an adversary before the key is released by\nthat sender. This is achieved through loose synchronization of the sender and\nreceiver. If the packet is legal, the receiver stores it; if it is spoofed, it is dropped.\nOnce the receiver verifies the key, it authenticates the packets with the key and\nreplaces that new key with the key it already has.\n5.3. Considerations for SPINS\nSPINS is one of the more efficient schemes for DSNs with advantages including\nsmaller code size, efficient performance, universal design, and low overhead. The\nscheme uses less of a sensor node’s memory; i.e., while crypto routines occupy 20\npercent of the space, µTESLA occupies 574 bytes and 2 Kbtes is the acceptable total\nused memory. The scheme’s performance is also efficient, as the bandwidth of the DSNs\nis adequate for the cryptographic primitives which SPINS uses. Additionally, most of\nthe SPINS design is universal and can be used in other networks of low-end devices.\nFinally, the communication costs for SPINS are small, with security properties like\ndata freshness, authentication and confidentiality only adding an overhead of 6 bytes\nin a 30-byte packet, which allows for inclusion in each and every packet. SPINS can\noffer even greater advantages when restrictions on bandwidth and memory are slightly\nrelieved.\nBroadcasting and authenticating data are not that easy for individual nodes, as\nstoring a one-way key in a node’s memory is not possible, computation of the keys\nusing a function generates much network overhead, and each node does not share a\ncommon key with every other node in the network. However, there are two solutions\nfor this problem. Firstly, the base station is used by a node to transmit all data that has\nto be broadcasted to other nodes. Secondly, the node broadcasts the data to the base\nstation while the base station generates the authenticating keys using the one-way key\nchain. It is efficient to implement the cryptographic primitives in a single block cipher\nas DSNs are resource-constrained and, therefore, can not afford additional overhead\nfor security. Yet, a strong cryptographic base is necessary for SPINS.\nBlock Cipher: Using RC5 can be very efficient in DSNs because of its small size\nand high efficiency. Moreover, as an algorithm it has been subject to scrutiny\nunder many attacks. Using TEA could also work for block ciphers, but it is not\nsubject to cryptanalysis scrutiny. DES and other algorithms are not usable for\nblock ciphers due to their large size and high computation requirements that\ncannot be met in DSNs.\nEncryption function: The counter (CTR) mode of block ciphers can use the\nsame function for both encryption and decryption, and the size of the cipher\ntext is the same as the data in this mode. These two properties make this mode\nvery useful while working in the encryption function of SPINS. Also, CTR\nmode offers semantic security, which is a strong cryptographic property already\ndiscussed. To use the CTR mode, both the sender and receiver nodes must\n"
        },
        {
            "page_number": 366,
            "text": "366\nVENKATA KRISHNA RAYI et al.\nmaintain counters in their memory and possess an efficient way to synchronize\nthe counters if needed. One advantage of maintaining a counter at both ends is\nthat the messages now will not have an overhead of carrying the counter with\nthem.\nFreshness: Using a counter and incrementing it every time a message is sent\nautomatically provides weak freshness. For strong freshness, the sender must\ncreate a nonce and should include it in the request message to the receiver.\nSPINS uses a MAC function for generating random numbers and a counter is\ncreated to keep track of those created.\nMessage authentication: Not only is a good encryption function necessary for\ndata, but also a secure MAC is needed. As the block cipher is used more than\nonce, CBC-MAC is used for MAC. An efficient way of message construction\nmust be used to achieve authentication and message integrity. The construction\n{M}k, MAC(k′, {M}k), in which M is the data, K is the encryption key, and\nK′ is the MAC key, is secure and protects the nodes from decrypting erroneous\nciphered text.\n6.\nLEAP: EFFICIENT SECURITY FOR LARGE-SCALE DSNS\nOne of the important mechanisms in sensor networks, in-network processing, is\nnot considered in the previous schemes. This critical issue must be handled while\ndealing with the resource-constrained property of DSNs. Most of the data has to be\ncollected by an aggregator node and then passed on to other nodes in a DSN; however,\ndata fusion through in-network processing can be used to save net- work energy and\nreduce communication overhead. The key establishment techniques that have been\ndiscussed so far do not support an In-Network Processing approach because the nodes\nin this method are unable to communicate with each other before transmitting data.\nPassive Participation is a form of In-Network Processing in which a sensor node takes\ncertain actions based on messages from other nodes. Zhu, Setia, and Jajodia [6] devised\na scheme called LEAP that would allow for data fusion, In-Network Processing and\nPassive Participation.\nBesides offering basic requirements like confidentiality and authentication, LEAP\nsupports various communication patterns, including unicast (addressing a single node),\nlocal broadcast (addressing a group of nodes in a neighborhood), and global broadcast\n(addressing all the nodes in a DSN). Sometimes DSNs are deployed in an adversary’s\narena and, where most of the time compromised nodes are undetected, LEAP provides\nsurvivability such that compromising of some nodes does not cede the entire network.\nLEAP is energy efficient since it supports techniques like In-network Processing and\nPassive Participation that greatly reduce network communication overhead and, in turn,\nincrease node battery life. Furthermore, LEAP ensures that messages transferred are\nnot fragmented, which would increase packet losses in transmission as well as make\nprotocol implementation more complex and difficult.\n"
        },
        {
            "page_number": 367,
            "text": "WIRELESS NETWORK SECURITY\n367\nLEAP is based on the theory that different types of messages exchanged between\nnodes need to satisfy different security requirements. All the packets transferred in a\nsensor network need to always be authenticated where a sensor node knows the sender\nof the data since an adversary may attack a DSN with false data at any time. On the\nother hand, confidentiality, like encryption of packets carrying routing information, is\nnot always needed. Different keying mechanisms are necessary to handle the different\ntypes of packets. For this, Zhu, Setia, and Jajodia [6] establish LEAP with four types of\nkeys that must be stored in each sensor: individual, pairwise, cluster, and group. Each\nkey has its own significance while transferring messages from one node to another in\na DSN and by using these keys, LEAP offers efficiency and security with resistance to\ncopious attacks such as the worm hole and the sybil.\nIndividual Key: This is a unique key that is shared between the base station and\neach sensor node. Sensor nodes use this key to calculate the MACs on their\nmessages to the base station like alert signals (reports on abnormal nodes). In\nthe same way, a base station can use an individual key to send messages to each\nand every node in the network.\nPairwise Shared Key: This is a unique key that is shared between each node and\nits neighboring node in the network. A node can use it to transfer individual\nmessages like sharing a cluster key or sending data to an aggregator node.\nCluster Key: This is a key that is shared between a node and its neighboring\nnodes, and is very important since it supports In-network Processing and Passive\nParticipation. A node may elect not to send a message to the base station if its\nneighboring node is sending the same message with a better signal, a discovery\nthat is only possible to implement if a node shares a common key with its\nneighboring nodes. With such a cluster key, a node can select which messages\nto transfer, thereby reducing the system communication overhead.\nGroup key: The base station shares this key with all the nodes in the network to\nsend queries to them. Group key used requires an efficient rekeying mechanism\nfor updating it as there is a chance for an adversary to know the key whenever\na node is compromised.\n6.1. Efficiently Establishing LEAP[6]\nEstablishing Individual Keys\nEvery node in a DSN shares a unique key with the base station that is preloaded\ninto each node’s memory before being deployed. The individual key Km\nu for node U is\ncalculated as Km\nu = fKm(u). For this, f is a pseudo-random function and Km is the\nmaster key known only to the controller. There is no need for the base station to store\nall the individual keys, because the base station generates them on the fly whenever it\nattempts to communicate with a node\n"
        },
        {
            "page_number": 368,
            "text": "368\nVENKATA KRISHNA RAYI et al.\nEstablishing Pairwise Shared Keys\nThe most common key used in a DSN is the pairwise that is shared between\neach node and its neighboring node. Sensor nodes are randomly scattered in an area;\ntherefore, the key establishment technique used should guarantee that nodes discover\nneighboring nodes when deployed. Because sensor nodes are static, the key establish-\nment technique does not have to consider deployment knowledge of others before node\ndeployment. When an adversary obtains a sensor node, it is assumed that the node\ncannot be compromised before time tmin. Whenever a node is deployed in a DSN,\nit requires some minimum time to identify neighbors and establish keys with them,\nwhich will be test. It is expected that tmin > test; otherwise, the adversary could\neasily capture all the nodes in the DSN and effectively take over the entire system.\nThe process of establishing keys when nodes are already deployed is similar to the\nprocess of key establishment when a new node is added to the network. There are four\nstages that represent the key establishment of new node U deployed in the network:\nkey predistribution, neighbor discovery, pairwise key establishment, and key erasure.\nDuring the initial stage of key predistribution, node U is loaded with the key Ki by the\ncontroller and derives the master key Ku using it. For neighbor discovery, node U first\ninitializes a timer to activate at time tmin, then starts communicating with its neighbors\nby broadcasting a HELLO message containing its ID. Node V responds to this message\nwith a reply containing its ID. The ACK of V is then authenticated using its master key\nKv derived from Ki. Node U verifies the authentication of V by generating the master\nkey Kv as node V shares Ki with it: U →∗: U and V →U : V, MAC(Kv, U|V ).\nFor the third stage of pairwise key establishment, node U computes the pairwise\nkey Kuv with node V using V’s identity. Node V can also do the same thing with U.\nThere is no need for authenticating node U to V as any future messages authenticated\nwith Kuv will prove node U’s identity. In the fourth and final stage, key erasure, node\nU erases Ki and all the master keys of the other nodes after the time expires. Then\nnode U will not be able to establish pairwise keys with any other nodes in the DSN so\nthat, though an adversary captures a node, the communications between it and another\nnode cannot be decrypted without the key Ki.\nPairwise Shared Keys do have computational overhead since each node U in the\nnetwork must verify the MACs generated by neighboring nodes and each must reply\nwith a message including its identity and an MAC. Each node must also generate a\npairwise key between every other neighboring node in the network. Because a HELLO\nmessage includes only a node ID and an ACK message has only an ID and an MAC,\nboth can be adjusted in a single packet. Also the space required for storing a preloaded\nis only one key Ki; therefore, the communication and storage overheads are small.\nThe HELLO message in our scheme is not authenticated, so an adversary may\ntry to attack the network by constantly sending these messages, which will drain a\nDSN’s resources. There are two solutions for this attack: the controller may try to\nload each new node with the group key of the network so that the nodes can verify the\nauthentication of the message by verifying the group key in the message, or else the\ncontroller might try to add some randomness into the IDs of the newly added nodes\n"
        },
        {
            "page_number": 369,
            "text": "WIRELESS NETWORK SECURITY\n369\nsuch that false ones will be detected and dropped. The assumption made here is that the\nsensor nodes are able to permanently eliminate the master key Ki from their memory,\nwhich may not be possible in all cases. One of the unique advantages of the scheme\nabove is that once pairwise keys are established between neighboring nodes in an area of\na DSN, they cannot be established again, which protects the network from clone attacks.\nIn clone attacks, an adversary tries to attack the network by installing a number of nodes\nwith keys acquired from compromised nodes, which then establish pairwise keys with\nother nodes in the network and compromises the entire DSN with just a few nodes.\nThe scheme stated above restricts this kind of attack to a local area as the cloned nodes\ncannot establish pairwise keys with other nodes in the network that are not neighboring\nnodes of the one compromised.\nThe security of the above scheme can be even increased by regularly changing the\nmaster key Ki. If an adversary compromises the sensor node before the establishment\ntime, the master key Ki can be obtained and then the whole network can be compro-\nmised. By changing the master key regularly, not only is this attack averted, but also\nattacks caused later by the same adversary are averted, as the master key could still be\nacquired by compromising a node and deriving the key from its memory. There is one\nmore security threat that must be addressed here: if an adversary attacks the DSN and\nsucceeds in compromising the nodes before key establishment time test, the network\ncan then be attacked through new nodes added into the network using the correct mas-\nter key. This problem, however, can be solved. Suppose that the controller wants to\nadd Ni nodes into the network in the time interval Ti, Ni ID’s are generated for the\nnodes based on a random seed Si and each of the Ni nodes is loaded with a unique\nID. The nodes can now establish pairwise keys as stated and when the controller later\nbroadcasts Ni and Si into the network using a broadcasting scheme like µTESLA, the\nnodes verify whether those attempting communications are valid or not based on Ni\nand Si. The pairwise keys of nodes that are not valid are then deleted from the memory\nof all the nodes. Compared to other approaches, this method offers greater efficiency\nand smaller overhead, while also protecting the network from clone attacks and other\nserious attacks.\nEstablishing Cluster Keys\nThe cluster key establishment is based on the pairwise key establishment. If node\nU wants to establish a cluster key with its neighbors v1, v2, v3, . . ., vn, first it generates\na key Kc and then encrypts that key using the pairwise key which it shares with each\nneighbor. Node U then transmits this encrypted message to its neighbors. Node v1\ndecrypts the key using the pairwise key which it shares with U, and then stores the key\nin a buffer. Next it sends back its own cluster key to node U. When any of the nodes\nare revoked, node U generates a new cluster key in the same way and transmits the key\nto all remaining nodes.\nEstablishing Group Keys\nA group key, which is shared between a base station and all the nodes in a DSN,\nis needed when the base station wants to send a message or query to all the nodes of\n"
        },
        {
            "page_number": 370,
            "text": "370\nVENKATA KRISHNA RAYI et al.\nthat DSN. One way of achieving this is using the hop-by-hop method in which the\nbase station encrypts messages using the cluster key which it has and then broadcasts\nthe message to all the nodes in its neighborhood. The nodes would decrypt the message\nand then encrypt it using the cluster key which they share with their neighbors. In this\nway, the message can be received by all the nodes in the network. This is efficient, but\nhas an overhead of encryption and decryption at every node.\nA simple method to establish a group key is to preload each node with the group\nkey before deployment, but this is still within the scope for rekeying the group key\nwhich will be necessary. Unicast-based group rekeying can also be considered for\nwhich the base station needs to send the group key to each node in the network, but\nthis involves much communication overhead. However, Zhu, Setia, and Jajodia [6]\nproposed an efficient scheme based on cluster keys in which the transmission cost will\nonly be one key. In DSNs, all messages sent by the base station must be authenticated\nor an adversary may impersonate it. The group key must be updated every time when\na node is revoked. Therefore the first issue to consider is how node revocation can be\ndone in this scheme. µTESLA, based on a one-way key chain and delayed disclosure of\nkeys, is an efficient method to broadcast messages into a DSN, as previously discussed.\nTo bootstrap µTESLA, each node should be preloaded with the commitment of the key\nchain. If Kg is the new group key and U is the node to be revoked, the base station\nbroadcasts the following message:\nM : Controller →∗: u, fK′\ng(0), MAC(kT\ni , u|fK′\ng(0))\n(16)\n, where fK′\ng(0) is the key that enables the node to verify the authentication of the group\nkey. The server then distributes the MAC key kT\ni after one µTESLA interval. After\na node V receives the message M, it verifies the authenticity of the message using\nµTESLA. If node V is neighbor of U, V will remove its pairwise key shared with U\nand update its cluster key. This process can also be used for updating the group key.\nFor secure key distribution, this scheme uses a protocol that is the same as the\nbeaconing protocol for which all the nodes are organized in a breadth-first spanning\ntree where each node not only remembers the parent and children of a spanning tree,\nbut also the other neighbors. The new group key K′\ng is distributed to all the nodes\nin the network using the spanning tree established by the routing protocol. The base\nstation initiates the process by sending the group key to all its neighbors in the network.\nThe nodes that receive the message verify its authentication by calculating fKg′ and\nby checking whether it is the same as the verification key received earlier in the node\nrevocation message. The algorithm continues recursively down the spanning tree with\nthe help of each node, which transmit the group key to neighbors while encrypting\nthe message using their cluster keys. As discussed earlier, this hop-by-hop scheme\ndoes not involve much overhead as only one key is encrypted and decrypted, and as\nrekeying of the group key is infrequent. However, it is desirable to change the group\nkey more often or an intruder may compromise the entire DSN by obtaining one node\nand, thereby, deriving the group key.\n"
        },
        {
            "page_number": 371,
            "text": "WIRELESS NETWORK SECURITY\n371\n6.2. Local Broadcast Authentication\nLocal broadcast is different from global broadcast in that in local broadcast a node\ngenerally does not know what packet it is going to generate next and messages generally\nconsist of aggregated sensor readings or routing protocols. µTESLA is not suitable\nfor local broadcasting because µTESLA does not provide authentication immediately,\nwhich is needed in some local broadcast cases. Also, in µTESLA nodes need to keep the\npackets in their buffers until the authenticating key arrives, which increases the storage\nspace required. A packet that has to travel L nodes will at least need L µTESLA\nintervals, thereby affecting latency of the network. Pairwise keys cannot be used for\nlocal broadcast because, if a node has n neighbors, the approach requires the sender\nnode to calculate n MACs for each message. Local broadcast needs a method where a\nnode can broadcast a message to all its neighbors using a single MAC and cluster keys,\nwith a problem as follows. If an adversary can compromise a node, the cluster key\nfrom that node is available and can be used to attack the network by impersonating that\nnode or a neighboring node. If nodes X, U, and V are three vertices of a triangle, X is\ncompromised, and U wants to send messages to X and V, X can use node U’s cluster\nkey to impersonate it and send false messages to V.\nFortunately, Zhu, Setia, and Jajodia [6] have designed a scheme called One-Way\nKey Chain-Based Authentication for defeating this attack totally. This scheme is based\non µTESLA in that each node generates a one-way key chain and sends the commitment\nof it to their neighbors. This transferring is done using the pairwise keys already shared\nwith neighbors. If a node wants to send a message to its neighbors, it attaches the next\nauthorization key from its key chain to the message. The receiving node can verify the\nvalidation of the key based on the commitment it has already received. The One-Way\nKey Chain-Based Authentication is designed based on two observations: a node only\nneeds to authenticate to its neighbors and that a node V will receive a packet before a\nneighboring X receives it and resends it to V. This observation is true because of the\ntriangular inequality among the distances of nodes involved. An adversary may still\ntry to attack the nodes by shielding node V while U is transmitting a message, and then\nlater send a modified packet to V with the same authorization key; but this attack can\nbe prevented by combining the authorization keys with the cluster keys. When this is\ndone, the adversary does not have the cluster key and so cannot impersonate node U.\nHowever, this scheme does not provide a solution for attacks from inside where the\nadversary knows U’s cluster key.\n6.3. LEAP Performance Evaluation\nOverhead\nSince the performance of the pairwise key establishment have already been dis-\ncussed, here we review other factors of performance like the computational cost, com-\nmunication cost and storage requirement of this approach. As mentioned, a cluster key\nis established based upon the pairwise keys of a node. Let us suppose that the number\nof neighbors to a node is n; if the cluster key has to be updated the scheme must perform\n"
        },
        {
            "page_number": 372,
            "text": "372\nVENKATA KRISHNA RAYI et al.\nn encryptions, which is computationally expensive. The value of n depends upon the\ndensity of the scheme; the computational cost increases as the network is denser. While\nfor securing distribution of a group key, the number of decryptions is equal to the size\nof the network. The total number of encryptions is also equal to the size of the network;\nso if the size of the network is M, the total number of symmetric operations will be\n2M. From these derivations, the computational cost of the scheme is dependent upon\nthe density of the network d. Zhu, Setia, and Jajodia [6] stated that the average number\nof symmetric operations of the scheme is about 2(d −1)2/(M −1) + 2. If the density\nof the network is reasonable, the computational cost may not be a bottleneck to the\nscheme.\nAlso, the cost decreases with the increases of M. The communication cost of the\nscheme is the same as the computational cost. The average number of keys a node\nhas to transfer for updating keys due to revocation is (d −1)2/(N −1) + 2. Just\nlike computational cost, communication cost increases with a increase in the density\nof the network and decreases with an increase in the size of the network. The storage\nrequirement of this scheme is a bit high because each node must store four types of\nkeys in it. Considering the degree of node to be d, a node has to store one individual\nkey, d pairwise keys, d cluster keys, and one group key. Also, a node must store a\none-way key chain and a commitment for each neighbor for local broadcast. If L is the\nnumber of keys stored in a key chain, the total number of keys the node has to store\nin this scheme will be 3d + 2 + L. Again, the storage requirement of LEAP depends\nupon the density of the network.\nResilience to Attack\nAn adversary might launch a selective forwarding attack in which a compromised\nnode drops the packets containing the routing information of selected nodes and for-\nwards the other packets normally. LEAP can minimize the affects of the scheme by\nminimizing this problem to a local area. As LEAP uses local broadcast, the attack’s\neffects will not transfer to more than 2-hops, which will result in defeating the purpose\nof such an attack. LEAP can also prevent a HELLO attack in which an adversary\nattacks the network by repeatedly transmitting HELLO messages and thereby depletes\nthe network’s resources. This attack is averted since the nodes in a LEAP scheme ac-\ncept packets only from authenticated neighbors. The sinkhole and wormhole attacks,\nhowever, are difficult to solve. In the sinkhole attack, a compromised node attracts\npackets by advertising information like high battery power, etc., then later drops all\nthe packets. In the wormhole attack an adversary launches two nodes in the network,\none near the target of interest and the other near the base station. The adversary then\nconvinces the nodes near the target, which would generally be multiple hops away from\nthe base station, that they are only two hops away thereby creating a sinkhole. Also,\nnodes that are far away think that they are neighbors because of the wormhole created.\nIn LEAP an adversary cannot launch a wormhole attack after key establishment as at\nthat point every node has knowledge about its neighbors so it is not easy to convince a\nnode that it is near a particular compromised node. An insider node must then succeed\n"
        },
        {
            "page_number": 373,
            "text": "WIRELESS NETWORK SECURITY\n373\nin compromising two nodes for creating a wormhole and those nodes must be near the\ntarget of interest and the base station after the key establishment phase is complete.\nAlthough an adversary may try, it is difficult to create an attractive sinkhole without\nbeing detected.\nLEAP includes efficient protocols for supporting four types of key schemes for\ndifferent types of messages broadcasted in DSNs and includes an efficient scheme for\nlocal broadcast authentication. LEAP is an efficient scheme for key establishment that\nresists many types of attacks on the network, including the sybil, sinkhole, wormhole,\nand so on. LEAP also provides efficient schemes for node revocation and key updating\nin DSNs.\n7.\nA KEY MANAGEMENT SCHEME FOR WIRELESS DSNS USING\nDEPLOYMENT KNOWLEDGE\nThroughout the discussion in this chapter, a significant piece of information re-\ngarding DSNs has not yet been mentioned, i.e., the deployment knowledge of these\nnetworks. As sensor nodes are randomly deployed in an area, it is difficult to obtain\ndeployment knowledge. Some information on deployment knowledge is achievable\nif deployment followed a particular order. For example, if sensor nodes are scattered\nusing an airplane pattern, these nodes might be grouped or placed in a particular or-\nder before deployment and, based on this pattern, an approximate knowledge of node\nposition can be acquired.\nDeployment knowledge offers numerous advantages when used in DSNs such as\nachieving better storage, better resilience to node capture and more. In their study of\nkey establishment techniques in sensor networks, Du, Deng, Han, Chen, and Varshney\n[8] propose a scheme using deployment knowledge that is based on the Basic Scheme,\nwhich has already been discussed earlier in this chapter. Deployment knowledge in\nthis scheme is modeled using probability density functions (pdfs). All the schemes\ndiscussed until now considered the pdf to be uniform; and when uniform, knowledge\nabout the nodes can not be derived from it. In LEAP, Du et al. consider non-uniform\npdfs, which means that they assume the position of sensor nodes to be at certain areas.\nTheir method first models node deployment knowledge in a DSN and then develops a\nkey predistribution scheme based on this model.\n7.1. Modeling of the Deployment Knowledge\nThe deployment point and the resident point are two terms that must be briefly\nunderstood when discussing the Deployment Model. Deployment point of a sensor\nnode is the point at which the sensor node is actually deployed; i.e., the node is dropped\nwhere the deployment is done through an airplane deployment point. Resident point\nis the point at which the sensor actually resides after deployment. Let us assume\nthe deployment area to be a 2-dimensional region X × Y . The pdf for node I, for\nI = 1, ..., N over the two-dimensional area is found by fi(x, y), where x ∈[0, X]\nand y ∈[0, Y ]. Generally nodes are deployed in groups, therefore the pdfs of the final\n"
        },
        {
            "page_number": 374,
            "text": "374\nVENKATA KRISHNA RAYI et al.\nresident points of all the sensors in a group is the same as the group of sensors deployed\nin a single deployment point. The group deployment model is designed as following\nin Du et al. [8]:\nN sensor nodes that are deployed in a place are divided into t × n equal size\ngroups. Each group Gi,j for i = 1, ..., t and j = 1, ..., n is from the deployment\npoint with index (i, j); and (xi, yj) is the deployment point for this group.\nThe Deployment Model follows a grid-based approach with all deployment\npoints arranged in a grid.\nThe pdf of the resident points for node K in group Gi,j is f ij\nK(x, y|K ∈Gi,j) =\nf(x −xi, y −yi).\nTwo groups that are deployed close together share some common keys. The amount\nof key overlap decreases as the deployment distance between the groups increases.\nWhen using the Basic Scheme, keys are drawn from the same key pool S; but, using\nthe Deployment Model, different key pools are allowed for different groups so that the\nkey pool can be divided into sub-key pools of |Sc| keys each. The combination of all\nthe sub-key pools still yields S.\nSensor nodes can be deployed in many different ways such as deployment through\nan airplane, using a vehicle, etc. In this scheme, deployment is considered as a Gaussian\ndistribution, which is widely studied and practiced. The deployment distribution for\nany node k in group Gi,j follows a two dimensional Gaussian distribution. The pdf of\nthe resident points for the node k in group Gi,j is [22]\nf ij\nk (x, y|k ∈Gi,j) = f(x −xi, y −yj)\n(17)\nWhen f(x, y) is uniform, we cannot determine which nodes are close together\nprior deployment as the resident points of the nodes are uniformly distributed over the\nregion. When f(x, y) is random we can tell which nodes are close together. Though the\ndistribution function is not uniform, the sensor nodes still need to be deployed evenly\nthrough the entire region. By selecting an appropriate distance between deployment\npoints, the probability of finding a node in each small region can be made equal.\n7.2. Key Predistribution Using Deployment Knowledge\nIn a key pre-distribution scheme based on the Deployment Model, it is assumed\nthat N sensor nodes are deployed in a place (point) and are divided into t × n equal\nsize groups, each group Gi,j for i = 1, ..., t and j = 1, ..., n. It is also assumed that\nthe deployment points are arranged in a grid.\nAs with the Basic Scheme, key predistribution in the Deployment Model also\nconsists of three phases: key predistribution, shared key discovery, and path key es-\ntablishment. This scheme differs only in the first stage while the other two stages are\nsimilar to that of the basic scheme.\n"
        },
        {
            "page_number": 375,
            "text": "WIRELESS NETWORK SECURITY\n375\nKey Predistribution: The most important step in this phase is to divide the key\npool into t × n key pools. The goal of dividing the key pools is to ensure\nthat neighboring key pools have more keys in common. Two key pools are\nneighbors if their deployment groups have nearby resident points. The concept\nof dividing the key pool is discussed briefly in the next section. After the key\npool is divided, each node in a group is selected and keys are installed from\ncorresponding subset key pools.\nShared Discovery Phase: In this phase each node must find its common keys\nshared with neighbors. There are many ways for doing this, but the simplest\nis to make the nodes broadcast their identifiers list to other nodes. If the nodes\ndiscover that they share a common key with other nodes, this key can be used as\ntheir communication link. When disclosing the nodes’ identities is not desired,\nMerkle’s Challenge Response Technique can be employed [9] in which each\nnode sends a puzzle to neighboring nodes for each stored key and, if those\nnodes share a key in common with the source node, they will respond with the\ncorrect solution creating a key link for secure communication.\nPath Key Establishment: When two neighboring nodes do not share a common\nkey, they can discover one using Path Key Establishment. If node U wants to\ncommunicate with V and the two do not share a common key, node U must\ncommunicate with its neighbor I, saying that it wants to communicate with V.\nNode U then sends its ID and a secret key to node I and, if I shares a common\nkey with V, it sends the message to V encrypted with that shared key. Through\nthis path U →I →V or V →I →U, both nodes U and V can communicate\nwith each other using a secret key.\n7.3. Creating Key Pools\nKey pools that are deployed nearby should share certain keys in common. To\nassign keys to each key pool Si,j for i = 1, ..., t and j = 1, ..., n, it is assumed that the\npools are deployed in a grid: (a) key pools that are horizontal or vertical share a|Sc|\nkeys, where 0 ≤α ≤0.25; (b) Key pools that are diagonal share b|Sc| keys, where\n0 ≤b ≤0.25 and 4a + 4b = 1; (c)Two non-neighboring key pools share no keys.\nHere, (a) and (b) are overlapping steps and to achieve the properties stated, the key\npool is divided into eight total partitions, each with keys that are shared by the other\nnodes. Du et al. [8] developed a method to select keys for each subset key pool Si,j,\nconsidering a grid scheme and given a global key pool S (the subset of key pools for\neach deployment point). The keys for the first subgroup (the group placed in the first\nrow and first column) S1,1 are selected from the global key pool S, and then keys for the\nsecond group in the same row are selected from the row left to it and S. This process\ncontinues for each row from left to right:\n"
        },
        {
            "page_number": 376,
            "text": "376\nVENKATA KRISHNA RAYI et al.\nSelect |Sc| keys for the group placed in S1,1 and remove those keys from S;\nSelect a|Sc| keys for group S1,2 from the key pool S1,1, and the remaining keys\nw from the global key pool S, and then remove the selected w keys from S.\nSelect a|Sc| keys for group S2,1 from the key pools S1,1, S3,1, S2,2 and select\nb|Sc| from the key pools S1,2 and S3,2. Then select and remove the remaining\nw keys from the global key pool S.\nIf w is the remaining keys that are to be selected from S, and Si,j be the group\nnumber in a grid, the selection procedure for different groups will be:\nW =\n⎧\n⎪\n⎪\n⎨\n⎪\n⎪\n⎩\n[1 −(a + b)] × |Sc|, forj = 1\n[1 −2 (a + b)] × |Sc|, for2 <j ≤n −1\n[1 −(2a + b)] × |Sc|, forj = n\n(18)\nSelecting the size of the key pool |Sc| from S is also critical in this scheme. To ensure\nthat no key in the network is shared between more than two nodes, a rule must be\nestablished: if a group G1 selects the required keys from its neighbor G2, no other\ngroup is allowed to select those keys. Since each group selects distinct keys from their\nneighbors, the size of |Sc| is equal to the sum of all those keys and is calculated as\n|Sc| =\n|S|\ntn −(2tn −t −n)a −2(tn −t −n + 1)b\n(19)\nIn their paper, Du et al. [8] analyzed the performance of their scheme based upon\nthe connectivity, communication overhead, and resilience against node capture of the\nnetwork. Two parameters for connectivity called the Global Connectivity and Local\nConnectivity are defined for analyzing the performance of connectivity in the Deploy-\nment Model. Global Connectivity is the ratio between the nodes that are isolated and\nthe total number of nodes in the network. Local Connectivity refers to the probability\nof any two nodes sharing at least one key. Both Global and Local Connectivity are\naffected by the key predistribution scheme where B(ni, nj) is the event that two nodes\nshare at least one key, and A(ni, nj) is the event that both the nodes are neighbors,\nhence\nplocal = Pr(B(ni, nj)|A(ni, nj))\n(20)\nLet λ be the ratio of the shared key pool between two nodes and |Sc|. Du et al. [8]\ncalculated the probability p(λ) that two nodes share at least one key when they have\nλ|Sc| in common as\np(λ) = 1 −Pr(two nodes do not share any key)\n(21)\nGlobal Connectivity is also required for better efficiency. The key sharing in a wireless\nDSN the using Deployment Model is not uniform, so uniform distribution techniques\n"
        },
        {
            "page_number": 377,
            "text": "WIRELESS NETWORK SECURITY\n377\nlike Erdos’ Random Graph Theory cannot be applied. Shakkottai et al. [23] determined\nthe connectivity of a wireless network with unreliable nodes that can be used for random\ndistribution. The authors simulated the results for Global Connectivity using m for the\nnumber of nodes in a node’s key ring, which showed that when m = 200, no nodes are\nwasted due to a lack of security links while only 0.12 percent of nodes are wasted when\nm = 100. The overlapping factors a and b has to be selected carefully, when a = 0.25\nand b = 0, each group shares keys only with horizontal and vertical neighbors only;\nwhen a = 0 and b = 0.25, each group shares keys only with diagonal neighbors only.\nThe values of a and b depend upon many factors like the size of the nodes key ring m,\ndifferent types of neighbors, etc.\n8.\nCONCLUSIONS\nKey management for Distributed Sensor Networks (DSNs) is a critical issue that\nhas been addressed through several proposed schemes presented in various papers.\nThis chapter provides an overview of these techniques, each of which offers different\nadvantages and disadvantages. A balance between the requirements and resources of\na DSN determines which key management scheme should be employed. A DSN used\nin a battlefield demands more security than one used in places like shopping centers;\nalso, the former can be made more costly but the later needs to be as cheap as possible.\nDecisions regarding the key management scheme to be used must be based on these\nrequirements for efficiency.\n8.1. Summary of Schemes Pros and Cons\nThe Basic Scheme\nPros: flexible, efficient, and fairly simple to employ, while also offering good\nscalability.\nCons: cannot be used in circumstances demanding heightened security and\nnode-node authentication.\nThe Q-Composite Scheme\nPros: provides better security than the Basic Scheme by requiring more keys\nfor two nodes to share one for communication, which makes it difficult for an\nadversary to compromise a node.\nCons: vulnerable to breakdown under large-scale attacks, does not satisfy scal-\nability requirements.\nMultipath Key Reinforcement\nPros: offers better security than the Basic Scheme or the Q-Composite.\n"
        },
        {
            "page_number": 378,
            "text": "378\nVENKATA KRISHNA RAYI et al.\nCons: creates communication overhead that can lead to depleted node battery\nlife and to the chance for an adversary to launch DOS attacks.\nRandom Pairwise Scheme\nPros: offers the best security of all the schemes with perfect resilience against\nnode capture as the keys used by each node are unique, and also provides\nresistance against node replication.\nCons: does not support networks of large size, and does not satisfy scalability\nrequirements.\nPolynomial Pool-Based Key Predistribution\nPros: allows the network to grow to a larger size after deployment.\nCons: t-collision resistant (compromising more than t polynomials leads to\nnetwork compromise.\nPolynomial Grid-Based Key Predistribution\nPros: offers low communication and computation overhead on the network,\nguarantees a total connected graph when there are no compromised nodes and\nall the nodes are in transmission range.\nCons: overhead of node storage as each must not only share the polynomial\nkeys but also the ID’s of compromised nodes to avoid attack.\nSPINS: SNEP and µTESLA\nPros: one of the best memory efficient schemes discussed, provides strong\nsecurity features with less complexity, universal design allows use in many\nlow-end devices, incurs low communication cost, and offers authentication and\nstrong data freshness with a minimum overhead.\nCons: µTESLA overhead from releasing keys after a certain delay, possible\nmessage delay.\nLEAP\nPros: offers efficient protocols for supporting four types of key schemes for\ndifferent types of messages broadcasted, reduces battery usage and communi-\ncation overhead through In-Network Processing, and uses a variant of µTESLA\nto provide local broadcast authentication.\nCons: requires excessive storage with each node storing four types of keys and a\none-way key chain, computation and communication overhead dependant upon\nnetwork density (the more dense a network, the more overhead)\n"
        },
        {
            "page_number": 379,
            "text": "WIRELESS NETWORK SECURITY\n379\nWireless DSNs Using Deployment Model\nPros: only to consider deployment knowledge that can minimize the number\nof keys and help increase resilience or resistance to node capture and reduce\nnetwork overhead, increases overall connectivity of the network graph, and\noffers same pros as the Basic Scheme on which it is based\n8.2. Future Work for DSNs\nMore schemes should be developed to make efficient use of sensor nodes’ lim-\nited resources. Greater emphasis should be given to the security in key management\nschemes, particularly as a majority of sensor node deployment is in hostile environ-\nments where providing strong security features is a must. Though receiving much\nattention recently, there are many problems to be addressed in DSNs such as finding\nthe compromised nodes in a network, making good use of deployment knowledge,\nmaking nodes tamperproof without much overhead, decreasing the bootstrapping time\nrequired for the network, and so on. Future research should especially seek techniques\nfor compromised node discovery and efficient methods to revoke compromised nodes.\nNumerous lives can be saved in wars with the data collected by sensors, but DSNs\nin the near future will offer many surprises for humans as they come to be used in\ndaily household matters like locking doors and switching off electronics, or controlling\ntraffic in high-volume areas. Sensors installed in big malls and shopping centers can\nguide people to their required products easily while those in hospitals can monitor\npatientconditionandthoseinforestscanprovideimmediateknowledgeaboutdisastrous\nhazards like wildfire. These advantages are only a small fraction of what DSNs could\npotentially offer when deployed more commonly. Future studies can prove DSNs to\nbe useful in a wider variety of environments.\n9.\nREFERENCES\n1. L. Eschenauer and V. D. Gligor. “A key management scheme for distributed sensor networks\", Proc. of\nthe 9th ACM Conference on Computerand Communication Security.\n2. H. Chan, A. Perrig, and D. Song, “Random Key Predistribution Schemes for Sensor Networks\", Proc.\nof the 2003 IEEE Symposium on Security and Privacy, May 11-14, pp. 197 - 213.\n3. D. Liu and P. Ning, “Establishing pairwise keys in distributed sensor networks,\" Proc. of the 10th ACM\nConference on Computer and Communications Security (CCS ’03), pp. 52-61. 2003.\n4. W. Du, J. Deng, Y. S. Han, and P. K. Varshney, “A pairwise key pre-distribution scheme for wireless sen-\nsor networks,\" Proc. of the 10th ACM Conference on Computer and Communications (SecurityCCS’03),\npp. 42-51, 2003.\n5. A. Perrig, et. al, “SPINS: Security Protocols for Sensor Networks,\"Proc. of ACM MOBICOM, 2001.\n6. S. Zhu, S. Setia, and S. Jajodia, “LEAP: Efficient Security Mechanisms for Large-Scale Distributed\nSensor Networks,\" Proc. of The 10th ACM Conference on Computer and Communications Security\n(CCS ’03), Washington D.C., October, 2003.\n7. D. Carman, P. Kruus, and B. Matt, “Constraints and approaches for distributed sensor network security,\"\nNAI Labs Technical Report No. 00-010, September 2000.\n"
        },
        {
            "page_number": 380,
            "text": "380\nVENKATA KRISHNA RAYI et al.\n8. W. Du, J. Deng, Y. S. Han, S. Chen and P. K. Varshney, “A Key Management Scheme for Wireless\nSensor Networks Using Deployment Knowledge,\" Proc. of IEEE INFOCOM 2004.\n9. R. Merkle, “Secure communication over insecure channels,\" Communications of the ACM, Vol. 21, No.\n4, pp.294-299, 1978.\n10. R. Anderson and A. Perrig, “Key infection: Smart trust for smart dust,\" Unpublished Manuscript,\nNovember 2001.\n11. R. Merkle, “Protocols for public key cryptosystems \" Proc. of 1980 IEEE Symposium on Security and\nPrivacy, 1980.\n12. W. Diffie and M. E. Hellman, “New directions in cryptography,\" IEEE Trans. Inform. Theory, IT-22,\npp. 644-654, November 1976.\n13. C. Karlof and D. Wagner, “Secure Routing in Sensor Networks: Attacks and Countermeasures,\" First\nIEEE International Workshop on Sensor Network Protocols and Applications, May 11, 2003.\n14. A.D. Wood and J.A. Stankovic, “Denial of service in sensor networks,\" Computer, Vol.35, No. 10, Oct.\n2002, pp. 54 - 62.\n15. C. Karlof, N. Sastry, and D. Wagner, “TinySec: A Link Layer Security Architecture for Wireless Sensor\nNetworks,\" Proc. of the Second ACM Conference on Embedded Networked Sensor Systems (SenSys\n2004). November 2004.\n16. F. Ye, H. Luo, S. Lu, and L. Zhang, “Statistical En-route Detection and Filtering of Injected False Data\nin Sensor Networks,\" Proc. of IEEE INFOCOM 2004.\n17. D.W. Carman, B.J. Matt, and G.H. Cirincione, “Energy-efficient and Low-latency Key Management\nfor Sensor Networks,\" Proc. of 23rd Army Science Conference, 2002.\n18. M. Chen, W. Cui, V. Wen, and A. Woo, “Security and Deployment Issues in a Sensor Network,\" Ninja\nProject, A Scalable Internet ServicesArchitecture, Berkeley,\n19. W. Zhang and Guohong Cao, “Group Rekeying for Filtering False Data in Sensor Networks: A Predis-\ntribution and Local Collaboration-Based Approach,\" Proc. of INFOCOM 2005.\n20. D. Liu, P. Ning, K. Sun, “Efficient self-healing group key distribution with revocation capability,\" Proc.\nof the 10th ACM conference on Computer and communication security, 2003.\n21. J. Lee and D. R. Stinson, “Deterministic key pre-distribution schemes for distributed sensor networks,\"\nTo appear in Lecture Notes in Computer Science (SAC 2004 Proceedings), 2004.\n22. A. Leon-Garcia, Probability and Random Processes for Electrical Engineering, 2nd ed. Reading, MA:\nAddison-Wesley Publishing Company Inc., 1994.\n"
        },
        {
            "page_number": 381,
            "text": "15\nSECURE ROUTING IN AD HOC\nAND SENSOR NETWORKS\nXu (Kevin) Su\nDepartment of Computer Science\nThe University of Texas at San Antonio\nE-mail: xsu@cs.utsa.edu\nYang Xiao\nComputer Science Department\nUniversity of Alabama\n101 Houser Hall\nBox 870290\nTuscaloosa, AL 35487-0290 USA\nE-mail: yangxiao@ieee.org\nRajendra V. Boppana\nDepartment of Computer Science\nThe University of Texas at San Antonio\nE-mail: boppana@cs.utsa.edu\nBoth mobile ad hoc networks and wireless sensor networks are gaining popularity due to the\nfact that they promise viable solutions to a variety of applications. A Mobile Ad hoc NET-\nwork (MANET) consists of a collection of wireless hosts that are capable of communicating\nwith each other without use of a network infrastructure or any centralized administration. A\nwireless sensor network (WSN) consists of tiny senor nodes with sensing, computation, and\nwireless communication capabilities. In sensor networks, hundreds or thousands of nodes\nwith low-power, low-cost, and possibly mobile but more likely at fixed locations, collec-\ntively monitor an area. However, both MANETs and WSNs are vulnerable to different\nattacks due to their fundamental characteristics such as open medium, dynamic topology,\nabsence of central administration, distributed cooperation, and constrained capability. Be-\nfore mobile ad hoc and sensor networks can be successfully deployed, security issues must\nbe addressed. In this chapter we introduce security attacks on MANETs and WSNs and\nreview the recent approaches of secure network routing protocols in both mobile ad hoc\nand sensor networks.\n"
        },
        {
            "page_number": 382,
            "text": "382\nXU SU et al.\n1.\nINTRODUCTION\nBoth mobile ad hoc networks and wireless sensor networks are gaining popularity\ndue to the fact that they promise viable solutions to a variety of applications. A Mobile\nAd hoc NETwork (MANET) consists of a collection of wireless hosts that are capable\nof communicating with each other without use of a network infrastructure or any cen-\ntralized administration. In such a network, each mobile node can operate not only as a\nhost but also as a router, forwarding packets for other mobile nodes which may not be\nwithin direct wireless transmission range of each other. Since MANETs can be easily\nand inexpensively set up as needed, they have a wide range of applications, especially\nin situations with geographical or terrestrial constrains such as battlefields, military\napplications, and other emergency and disaster situations. A wireless sensor network\nconsists of tiny senor nodes with sensing, computation, and wireless communication\ncapabilities. In sensor networks, hundreds or thousands of nodes with low-power, low-\ncost, and possibly mobile but more likely at fixed locations, collectively monitor an\narea. These large sensor networks generate a substantial amount of data and send the\ndata to one or more points of centralized control called base stations or sinks. A base\nstation is typically a gateway to another network, and it can be a powerful data process-\ning or storage center, or an access point for human interface. Networking unattended\nsensor nodes are increasingly popular to provide economical solutions to both military\nand civil applications such as real-time traffic monitoring, wildfire tracking, wildlife\nmonitoring, weather monitoring, security and tactical surveillance, target filed imaging,\nand disaster management etc.\nHowever, recent wireless research work indicates that the ad hoc networks and sen-\nsor networks are more vulnerable than conventional wired and wireless networks due to\ntheir fundamental characteristics such as open medium, dynamic topology, absence of\ncentral administration, distributed cooperation, and constrained capability (power and\ncomputation constraints). Use of wireless links makes ad hoc and sensor networks more\nprone to physical security threats than wired networks, ranging from passive eavesdrop-\nping to active interference. Without proper security, mobile hosts and sensor nodes are\neasily captured, compromised and hijacked by malicious nodes. The adversary might\nlisten to or/and modify the messages in the wireless communication channel, inject\nerroneous messages, delete messages, and even impersonate other nodes. Some or all\nof the nodes in a MANET might rely on batteries and all sensor nodes in a WSN have\nlow battery power. The adversary could force a node to replay packets to exhaust its\nenergy.\nBefore ad hoc and sensor networks can be successfully deployed, security issues\nmust be addressed. Recently significant progress has been made in securing ad hoc and\nsensor networks via the development of secure routing protocols.\nTo assess the security needs of ad hoc and sensor networks effectively and to\nevaluate and choose various policies, we need some systematic way of defining the\nrequirements for security and characterizing the approaches to satisfying those require-\nments. Usually the following three aspects of information security are considered:\n"
        },
        {
            "page_number": 383,
            "text": "WIRELESS NETWORK SECURITY\n383\nSecurity attack: any action to compromise the security of information in the\nsystem.\nSecurity mechanism: a mechanism that is designed to detect, prevent, or\nrecover from a security attack.\nSecurity service: a service that enhances the security of the data processing\nsystems and information transfers of an system. These services are used to\ncounter security attacks, and they make use of one or more security mechanisms\nto provide the service.\nOur purpose in this chapter is to review the recent approaches to secure routing\nprotocols in ad hoc and sensor networks. The rest of the chapter is organized as follows.\nWe first give an overview of current routing protocols without security enforcement\nand describe several often used routing protocols on which most recent secure routing\nprotocols are based in Section 3. In Section 4, different security services are introduced.\nIn Section 5, security attacks on ad hoc and sensor networks are classified and some\nwell-known attacks are given. We then give an review on current security mechanisms\nto secure routing protocols in ad hoc and sensor networks in Section 6. Finally, Section\n7 contains concluding remarks.\n2.\nROUTING PROTOCOLS\nDuring the last decade, extensive research works have been conducted on designing\nrouting protocols in both mobile ad hoc networks and wireless sensor networks, and\nmany efficient routing protocols have been proposed.\nIn MANETs, the routing protocols can be categorized by the timing of acquisition\nof routing information and the methods by which the routes are maintained into three\ncategories: reactive protocol, proactive protocol, and hybrid protocol. In the reactive\n(on-demand) protocols, such as Ad hoc On-demand Distance Vector (AODV) [39],\nDynamic Source Routing (DSR) [22], and Temporally Ordered Routing Algorithm\n(TORA) [37], a route is created only when desired by the source node. In the proactive\nprotocol, such as Destination Sequence Distance Vector (DSDV) [38] and Adaptive\nDistance Vector (ADV) [3] etc., each node maintains one or more routing tables to\nstore routing information, and any changes in the network topology need to be reflected\nby propagating updates throughout the network. There are other hybrid protocols, such\nas Zone-based Routing Protocol (ZRP) [14], which employ both mechanisms.\nIn WSNs, routing can be divided into flat-based routing, hierar-chical-based rout-\ning, and location-based routing depending on the network structure. In flat-based rout-\ning protocols, such as Sensor Protocols for Information via Negotiation (SPIN) [15, 27],\nDirected Diffusion [21], Rumor Routing [4], Minimum Cost Forwarding Algorithm\n(MCFA) [48] etc., all nodes are typically assigned equal roles or functionality. While in\nhierarchical-based routing protocols, such as LEACH protocol [16], Power-Efficient\nGathering in Sensor Information Systems (PEGASIS) [28], and Threshold-sensitive\nEnergy Efficient Sensor Network Protocol (TEEN) [30] etc., nodes play different roles.\n"
        },
        {
            "page_number": 384,
            "text": "384\nXU SU et al.\nFor example, higher energy nodes can be used to process and send the information\nwhile low energy nodes can be used to perform the sensing in the proximity of the\ntarget. In location-based routing protocols, such as GPSR [26], GEAR [51], SPAN [9],\nGAF [45], and CEC [46] etc., sensor nodes’ positions are exploited to route data in the\nnetwork. Furthermore, these protocols can be classified into multipath-based, query-\nbased, negotiation-based, QoS-based, or coherent-based routing approaches depend-\ning on the protocol operation.\nSince most recent work to secure ad hoc and sensor network routing protocols\nare based on the several main routing protocols: AODV, DSR, DSDV, and Directed\nDiffusion, we describe these routing protocols as follows.\n2.1. Destination Sequence Distance Vector – DSDV\nIn DSDV [38] routing, each router maintains a routing table giving the distance\nfrom itself to all possible destinations. Each routing table entry consists of destination\nIP address, the distance to it and the next node in the path. Each router periodically\nbroadcasts this table information to each of its neighbor routers, and uses similar routing\nupdates received from neighbors to update its table. DSDV solves the looping prob-\nlem by attaching sequence numbers to routing entries. A node increments its current\nsequence number and includes it in the updates originated at that node. Along with\ndistance information this sequence number is propagated. Any node that invalidates\nits entry to a destination because of loss of next hop node, increments the sequence\nnumber and uses the new sequence number in its next advertisement of this route. A\nnode invalidates or modifies its routing entry if a neighbor broadcasts a routing entry\nto the same destination with a higher sequence number. An invalid entry can become\nvalid when the node receives an advertisement of this route with the same sequence\nnumber (as the one it has) and better metric or higher sequence number. The rout-\ning table entries in all the nodes for a given destination collectively specify a virtual\ndestination-based tree to send packets to that destination. A simplistic view of DSDV\nis that it maintains one such destination tree for each node in a distributed manner. To\nkeep up with network changes, DSDV algorithms use periodic and triggered routing\nupdates. Periodic updates include full routing table and occur once in 30-90 seconds.\nTriggered updates occur in between periodic updates if enough number of routing en-\ntries has changed and often include only newly modified entries. ADV [3] is another\ndistance vector algorithm, which is similar to DSDV but reduces the routing overhead\nby varying the size and frequency of routing updates in response to traffic variance and\nnode mobility.\n2.2. Ad hoc On-demand Distance Vector – AODV\nAODV [39] is based upon the distance vector algorithm. The difference is that\nAODV is reactive, as opposed to proactive protocols like DSDV, i.e., AODV requests a\nroute only when needed and does not require nodes to maintain routes to destinations\nthat are not actively used in communications.\n"
        },
        {
            "page_number": 385,
            "text": "WIRELESS NETWORK SECURITY\n385\nRoute discovery: A node broadcasts a RREQ when it determines that it needs\na route to a destination and does not have one available. This can happen if the\ndestination is previously unknown to the node, or if a previously valid route\nto the destination expires. To prevent unnecessary broadcasts of RREQs the\nsource node uses an expanding ring search technique as an optimization. In the\nexpanding ring search, increasingly larger neighborhoods are searched to find\nthe destination. The search is controlled by the time-to-live (TTL) field in the\nIP header of the RREQ packets. If the route to a previously known destination\nis needed, the prior hop-wise distance is used to optimize the search.\nRoute maintenance: Every routing table entry maintains a route expiry time\nwhich indicates the time until which the route is valid. Each time that route\nis used to forward a data packet, its expiry time is updated to be the current\ntime plus ACTIVE ROUTE TIMEOUT. A routing table entry is invalidated if\nit is not used within such expiry time. AODV uses an active neighbor node list\nfor each routing entry to keep track of the neighbors that are using the entry to\nroute data packets. These nodes are notified with route error (RERR) packets\nwhen the link to the next hop node is broken. Each such neighbor node, in turn,\nforwards the RERR to active neighbors in its own list, thus invalidating all the\nroutes using the broken link.\n2.3. Dynamic Source Routing – DSR\nA routing entry in DSR [22] contains all the intermediate nodes to be visited by\na packet rather than just the next hop information maintained in DSDV and AODV. A\nsource puts the entire routing path in the data packet, and the packet is sent through\nthe intermediate nodes specified in the path. If the source does not have a routing\npath to the destination, then it performs a route discovery by flooding the network with\na route request (RREQ) packet. The RREQs record route information as they visit\nintermediate nodes on the way to the destination. Any node that has a path to the\ndestination in question can reply to the RREQ packet by sending a route reply (RREP)\npacket. The reply is sent using the route recorded in the RREQ packet. A node that\nreceives a RREQ can use the path recorded to improve its path to the source. To reduce\nthe cost of route discovery, each node maintains a cache of source routes which it has\nlearned or overheard. It aggressively uses to limit the frequency and propagation of\nRREQs. When an intermediate node discovers that a source route is broken, the source\nnode is notified with a route error (RERR) packet. The source node can then attempt\nto use any other route to destination already in its cache or can invoke route discovery\nagain to find a new route. To limit the need for route discovery, DSR also allows nodes\nto operate their network interfaces in promiscuous mode and snoop all (including data)\npackets sent by their neighbors. Since complete paths are indicated in data packets,\nsnooping can be very helpful in keeping the paths in the route cache fresh. To further\nreduce the cost of route discovery, the RREQs are initially broadcasted to neighbors\n"
        },
        {
            "page_number": 386,
            "text": "386\nXU SU et al.\nonly (zero-ring search), and then to the entire network if no reply is received. Another\noptimization feasible with DSR is the gratuitous route replies; when a node overhears\na packet containing its address in the unused portion of the path in the packet header,\nit sends the shorter path information to the source of the packet. Also, an intermediate\nnode may replace a packet’s current path specification with an alternate path if it is\nunable to send the packet to the original next hop node.\n2.4. Directed Diffusion\nDirected Diffusion protocol is a data-centric routing protocol in which all data\ngenerated by sensor nodes is named by attribute-value pairs. This protocol can be\ndivided into four phases as follows,\nInterest Propagation. Interest describes a task required to be done by the\nnetwork. It can be defined by a list of attribute-value pairs such as name of\nobjects, interval, duration, geographical area, etc. For each active task, the sink\nperiodically broadcasts an interest message to each of its neighbors. Every node\nmaintains an interest cache. When a node receives an interest, it first checks\nto see if the interest exists in the cache. If no matching entry exists, the node\ncreates an interest entry and save the parameters from the received interest.\nEach sensor that receives the interest setups a gradient toward the sensor nodes\nfrom which it receives the interest. This process continues until gradients are\nsetup from the sources back to the sink or Base Station (BS). More generally,\na gradient specifies an attribute value and a direction.\nLow-rate data Propagation and routing setup. In this phase, when a sensor\nnode detects a target, it searches its interest cache for a matching interest entry.\nIf matching entry exists, the node sends low-rate data to the nodes for whom it\nhas a gradient. A node that receives a data message from its neighbor attempts to\nfind a matching interest entry in its cache. If no match exists, the data message\nis silently dropped. If a match exists, the node adds the received message to\nthe data cache and resends it to its neighbors.\nReinforcement. After the low-rate data reaches the sink or BS, the sink selects\nand reinforces one particular (e.g., the best) path in order to draw down higher\nquality events (e.g., higher-rate data).\nData Propagation. In this phase, the source node computes the highest re-\nquested event rate among all its outgoing gradients and sends them to its neigh-\nbors. The node which receives the message examines the matching interest\nentry’s gradient list. If there is a lower data rate than the received data rate, it\nmay down-convert the data to the appropriate gradient. And it also does some\nin-network data aggregation before resending the message in order to make the\ndata aggregation more efficient.\n"
        },
        {
            "page_number": 387,
            "text": "WIRELESS NETWORK SECURITY\n387\n3.\nSECURITY SERVICE\nHighly level security requirements for ad hoc and sensor networks are basically\nidentical to security requirements for any other communication systems. They can be\nclassified as follows:\nConfidentiality: Ensures that the information is accessible only to authorized\nentities. Transmission of sensitive information (data packets) in ad hoc and\nsensor networks requires confidentiality. Disclosure of such information to\nenemies could cause devastating consequences. Routing information (control\npackets) must also remain confidential to certain extent, since such information\ncan be used by the enemies to identify and locate their attacking targets.\nAuthentication: Ensures that the origin of a message is correctly identified\nand its identity is not forged. Without authentication, the malicious node could\nimpersonate other node to gain unauthorized access to resource and sensitive\ninformation.\nIntegrity: Ensures that only authorized identities are able to modify system\nassets and transmit information. A message could be corrupted because of link\nfailure, or malicious attack.\nData Freshness: Ensures the freshness of messages.\nNonrepudiation: Ensures that the origin of a message can not deny having\nsent the message. It is very useful to detect and isolate compromised nodes.\nAccess control: Requires that access to information resources may be con-\ntrolled by or for the target system. To achieve this control, each entity trying to\ngain access must be first identified, or authenticated, so that access rights can\nbe tailored to the individual.\nAvailability: Ensures the survivability of network service despite malicious\nattacks (e.g., denial-of-service attacks).\n4.\nSECURITY ATTACK\nExisting routing protocols are subject to a lot of security attacks. The attacks can be\nclassified in different ways. One way is to divide attacks into four categories according\nto where the attacker deploys the attack in the flow of information from a source to a\ndestination shown in Figure 1:\nInterruption: An asset of the network is destroyed or becomes unavailable\nor unusable.\nThis is an attack on availability.\nExamples include silently\ndiscarding control or data packets.\n"
        },
        {
            "page_number": 388,
            "text": "388\nXU SU et al.\nSource\nInformation\nInformation\ndestination\n(a) Normal flow\n(b) Interruption\n(c) Interception\n(d) Modification\n(e) Fabrication\nFigure 1. Security Threats.\nInterception: An unauthorized node gains access to an asset of the network.\nThis is an attack on confidentiality. Examples include eavesdropping control\nor data packets in the networks.\nModification: An unauthorized node not only gains access to but tampers with\nan asset. This is an attack on integrity. Examples include modifying control\nor data packets.\nFabrication: An unauthorized node inserts counterfeit objects into the system.\nThis is an attack on authenticity. Examples include inserting false routing\nmessages into the network or impersonating other node.\nA more useful categorization of these attacks is in terms of passive attacks and\nactive attacks [43, 10], shown in Figure 2:\nPassive attacks: A passive attack does not disrupt the operations of a routing\nprotocol, but only attempts to discover valuable information by eavesdropping,\nor silently discard messages received. Three types of passive attacks are release\nof message contents, traffic analysis, and message dropping.\n–\nThe release of message contents is easily understood. The malicious\nnode may leak confidential information to unauthorized users in the net-\nwork, such as routing or location information. We would like to prevent\nthe enemies from learning the contents of the sensitive data.\n"
        },
        {
            "page_number": 389,
            "text": "WIRELESS NETWORK SECURITY\n389\nTraffic\nAnalysis\nMessage\nDropping\nRelease of \nMessage Content\n(authenticity)\nModification\n(authenticity)\nDenial−of−Service\n  (availability)\n   Replay\n  (integrity)\n Masquerade\nPassive Attacks\nInterception (secrecy)\nActive Attacks\nFigure 2. Active and Passive Security Threats.\n–\nThe second passive attack, traffic analysis, is more subtle. Suppose that\nwe use encryption to mask the contents of messages, the enemies might\nstill be able to observe the pattern of the messages and determine the\nlocation and identity of communication hosts.\n–\nThe third passive attack, message dropping, is to silently discard the\nmessage carried on by an intermediate host.\nFor example, in AODV\nand DSR, when a malicious node receives a control packet (e.g., RREQ,\nRREP) or a data packet, instead of looking up route table and forwarding\na packet, the malicious node drops the packet silently.\nIt is usually hard to distinguish passive attacks from Byzantine failures [8, 2] in\nad hoc and sensor networks because passive attacks do not involve any alteration\nofcontentsofthemessages. Messagelosscanoccurbecauseoftopologychange\nor unreliable wireless media. However, it is feasible to prevent the success of\nthese attacks. Thus, the emphasis in dealing with passive attacks should be put\nmore on prevention rather than detection.\nActive attacks: An active attack involves modification of the contents of mes-\nsages or creation of false messages. It can be subdivided into four categories:\nmasquerade, replay, modification of messages, and denial of service.\n–\nA masquerade takes place when one node pretends to be a different node.\nIt is usually combined with other active attacks by the adversary to deploy\nsecurity attacks. For example, a malicious node may impersonate another\nnode while sending the control packets to create an undesirable update in\nthe routing table.\n–\nReplay involves the passive capture of a valid message and retransmission\nto create an unauthorized effect.\n–\nModification of message means that some portion of a legitimate message\nisaltered, orthatmessageisdelayed, reorderedtoproduceanunauthorized\neffect.\n"
        },
        {
            "page_number": 390,
            "text": "390\nXU SU et al.\n–\nDenial of service(DoS) prevents the normal use of communication facil-\nities. In ad hoc and sensor network routing, DoS attacks can be classified\ninto two categories: DoS attack on routing traffic and DoS attack on data\ntraffic. An attacker can launch DoS attacks against a network by dis-\nseminating false routing information so that established routes for data\ntraffic transmissions are invalid. For example, an attacker can create a\nrouting loop, causing packets to travel nodes in a cycle without reaching\ntheir destinations, or partition the network by injecting malicious routing\npackets to prevent one set of nodes from reaching others. An attacker can\nalso launch DoS attacks on data traffic by ejecting a significant amount of\ndata traffic into the network to clog the network. Both of these two types\nof attacks might be used to consume valuable network resources such as\nbandwidth, or to consume node resources such as memory or computation\npower.\nActive attack can be also further divided into external attacks and internal\nattacks. An external attack is one caused by nodes that do not belong to the\nnetwork. An internal attack is the one launched from compromised or hi-\njacked nodes that belong to the network. Internal attacks are often more severe\ncompared to external attacks.\nNext, we describe some well-known active attacks against ad hoc and sensor net-\nworks such as Black Hole and Gray Hole Attacks , Wormhole Attacks, Rushing Attacks,\nJellyfish Attacks, Sinkhole Attacks, and Sybil Attacks.\nBlack Hole and Gray Hole Attacks: The malicious node can easily deploy\nblack hole attack [17] by advertising shorter distance (e.g., smaller hop count)\nor more fresh routes (e.g., a larger destination sequence number) to the desti-\nnation or both to make the source node think that the route discovery process\nis complete and the route which is passing the malicious node is the best one.\nAs a result, all the packets through the malicious node are simply consumed\nor lost. The malicious node could be said to form a black hole in the network,\nand this is called black hole attack. As a special case of black hole attack, the\nmalicious node could create a gray hole, in which it selectively drops some, not\nall of packets that it receives.\nWormhole attacks: In a wormhole attack [19], two attacker nodes collude\ntogether. One attacker node receives packets at one point and “tunnel\" them to\nanother attacker node via a private network connection, and then replays them\ninto the network. The wormhole puts the attacker nodes in a very powerful\nposition compared to other nodes in the network.\nFor instance in reactive\n(on-demand) routing protocols such as AODV or DSR, the attackers can tunnel\neach route request RREQ packets to another attacker which is near to destination\nnode of the RREQ. When the neighbors’ of the destination hear this RREQ,\n"
        },
        {
            "page_number": 391,
            "text": "WIRELESS NETWORK SECURITY\n391\nthey will rebroadcast this RREQ and then discard all other received RREQs\nin the same route discovery process. This type of attack prevents other routes\ninstead of the wormhole from being discovered, and thus creates a permanent\nDenial-of-Service attack by dropping all the data, or selectively discarding or\nmodifying certain packets as needed. The wormhole attack is very dangerous\nagainst ad hoc networking routing protocols and is harder to detect than other\nattacks since there is collusion between attackers.\nRushing attacks: Rushing attack [20] is another type of DoS attack against all\ncurrently reactive (on-demand) routing protocols in MANETs. An attacker can\nforward route request packets (e.g., RREQs in AODV and DSR) more quickly\nthan legitimate nodes, and thus increase the chance that routes which include\nthe attacker will be discovered rather than other valid routes. After the attacker\nincludes itself into the routes, it can launch different attacks such as dropping\nthe packets that it receives, or modifying the content of the packets etc.\nJellyfish attacks: Jellyfish (JF) attack [1] is targeted against closed-loop flows\nsuch as TCP which apply congestion control. In general, there are three kind of\nJellyfishattacks: JFReorderAttack, JFPeriodicDroppingAttack, andJFDelay\nVariance Attack. TCP uses “ACK-N\" to indicate that all segements 1, . . . , N has\nbeen received. In JF Reorder Attack, before delivering all packets, the attacker\nplace the packets received in a re-ordering buffer rather than an FIFO buffer.\nSuch persistent re-ordering of packets will result in near zero goodput, even if\nall the packets will be delivered. In JF Periodic Dropping Attack, the attacker\nchooses a good dropping pattern to drop packets for only a small fraction of\ntime, it can cause repeated timeout phases for other nodes, which might result\nin near-zero throughput. In JF Delay Variance Attack, the attacker still place\npackets into an FIFO buffer but wait a random time before delivering each\npacket, thus significantly increasing delay variance. Such high delay variance\ncan (1) cause TCP to send traffic in bursts and lead to increased collisions and\nloss, (2) cause incorrect estimation of available bandwidth, and (3) result in an\nextremely high retransmission timeout (RTO) value.\nSinkhole attacks: In a sinkhole attack for sensor networks, the attacker tries\nto lure nearly all the traffic from a particular area through a compromised node,\ncreating a metaphorical sinkhole with the attacker at the center. Like black\nhole attacks in ad hoc networks, sinkhole attacks typically work by making a\ncompromised node look especially attractive to surrounding nodes with respect\nto the routing algorithm.\nSybil attacks: In a Sybil attack [13], the attacker presents multiple identities\nto other nodes in the network. The Sybil attack can significantly reduce the\neffectiveness of fault-tolerant distributed storage systems, routing algorithms,\ndata aggregation, voting, and fair resource allocation etc.\n"
        },
        {
            "page_number": 392,
            "text": "392\nXU SU et al.\nThere are also many other attacks. In [24], Jakobsson et al. describe several\nstealth attacks against ad hoc networks. Wireless physical layers are often vulnerable\nto jamming.\nMedia Access Control protocols are also vulnerable to attacks.\nFor\nexample, in IEEE 802.11, an attacker can suppress its neighbor nodes from sending\npackets by periodically sending Clear-To-Send (CTS) frames in which “Duration\" field\nis set a value greater than or equal to the interval between such frames. We will not\ndiscuss those in details since they are beyond the scope of this chapter.\n5.\nSECURITY MECHANISMS\nBefore mobile ad hoc networks and sensor networks can be successfully deployed,\nsecurity issues must be addressed first. Recently significant progress has been made\nin securing mobile ad hoc networks and sensor networks via the development of se-\ncure routing protocols. We divide these works into three categories based on the main\ncharacteristics of their security mechanisms: Prevention Mechanisms, Detection Mech-\nanisms, Other Mechanisms.\nMost of these mechanisms apply either asymmetric cryptography or symmetric\ncryptography or both. In asymmetric (public key) cryptography, there are RSA, El-\nliptic Curve Cryptography (ECC) etc. Digital signature is one way using asymmetric\ncryptography to authenticate the origin of messages. For symmetric cryptography, hash\nfunctions are used, such as HMAC, MD5, SHA-1 etc. Hash-chain and hash-tree use\none-way hash function to encrypt messages. Asymmetric cryptography is generally\nmuch slower and more power-consumable compared to symmetric cryptography, and\nthus symmetric cryptography is preferred in MANETs, especially for sensor networks\nwith less power and less computational capability.\nNo matter whether we use asymmetric cryptography or symmetric cryptography,\nwe need to distribute the public key (in asymmetric cryptography) or shared key (in\nsymmetric cryptography). Key distribution and management is another big issue in\nmobile ad hoc and sensor networks. Since this is quite beyond from our topic, we do\nnot discuss them here.\n5.1. Prevention Mechanisms\nARAN\nSanzgiri and Dahil [41] propose ARAN, an on-demand routing protocol for ad\nhoc networks. In ARAN, each node that forwards a route discovery or a route reply\nmessage must sign it. When a source node, S, needs to send data to a destination D, it\nbroadcasts a route discovery packet (RDP):\nS →brdcast : [RDP, IPD, certS, NS, t]KS−\nThe RDP includes a packet type identifier “RDP\", the IP address of the destination IPD,\nS’s certificate certS, a nonce NS, and the current time t, all signed with S’s private\n"
        },
        {
            "page_number": 393,
            "text": "WIRELESS NETWORK SECURITY\n393\nkey. When a node B, the neighbor of S, receives the RDP broadcast, it rebroadcasts:\nB →brdcast :\n\u001e\n[RDP, IPD, certS, NS, t]KS−\n\u001f\nKB−,certB\nUpon receiving the RDP, B’s neighbor C validates the signature with the given certifi-\ncate, then C removes B’s certifcate and signature, records B as its predecessor, signs\nthe contents of the message originally broadcast by S, appends its own certificate, and\nbroadcasts the message as follows:\nC →brdcast :\n\u001e\n[RDP, IPD, certS, NS, t]KS−\n\u001f\nKC−,certC\nAll other nodes follow the same procedure if they receive RDP broadcasts, and they\ncan use (NS, IPD) to check if it received the RDP before.\nWhen the first RDP message arrives at the destination, D, the destination unicasts\na Reply (REP) packet back along the reverse path to the source using digital signature\nto sign the packet, the same way as in the route discovery procedure.\nDuring the route maintenance, all the Error (ERR) messages must be signed, and\nthey are forwarded along the path toward the source without modification.\nThis scheme involves too expensive computation since every message is signed in\na point-to-point manner, and it is prone to attacks using fabricated ERR messages.\nSEAD\nIn SEAD [18], hash chains are used to secure Destination-Sequenced Distance-\nVector routing protocol (DSDV). A one-way hash chain is built on a one-way hash\nfunction H which maps an input of any length to a fixed-length bit string. It can be\ndenoted as: H : {0, 1}∗→{0, 1}ρ, where ρ is the length in bits of the output of the\nhash function. For each node, to create a one-way hash chain, it chooses a random\nnumber x ∈{0, 1}ρ and computes a list of values:\nh0, h1, h2, . . . , hn\nwhere h0 = x, and hi = H(hi−1), 0 < i ≤n, for some n. In SEAD, it assumes\nthat there is some mechanism for a node to distribute an authentic element hn from its\nhash chain. It assumes that all metrics in any routing update are less than m, and n is\ndivisible by m. Let k = n\nm −i, where i is the sequence number in some routing update\nentry. An element from the following group:\nhkm, hkm+1, . . . , hkm+m−1\nis used to authenticate the entry. For example, if the metric value (e.g., hop count) is\nj, 0 ≤j < m, then hkm+j is used to authenticate the routing update entry. If the\nsequence number in a route reply message is i and metric value is j, we can check if\nHn−km−j(hkm+j) = hn to know whether the this route reply message is valid or not.\n"
        },
        {
            "page_number": 394,
            "text": "394\nXU SU et al.\nSEAD can prevent any node from advertising a route to some destination claiming\na higher destination sequence number or smaller metric. However, it can not prevent the\nmalicious node from advertising a route with the same metric as in the advertisement it\nreceived, and after using all the elements of the hash chain, a new one must be computed\nand distributed to all other nodes.\nAriadne\nAriadne [17] is proposed to secure DSR protocol by the same authors of SEAD.\nIt authenticates routing messages using one of three schemes: shared secrets between\neach pair of nodes, shared secrets between end-to-end nodes combined with broadcast\nauthentication TESLA [35], or digital signatures. However, it requires clock synchro-\nnization, which would not be realistic requirement for mobile ad hoc networks.\nSAODV\nZapata and Asokan [52] propose a security mechanism to protect AODV’s rout-\ning information: Secure AODV (SAODV). Two mechanisms are used to secure the\nAODV messages (RREQs, RREPs, and RERRs): (i) digital signatures to authenti-\ncate the non-mutable fields of the messages (e.g., destination sequence number) in an\nend-to-end manner, (ii) hash chains to secure the mutable information of the message\n(hop count). The RREQ and RREP messages include four new fields Hash Function,\nMax Hop Count, Top Hash, and Hash. When a node originates an RREQ or an RREP\nmessage, it performs the following operations:\n1. Generate a random number seed and set Hash field to the seed value:\nHash = seed.\n2. Set the Max Hop Count field to the TimeToLive value in the IP Header:\nMax Hop Count = TimeToLive.\n3. Set the Hash Function field to the identifier of the hash function h that will be\nused. If a fixed hash function is used throughout the network, such as MD5 or\nSHA1 etc., the Hash Function field is not needed anymore.\n4. Set Top Hash field value by hashing seed MAX HOP Count times:\nTop Hash = hMAX Hop Count(seed).\nWhen a node receives an RREP or an RREP message, it can easily verify the hop count\nby comparing if\nTop Hash = hMAX Hop Count−Hop Count(Hash),\n"
        },
        {
            "page_number": 395,
            "text": "WIRELESS NETWORK SECURITY\n395\nwhere Hop Count is the hop count value in the message. Since all other fields are signed\nusing digital signature by either source or destination node, current node can verify if\nother fields have been modified in the middle or if the message is fabricated. Before\nrebroadcasting an RREQ or relaying an RREP, the node applies the hash function to\nthe value in the Hash field and write the new value back to the Hash field. There is\na problem in applying digital signatures on RREP messages because AODV allows\nintermediate nodes to reply RREQ messages if they have a “fresh enough” route to the\ndestination. To solve this problem, the authors propose two mechanisms. The first one,\nalso the obvious one, is that intermediate nodes can not initiate an RREP even if they\nhave a fresh route to the destination. The second one, aslo more complex one, is that,\nevery time a node S generates an RREQ message, it also includes the RREP flags, the\nprefix size and the signature that can be used by any intermediate node I that creates\na reverse route to the originator of the RREQ, S, to reply other RREQs in which the\ndestination is S. The intermediate node should include two lifetime: the old one which\nis needed to verify the signature of the route destination and the new lifetime of this\nroute which is signed by itself.\nConsidering that RERR messages have too much mutable information, in their\nscheme, every node will use digital signatures to sign the whole message and any node\nthat receives it can verify the signature. Since the destination sequence numbers are\nnot signed by the corresponding node, a node will not update any destination sequence\nnumber in its routing table based on RERR messages.\nRAP\nHu et al. [20] present a new attack, rushing attack that results in DoS when used\nagainst on-demand ad hoc networking routing protocols. They proposed Rushing At-\ntack Prevention (RAP) against rushing attacks for on-demand routing protocols. RAP\nconsists of three generic mechanisms: secure Neighbor Detection, secure Route Dele-\ngation, and randomized Route Request forwarding. Secure Neighbor Detection allows\neach node to verify that its neighbor is within a given maximum transmission range.\nAfter a node A forwarding a Route Request message determines that the node B is its\nlegitimate neighbor, it signs a Route Delegation message, allowing node B to forward\nthe Route Request message. Once node B verify that node A is within its allowable\nrange, it signs an Accept Delegation message. Randomized Route Request forwarding\nis used to make sure that arriving Route Requests with low latency are only slightly\nmore likely to be forwarded than others.\nSRP\nPapadimitratos and Haas [33] propose the Secure Routing Protocol (SRP) which\nfocuses on multiple routes discovery for on-demand source routing protocols such as\nDSR[22] or ZRP[14]. SRP relies exclusively on the mutual authentication of the end\nnodes only (source and destination). A security association (SA) between the source\nnode S and destination node D is assumed. There is a shared key KS,D between\n"
        },
        {
            "page_number": 396,
            "text": "396\nXU SU et al.\nthem. When a source node S starts the route discovery, it broadcasts a route request\nmessage which includes a query sequence number and a random query identifier and\nMessage Authentication Code (MAC) using KS,D. The source, destination, the query\nsequence number and the random query identifier are inputs for the calculation of MAC.\nThe identifier of the traversed intermediate nodes are accumulated in the route request\npacket. When a route request arrives at the destination D, D will verify the route\nrequest and calculate an MAC covering the contents of the route reply and unicast\nroute reply back to S along the reverse of the route accumulated in the route request.\nThe destination responds to one or more request packets of the same query, so that the\nsource can get multiple paths to the destination. When the source node S receives the\nroute reply packet, it will verify that the packet is originated from the destination.\nSince SRP does not require any authentication of the relay nodes, it makes the\nprotocol more light-weight, but also more prone to attacks.\nSMT\nPapadimitratos and Haas [34] devise a Secure Message Transmission protocol\n(SMT), based on SRP [33], to safeguard the data transmission against malicious behav-\niors. The principle of SMT is that both source and destination make use of an “Active\nPath Set\" which consists of diverse, preferably link disjoint paths that are assumed valid\ninitially. The source disperses each outgoing packet into a number of pieces. This op-\neration introduces redundancy such that the destination can successfully reconstruct\nthe dispersed message if sufficient number of pieces are received. Each dispersed piece\nis transmitted across a different route chosen from the “Active Path Set\" and it carries\na Message Authentication Code, so that the destination can verify its integrity and the\nauthenticity of its origin.\nSNEP and µTESLA:\nIn [36], two secure building blocks: SNEP and µTESLA are proposed for sensor\nnetworks. SNEP, which stands for Sensor Network Encryption Protocol, provides data\nconfidentiality, two-party data authentication, integrity, and freshness. In SNEP, data\nconfidentiality is achieved through encryption using the shared key between source and\nsink. To achieve two-party authentication and data integrity, a message authentication\ncode (MAC) is used. Using secret shared counter between the sender and the receiver,\nweak freshness is assured. µTESLA is a protocol for efficient, authenticated broadcast\nand flooding that uses only symmetric key cryptography and requires minimal packet\noverhead. µTESLA achieves the asymmetric necessary for authenticated broadcast and\nflooding by using delayed key disclosure and one-way key chains constructed with a\npublicly computable cryptographically secure hash function. Replay can be prevented\nsince messages authenticated with previously disclosed keys are ignores. Like TESLA,\nit requires loose time synchronization too.\n"
        },
        {
            "page_number": 397,
            "text": "WIRELESS NETWORK SECURITY\n397\nINSENS\nIn [11], Deng et al. propose an intrusion-tolerant routing protocol for wireless\nsensor networks (INSENS). INSENS constructs forwarding tables at each node to\nfacilitate communication between sensor nodes and a sink or base station.\nRoute\ndiscovery consists of three phases. First, the base station floods a request message\nto all reachable sensor nodes in the network. Second, sensor nodes send their (local)\ntopology information using a feedback message to the base station. Finally, the base\nstation computes the forwarding tables for each sensor node based on the information\nreceived in the second phase and sends them to the corresponding nodes using a routing\nupdate message. To counter spurious request message, a scheme similar to µTESLA\nwhere one-way key chains are used to identify a message originating from the base\nstation. Moreover, multipath routing is built into INSENS to address the notion of\ncompromised nodes.\nSIA\nPrzydatek et al. [40] propose a novel framework for secure information aggrega-\ntion in large sensor networks. They focus on defending stealthy attacks. In this type of\nstealthy attacks, the attackers’ goal is to make the base station accept false aggregation\nresults, which is significantly different from the true results determined by the mea-\nsured values. A general approach, called aggregate-commit-prove, is proposed. This\ntechnique consists of three parts. In the first part, the aggregator collects data from\nthe sensors and computes the aggregation result. In the second part, the aggregator is\nresponsible for committing to the collected data. This commitment ensures that the\naggregator actually uses the data collected from the sensors. One way to perform this\ncommitment is to use Merkle hash-tree construction. In the final part, the aggregator\nuses an interactive proof to prove the correctness of the results. This interactive proof\nusually contains two steps. In the first step, the base station checks to ensure that the\ncommitted data is a good representation of the true data in the sensor networks. The\nsecond step, the base station checks if the aggregator is cheating. This can be done by\nchecking whether the aggregation result is close to the committed result.\nSDD\nIn [44], Wang et al. propose a new secure directed diffusion protocol (SDD)\nwhich provides a simple scheme to securely diffuse data in sensor networks. SDD is\nalso divided into four phases as in directed diffusion protocol. In the first phase of\ninterest propagation, modified authentication TESLA protocol is used to ensure the\ninterest is from the sink. In the low-rate data propagation and secure route setup phase,\nmodified immediate authentication TESLA protocol is used to ensure that data from the\nsource node could be read by the intermediate nodes and have not been modified. Each\nintermediate node also adds its own identity to a node list in the forwarded message\nand adds its “signature\" to the message by encrypting a nonce using its shared key\nbetween itself and the base station. In the third phase of reinforcement, the base station\n"
        },
        {
            "page_number": 398,
            "text": "398\nXU SU et al.\nselects a path and reinforces it. This reinforcement is protected in the same way as an\ninterest is protected in the first phase. In the last phase of data propagation, the data\nis protected the same way as in the second phase for the low-rate data. However, in\nSDD, in-network aggregation, which can make the data propagation more efficient, is\nforbidden,\n5.2. Detection Mechanisms\nWatchdog and Pathrater\nIn order to mitigate attacks on packet forwarding, Marti et al. [31] propose watch-\ndog to identfy malicious nodes and pathrater to help routing protocol to avoid these\nnodes. Each node is set to promiscuous mode. After a node forwards a packet, the\nnode’s watchdog verifies if the next node in the path also forwards the packet by promis-\ncuously listening to the next node’s transmissions. If the next node does not forward\nthe packet as it is supposed to do, it is marked as a malicious node. The pathrater uses\nthis knowledge of misbehaving nodes to choose the network path that is most likely\nto deliver packets. Although it is promising approach, there are three key limitations.\nFirst, it cannot be deployed in combination with dynamic power management where\neach node can reduce its power level while transmitting and thus this transmission\nmight not be heard by its upstream node. Second, watchdog assumes that all nodes\nuse omnidirectional antenna. If a malicious node use unidirectional antenna, then it\ncan fool its upstream node by sending packet back toward the upstream node. Third, a\nnode can also uses variable power transmission to cheat the upstream node.\nIdentification\nIn some approaches, once a victim has detected a DoS attack, it will try to establish\na new route. A more sophisticated approach is trying to identify the malicious node(s)\non the route. Awerbuch et al. [2] propose a mechanism to detect a malicious link\non the route using an adaptive probing technique. This approach is based on using\nacknowledgements (acks) of the data packets from the destinations. If a source detects\nthat the number of lost packets (without acks) is larger than a given threshold, it performs\na binary search on the path to identify the malicious link. In order to realize that, it\npolls specific nodes via “probes\" and waits for replies. The protocol assumes that the\nmalicious node can not distinguish between probing packets and normal ones, and is\nalso unable to know when the source has started a probing process.\nAlthough it is a good approach, how to select the threshold and the performance\nof this approach still need more investigation.\nExamination\nJust et al. [23] propose a distributed probing technique to detect and mitigate\nmalicious packet dropping attacks in ad hoc networks. In this approach, every node\nperiodically monitors the forwarding behavior of other nodes. Instead of using promis-\n"
        },
        {
            "page_number": 399,
            "text": "WIRELESS NETWORK SECURITY\n399\ncuous listening in [31], a node A sends a probe message (indistinguishable from normal\ndata) to another node C one or more hops away from the node to be checked, B, and\nwaits for an acknowledgement from C within a certain time period. If A can not receive\nthe acknowledgement in the time period, then B is suspected. The best thing of this\napproach is that it can also be implemented in the application layer and does not require\nthe modification of routing protocols. However, the overhead of this approach is too\nmuch and might not work well in mobile ad hoc networks with high mobility.\nPacket Leashes\nWormhole attack [19] is very dangerous against ad hoc network routing protocols\nsince it can create a permanent DoS attack. There are some ways to prevent wormhole\nattacks such as using a secret method for modulating bits over wireless transmission, or\nusing RF watermarking, modulating the RF waveform in a way known only to autho-\nrized nodes, to authenticate a wireless transmission, etc. However, these approaches\nsuffer from certain problems as follows. Once a node is compromised, the secret\nmethod for modulating bits is likely to fail unless the radio is kept inside a tamper-\nresistant hardware. When a waveform is exactly captured, RF watermarking fails too.\nIn contrast to these approaches, Hu et al. [19] propose a general approach, packet leash.\nThe principle is to add information (leash) to a packet designed to restrict the packet’s\nmaximum allowed transmission distance. They propose two methods: geographical\nleashes and temporal leashes. A geographical leash ensures that the recipient of the\npacket is within a certain distance from the sender. The temporal leash ensures that a\npacket has an upper bound on its lifetime, which restricts its maximum travel distance.\nEach type of leash can prevent wormhole attacks. However, since in geographical leash\neach node must know its own location, and all nodes must have tightly synchronized\nclock in temporal leash, these requirements restrict their applicability.\nMitigation\nMalicious route request flooding attack targets at the overall network performance.\nThese malicious nodes behave like normal nodes except that they do route discoveries\nmuch more frequently than normal ones. Desilva and Boppana [12] propose a statistical\nprofiling technique to detect the malicious nodes and an adaptive statistical packet\ndropping mechanism to mitigate this type of attack, and the proposed scheme reduces\nthe loss of throughput. In this approach, each node monitors the route requests which\nit receives and maintains a count of RREQs received for each RREQ sender. For node\nA, when the number of RREQs received from RREQ sender M in a period is larger\nthan some threshold which is related to the average RREQs sent for each node in the\nwhole network, then M is marked as a malicious node. This approach is simple and\ndoes not use any additional network bandwidth.\nValidation\nTo defend against the Sybil attack, Newsome et al. [29] describe two methods to\nvalidate identities: direct validation and indirect validation. In direct validation, a\n"
        },
        {
            "page_number": 400,
            "text": "400\nXU SU et al.\ntrusted node directly tests whether another node identity is valid. In indirect validation,\nother trusted nodes are allowed to vouch for (or against) the validity of a joining node.\nNewsome et al. present a novel approach, radio test, to direct validation. This approach\nassumes that any physical device has only one radio and the radio is incapable of\nsimultaneously sending or receiving on more than one channel. In the radio test, a\nnode assigns each of its neighbors a different channel on which to broadcast some\nmessages. The node then randomly chooses a channel and listens. If the neighbor that\nwas assigned that channel is legitimate, it should hear the message. Another technique\nto defend against the Sybil attack proposed in [29] is to use random key predistribution\ntechniques. The principle of this technique is that with a limited number of keys in a\nkey pool, a node that randomly generates identities will not possess enough keys to take\non multiple identities and thus will not be able to exchange messages in the network.\n5.3. Other Mechanisms\nResearchers have studied approaches to enforce node cooperation in mobile ad hoc\nnetworks. Buchegger et al. [5] propose a protocol, called CONFIDANT, for making\nmisbehavior unattractive. It aims at not only detecting and avoiding, but also isolating\nmisbehaving nodes, thus making it unattractive to deny cooperation. CONFIDANT\nconsists of four components in each node: a Monitor which identifies deviations from\nthe normal routing behavior, a Reputation System which rates other nodes according to\nthe observed behavior, a Path Manager which maintains path rankings and performs\nspecific actions upon receiving routing messages, and a Trust Manager which sends\nand receives alarm messages to and from other trust managers. Similarly, Michiardi\nand Molva propose CORE [32], a generic mechanism based on reputation to enforce\ncooperation among nodes in mobile ad hoc networks.\nButtyan and Hubaux propose the usage of a virtual currency, nuglet [6], to stimulate\npacket forwarding. They propose two payment models: Packet Purse Model and Packet\nTrade Model. In the Packet Purse Model, the source of a packet pays by loading some\nnuglets in the packet before sending it, the intermediate nodes can acquire some nuglets\nafter forwarding the packet. In the Packet Trade Model, a packet does not carry nuglets.\nEach intermediate node “buys\" the packet from the previous one with some nuglets, and\n“sells\" it to the next one for more nuglets. Later they propose a new approach [7] which\nis based on a counter, called nuglet counter. Packets do not need to carry nuglets, when\nthe node sends its own packet, the nuglet counter decreases, and the counter increases\nwhen the node forwards a packet.\nZhou et al. propose Sprite [53], a simple, cheat-proof, credit-based system for\nstimulatingcooperationamongselfishnodesinmobileadhocnetworks. Spriteprovides\nincentive for mobile nodes to cooperate and report actions honestly, and thus stimulates\ncooperation among selfish mobile nodes. Moreover, Sirinivasan et al. [42] address the\nissue of user cooperation in ad hoc networks, where each node is constrained by finite\nenergy capacity so that the network has a limited life time.\nMore recently, Yu et al. propose HADOF [50], a set of mechanisms to protect\nmobile ad hoc networks against routing disruption attacks by insider attackers, which\n"
        },
        {
            "page_number": 401,
            "text": "WIRELESS NETWORK SECURITY\n401\ncanbeimplementedupontheexistingsourceroutingprotocolssuchasDSR.InHADOF,\neach node monitors the behavior of the valid routes in its route cache and collect the\npacket forwarding statistics submitted by the nodes on this route. Each node keeps\ncheating records for other nodes, if a node is detected as dishonest, this node will be\nexcluded from future routes. Moreover, each node tries to build friendship with others\nto speed up malicious node detection. Each source node also discovers multiple routes\nto the destination, and adaptive route rediscovery is used to determine when new routes\nshould be discovered. HADOF can handle various attacks with little overhead, such as\nblack hole attacks, gray hole attacks, rushing attacks, and wormhole attacks, etc.\nYi et al. [49] propose a routing protocol, called Security-Aware ad hoc Routing\n(SAR), which incorporates security attributes as parameters into ad hoc route discov-\neries, to improve QoS of routing protocols. Yang et al. [47] present a new secure\nenhancement scheme (SES), which incorporates a new metric for routing. The new\nmetric takes into account both the distribution of possible existing attacks and the\npossibility of future potential attacks.\nIn [1], Aad et al. study two types of attacks: JellyFish attack and Black Hole attack.\nThis is the first paper that provides an analytical model to quantify the scalability of\nDoS attacks as a function of performance parameters such as mobility, system size,\nnode density, and counter-DoS strategy, etc.\nIn [25], Karlof et al. give a good discussion on attacks on routing protocols in\nwireless sensor networks. Some countermeasures and design considerations for routing\nin sensor networks are suggested.\n6.\nCONCLUSION\nBoth mobile ad hoc networks and wireless sensor networks are gaining popu-\nlarity due to the fact that they promise viable solutions to a variety of applications.\nSince MANETs can be easily and inexpensively set up as needed, they have a wide\nrange of applications, especially in situations with geographical or terrestrial constrains\nsuch as battlefields, military applications, and other emergency and disaster situations.\nNetworking unattended sensor nodes are increasingly popular to provide economical\nsolutions to both military and civil applications such as real-time traffic monitoring,\nwildfire tracking, wildlife monitoring, weather monitoring, security and tactical sur-\nveillance, target filed imaging, and disaster management, etc. However, both MANETs\nand WSNs are vulnerable to different attacks due to their fundamental characteristics\nsuch as open medium, dynamic topology, absence of central administration, distributed\ncooperation, and constrained capability (power and computation constraints). In order\nto make mobile ad hoc and sensor networks work properly, security issues must be\naddressed first. Recently significant progress has been made in securing mobile ad hoc\nand sensor networks via the development of secure routing protocols.\nIn order to assess the security needs of MANETs and WSNs effectively and to eval-\nuate and choose various policies, three aspects of information security are considered:\nsecurity service, security attack, and security mechanism.\n"
        },
        {
            "page_number": 402,
            "text": "402\nXU SU et al.\nIn this chapter, we first go over what kind of security services are needed in ad\nhoc and sensor networks, then introduce different attacks, and finally review the recent\napproaches to secure routing protocols in both mobile ad hoc networks and sensor\nnetworks. We divide current approaches into three groups: prevention mechanism and\ndetection mechanism, and all other related works as the third group.\nTo secure routing in ad hoc and sensor networks, both control packets and data\npackets must be protected from attacking. Although significant progress has been made\nto detect and prevent known security attacks, there still will be new potential attacks,\nand more research work must be done to make routing in ad hoc and sensor networks\nmore secure and practically deployable.\nSince it is usually hard to distinguish passive attacks from Byzantine failures,\nthe emphasis in dealing with passive attacks should be put more on prevention rather\nthan detection. While for active attacks, it is better to put more efforts on detection\nmechanisms and reputation-based approaches.\n7.\nREFERENCES\n1. I. Aad, J. Hubaux, and E. Knightly, Denial of Service Resilience in Ad Hoc Networks, Proceedings of\nACM MOBICOM Wireless Security Workshop (WiSe), 2004.\n2. B. Awerbuch, D. Holmer, C. Nita-Rotaru, and H. Rubens, An On-demand Secure Routing Protocol\nResilient to Byzantine Failures, Proceedings of ACM MOBICOM Wireless Security Workshop (WiSe),\n2002.\n3. R. V. Boppana and S. Konduru, An adaptive Distance Vector Routing Algorithm for Mobile, Ad Hoc\nNetworks, Proceedings of IEEE Infocom, 2001.\n4. D. Braginsky and D. Estrin, Rumor Routing Algorithm for Sensor Networks, Proceedings of the First\nWorkshop on Sensor Networks and Application (WSNA), 2002.\n5. S. Buchegger and J. Boudec, Performance Analysis of the CONFIDANT Protocol: Cooperation Of\nNodes Fairness In Dynamic Ad-hoc Networks, Proceedings of IEEE/ACM Symposium on Mobile Ad\nHoc Networking and Computing (MobiHOC), 2002.\n6. L. Buttyan and J. Hubaux, Enforcing Service Availability in Mobile Ad-Hoc WANs, Proceedings of\nthe First IEEE/ACM Workshop on Mobile Ad Hoc Networking and Computing (MobiHOC), 2000.\n7. L. Buttyan and J. Hubaux, Stimulating Cooperation in Self-Organizing Mobile Ad Hoc Networks,\nMobile Networks and Applications, 2003.\n8. M. Castro and B. Liskov, Practical Byzantine Fault Tolerance, Proceedings of the Third Symposium on\nOperating Systems Design and Implementation, 1999.\n9. B. Chen, K. Jamieson, H. Balakrishnan, and R. Morris, SPAN: An Energy-Efficient Coordination\nAlgorithm for Topology Maintenance in Ad Hoc Wireless Networks, Wireless Networks, Vol. 8, No. 5,\npp. 481-494, 2002.\n10. H. Deng, W. Li, and D. Agrawal, Routing Security in Wireless Ad Hoc Networks, IEEE Communication\nMagazine, (2002) pp. 70-75.\n11. J. Deng, R. Han, and S. Mishra, INSENS: Intrusion-tolerant Routing in Wireless Sensor Networks,\nTechnical Report CU-CS-939-02, Department of Computer Science, University of Colorado, 2002.\n12. S. Desilva and R. V. Boppana, Mitigating Malicious Control Packet Floods in Ad Hoc Networks, IEEE\nWireless Communications and Networking Conference, March 2005.\n13. J. R. Douceur, The Sybil Attack, The 1st International Workshop on Peer-to-Peer Systems(IPTPS’02),\n2002.\n"
        },
        {
            "page_number": 403,
            "text": "ABOUT THE EDITORS\nYang Xiao worked at Micro Linear as an MAC (Medium Access Control) architect\ninvolving the IEEE 802.11 standard enhancement work before he joined Department\nof Computer Science at The University of Memphis in 2002. Dr. Xiao is the direc-\ntor of W4-Net Lab, and was with CEIA (Center for Information Assurance) at The\nUniversity of Memphis. He is currently with Department of Computer Science at The\nUniversity of Alabama. He is an IEEE Senior member. He was a voting member of\nIEEE 802.11 Working Group from 2001 to 2004. He currently serves as Editor-in-\nChief for International Journal of Security and Networks (IJSN) and for International\nJournal of Sensor Networks (IJSNet). He serves as an associate editor or on editorial\nboards for the following refereed journals: (Wiley) International Journal of Commu-\nnication Systems, (Wiley) Wireless Communications and Mobile Computing (WCMC),\nEURASIP Journal on Wireless Communications and Networking, and International\nJournal of Wireless and Mobile Computing. He serves as a guest editor for IEEE\nWireless Communications, special issue on “Radio Resource Management and Proto-\ncol Engineering in Future Broadband and Wireless Networks in 2006, a (lead) guest\neditor for International Journal of Security in Networks (IJSN), Special Issue on “Se-\ncurity Issues in Sensor Networks in 2005, as a (lead) guest editor for EURASIP Journal\non Wireless Communications and Networking, Special Issue on “Wireless Network\nSecurity\" in 2005, as a (sole) guest editor for (Elsevier) Computer Communications\njournal, special Issue on “Energy-Efficient Scheduling and MAC for Sensor Networks,\nWPANs, WLANs, and WMANs\" in 2005, as a (lead) guest editor for (Wiley) Jour-\nnal of Wireless Communications and Mobile Computing, special Issue on “Mobility,\nPaging and Quality of Service Management for Future Wireless Networks\" in 2004,\nas a (lead) guest editor for International Journal of Wireless and Mobile Computing,\nspecial Issue on “Medium Access Control for WLANs, WPANs, Ad Hoc Networks,\nand Sensor Networks\" in 2004, and as an associate guest editor for International\nJournal of High Performance Computing and Networking, special issue on “Paral-\nlel and Distributed Computing, Applications and Technologies\" in 2003. He serves as\neditor/co-editor for ten edited books: WiMAX/MobileFi: Advanced Research and Tech-\nnology, Security in Distributed and Networking Systems, Security in Distributed, Grid,\nand Pervasive Computing, Security in Sensor Networks, Wireless Network Security,\n"
        },
        {
            "page_number": 404,
            "text": "404\nABOUT THE EDITORS\nAdaptation Techniques in Wireless Multimedia Networks, Wireless LANs and Blue-\ntooth, Security and Routing in Wireless Networks, Ad Hoc and Sensor Networks, and\nDesign and Analysis of Wireless Networks. He serves as a referee/reviewer for many\nfunding agencies, as well as a panelist for US NSF and a member of Canada Foundation\nfor Innovation (CFI)’s Telecommunications expert committee. He serves as TPC for\nmore than 70 conferences such as INFOCOM, ICDCS, ICC, GLOBECOM, WCNC,\netc. His research areas are wireless networks, mobile computing, and network secu-\nrity. He has published more than 140 papers in major journals and refereed conference\nproceedings related to these research areas. E-mail: yangxiao@ieee.org\nXuemin (Sherman) Shen received a B.Sc. (1982) degree from Dalian Maritime\nUniversity, China, and M.Sc. (1987) and Ph.D. (1990) degrees from Rutgers University,\nNew Jersey, USA, all in electrical engineering. Currently, Dr. Shen is with the Depart-\nment of Electrical and Computer Engineering, University of Waterloo, Canada, where\nhe is a Professor and the Associate Chair for Graduate Studies. His research focuses\non mobility and resource management in interconnected wireless/wireline networks,\nUWB wireless communications systems, wireless security, and ad hoc and sensor net-\nworks. He is a co-author of three books, and has published more than 200 papers and\nbook chapters in wireless communications and networks, control, and filtering. He\nwas Technical Co-Chair for the IEEE GLOBECOM’03, ISPAN’04, QShine’05, IEEE\nBroadnets’05, and WirelessCom’05, and is Special Track Chair of the 2005 IFIP Net-\nworking Conference. He serves as an Editor/Associate Editor for IEEE Transactions\non Wireless Communications, IEEE Transactions on Vehicular Technology, Computer\nNetworks, ACM/Wireless Networks, Wireless Communications and Mobile Comput-\ning (Wiley), and International Journal Computer and Applications. He has also served\nas Guest Editor for IEEE JSAC, IEEE Wireless Communications, and IEEE Commu-\nnications Magazine. He received the Outstanding Performance Award in 2004 from\nthe University of Waterloo, the Premier’s Research Excellence Award (PREA) in 2003\nfrom the Province of Ontario, Canada, for demonstrated excellence of scientific and\nacademic contributions, and the Distinguished Performance Award in 2002 from the\nFaculty of Engineering, University of Waterloo, for outstanding contributions in teach-\ning, scholarship, and service.\nDing-Zhu Du received his M.S. degree in 1982 from Institute of Applied Mathemat-\nics, Chinese Academy of Sciences, and his Ph.D. degree in 1985 from the University of\nCalifornia at Santa Barbara. He worked at Mathematical Sciences Research Institute,\nBerkeley in 1985-86,at MIT in 1986-87, and at Princeton University in 1990-91. He was\nan associate-professor/professor at Department of Computer Science and Engineering,\nUniversity of Minnesota in 1991-2005, a Program Director for Theory of Computing at\nNational Science Foundation in 2002-2005, a professor at City University of Hong Kong\n"
        },
        {
            "page_number": 405,
            "text": "ABOUT THE EDITORS\n405\nin 1998-1999, and a research professor at Institute of Applied Mathematics, Chinese\nAcademy of Sciences in 1987-2002. Currently, he is a professor at Department of Com-\nputer Science, University of Texas at Dallas and a Dean of Science at Xi’an Jiaotong\nUniversity. His research interests include combinatorial optimization, communication\nnetworks, and theory of computation. He has published more than 160 journal papers\nand 40 books. He is the editor-in-chief of Journal of Combinatorial Optimization and\nbook series on Network Theory and Applications. He is also in editorial boards of more\nthan 15 journals. He is well-known for proving the Gilbert-Pollak conjecture on the\nSteiner ratio, the Derman-Leiberman-Ross conjecture on optimal consecutive 2-out-\nof-n systems in reliability, and the global convergence of Rosen gradient projection\nmethod in nonlinear programming. In 1998, he received received CSTS Prize from\nINFORMS (a merge of American Operations Research Society and Institute of Man-\nagement Science) for research excellence in the interface between Operations Research\nand Computer Science. In 1996, he received the 2nd Class National Natural Science\nPrize in China. In 1993, he received the 1st Class Natural Science Prize from Chinese\nAcademy of Sciences. In 1992 the proof of Gilbet-Pollak conjecture was selected by\n1992 Year Book of Encyclopaedia, Britannica, as the first one among six outstanding\nachievements in mathematics in 1991.\nYang Xiao\nDepartment of Computer Science\nThe University of Alabama\n101 Houser Hall\nBox 870290\nTuscaloosa, AL 35487-0290 USA\nE-mail: yangxiao@ieee.org\nXuemin (Sherman) Shen\nDepartment of Electrical and Computer Engineering\nUniversity of Waterloo Waterloo\nWaterloo, Ontario, Canada N2L 3G1\nE-mail: xshen@bbcr.uwaterloo.ca\nDing-Zhu Du\nDepartment of Computer Science\nUniversity of Texas at Dallas\nRichardson, TX 75083, USA\nE-mail: dzdu@utdallas.edu\n"
        },
        {
            "page_number": 406,
            "text": "INDEX\nAAA (Authentication authorization and\naccounting), 247\naccess control and, 276–77, 277f\nbenefits of, 277–78\noverview of, 276–78\nAAA brokers, 279\nnetwork structure of, 279f\nAAKE. See Authentication and\nauthenticated key establishment\nAAKE protocols, implementation of, 67\nlevel of resource constraints in, 72\nproblems in, 70–73\nsecurity of communication channels in, 72\nstages in, 68–69\nTTP availabilities in, 70–72, 71f\nABC’s. See Anomalous Binary Curves\nAccess control\nAAA and, 276–77\nsecurity and, 119, 387\nAccounting, 276\nAction agents, 167\nActive Names, 44\nActive Networks, 59\nActive security attacks, 105–6, 105t,\n388–92, 389f\nActive Services, 59\nAd hoc networks, vii–viii, 65–180\nAMs in, 78–80, 79t\nattacks and countermeasures in mobile,\n103–35\nclient-server networks v., 66\ndesign goals in, 69–70\nhierarchical v. flat, 73\nidentity-based key management in\nwireless, 83–102\nintrusion detection in mobile, 159–80\nMANETs v. smart sensor, 66\nnetwork phases in, 68\nPAMs in, 74–78, 75t\nproperties of, 66\npurpose of, 65–66\nsecure routing in wireless, 137–58\nsecurity challenges in, 66–67\nsecurity framework in, 67–70\nsecurity in, vii–viii, 65–180\nself-organizing property of, 66\nsystem configurations in, 70–73\nwireless networks v., 66\nAd hoc On-demand Distance Vector\n(AODV), 100\nattacks on, 112\ncharacteristics of, 384–85\nrole of, 140\nAdaptation model, of Adaptive Encryption\nProtocol, 50–56\nadaptation path search algorithm in, 53–56\nPAT and PADs in, 50–51, 52f\nAdaptation proxy, 46, 47f\ndistribution manager in, 49\nnegotiation manager in, 48–49\nstructure of, 48–49, 48f\nAdaptive anomaly detection\nentropy-based approach in, 204\nfeedback-based approach in, 204\nmetric selection in, 204–5\nAdaptive encryption protocol, 43–62\nadaptation model of, 50–56\nadaptation proxy in, 46, 47f, 48–49, 48f\nAES in, 46\nCDN in, 45\nDES/Triple DES in, 44–45\ndistributed adaptation and, 59\nflexibility of, 60–61\nINP in, 49–50, 51f\nmobile code and, 44–45, 60\nperformance evaluation and analysis,\n56–57\nplatform for, 46–50, 47f\nprotocol adaptation and, 45, 59–60\nRC4 stream cipher in, 46\n"
        },
        {
            "page_number": 407,
            "text": "Adaptive normal profile, algorithm of, 206,\n207f\nAES (Advanced Encryption Standard), 46\nAffine projective point system, 22t\nAggregation principle, 332–33\nAlgorithm 1: group-level multiplication, 7\nAlgorithm 2: inversion by square and\nmultiply, 7\nAlgorithm 3: scalar multiplication by double\nand add method, 9\nAlgorithm 4: efficient group level\nmultiplication, 11\nAlgorithm 5: optimized inversion by square\nand multiply, 18\nAlgorithm 6: scalar multiplication for signed\nbinary representation, 23\nAlgorithm 7: generation of Binary NAF, 24\nAlgorithm 8: scalar multiplication for τ-adic\nintegers, 25\nAlgorithm 9: generation of τ-adic NAF, 26\nAlgorithm 10: reduction mod τ m, 27\nAlgorithm 12: state machine algorithm, 36\nAM-A1: Challenge-Response, 79, 79t\nAM-A2: Key Chain Model, 79–80, 79t\nAM-H1: Password Model, 78–79, 79t\nAMs. See Authentication models\nAM-S1: Challenge-response, 78, 79t\nAnomalous Binary Curves (ABC’s), 24\nAnomaly detection algorithm, 198\nAnomaly-based IDSs, 188–89\nChi-square Test and, 189\nexpert systems and, 189\nimmunology and, 189\nneural networks as, 188\nstatistics type of, 188\nAntiBlaster, 226\nAODV. See Ad hoc On-demand Distance\nVector\nApplication layer attacks, 106t, 115\ndefense against, 126\nApplication layer defense, 126\nApplication Metadata (AppMeta), 48f, 49\nApplication servers, 46, 47f\nApplication-transparent adaptation, 59, 60\nAppMeta. See Application Metadata\nARAN. See Authenticated Routing for\nAd-hoc Networks\nARAN Routing Discovery Authentication at\nEach Hop, 124, 125f\nAriadne routing scheme, 99–100, 102\ndescription of, 148–50, 394\nARM Integrator/ARM7TDMI, 37\nARM Integrator/LM-XCV600E+, 37\nAS. See Authentication server\nAttacks, on smart phones, 213, 214t, 215\nAUTH. See Authenticator\nAuth-chal message, 254\nAuthenticated broadcasts, in\n{symbol}mTESLA, 363–65\nAuthenticated encryption, 94\nAuthenticated Routing for Ad-hoc Networks\n(ARAN), 100, 105, 124\ndescription of, 150–53, 392–93\nkey revocation in, 152–53\nnotations used in, 151t\nroute maintenance in, 152\nAuthenticating Broadcast Packets, 364–65\nAuthentication\nAAA and, 276\nfor Internet applications, 283–84, 283f\nsecurity and, 118, 387\nAuthentication and authenticated key\nestablishment (AAKE), 67\nAuthentication authorization and\naccounting. See AAA\nAuthentication forging, in security policies,\n318\nAuthentication latency\ndistributions of, 265f\nend-to-end domain distance v., 266f\nvariations of, 264f\nAuthentication mechanism, IEEE 802.11i,\n245, 246\nBAN logic in, 250–51\ncross-domain-related protocols in, 247–49\ndesign requirements for, 249–50\nmessage exchanges in, 247f\noverview and requirements of, 246–51\nAuthentication models (AMs), 78–80, 79t\nasymmetric solutions in, 79–80\nhybrid solutions in, 78–79\nsymmetric solutions in, 78\nAuthentication phases, 68, 69\nAuthentication server (AS), 247, 247f, 248\nAuthenticator (AUTH), 247, 257\nAuthenticity, in sensor networks, 342\nAuthorization, 276\nAuth-req message, 254\nAuth-res message, 255\nAV-1, 71–72, 71f\nAV-2, 71f, 72\nAV-3, 71f, 72\nAV-4, 71f, 72\nAvailability, security and, 119, 387\nBAN logic, 250–51\n408\nINDEX\n"
        },
        {
            "page_number": 408,
            "text": "Bandwith, in sensor networks, 343–44\nBARWAN/ Daedalus, 59\nBase Stations (BSs), 185, 185f\nBasic Scheme, 345–48\nanalysis of, 347–48\nkey distribution in, 345–47\npros/cons of, 377\nQ-Composite Random Key\nPredistribution Scheme v., 348–51\nBattery life, in sensor networks, 343\nBayes decision rule, 189\nBDHP. See Bilinear Diffie-Hellman problem\nBerkeley Motes, 46\nBF. See Boneh and Franklin\nBF-IBE scheme, 89, 91\nencryption and decryption in, 93, 93f\nenergy-constrained wireless ad-hoc\nnetworks and, 97\nLQ-IBSC scheme and, 94\nBilinear Diffie-Hellman problem (BDHP),\n89\nBinary BAF\nAlgorithm 7: generation of, 24\nscalar multiplication and, 23–24\nBinary finite fields, arithmetic over, 5–7\nBlack hole attacks, 100, 106t, 112, 113, 390\ncooperative IDS and, 176\ndefense against, 123–24\nrouting protocols and, 142–43\nBlaster, 226\nBlind signature, 190\nBlock Cipher, in SPINS, 365\nBLS. See Boneh, Lynn and Shacham\nBLS-IBS scheme, 94, 95, 96, 97\nBluetooth protocol, 68, 74\nBluetooth Special Interest Group (SIG), 74\nBoneh and Franklin (BF), 89\nBoneh, Lynn and Shacham (BLS), 94\nBootstrapping New Receivers, 364\nBroadcasting Authenticated Packets, 364\nBSs. See Base Stations\nByzantine attacks, 106t, 112, 113\ncoping with, 153–55\nByzantine fault detection, 154–55\nCA. See Certification authority\nCabir worm, 213\nCanadian Microelectronics Corporation\n(CMC), 37\nCANS, 44, 59\nCarrier sense multiple access/collision\navoidance protocol (CSMA/CA), 108\nCase study of IDS. See IDS case\nstudy/example\nCDN. See Content distribution networks\nCDN edgeservers, 46, 47f\nCell phone cloning, 193\nCellular mobile networks, viii, 183–236\nbackground knowledge on, 185\nexample of, 185f, 194f\nintrusion detection in, 183–210\nmodeling of, 193–95\nsecurity in, viii, 183–236\nsmartphones, epidemics in and, 211–41\nCertificate revocation lists (CRLs), 70\nCertification authority (CA), 73\nChallenge-Response (AM-A1), 79, 79t\nChallenge-Response (AM-S1), 78, 79t\nChal-res message, 255\nChannel Contention, in 802.11 MAC, 109f\nChi-square Test, anomaly-based IDSs and,\n189\nClients, 46, 47f\nCluster keys, in LEAP, 367, 369\nCMC. See Canadian Microelectronics\nCorporation\nColored Petri Nets (CPNs), 187\nmisused-based IDSs and, 187\nCOMET, 59\nCommunication channels, security of, 72\nCommunication framework, 165, 166f\nComparator/Adder, 19–20, 20t\nConductor, 59\nCONFIDANT (Cooperation Of Nodes,\nFairness In Dynamic Ad-hoc\nNeTworks), 172–73, 173f, 400\nCORE v., 174\nIDSs comparison with, 177t\nConfidentiality\nsecurity and, 118, 387\nsensor networks and, 342, 387\nConfinement principle, 332\nContent distribution networks (CDN), 44, 45\nControllers, 73\nCooperation enforcement, 130–31\nCooperative IDSs, 175–77\nblack hole in, 176\nIDSs comparison with, 177t\nmalicious flooding in, 176–77\npacket dropping in, 176\nCo-processor architecture, for ECC scalar\nmultiplication, 28–41\ncomparison of published results in, 40t\ndata path for, 30, 31f\nefficient Frobenius mapping in, 33f\nINDEX\n409\n"
        },
        {
            "page_number": 409,
            "text": "field and curve operations, performance\nof in, 40t\nfield arithmetic units in, 35, 37\nfield element storage in, 30–32, 31f\nFPGA prototype in, 37\nhierarchical control path of, 29–30, 30f\nmicro-sequencer instruction set in, 32–33\nmicro-sequencer routines in, 33–34\nperformance and cost results in, 40t\ntop level control in, 34–35\nτQ, computation of in, 32\nCORE, 173–74, 400\nCONFIDANT v., 174\nIDSs comparison with, 177, 177t\nCost analysis, of security policies, 308–10\nauthentication time and, 308–9\ncrytographic cost in, 309–10\nthroughput (bits/second) in, 310\nCountermeasures to security attacks, 118–31\napplication layer defense in, 126\nkey management attacks, defense against\nin, 128\nlink layer defense in, 121–22\nMANET IDS in, 129–31\nmulti-layer attacks, defense against in, 127\nnetwork layer defense in, 122–25\nphysical layer defense in, 120–21\nsecurity attributes in, 118–19\nsecurity mechanisms in, 119–20, 383,\n392–401\ntransport layer defense in, 125–26\nCPNs. See Colored Petri Nets\nCPU utilization, in MAP simulation, 267f\nCredit-based mechanism, 131\nCRLs. See Certificate revocation lists\nCross-domain mobility-adaptive\nauthentication, 245–71. See also\nMobility-adjusted Authentication\nProtocol\nauthentication mechanism in, 246–51\nMAP and, 251–58\nMAP performance evaluation in, 261–68\nrelated work and, 268–69\nsecurity considerations in, 258–61\nCross-domain-related protocols, 247–49\nCryptographic cost, security policies and,\n309–10\nin roaming/non-roaming scenarios,\n314–16, 315t\nCryptographic primitive attacks, 107t\nCryptography related attacks, 106\nCSMA/CA. See Carrier sense multiple\naccess/collision avoidance protocol\nData acquisition, in wireless LAN testbed,\n310–11\nData collection module, 169\nData compression, 196\nData Encryption Standard. See DES\nData forwarding phase attacks, 112\nData Propagation, in Directed Diffusion, 386\nDCF. See Distributed coordination function\nDecision agents, 167\nDecryption and encryption, 93, 93f\nDefined keys, 253–54, 254f\nDefined messages, 254–55\nDenial-of-service (DoS) attacks, 84, 105,\n105t, 106, 106t, 116, 389f, 390\ndefense against, 127\nsecurity policies and, 319–20\nDeployment knowledge, 373\nmodeling of, 373–74\nDeployment Model, 373–77\nkey pools, creation of in, 375–77\nkey predistribution in, 374–75\npros/cons of, 379\nDeployment point, 373–74\nDES (Data Encryption Standard), 45–46\nDesign requirements for authentication\nmechanism, 249–50\nDestination Sequence Distance Vector\n(DSDV), 139, 145\ncharacteristics of, 384\nDetection engine, 169\nDetection security mechanisms, 392,\n398–400\nDevice Metadata (DevMeta), 48f, 49\nDevMeta. See Device Metadata\nDigital signature attacks, 117\nexamples of, 107t\nDigital signature standard (DSS), 107\nDirect sequence spread spectrum (DSSS),\n120–21, 121t\nDirected Diffusion protocol, 386\nDirectional antennas, 123\nDisruption MAC (802.11), 106t, 108, 110\nDistance effect routing algorithm for\nmobility (DREAM), 141, 167\nDistributed adaptation, 59\nDistributed and cooperative IDSs, 162, 163f,\n164–65\nDistributed CA Model (PAM-A4), 75t, 77,\n81\nDistributed coordination function (DCF), 108\nDistributed IDS, with multiple sensors,\n166–67, 167f\n410\nINDEX\n"
        },
        {
            "page_number": 410,
            "text": "action agent in, 167\ndecision agent in, 167\nmonitoring agent in, 167–68\nDistributed Sensor Networks (DSNs), 342\nkey management for, 341–80\nDistribution manager, 49\nDK. See Domain key\nDomain key (DK), 253, 254, 254f\nDoS attacks. See Denial-of-service attacks\nDREAM. See Distance effect routing\nalgorithm for mobility\nDSDV. See Destination Sequence Distance\nVector\nDSNs. See Distributed Sensor Networks\nDSR. See Dynamic Source Routing\nDSR protocol, 85\nDSS. See Digital signature standard\nDSSS. See Direct sequence spread spectrum\nDucklings, 73\nDynamic Source Routing (DSR), 85\nattacks on, 112\ncharacteristics of, 140–41, 385–86\nroute discovery in, 95–96, 385\nroute maintenance in, 96–97, 385\nEAP. See Extensible Authentication Protocol\nEavesdropping, 105, 105t, 106, 107\nEC Diffie Hellman, 4\nEC El-Gammal, 4\nECC. See Elliptic curve cryptography\nECC scalar multiplication, 20–28, 29t\nBinary BAF in, 23–24\nco-processor architecture for, 28–41\nprojective coordinates in, 20–22, 22t\nrecoded integers in, 22–23\nrun times in, 29t\nτ-NAF in, 24–27\nECDSA. See Elliptic Curve Digital\nSignature Algorithm\nEEA. See Extended Euclidean Algorithm\nEfficient group level multiplication,\nAlgorithm 4, 11\n802.1x policies, 302\n802.11i authentication mechanism. See\nAuthentication mechanism, IEEE\n802.11i\nELGamal signature, 107t\nElliptic curve cryptographic co-processors,\n3–42\nElliptic curve cryptography (ECC), 4\nElliptic Curve Digital Signature Algorithm\n(ECDSA), 4\nElliptic curve group, arithmetic over, 7–9\nElliptic curve scalar multiplication, 4\narithmetic over binary finite fields in, 5–7\narithmetic over elliptic curve group in, 7–9\nEMERALD. See Event Monitoring Enabling\nResponses to Anomalous Live\nDisturbances\nEncryption and decryption, 93, 93f\nEncryption function, in SPINS, 365–66\nEntropy, 191\nEpidemic fast die out, 211\ncritical condition for, 224–25\nEpidemic Spread on Smartphones (ESS), 215\nEpidemics, 211–12\ndefinition of, 212n\nsmartphones and, 211–41\nEpidemics, smartphone\npropagation method of, 218–19\nspread model of, 220, 222–24\nESS. See Epidemic Spread on Smartphones\nESS model, 215, 216, 217\nanalysis of, 224–28\nbasic, 220, 222\neffectiveness and accuracy of, 240\nepidemic fast die out in, 211, 224–25\nextended, 222–24\nnomadism’s influence on, 222, 223–24\nnotions for, 221 t\nparameters’ influence on, 225–26, 227t,\n230\nperformance evalutation of, 228–38\nsimulation result v., 228–31\ntopology’s influence on, 222–23\nWaxman random model v., 229, 231, 231f\nEvent Monitoring Enabling Responses to\nAnomalous Live Disturbances\n(EMERALD)\nanomaly-based IDSs and, 188\nmisuse-based IDSs and, 187\nEWMA. See Exponentially Weighted\nMoving Average\nEWMA-based mobility trie, 195–96, 196f,\n198–200\ncomputation of entropy of, 205–6\nexample of, 199f, 205f\nSimilarity Measure in, 200–201\nExamination detection mechanism, 398–99\nExperimental study on security protocols in\nWLANs, 295–322. See also Wireless\nLAN testbed\nExpert systems\nanomaly-based IDSs and, 189\nmisuse-based IDSs and, 187\nrole of, 187\nINDEX\n411\n"
        },
        {
            "page_number": 411,
            "text": "Exponentially Weighted Moving Average\n(EWMA), 191\nExtended Euclidean Algorithm (EEA), 19\nExtensible Authentication Protocol (EAP),\n297, 298\nExternal security attacks, 106, 390\nFabrication attacks, 143–44, 388, 388f\nservice policies and, 319\nFault Detection, 154f\nFeature selection, IDSs and, 190–91\nFHSS. See Frequency hopping spread\nspectrum\nField programmable gate array (FPGA), 5\nprototype, 37\nFinite field arithmetic, high performance,\n9–20, 21t\nComparator/Adder in, 19–20, 20f\ndigit size, choice of in, 13, 15, 16t\ninversion in, 17–19, 19t\nmultiplication algorithm in, 10–11\nmultiplier’s data path in, 13, 15f\nR(x)W(x) mod F(x), computation of in,\n11–13, 13f, 14f\nsquaring in, 15–17, 17f\nxiW(x) mod F(x), generation of in, 12f\nFinite field units, utilization of\nfor point addition, 38f\nfor point doubling, 39f\nFlat ad hoc networks, 73\nFlexibility, in sensor networks, 343\nFlooding attacks, 106t\nForgery attacks, 261\nForwarding misbehaviors, 156\nFPGA. See Field programmable gate array\nFractal framework, 43, 44, 46, 56\nadaptive path search algorithm and, 53, 54\nCDN and, 45\nFrequency hopping spread spectrum\n(FHSS), 120, 120f, 121\nFreshness\nsecurity service and, 387\nin SPINS, 366\nFrobenius mapping, 24\nin co-processor architecture, 33f\nGACE module. See Global aggregation and\ncorrelation module\nGalileo satellite navigation system, 73\nGentry and Silverberg (GS), 91\nGlobal aggregation and correlation (GACE)\nmodule, 170, 170f\nGPS chips, 73\nGPSR. See Greedy Perimeter Stateless\nRouting\nGray holes, 143\nGreedy Perimeter Stateless Routing (GPSR),\n141, 168\nGrid-Based Key Predistribution, 358–61\npros/cons of, 378\nGroup keys, in LEAP, 367, 369–70\nGroup-level multiplication, Algorithm 1, 7\nGS. See Gentry and Silverberg\nGS-HIBC, 91\nGSM, 73\nHADOF, 400–401\nHandoff pattern, 263\nHandoffs, simulation model for, 263f\nHash chains, 79–80\nHash collision attacks, 117\nexamples of, 107t\nHash message authentication code (HMAC),\n94, 97, 251\nHealthcare sensor networks. See Sensor\nnetworks, in healthcare\nHELLO attack, 372\nHIBC. See Hierarchical PKG structure\nHierarchical ad hoc networks, 73\nHierarchical IDSs, 163\nHierarchical PKG structure (HIBC), 91\nHigh layer protocols, 298\nHLR. See Home Location Register\nHMAC. See Hash message authentication\ncode\nHome Location Register (HLR), 185, 185f\nHorizontal roaming architecture, 280–82,\n280f\nHybrid routing protocols, 141\nHybrid security policies, 302–3\nHyperview, 188\nIBAE. See Identity-based authenticated\nencryption\nIBC. See Identity-based cryptography\nIBC-based encryption (IBE), 89\nIBC-based schemes, 100\nIBC-based short signature (IBS), 94\nIBC-based signcryption (IBSE), 97\nIBE. See IBC-based encryption\nIBS. See IBC-based short signature\nIBSE. See IBC-based signcryption\nIBSE scheme, 97\nIDAMN. See Intrusion Detection\nArchitecture for Mobile Networks\nIdentification detection mechanism, 398\n412\nINDEX\n"
        },
        {
            "page_number": 412,
            "text": "Identity-based authenticated encryption\n(IBAE), 94\nIdentity-based cryptography (IBC), 85,\n88–89, 89f\nPKC v., 85, 88, 89\nIdentity-based key management schemes,\nwireless ad-hoc networks and, 90–92,\n97–100\ninformation exchange and, 98–99\nkey maintenance in, 91–92\nlimitations of, 98\nmessage routing and, 99–100\npeer keying in, 90–91\npractical extensions of, 97\nsystem setup in, 90\nIdentity-Based Model (PAM-A2), 75t,\n76–77, 81\nIdentity-based secure communications,\nwireless ad-hoc networks and, 93–97\nauthenticated encryption in, 94\nencryption and decryption in, 93, 93f\nmessage routing in, 94–97\nsigned encryption in, 94\nIDES. See Intrusion Detection Expert System\nIDIOT. See Intrusion Detection In Our Time\nIDS case study/example, 191–206\nadaptive algorithm in, 206, 207f\nadaptive anomaly detection in, 203–6\nassumptions in, 192–93\nLZ-based anomaly detection in, 195,\n196–201, 196f\nMarkov-based anomaly detection in, 195,\n196, 201–3\nmobility model in, 195\nnetwork model in, 193–95\nthreat model in, 193\nIDS, in MANETs, 159–80\narchitectures for, 162–64\nfuture directions of, 178\nnode cooperation and, 171–77, 177t\nsample, 164–71\nIDS sample, for MANETs, 164–71\ndistributed and cooperative, 164–65\ndistributed IDS with multiple sensors in,\n166–67\ndynamic hierarchical intrusion detection\narchitecture in, 168–69\nLIDS in, 165–66\nZBIDS in, 169–71, 170f\nIDSs. See Intrusion detection systems\nIEEE 802.11 wireless LAN roaming. See\nWireless LAN roaming, IEEE 802.11\nIEEE 802.11i authentication mechanism.\nSee Authentication mechanism, IEEE\n802.11i\nIEEE 802.15.1 (Bluetooth) v. IEEE\n802.15.4, 336–38\naccess delay in, 337\nbuffering in, 337\nevent detection reliability in, 338\nnoise errors in, 336\nsaturation regime and, 337\nIEEE 1073 Medical Device\nCommunications standards\norganization, 325\nIETF. See Internet Engineering Task Force\nIFS. See Interframe space\nIMEI. See International Mobile Equipment\nIdentifier\nImmunology, anomaly-based IDSs and, 189\nImpersonation attacks, 106t, 116\ndefense against, 124, 125f\nMAP and, 261\nrouting protocols and, 144\nImprinting, 68\nIndividual authentication time, 312t\nIndividual keys, in LEAP, 367\nIndividual security policies, 301–2\nIndustrial, Scientific and Medical (ISM), 336\nInfoPyramid, 59\nInformation exchange, in wireless ad-hoc\nnetworks, 98–99\nInitialization, 68\nInitialization vector (IV), 107t, 110\nINP. See Interactive negotiation protocol\nINSENS routing protocol, 397\nIntegrity\nsecurity and, 118–19, 387\nsensor networks and, 343, 387\nInteractive negotiation protocol (INP), 49–50\nstructure of, 51f\nInterception attacks, 388, 388f\nInterest Propagation, 386\nInterference, 107\nInterframe space (IFS), 108\nInternal security attacks, 106, 390\nInternational Mobile Equipment Identifier\n(IMEI), 193\nInternet Engineering Task Force (IETF), 104\nInterruption attacks, 387, 388f\nIntrusion Detection Architecture for Mobile\nNetworks (IDAMN), 189–90\nIntrusion Detection Expert System (IDES),\n187\nINDEX\n413\n"
        },
        {
            "page_number": 413,
            "text": "Intrusion Detection In Our Time (IDIOT),\n187\nIntrusion detection systems (IDSs), 105,\n161, 183–210. See also IDS case\nstudy/example\nadaptability of, 191\nanomaly-based, 188–89\ncase study/example of, 191–206\ncellular mobile networks and, 189–90\ndefinition of, 186\ndistributed/cooperative, 162, 163f, 164–65\nfeature selection, importance of in, 190–91\nhierarchical, 163\nMANETs and, 159–80\nmisuse-based, 161, 186–88\nprevention-based approaches v., 184\nspecification-based, 189\nstand-alone, 162\nwired networks and, 186–89\nIntrusions, definition of, 186\nInversion by square and multiply, Algorithm\n2, 7\nInversion, finite field arithmetic, 17–19, 19t\nIP security (IPsec), 297, 298\nIP security policies, 302\nIPsec. See IP security\nISM. See Industrial, Scientific and Medical\nIV. See Initialization vector\nJacobian projective point system, 22t\nJamming attacks, 105, 105t, 106t, 107\nJellyfish attacks, 391\nKerberos protocol, 247–49\nMAP v., 245, 246, 268f\nmessage exchanges for remote\nauthentication in, 248, 248f\nKey Chain Model (AM-A2), 79–80, 79t\nKey distribution\nin Basic Scheme, 345–47\nin Deployment Model, 374, 375\nKey generation center (KGC), 77\nKey maintenance, in key management,\n91–92\nKey management for DSNs, 341–80\nBasic Scheme in, 345–48, 348–51, 377\nDeployment Model in, 373–77, 379\nfuture work in, 379\nLEAP scheme in, 366–73, 378\nMultipath Key Reinforcement Scheme in,\n348, 351–52, 377–78\nPolynomial Pool-Based Key\nPredistribution Scheme in, 355–61,\n378\nQ-Composite Random Key\nPredistribution Scheme in, 348–51,\n377\nRandom Pairwise Key Scheme in, 348,\n352–54, 378\nsingle network-wide key in, 344–45\nKey management, MANET and, 132\nKey management vulnerability, 117–18\ndefense against, 128\nKey predistribution stage\nin Basic Stage, 345–46, 347\nin Deployment Model, 374, 375\nKey recovery attacks, 261\nKGC. See Key generation center\nKoblitz curves, 24\nLA. See Location Area\nLACE module. See Local aggregation and\ncorrelation module\nLAR. See Location-aided routing\nLEAP, 366–73\nadvantages of, 366–67, 378\ncluster keys in, 367, 369\ngroup keys in, 367, 369–70\nindividual keys in, 367\nlocal broadcast authentication in, 371\noverhead in, 371–72\npairwise shared keys in, 367, 368–69\nperformance evaluation of, 371–73\npros/cons of, 378\nresilience to attack of, 372–73\nLempel-Ziv algorithms, 196, 197\nLibert and Quisquarter (LQ), 94\nLimited Global Payoff Requirement, 350–51\nLink layer attacks, 106t, 108–10\ndefense against, 121–22\ndisruption on MAC DCF and backoff\nmechanism in, 106t, 108, 110\nWEP(801.11), weakness of in, 106t, 110\nLink layer defense, 121–22\nLink weight management, 155\nLocal aggregation and correlation (LACE)\nmodule, 169, 170f\nLocal Intrusion Detection System (LIDS),\n165–66\narchitecture of, 165, 166f\nLocal LIDS Agent, 165, 166f\nLocal MIB Agent, 165, 166f\nLocation Area (LA), 185\nin Cellular Mobile Networks, 194f\n414\nINDEX\n"
        },
        {
            "page_number": 414,
            "text": "Location disclosure attacks, 106t, 112, 113\nLocation spoofing attacks, 105, 105t, 106\nposition-based routing and defense\nagainst, 155–56\nrouting protocols and, 144\nLocation-aided routing (LAR), 141, 168\nLocation-Limited Model (PAM-A1), 75t,\n76, 80\nLópez & Dahab projective point system, 22t\nLow-rate data Propagation and routing\nsetup, 386\nLQ. See Libert and Quisquarter\nLQ-IBSC scheme, 94\nLZ-based anomaly detection, 195, 196–201,\n196f\nanomaly detection algorithm in, 198\nprobability calculation in, 198\nMA. See Mobile agents\nMAC. See Medium access control\nMAC algorithms, 253\nMAC layer protocols, 297\nMalicious code attacks, 115\nMalicious Flooding, 176–77\nManagement Information Bases (MIBs), 165\nMANET IDSs, 129–31\nagent conceptual architecture in, 129\ncooperation enforcement in, 130–31\nrouting misbehavior, detection of in,\n129–30\nMANETs. See Mobile ad hoc networks\nMan-in-the-middle attacks (MITM), 106t,\n116\nsecurity policies and, 318–19\nMAP. See Mobility-adjusted Authentication\nProtocol\nMAP client module, 258\nMAP SC module, 257\nMAP server module, 256\nMarkov Chain, 189, 202\nMarkov model, of event transitions, 189\nMarkov predictors, 201\nMarkov-based anomaly detection, 195, 196,\n201–3\nMasquerader, 193\nMasquerades, 389, 389f\nMean degree and Topology Factor, 237f\nMedium access control (MAC), 108\nMemory, in sensor networks, 344\nMessage authentication, in SPINS, 366\nMessage dropping, 389, 389f\nMessage exchanges, in MAP, 255–56\nMessage receivers\non desktop client host, 56, 57, 58f\non laptop client host, 56, 57, 58f\non PocketPC client host, 56, 57, 58f\nMessage routing\nroute discovery in, 95–96, 385\nroute maintenance in, 96–97, 385\nsecure communications and, 94–97\nwireless ad-hoc routing schemes and,\n99–100\nMetadata, definitions of, 48f\nMIBs. See Management Information Bases\nMicro-sequencer routines, 33–34\nMicro-sequencers, 32–33\nMisuse-based IDSs, 161, 186–88\nCPNs and, 187\nexpert systems and, 187\npattern recognition and, 187\nstate transition analysis and, 187–88\nMitigation detection mechanism, 399\nMITM. See Man-in-the-middle attacks\nMNS. See Modified NS protocol\nMobile ad hoc networks (MANETs), 66, 73\napplications of, 104, 160\nattacks, overview of in, 104–5, 160\nchallenges and future of, 131–32, 178\ncountermeasures to security attacks in,\n118–31\ndefinitions/characteristics of, 103–4,\n159–61, 382\nIDS Agent, conceptual model for in, 129f\nintrusion detection in, 159–80\nresearch topics in, 132\nrouting protocols in, 383–86\nsecurity attacks in, 105–18\nsmart sensor networks v., 66\nMobile agents (MA), 60, 166, 166f\nIDSs and, 163–62, 164f\nMobile Agents Place, 166, 166f\nMobile cellular networks. See Cellular\nmobile networks\nMobile code, 44–45, 60\nMobile IP handoff, 282–83\nMobile shortcuts, 211, 215, 216, 216f, 246\nregular network influenced by, 235, 236f\nsmall world networks influenced by,\n234–35, 235f\nMobile Switching Center (MSC), 185, 185f\nMobility trie. See EWMA-based mobility trie\nMobility-adjusted Authentication Protocol\n(MAP), 245, 251\narchitecture of, 252, 253f\nattacks in, 261\nauthentication in, 253\nINDEX\n415\n"
        },
        {
            "page_number": 415,
            "text": "authenticator in, 257\nclient module of, 258\ncommunication between SCRs in, 252–53\ndefined keys in, 253–54, 254f\ndefined messages in, 254–55\ndescription of, 246, 251–58\nKerberos v., 245, 246, 268f\nmessage exchanges in, 255–56\nMSN v., 268f\nother protocols v., 267, 268f\nperformance evaluation of, 261–68\nprotocol analysis in, 259–60\nrelated work and, 268–69\nSC module of, 257\nsecurity considerations in, 258–61\nserver module of, 256\nsimulation methodology and, 261–62\nsimulation model and, 262–64, 263f\nsimulation results in, 264–67\nstorage overhead in, 268\nModification attacks, 105, 105t, 106, 388,\n388f, 389, 389f\ndefense against, 125\nrouting protocols influenced by, 143\nModified NS protocol (MNS), 248\nMAP v., 268f\nMonitor, in CONFIDANT, 172–73, 173f\nMonitoring agents, 166–67\nMonitoring attacks, 105, 105t, 106, 106t\nMother duck, 73\nMSC. See Mobile Switching Center\nMulti-fence solution, MANET and, 132\nMulti-hop routing paths, 66\nMulti-layer attacks, 106t, 116\ndefense against, 127\nMultipath Key Reinforcement Scheme, 348,\n351–52\npros/cons of, 377–78\nMultiplication Algorithm, 10–11\nMultiSpace, 59\nNAS. See Network access server\nNAS/FA, 280–82\ninternal design of, 281f\nNAV. See Network allocation vector\nNeedham-Schroeder (NS), 246\nNegotiation manager, 48–49\nNetwork access server (NAS), 274\nNetwork allocation vector (NAV), 110\nNetwork initialization phase, 68\nNetwork layer attacks, 106t, 110–13\ndata forwarding phase and, 112\ndefense against, 122–25\nother advanced, 112–13\nparticular routing protocols and, 112\nrouting discovery phase and, 111–12\nrouting maintenance phase and, 112\nNetwork layer defense, 122–25\nNetwork Metadata (NtwkMeta), 48f, 49\nNetwork phases, 68\nNeural networks, IDSs and, 188\nNext Generation Intrusion Detection System\n(NIDES), 187\nNIDES. See Next Generation Intrusion\nDetection System\nNIST Recommended Finite Fields, 4t\nNomadism, ESS model influenced by, 222,\n223–24\nNonce, 107t\nNon-cryptography related attacks, 106\nNonrepudiation, security and, 119, 387\nNon-roaming network scenarios, 300–301,\n300f\nauthentication time v. SI in, 313f\nNon-stealthy attacks, 106\nNS. See Needham-Schroeder\nNS protocol, 246, 247–49\nmessage exchanges in, 248–49, 249f\nNtwkMeta. See Network Metadata\nNWSLite, 60\nOCEAN (Observation-based Cooperation\nEnforcement in Ad hoc Networks),\n174–75\nIDSs comparison with, 177t\nOdyssey, 44, 59\nOne-Way Key Chain-Based Authentication,\n371\nOnline predictors, 196\nOptimized inversion by square and multiply,\nAlgorithm 5, 18\nPacket dropping, 142–43\nPacket leashes, 123, 399\nPAD Metadata (PADMeta), 48f, 49\nPADMeta. See PAD Metadata\nPADs. See Protocol adaptors\nPairwise Key Pre-Distribution Model\n(PAM-S4), 75t, 76, 80\nPair-wise master key (PMK), 247, 247f\nPairwise shared keys, in LEAP, 367, 368–69\nPair-wise temporary key (PTK), 247, 247f\nPAKE protocols. See\nPassword-authenticated key exchange\nprotocols\n416\nINDEX\n"
        },
        {
            "page_number": 416,
            "text": "PAM-A1: Location-Limited Model, 75t, 76,\n80\nPAM-A2: Identity-Based Model, 75t,\n76–77, 81\nPAM-A3: Self-Certified Public Key Model,\n75t, 77, 81\nPAM-A4: Distributed CA Model, 75t, 77, 81\nPAM-A5: Trusted Path Model, 75t, 77–78,\n81\nPAMs. See Pre-authentication models\nPAM-S1: Secure Side Channel Model, 74,\n75t\nPAM-S2: PIN Model, 74, 75t, 80\nPAM-S3: Physical Contact Model, 74, 75t,\n80\nPAM-S4: Pairwise Key Pre-Distribution\nModel, 75t, 76, 80\nPAN. See Personal Area Network\nPassive security attacks, 105–6, 105t,\n388–92, 389f\nPassword Model (AM-H1), 78–79, 79t\nPassword-authenticated key exchange\n(PAKE) protocols, 78–79, 80\nPAT. See Protocol adaptation tree\nPath Manager, in CONFIDANT, 172–73,\n173f\nPath-key establishment stage\nin Basic Scheme, 345, 346, 347\nin Deployment Model, 374, 375\nPathrater, 171–72m 388\nIDSs comparison with, 177t\nPatient security processor (PSP), 328, 329,\n329f\nPattern recognition, misused-based IDSs\nand, 187\nP-BEST. See Production Based Expert\nSystem Toolset\nPCF. See Point coordination function\nPeer keying, in key management, 90–91\nPersonal Area Network (PAN), 328, 329,\n329f\nPhysical Contact Model (PAM-S3), 74, 75t,\n80\nPhysical layer defense, 120–21\nPhysical layer security attacks, 106t, 107\ndefense against, 120–21\nPIN Model (PAM-S2), 74, 75t, 80\nPK. See Primary key\nPKC. See Public-key cryptography\nPKC-based schemes, 100\nPKGs. See Private-key generators\nPlatform for adaptive encryption protocol,\n46–50, 47f\nPlugins, 60\nPMK. See Pair-wise master key\nPoint coordination function (PCF), 108\nPolynomial Pool-Based Key Predistribution\nScheme, 355–61\nfeatures of, 355–56\ngrid-based key assignment in, 358–61, 378\npros/cons of, 378\nrandom subset key assignment in, 356–58\nPosition-based routing, 155–56\ndefense against forwarding misbehaviors\nin, 156\ndefense against location spoofing in,\n155–56\ndefense against traffic abusing in, 156\nPosition-based routing calls, 141–42\nPower-law networks, 229, 233\nsimulations on, 230f, 233f\nPPM. See Prediction by Partial Matching\nPre-authentication models (PAMs), 74–78,\n75t\nasymmetric solutions in, 76–78\nsymmetric solutions in, 74–76\nPre-authentication phase, 67, 68, 69\nPrediction by Partial Matching (PPM), 198\nPre-patch, 211, 216, 218, 219, 222, 226, 228\nimportance of, 226, 240\nPrevention security mechanisms, 119,\n392–98\nPrimary key (PK), 253, 254, 254f\nPrivate-key generators (PKGs), 89, 90, 91,\n92, 98\nProactive routing protocols, 139\nProduction Based Expert System Toolset\n(P-BEST), 187\nProtocol adaptation, 45, 59–60\nProtocol adaptation tree (PAT), 50–51, 52f\nProtocol adaptors (PADs), 44, 46, 47f\nkey lengths and sizes of, 52t\nPAT and, 50–51, 52f\nProtocol boosters, 59\nProtocol stages, 68–69\nPseudorandom number attacks, 116–17\nexamples of, 107t\nPSP. See Patient security processor\nPTK. See Pair-wise temporary key\nPublic-key cryptography (PKC), 76, 84\nIBC v., 85, 88, 89\ninformation exchange and, 98–99\nmessage routing and, 99, 100\nQ-Composite Random Key Predistribution\nScheme, 348–51\nINDEX\n417\n"
        },
        {
            "page_number": 417,
            "text": "Basic Scheme v., 348–51\nLimited Global Payoff Requirement and,\n350–51\npros/cons of, 377\nresilience of, 349–50\nQoS. See Quality-of-Service\nQuality-of-Service (QoS), 246, 247\nRADIUS. See Remote Authentication\nDial-In User Service\nRandom Key Predistribution Scheme, 345.\nSee also Basic Scheme\nRandom Packet Dropping, 176\nRandom Pairwise Key Scheme, 348, 352–54\ndistributed node revocation in, 353–54\nefficiency of, 352\ninitialization stage of, 353\nnode replication and, 352\nnode-to-node authentication in, 352\npros/cons of, 378\nRandom Subset Key Predistribution, 356–58\nRandom walk model, 195\nRAP. See Rushing Attack Prevention\nRapid Prototyping Platform (RPP), 37\nRC4 stream cipher, 46\nReactive Location Service (RLS), 142, 168\nReactive routing protocols, 139–41\nReactive security mechanism, 119–20\nRealSecure, 187\nReauth-req message, 255\nReauth-res message, 255\nRecoded integers, scalar multiplication and,\n22–23\nRedhat Linux 9.0 kernel 2.4.20, 299\nReduction mod τ m, Algorithm 10, 27\nReinforcement, in Directed Diffusion, 386\nRelease of message contents, 388, 389f\nRemote Authentication Dial-In User Service\n(RADIUS), 278–79, 297, 298\nproxy of, 279\nReplay attacks, 100, 105, 105t, 106, 106t,\n389, 389f\nRepudiation attacks, 106t, 115\ndefense against, 124, 125f\nReputation System, in CONFIDANT,\n172–73, 173f\nReputation-based mechanism, 131\nRERR. See Route error\nRERR packets, 140\nResident point, 373–74\nResilience, in sensor networks, 344\nResistance, in sensor networks, 344\nResource constraints, level of, 72–73\nResource consumption attacks, 106t, 112,\n113\nRevocation, in sensor networks, 344\nRLS. See Reactive Location Service\nRoaming network scenarios, 300–301, 300f\nauthentication time v. SI in, 313f\nRoute cache poisoning, 144\nRoute discovery, 95–96, 385\nRoute Discovery with Fault Avoidance,\n153–54\nRoute error (RERR), 140\nRoute maintenance, 96–97, 385\nin ARAN, 152\nRoute maintenance phase attacks, 112\nRoute reply (RREP), 140\nRoute request (RREQ), 140\nRouting attack, illustration of, 111f\nRouting cache poisoning attack, 111\nRouting discovery phase attacks, 111\nRouting, in wireless ad-hoc networks,\n137–58\nattacks on, 142–44\nhybrid routing protocols in, 141\nposition-based routing protocols in,\n141–42\nproactive routing protocols in, 139\nreactive routing protocols in, 139–41\nsecure position-based routing in, 155–56\nsecure topology-based routing protocols\nin, 144–55\nRouting protocols, in MANETS and sensor\nnetworks, 383–86\nAODV in, 100, 112, 140, 384–85\nDirected Diffusion in, 386\nDSDV in, 139, 145, 384\nDSR in, 85, 95–97, 112, 140–41, 385–86\nsecurity attacks on, 383, 388–92\nsecurity mechanisms in, 119–20, 383,\n392–401\nsecurity services and, 383, 387\nRouting table overflow attack, 111, 144\nRover, 44, 59\nRPP. See Rapid Prototyping Platform\nRREP. See Route reply\nRREP packets, 140\nRREQ. See Route request\nRREQ packets, 140, 141\nRSA signature, 107t\nRunning system phase, 68, 69\nRushing Attack Prevention (RAP), 395\nRushing attacks, 100, 112, 113, 391\nRAP and, 395\n418\nINDEX\n"
        },
        {
            "page_number": 418,
            "text": "R(x)W(x) mod F(x), computation of, 11–13,\n13f, 14f\nSAODV. See Secure AODV\nSAR. See Security-aware adhoc routing\nSAR scheme, 100, 102, 401\nSB representation. See Signed binary\nrepresentation\nSC module, 257\nScalability, in sensor networks, 343\nScalar, example representations of, 35t\nScalar k, representation of, 35t\nScalar multiplication by double and add\nmethod, Algorithm 3, 9\nSCR configuration, MAP simulation and,\n263–64\nSCRs. See Security context routers\nSDD. See Secure directed diffusion protocol\nSEAD. See Secure Efficient Ad hoc\nDistance vector routing protocol\nSECTOR mechanism, 123\nSecure AODV (SAODV), 100, 102\ndescription of, 394–95\nSecure communications. See Identity-based\nsecure communications\nSecure directed diffusion protocol (SDD),\n397–98\nSecure Efficient Ad hoc Distance vector\nrouting protocol (SEAD), 100, 102, 105\ndescription of, 145–46, 393–94\nSecure Message Transmission protocol\n(SMT), 396\nSecure Routing Protocol (SRP), 99–100, 102\ndescription of, 146–48, 395–96\nSecure Side Channel Model (PAM-S1), 74,\n75t\nSecure topology-based routing protocols,\n144–55\nSecurity attacks, in MANETs, 383, 387–92\ncountermeasures to, 118–31\ncryptographic primitive attacks and, 107t,\n116–18\ncryptography related v. non-cryptography\nrelated, 106\ninternal v. external, 106, 390\nInternet model layers and, 106–16, 106t\npassive v. active, 105–6, 105t, 388–92,\n389f\nphysical layer, 106t, 107\nstealthy v. non-stealthy, 106\nSecurity attributes, 118–19\nSecurity context routers (SCRs), 245, 246\ncommunication between, 252–53\nSecurity mechanisms, 119–20, 383, 392–401\ndetection, 398–400\nother, 400–401\nprevention, 392––398\nSecurity model, in wireless ad-hoc networks,\n87–88\nSecurity policies\nauthentication forging in, 318\nauthentication time in, 311–14, 312t\ncost analysis associated with, 308–10\nDoS/traffic analysis in, 319–20\nfabrication of messages and, 319\nfeatures of, 303t\nhigh strength, 311\nhybrid, 302–3\nindividual, 301–2\nlow strength, 311\nmiddle strength, 311\nMITM and, 318–19\nsecurity index for, 304–8\nuser access control and, 319\nSecurity protocols, for WLANs, background\non, 297–98\nSecurity services, 383\ncomponents of, 387\nSecurity threats, 387–88, 388f\nSecurity-aware adhoc routing (SAR), 100,\n102, 401\nSelective Packet Dropping, 176\nSelf-Certified Public Key Model (PAM-A3),\n75t, 77, 81\nSemantic security, 362\nSender Setup, 364\nSensor networks, viii, 325–404\napplications of, 342\nclinical information systems and security\nin, 325–40\nconstraints of, 343–44\ndeployment knowledge in, 373–77\nevaluation metrics in, 344\nkey management schemes in, 341–80\nrequirements of, 342–43\nsecure routing in, 381–404\nsecurity attacks in, 388–92\nsecurity in, viii, 325–404\nsecurity protocols for, 361–66\nsecurity requirements for, 387\nsingle network-wide key in, 344–45\nSensor networks, in healthcare, 325–40. See\nalso WPAN technologies, in healthcare\naggregation principle and, 332–33\nconfinement principle and, 332\npatient privacy and, 330–31\nINDEX\n419\n"
        },
        {
            "page_number": 419,
            "text": "privacy and integrity rules, enforcement\nof in, 330–32\nresearch issues in, 326–27\nsecurity architecture of, 328–30, 329f\nsecurity policy for, 327–28\nsensed records, timestamping of in,\n331–32\nWPAN technologies and, 333–38\nService granting ticket (SGT), 248, 248f\nSession hijacking, 106t, 114–15\nSession key establishment, 69\nSGT. See Service granting ticket\nShared key discovery stage\nin Basic Scheme, 345, 346, 347\nin Deployment Model, 374, 375\nSIA, 397\nSIG. See Bluetooth Special Interest Group\nSigned binary (SB) representation, 23\nAlgorithm 6: scalar multiplication for, 23\nSigned encryption, 94\nSIM. See Subscriber Identity Module\nSimple Network Management Protocol\n(SNMP), 165\nSingle network-wide key, 344–45\nSinkhole attacks, 391\nSIR model. See\nSusceptible-Infected-Removed model\nSIS model, 217, 218\nSKC. See Symmetric-key cryptography\nSKC systems, 84\nSKC-based schemes, 100\nSmall world networks, 229, 232, 234–35,\n235f\nsimulations on, 230f, 232f\nSmart cards, 46\nSmart sensor networks, 66\nSmartphones, 212–13\nattacks on, 213, 214t, 215\ndefinition of, 212\nepidemics on, 211–41\nfeatures of, 212–13\nmobility of, 216f\nspread of epidemics on, 215–17\nSMT. See Secure Message Transmission\nprotocol\nSNEP, 361, 362–63, 396\nauthentication/freshness in, 362, 363\ndata confidentiality in, 362, 363\npros/cons of, 378\nSNMP. See Simple Network Management\nProtocol\nSpecification-based IDSs, 161\nSpecification-based techniques, 189\nSPINS protocol, for sensor networks, 361–66\nadvantages of, 365\ncryptographic base for, 365–66\ngoal of, 361\npros/cons of, 378\nrepresentations in, 361\nSNEP in, 361, 362–63, 378, 396\n{symbol}mTESLA in, 361, 363–65, 378\nSquaring, finite field arithmetic and, 15–17,\n17f\nSRI International, 187\nSRP. See Secure Routing Protocol\nStand-alone IDSs, 162\nStandard projective point system, 22t\nState Machine Algorithm, 36\nState Transition Analysis, misuse-based\nIDSs and, 187–88\nState Transition Diagrams, 187\nStatic shortcuts, 211, 215, 216, 246\nStatistics-based anomaly detection\ntechniques, 188\nStealthy attacks, 106\nSubscriber Identity Module (SIM), 193\nSusceptible-Infected-Removed (SIR) model,\n216, 217–18, 220, 225\nSybil attacks, 100, 102, 391–92\nrouting protocols and, 144\nSymmetric-key cryptography (SKC), 84\ninformation exchange and, 98–99\nmessage routing and, 99, 100\nSYN flooding attack, 106t, 114\nSynopsys FPGA Compiler II, 37\nSystem setup, in key management, 90\nSystem storage availability, 266f\nTACC, 59\nτ-adic integers, Algorithm 8: scalar\nmultiplication for, 25\nτ-adic NAF, Algorithm 9: generation of, 26\nTCP ACK Storm, 115f\nTCP Three-way Handshake, 114f\nTemporary key (TK), 253, 254, 254f\n{symbol}mTESLA, 148–49, 150, 361,\n363–65, 396\nauthenticated broadcasts and, 363–65\npros/cons of, 378\nTestbed, wireless LAN. See Wireless LAN\ntestbed\nTGS. See Ticket granting server\nTGT. See Ticket granting ticket\n32-bit/163-bit Address Map, 32f\nThroughput (bits/second), 310\nTicket granting server (TGS), 248, 248f\n420\nINDEX\n"
        },
        {
            "page_number": 420,
            "text": "Ticket granting ticket (TGT), 248, 248f\nTime of Flight (ToF), 155\nTK. See Temporary key\nτ-NAF, scalar multiplication and, 24–27\nToF. See Time of Flight\nToken-based mechanism, 130–31\nTopology, ESS model influenced by, 222–23\nTopology Factor, 226, 232, 232f, 233f\nτQ, computation of, 32\nTraffic abusing, 156\nTraffic analysis attacks, 105, 105t, 106,\n106t, 389, 389f\nsecurity policies and, 319–20\nTransformer tunnels, 59\nTransmission range, in sensor networks, 343\nTransport layer attacks, 106t, 113–15\ndefense against, 125–26\nTransport layer defense, 125–26\nTrie. See also EWMA-based mobility trie\ndefinition of, 201\nstorage of mobility information in, 201\nTriple DES, 44–45\nTrust Manager, in CONFIDANT, 172–73,\n173f\nTrust-based system, MANET and, 132\nTrusted Path Model (PAM-A5), 75t, 77–78,\n81\nTrusted Third Party (TTP), 70–72\navailabilities, 70–72, 71f\nroles of, in network, 70\nTTP. See Trusted Third Party\nUnauthorized participation, in security\npolicies, 319\nUnconditional Packet Dropping, 176\nUser access control, in security policies, 319\nUstat, 188\nValidation detection mechanism, 399–400\nVerilog HDL, 37\nVideo streaming, 245\nVirtex 2000E FPGA, 37\nVisitor Location Register (VLR), 185, 185f\nVLR. See Visitor Location Register\nVoice over IP (VoIP), 245, 246\nWatchdog, 171–72, 171f, 398\nIDSs comparison with, 177t\nWaxman random network, 229, 231, 233–34\nconfigurations of epidemic spread on, 246t\nnodes density’s influence on, 237f\nsimulation on, 231f, 234f\nWEP. See Wired equivalent privacy\nWEP policies, 302\nWEP weakness, 106t, 110\nWired equivalent privacy (WEP), 110, 297\nWired networks, intrusion detection for,\n186–89\nanomaly-based IDSs and, 188–89\nmisuse-based IDSs and, 186–88\nspecification-based techniques and, 189\nWireless ad hoc networks, 83–86\nIBC in, 88–89, 89f\nkey management in, 90–92\nnetwork model in, 86–87\nat recreation park, 86f\nsecure communications in, 93–97\nsecure routing in, 137–58\nsecurity model in, 87–88\nWireless Internet service providers (WISPs),\n274\nWireless LAN roaming, IEEE 802.11,\n278–83, 278f\nauthentication and key negotiation\ndemonstration in, 292, 292f\nhorizontal roaming architecture and,\n280–82, 280f\nmobile IP handoff, improvement of in,\n282–83\nradius proxy in, 279\ntransmission privacy in, 283–91\nWireless LAN testbed, 296–97, 299f\ndata acquisition in, 310–11\nexperimental results in, 311–18\nhardware configuration for, 298–99\nimplementation setup for, 298–303\nnetwork scenarios for, 300–301, 300f\nsoftware configuration for, 299–300\nWireless LANs (WLANs)\nAAA architecture and authentication for\nroaming in, 273–94\ncross-domain mobility-adaptive\nauthentication in, 245–71\nsecurity in, viii, 245–322\nsecurity protocols, experimental study on\nin, 295–322\nWireless personal area networks (WPANs),\n74, 326\nhealthcare sensor networks and, 333–38\nWireless transmission privacy, 283–91\nauthentication and key negotiation in,\n284–85, 286f\nauthentication of internet applications in,\n283–84, 283f\nprotocol for, 285–89, 289f\nsecurity analysis in, 289–91, 290f\nINDEX\n421\n"
        },
        {
            "page_number": 421,
            "text": "Wireless worm, 213\nWireless/mobile networks, security in, vii,\n3–62\nadaptive encryption protocol and, 43–62\nelliptic curve cryptographic co-processors\nand, 3–42\nWISPs. See Wireless Internet service\nproviders\nWitty worm, 226\nWLAN service integration architecture, 273,\n274f, 275\nWLANs. See Wireless LANs\nWormhole attacks, 100, 106t, 112, 390–91\ndefense against, 122–23\nrouting protocols and, 144\nWPAN technologies, in healthcare, 333–38.\nSee also IEEE 802.15.1 (Bluetooth) v.\nIEEE 802.15.4; Sensor networks, in\nhealthcare\nBluetooth v. IEEE 802.15.4 in, 336–38\nevent detection in, 334–35\nhealthcare applications and, 333–34\ninterconnection devices in, 334\npatient’s mobility and, 335\nresearch issues in, 326–27\nWPANs. See Wireless personal area\nnetworks\nXilinx Coregen tool, 31\nZBIDs. See Zone-Based Intrusion Detection\nSystems\nZone Routing Protocol (ZRP), 141\nZone-Based Intrusion Detection Systems\n(ZBIDSs), 169–71, 170f\nZRP. See Zone Routing Protocol\n422\nINDEX\n"
        },
        {
            "page_number": 422,
            "text": "SIGNALS AND COMMUNICATION TECHNOLOGY\n(continued from page ii) \nSDMA for Multipath Wireless Channels  \nLimiting Characteristics  \nand Stochastic Models  \nI.P. Kovalyov     ISBN 3-540-40225-X \nDigital Television  \nA Practical Guide for Engineers  \nW. Fischer     ISBN 3-540-01155-2 \nSpeech Enhancement \nJ. Benesty (Ed.) \nISBN 3-540-24039-X \nMultimedia Communication Technology  \nRepresentation, Transmission  \nand Identification of Multimedia Signals  \nJ.R. Ohm     ISBN 3-540-01249-4 \nInformation Measures  \nInformation and its Description in Science  \nand Engineering  \nC. Arndt     ISBN 3-540-40855-X \nProcessing of SAR Data \nFundamentals, Signal Processing,  \nInterferometry  \nA. Hein     ISBN 3-540-05043-4 \nChaos-Based Digital Communication Systems \nOperating Principles, Analysis Methods, and \nPerformance Evalutation \nF.C.M. Lau and C.K. Tse \nISBN 3-540-00602-8 \nAdaptive Signal Processing  \nApplication to Real-World Problems  \nJ. Benesty and Y. Huang (Eds.)  \nISBN 3-540-00051-8 \nMultimedia Information Retrieval and \nManagement Technological  \nFundamentals and Applications D. Feng, W.C. \nSiu, and H.J. Zhang (Eds.) \nISBN 3-540-00244-8 \nStructured Cable Systems  \nA.B. Semenov, S.K. Strizhakov,and I.R. \nSuncheley  \nISBN 3-540-43000-8 \nUMTS\nThe Physical Layer of the Universal Mobile \nTelecommunications System \nA. Springer and R. Weigel \nISBN 3-540-42162-9 \nAdvanced Theory of Signal Detection \nWeak Signal Detection in Generalized \nObeservations \nI. Song, J. Bae, and S.Y. Kim \nISBN 3-540-43064-4 \nWireless Internet Access over GSMand UMTS \nM. Taferner and E. Bonek \nISBN 3-540-42551-9 \n"
        }
    ]
}