Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Trust-based spectrum and energy efficient collaborative spectrum sensing in cognitive radio networks Mousavifar, Seyed Ali 2015

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2015_may_mousavifar_seyedali.pdf [ 1.68MB ]
Metadata
JSON: 24-1.0166185.json
JSON-LD: 24-1.0166185-ld.json
RDF/XML (Pretty): 24-1.0166185-rdf.xml
RDF/JSON: 24-1.0166185-rdf.json
Turtle: 24-1.0166185-turtle.txt
N-Triples: 24-1.0166185-rdf-ntriples.txt
Original Record: 24-1.0166185-source.json
Full Text
24-1.0166185-fulltext.txt
Citation
24-1.0166185.ris

Full Text

Trust-based Spetrum and Energy Eient Collaborative SpetrumSensing in Cognitive Radio NetworksbySeyed Ali MousavifarB. A. S., The University of British Columbia, 2006M. A. S., The University of British Columbia, 2009A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinThe Faulty of Graduate and Postdotoral Studies(Eletrial and Computer Engineering)THE UNIVERSITY OF BRITISH COLUMBIA(Vanouver)April 2015© Seyed Ali Mousavifar, 2015AbstratCognitive radio (CR) is a promising tehnology designed to improve the utilization oflightly used portions of the liensed spetrum while ensuring no undue interferenewith inumbent users (IUs). CR networks (CRNs) employ ollaborative spetrumsensing (CSS) methods to disover spetrum opportunities. Spetrum and energyoverhead osts play important roles in the eieny of CSS in CRNs.A trust-based energy eient CSS (EE-CSS) protool is proposed. The protoolahieves energy eieny by reduing the total number of sensing reports exhangedbetween the seondary users (SUs) and the fusion enter (FC) in the presene ofmisbehaving SUs (MSUs). The steady-state and transient behavior of the averagenumber of sensing reports and trust values of SUs in EE-CSS are analyzed and om-pared to those in traditional CSS (T-CSS). The impat of link outages on the globalfalse alarm (FA) probabilities, Qf , and the global miss detetion (MD) probabilities,Qmd, in EE-CSS and T-CSS is also analyzed.A entralized trust-based ollusion attak strategy, in onjuntion with integerlinear programming, is proposed to ompromise the deision of the FC in EE-CSS.The proposed strategy aims to attak only when it is likely to alter the deision ofthe FC. A mitigating sheme, based on the ross-orrelation of sensing reports, isproposed to identify SUs with abnormal behaviors and to eliminate them from thedeision making proess at the FC.We also propose a trust-based spetrum and energy eient CSS (SEE-CSS)iiAbstratsheme for the IEEE 802.22 standard wireless regional area network (WRAN). Theproposed sheme aims to redue the number of urgent oexistene situation (UCS) no-tiations transmitted from ustomer premise equipment (CPE) nodes to the WRANbase station (BS). The UCS messages inform the BS of the presene of ative IUson the liensed spetrum. We adapt the ollusion attak strategy for SEE-CSS andapply the ross-orrelation method at the BS to mitigate against the ollusion attak.The results show that while Qf and Qmd are kept the same in T-CSS and SEE-CSS,the SEE-CSS protool is more energy and spetrum eient.iiiPrefaeI hereby delare that I am the rst author of this thesis. Chapters 25 are based onworks under the supervision of Professor Cyril Leung. In addition, the work in Chap-ters 25 inludes ontributions that have been published or submitted for review, aslisted below.Publiations related to Chapter 2:ˆ S. A. Mousavifar and C. Leung, Energy Eient Collaborative SpetrumSensing Based on Trust Management in Cognitive Radio Networks, IEEETrans. Wireless Commun., vol. 14, no. 4, pp. 1927-1939, Apr. 2015.ˆ S. A. Mousavifar and C. Leung, Trust-Based Energy Eient SpetrumSensing in Cognitive Radio Networks, in Pro. IEEE VTC Fall, Las Vegas,NV, U.S.A., Sep 2013, pp. 1-6.Publiations related to Chapter 3:ˆ S. A. Mousavifar and C. Leung, Exat Transient Analysis in Cognitive RadioCollaborative Spetrum Sensing, IEEE Wireless Commun. Lett., Aepted forpubliation, Mar. 2015.ˆ S. A. Mousavifar and C. Leung, Transient Analysis in Cognitive Radio Col-laborative Spetrum Sensing, in Pro. IEEE WCNC, Istanbul, Turkey, Apr.2014, pp. 954-959.ivPrefaePubliations related to Chapter 4:ˆ S. A. Mousavifar and C. Leung, Centralized Collusion Attak in CognitiveRadio Collaborative Spetrum Sensing, in Pro. IEEE VTC Fall, Vanouver,Canada, Sep. 2014, pp. 1-6.Publiations related to Chapter 5:ˆ S. A. Mousavifar and C. Leung, Trust-based Spetrum and Energy EientCollaborative Spetrum Sensing in Cognitive Radio Networks, submitted Jan.2015.Other ontributions not presented in this dissertation are listed in Appendix A.vTable of ContentsAbstrat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPrefae . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viList of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvList of Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .xviiiList of Prinipal Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . xixAknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxviDediation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .xxviii1 Introdution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Bakground and Related Work . . . . . . . . . . . . . . . . . . . . . . 21.1.1 Sensing Methods and Collaborative Spetrum Sensing . . . . . 31.1.2 Data Fusion Tehniques . . . . . . . . . . . . . . . . . . . . . 31.1.3 MSUs in Collaborative Spetrum Sensing . . . . . . . . . . . . 41.1.4 Trust and Reputation Management Systems . . . . . . . . . . 51.1.5 Bandwidth and Energy Eient CSS . . . . . . . . . . . . . . 7viTABLE OF CONTENTS1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101.3 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111.4 Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131.5 Organization of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . 142 Trust-based Centralized EE-CSS . . . . . . . . . . . . . . . . . . . . . 162.1 Introdution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.2.1 Energy Detetion Method . . . . . . . . . . . . . . . . . . . . 192.2.2 Loal Missed Detetion and False Alarm Probabilities . . . . . 212.3 The Proposed EE-CSS Protool . . . . . . . . . . . . . . . . . . . . . 222.3.1 MAC Protool . . . . . . . . . . . . . . . . . . . . . . . . . . . 222.3.2 Trust Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242.3.3 Intermediate Deision . . . . . . . . . . . . . . . . . . . . . . . 242.3.4 Final Deision . . . . . . . . . . . . . . . . . . . . . . . . . . . 262.3.5 Maliious Seondary Users . . . . . . . . . . . . . . . . . . . . 262.3.6 Expliit Reports in EE-CSS . . . . . . . . . . . . . . . . . . . 292.3.7 Link Outages in EE-CSS . . . . . . . . . . . . . . . . . . . . . 302.4 Analysis of EE-CSS . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312.4.1 Derivation of T h, Tm, NH , and NM . . . . . . . . . . . . . . . 312.4.2 Energy Consumption in T-CSS and EE-CSS . . . . . . . . . . 342.4.3 Global False Alarm and Global Detetion Probabilities . . . . 372.5 Numerial Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493 Transient Analysis in EE-CSS . . . . . . . . . . . . . . . . . . . . . . . 513.1 Introdution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51viiTABLE OF CONTENTS3.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523.3 Derivation of Transient Average Trust Values and Number of Trans-mitted Sensing Reports . . . . . . . . . . . . . . . . . . . . . . . . . . 543.3.1 Evaluation of T h,k and Tm,k . . . . . . . . . . . . . . . . . . . 543.3.2 Evaluation of NHI ,k and NMI ,k . . . . . . . . . . . . . . . . . 553.3.3 Evaluation of NHII ,k and NMII ,k . . . . . . . . . . . . . . . . . 563.4 Numerial Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 604 A Collusion Attak in EE-CSS . . . . . . . . . . . . . . . . . . . . . . 614.1 Introdution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 614.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 634.3 Centralized Trust-based Collusion Attak in EE-CSS . . . . . . . . . 644.4 Cross-orrelation Filter Method in EE-CSS . . . . . . . . . . . . . . . 694.5 Numerial Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 724.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 765 Trust-based Centralized SEE-CSS . . . . . . . . . . . . . . . . . . . . 775.1 Introdution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 775.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 805.2.1 Network Model . . . . . . . . . . . . . . . . . . . . . . . . . . 805.2.2 Signal Propagation Model, Noise Model, and Sensing Method 815.2.3 The T-CSS Protool in the IEEE 802.22 WRAN . . . . . . . . 845.2.4 Deision Fusion and Trust Model . . . . . . . . . . . . . . . . 855.3 The Proposed SEE-CSS Protool . . . . . . . . . . . . . . . . . . . . 865.3.1 Intermediate Deision and Broadast . . . . . . . . . . . . . . 865.3.2 Partiipating CPEs in SEE-CSS and T-CSS . . . . . . . . . . 88viiiTABLE OF CONTENTS5.3.3 Spetrum and Energy Overheads in SEE-CSS and T-CSS . . . 935.4 Attak Strategies and a Mitigating Method . . . . . . . . . . . . . . . 975.4.1 Independent Attak . . . . . . . . . . . . . . . . . . . . . . . . 975.4.2 Centralized Collusion Attak in IEEE 802.22 WRAN . . . . . 985.4.3 Cross-orrelation Filter Method . . . . . . . . . . . . . . . . . 1015.5 Numerial Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1025.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1126 Conlusions and Topis for Future Researh . . . . . . . . . . . . . . 1146.1 Conlusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1146.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121Appendies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131A List of Other Publiations . . . . . . . . . . . . . . . . . . . . . . . . . 131B Eet of Delay in Communiating the BSI from PUBS to SUBS . 132C Evaluating Trust Based on the Loal Deision of the FC . . . . . . 136D Evaluating NH0hII , NH1hII , NH0mII , and NH1mII . . . . . . . . . . . . . . . . . . 142E Evaluating NHI ,k and NMI ,k . . . . . . . . . . . . . . . . . . . . . . . . . 144F Evaluating NHII ,k and NMII ,k . . . . . . . . . . . . . . . . . . . . . . . . 148G The Average Cardinality of CPE Sets . . . . . . . . . . . . . . . . . . 150ixList of Tables2.1 Independent Attak Poliy I . . . . . . . . . . . . . . . . . . . . . . . 272.2 Simulation Parameter Values I . . . . . . . . . . . . . . . . . . . . . . 433.1 Simulation Parameter Values II . . . . . . . . . . . . . . . . . . . . . 574.1 Simulation Parameter Values III . . . . . . . . . . . . . . . . . . . . . 725.1 Independent Attak Poliy II . . . . . . . . . . . . . . . . . . . . . . 985.2 Simulation Parameter Values IV . . . . . . . . . . . . . . . . . . . . . 103B.1 Simulation Parameter Values VI . . . . . . . . . . . . . . . . . . . . . 132C.1 Simulation Parameter Values V . . . . . . . . . . . . . . . . . . . . . 137xList of Figures1.1 A entralized CRN shares the spetrum with TV, Publi Safety, WiFi,and WiMAX networks. . . . . . . . . . . . . . . . . . . . . . . . . . . 21.2 Trust management system in a entralized CRN . . . . . . . . . . . . 62.1 A entralized overlay CRN with multiple HSUs and MSUs shares thespetrum with PUs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202.2 Components of the energy detetor . . . . . . . . . . . . . . . . . . . 212.3 The EE-CSS MAC protool: Phases I and II . . . . . . . . . . . . . . 232.4 Two-phase Deision making proess at the FC in EE-CSS . . . . . . . 272.5 Minimum number of sensing reports (or minimum number of parti-ipating SUs, i.e. eah SU transmits one sensing report to the FC)required from all SUs to satisfy Qmd < ǫmd and Qf < ǫf in the T-CSSversus EE-CSS protools. . . . . . . . . . . . . . . . . . . . . . . . . . 412.6 Average total energy onsumed to transmit and to reeive pakets asa funtion of θ for the T-CSS and EE-CSS protools. . . . . . . . . . 422.7 Probability of hoosing an honest or maliious SU in Phase I as afuntion of time, k (in time slots), with pout = pBr.out = 0.2, H = 2, andM = 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44xiLIST OF FIGURES2.8 NMII and NHII as a funtion of time, k (in time slots), with pout =pBr.out = 0.2, H = 4, and M = 4. The dashed horizontal lines show thesteady state values for the orresponding urves. . . . . . . . . . . . . 452.9 The steady-state average trust value as a funtion of time, k (in timeslots), with pout = pBr.out = 0.2, H = 4, and M = 4. The dashedhorizontal lines show the steady state values for the orrespondingurves. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462.10 The global FA and MD probabilities (Qf and Qmd, respetively) as afuntion of time, k (in time slots), with pout = 0.2, pBr.out = 0.2, H = 4,and M = 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462.11 ROC for EE-CSS and T-CSS when H = 3 and M = 1 for three outageprobability values, pout = 0, 0.5, 1. . . . . . . . . . . . . . . . . . . . . 472.12 Left Y axis: Qf and Qmd in EE-CSS and T-CSS when M = 5 as afuntion ofHM . Right Y axis:NHH in EE-CSS M = 5 as a funtion ofHM . In this gure pout = 0 . . . . . . . . . . . . . . . . . . . . . . . . 483.1 T h,k and Tm,k as a funtion of time slot k. Theoretial urves obtainedusing (3.8) and (3.9). . . . . . . . . . . . . . . . . . . . . . . . . . . 583.2 NHI ,k and NMI ,k as a funtion of time slot k. Theoretial urvesobtained using (E.9) and (E.10). . . . . . . . . . . . . . . . . . . . . . 593.3 NHII ,k and NMII ,k as a funtion of time slot k. Theoretial urvesobtained using (3.13) and (3.14). . . . . . . . . . . . . . . . . . . . . 594.1 A entralized overlay CRN with H HSUs, M MSUs, an FC, and anMFC, the MFC and MSUs ollaborate in a ollusion attak strategy. 634.2 The operations of the FC and MFC during the two-phase entralizedollusion attak in EE-CSS . . . . . . . . . . . . . . . . . . . . . . . . 68xiiLIST OF FIGURES4.3 Pseudoode for the ross-orrelation lter Method . . . . . . . . . . . 714.4 The global FA and MD probabilities (Qf and Qmd, respetively) as afuntion of time slot k for the CRN with independent and ollusionattaks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 734.5 The average ross-orrelation of SU pairs as a funtion of time slot k. 744.6 The global FA and MD probabilities (Qf and Qmd, respetively) as afuntion of time slot k for CRN with ollusion attak and CCF method 754.7 The average suess-per-attak ratio (SAR) as a funtion of time slot k. 755.1 Network model in a WRAN . . . . . . . . . . . . . . . . . . . . . . . 815.2 The SEE-CSS protool in CSSP k (k = ⌊ i+12 ⌋) . . . . . . . . . . . . . 875.3 The global false alarm probability in a WRAN as a funtion of k withollusion attak (with and without CCF) and independent attak usingOR and TW deisions at the BS with M = 10, H = 5, Ns = 1000 . . 1045.4 The global detetion probability in a WRAN as a funtion of k withollusion attak (with and without CCF) and independent attak usingOR and TW deisions at the BS with M = 10 and H = 5 . . . . . . . 1055.5 The normalized number of UCS and UCS-D notiations in the T-CSSand SEE-CSS protools, respetively, by HCPEs and MCPEs in Framei+ 1 as a funtion of Ns for H = 25, 250 (M = 25, 250) . . . . . . . . 1075.6 Left Y axis: the throughput eieny ratio as a funtion of Ns. RightY axis: the energy eieny ratio as a funtion of Ns . . . . . . . . . 1085.7 The probabilities pCPE∗ and pBS (pBS , (1− pCPE∗)) as a funtion ofpus,i and psucc,i for H = 25, 250 (M = 25, 250) and Ns = 2000 . . . . . 109xiiiLIST OF FIGURES5.8 The normalized number of UCS and UCS-D notiation reports in T-CSS and SEE-CSS by HCPEs in Frames i and i + 1 as a funtion ofpus,i and psucc,i for H = 25, 250 (M = 25, 250) and Ns = 2000 . . . . . 1105.9 Left Y axis: the throughput eieny ratio as a funtion of pus,i andpsucc,i for H = 25, 250 (M = 25, 250) and Ns = 2000. Right Y axis:the energy eieny (right Y-axis) ratio as a funtion of pus,i and psucc,ifor H = 25, 250 (M = 25, 250) and Ns = 2000 . . . . . . . . . . . . . 112B.1 The average trust value of HSUs and MSUs as a funtion of time slotk for several Nd values. The dotted at line shows the steady-stateaverage trust value of HSUs and MSUs. . . . . . . . . . . . . . . . . . 133B.2 The probability of HSU or MSU hosen in Phase I as a funtion oftime slot k for several Nd values . . . . . . . . . . . . . . . . . . . . . 133B.3 The average number of reports transmitted in Phase II as a funtionof time slot k for several Nd values . . . . . . . . . . . . . . . . . . . 134B.4 The global FA and MD probabilities as a funtion of time slot k forseveral Nd values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135C.1 The steady-state average trust value of HSUs and MSUs as a funtionof time slot k evaluated using BSI or the loal deision of the FC . . 138C.2 The probability of HSU or MSU hosen in Phase I as a funtion oftime slot k for trust evaluation using BSI and the FC loal deision. . 139C.3 The average number of reports transmitted in Phase II as a funtionof time slot k for trust evaluation using BSI and the FC loal deision. 140C.4 The global FA and MD probabilities (Qf and Qmd, respetively) as afuntion of time slot k for trust evaluation using BSI and the FC loaldeision. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141xivList of AbbreviationsATSC Advaned Television Systems CommitteeAWGN Additive White Gaussian NoiseBLM-REP Bulk Measurement ReportsBS Base StationBSI Band State InformationCCF Cross-orrelation FilterCDF Cumulative Distribution FuntionCH Cluster HeadCHS-REQ Channel Swith RequestCM Cluster MemberCPE Customer Premise EquipmentCR Cognitive RadioCRN Cognitive Radio NetworkCSS Collaborative Spetrum SensingCSSP Collaborative Spetrum Sensing PeriodDF Deision FusionDS DownstreamDSAA Dynami Spetrum Aess AttakDSMA Dynami Spetrum Multiple AessEE-CSS Energy Eient Collaborative Spetrum SensingxvList of AbbreviationsERP Eetive Radiated PowerFA False AlarmFC Fusion CenterFCH Frame Control HeaderGMH Generi MAC headerHCPE Honest Customer Premise EquipmentHDC Hard Deision CombiningHSU Honest Seondary Usersi.i.d. Identially and Independent DistributionIEEE Institute of Eletrial and Eletronis EngineersITU-R International Teleommuniation Union Radio-ommuniationIU Inumbent UserMANET Mobile Ad Ho NetworksMBS Misbehaving Base StationMCPE Misbehaving Customer Premise EquipmentMD Miss DetetionMFC Misbehaving Fusion CenterMSU Misbehaving Seondary UserNF Noise Figurepmf probability mass funtionPU Primary UsersPUBS Primary User Base StationRHS Right Hand SideROC Reeiver Operating Charateristirv Random VariablexviList of AbbreviationsSDC Soft Deision CombiningSEE-CSS Spetrum and Energy Eient Collaborative Spetrum SensingSNR Signal-to-Noise-RatioSSDF Spetrum Sensing Data FalsiationSU Seondary UsersSUBS Seondary User Base StationT-CSS Traditional Collaborative Spetrum SensingTMS Trust Management SystemsTRMS Trust and Reputation Management SystemsTW Trust-WeightedUCS Urgent Coexistene SituationUCS-D Urgent Coexistene Situation DisagreementUS UpstreamWiFi Wireless FidelityWiMAX Worldwide Interoperability for Mirowave AessWRAN Wireless Regional Area NetworkWSN Wireless Sensor NetworksxviiList of Notations(·)∗ Complex onjugate| · | Absolute value of a omplex number or ardinality of a setB(a, b, c) Bernoulli Binomial operation ∑bl=a(bl)(c)l(1− c)b−lN (µ, σ2) A Gaussian random variable with mean µ and variane σ2Γ(x) Gamma Funtion (∫∞0 e−ttx−1dt)Γ(a, x) Gamma Inomplete Funtion ( 1Γ(a)∫ x0 e−tta−1dt)E{·} Statistial expetation of a random variableFX(x) Commutative distributed funtion of random variable xN The set of natural numbersPr{·} Probability of an eventQ(x) Gaussian Qfuntion ( 1√2pi∫∞x e−u22 du)Qu(a, b) Generalized Marum Q-funtion ( 1au−1∫∞b xuexp(−(x2+a2)2 )Iu−1(ax)dx),where Iu−1 Modied Bessel Funtion of the rst kind of order u− 1X Average value of a random variableX̂ Estimated or observed value of a random variablexviiiList of Prinipal SymbolsThe list of symbols and notations are sorted based on the order predened in "nomenl"pakage of LATEX, i.e.1. Caligraphed fonts2. Lower-ased and upper-ased Greek letters (for a letter, the priority is given tothe lower-ased letter)3. Lower-ased and upper-ased English letters (for a letter, the priority is givento the lower-ased letter).The list of symbol is presented as follows.Ak The set of ative CPEs in CSSP kAk,a The set of ative a ∈ {H,M} denoting HCPEs (HSUs) and MCPEs (MSUs),respetively, in CSSP kBk,a|c The set of ative a ∈ {H,M} in CSSP k whih have deided that thehannel is busy given hypothesis c ∈ {H0,H1}Di+1,a|c The set of ative a ∈ {H,M} whih disagree with IDBS,kin Frame i+ 1 given hypothesis c ∈ {H0,H1}Gi,a|c The set of ativea ∈ {H,M} with US resoures whih have transmittedUCS notiations in Frame i given hypothesis c ∈ {H0,H1}H0 Idle hannel hypothesisxixList of Prinipal SymbolsH1 busy hannel hypothesisIi|c The set of ative CPEs whih have suessfully transmittedtheir UCS notiation to the BS in Frame i given hypothesis c ∈ {H0,H1}PS i|a The set of ative a ∈ {H,M} whih an suessfully transmitted their UCSnotiation to the BS in the ontention-based period in Frame i withprobability psucc,iSi,a|c The set of ative a ∈ {H,M} whih have suessfully transmitted their UCSnotiation to the BS in the ontention-based period in Frame i,given hypothesis c ∈ {H0,H1}Si+1,a|c The set of ative a ∈ {H,M} whih have suessfully transmitted their UCSnotiation to the BS in the ontention-based period in Frame i+ 1,given hypothesis c ∈ {H0,H1}αpl Path loss exponentǫ Target global error (ǫmd + ǫf )ǫup Upper bound error in the in Chebyshev's inequalityǫf Target global FA probabilityǫmd Target global MD probabilityηi+1 The throughput eieny ratio in Frame i+ 1ηE,k The energy eieny in the kth CSSPγj The SNR in Watt at the jth sensing nodeγj,dBm The SNR in dB at the jth sensing nodeλn The energy deision threshold at nth SUµj,dB The average power reeived at the jth CPE without shadow fadingψth,k The ross-orrelation ondene interval at time slot (CSSP) kσ2dBm Shadow fading gain varianexxList of Prinipal Symbolsσ2hn The hannel gain variane at the nth SUσ2n The noise variane at the nth SUτth,b The trust threshold in Collusion attah where b ∈ {BS,FC,MFC,MBS}θ Reeive-to-transmit energy ratiobblm The number of bits in BLM-REPbdata∗ The number of bits in GMH plus payloadbSEE,i+1|H0 The number of bits transmitted from the CPEs to the BS in Frame i+ 1given that the hannel is idle in SEE-CSSbSEE,i+1|H1 The number of bits transmitted from the CPEs to the BS in Frame i+ 1given that the hannel is busy in SEE-CSSbT,i+1|H0 The number of bits transmitted from the CPEs to the BS in Frame i+ 1given that the hannel is idle in T-CSSbT,i+1|H1 The number of bits transmitted from the CPEs to the BS in Frame i+ 1given that the hannel is busy in T-CSSbucs The number of bits in GMH of the UCS notiationB Channel bandwidthd0 Referene distanedj The distane from the DTV broadast antenna to the jth CPEDe,k The loal deision of e ∈ {BS,FC, h, ID, n, n∗, m}at time slot k,Dth Fusion deision thresholdEEE Total energy onsumed in steady-state in EE-CSSESEE,k The total energy onsumed in SEE-CSS at the kth CSSPErx Energy onsumed to reeived a paketErx,g The energy onsumed to reeived a paket at g ∈ {FC, h, SUn∗}ET Total energy onsumed in T-CSSxxiList of Prinipal SymbolsET,k The total energy onsumed in T-CSS at the kth CSSPEtx The energy onsumed to transmit a paketEtx,g The energy onsumed to transmit a paket at g ∈ {FC, h, SUn∗}FDb,k The nal deision at b ∈ {FC,BS,MFC,MBS} at the k CSSP (or time slot)h The hth HSUhn The hannel gain from PUBS to the nth SUH The number of honest SUs or CPEsIDFC,k The intermediate deision at the FC at the kth CSSP (or time slot)IDMFC,k The intermediate deision at the MFC at the kth CSSP (or time slot)kB Boltzmann onstant (1.381×10-23 J/k)m The mth MSUM The number of misbehaving SUs or CPEsn The nth SUn∗ The hosen SU or CPE in Phase IN Number of sensing entities inluding the FCNF Noise gureN cal,k The number of sensing reports transmitted by a ∈ {H,M, h,m} inPhase l ∈ {I, II} of the kth time slot for hannel hypothesis c ∈ {H0,H1}The subsript k for the steady-state values is dropped.The supersript a for the addition of both hypothesis is dropped.The subsript b for the addition of both Phases I and II is dropped.The subsript H with M for MSUs is replaed.The subsript h and m orrespond to hth and mth SU.Ns Number of samples at eah sensorpα0 The probability that MSU reports dishonestly, if it has deided H0xxiiList of Prinipal Symbolspα1 The probability that MSU reports dishonestly, if it has deided H1pα0,0 The probability that MSU reports dishonestly, if it has deided H0and the intermediate deision is H0pα0,1 The probability that MSU reports dishonestly, if it has deided H1and the intermediate deision is H0pα1,0 The probability that MSU reports dishonestly, if it has deided H0and the intermediate deision is H1pα1,1 The probability that MSU reports dishonestly, if it has deided H1and the intermediate deision is H1pact,k The probability that eah CPE remains ative in the kth CSSPpd,r The loal detetion probability at r ∈ {BS,FC, h, ID, n,m, }pf,r The loal FA probability at at r ∈ {BS,FC, h, ID, n,m, }pBr.f,m,FC The FA probability of the mth MSU seen at the FC given thatthere is a link outage from the FC to the MSUpH0 The probability that the hannel is idlepH0 The probability that the hannel is busypmd,r The loal MD probability at r ∈ {BS,FC, h, ID, n,m, }pmd,m,FC The MD probability of the mth MSU seen at the FCpBr.md,m,FC The MD probability of the mth MSU seen at the FC given thatthere is a link outage from the FC to the MSUpBr.out,h The outage probability of the link from the FC to the hth HSUpBr.out,m The outage probability of the link from the FC to the hth HSUpout,h The outage probability of the link from the hth HSU to the FCpout,m The outage probability of the link from the mth MSU to the FCpsucc,i The probability that CPEs an transmit UCS notiationsxxiiiList of Prinipal Symbolssuessfully in frame ipus,H,i The probability that the BS alloates US resoures to HCPEs in Frame ipus,M,i The probability that the BS alloates US resoures to MCPEs in Frame iPj Reeived signal power at the jth CPE in WattPn0,dBm The average noise power at sensing nodes in dBmPDTV,dBm The transmit power of DTV signal in dBmQd The global detetion probabilityQf The global FA probabilityQmd The global MD probabilityrn,k Sum of the reward from the rst time slot to the kth time slotRk,j1j2 Cross-orrelation of the reports for CPEj1 and CPEj2 at the kth CSSPs(t) The transmitted signal from PUBSSdB Shadow fading power gainT Observation periodTc The temperature onstant, i.e. 300kTCC The period at whih ross-orrelation proedure takes plaeTu,k The trust at u ∈ {FC, h, j∗, n, n∗, m,MBS,MFC} at time slot k,We drop the subsript k for the steady-state values.Un,k Sensed energy at the nth SU in time slot kV Number of spetrum bandswn(t) AWGN at the nth SU at time tW Sampling frequenyXv,k Sensing report reeived from v ∈ {FC, h, j,m, n, n∗} at the FC (or BS) at timeslot (or CSSP) kyn(t) Signal reeived at the nth SU at time txxivList of Prinipal Symbolsyn,k[i] ith sample of the reeived signal at nth SU in time slot kYj,k Sensing reports of jth MCPE reeived at the MBS during the kth CSSPYMBS,k Loal deision of the MBS in the kth CSSPZj,k Solution of the ILPO problemxxvAknowledgmentsI would like to express my deep and sinere gratitude to my advisor, Professor CyrilLeung, for aepting me as his student, in the Masters and PhD programs at UBCand for his invaluable support and guidane over the past eight years. His knowl-edge, experiene, and onstrutive ritiism, both on the professional and personallevels, have been invaluable. Professor Leung is a dediated mentor and his aademiprofessionalism is a onstant soure of motivation for all his students, inluding me.He will ontinue to be a role model in my aademi and personal areer.I would like to thank my supervisory ommittee members, Professor Jane Wangand Professor Vitor Leung, for evaluating my work and providing valuable feedbakand suggestions.I would also like to express my gratitude to NSERC Canada and Bell Canada.This work was supported in part by the Natural Sienes and Engineering ResearhCounil (NSERC) of Canada under Grants CRDPJ 395689-09 and RGPIN 1731-2013,and by a ontrat from Bell Canada.I would like to thank every member in our wireless ommuniation group fortheir friendship, enouragement, and help. I have learned a lot from eah one ofthem, and for that I am deeply grateful. Speial thanks to the past and urrentexeutives of the Eletrial and Computer Engineering (ECE) Graduate StudentAssoiation for dediating their times and volunteering endless hours to enhane thesoial, professional, and aademi aspets of the lives of ECE graduate students.xxviAknowledgmentsI would like to thank my ollaborators and oauthors over the ourse of my Mas-ter's and PhD program, Miss Y. Deng, Mr. A. Jassal, Dr. T. Q. Duong, Dr. M. Elka-shlan, Dr. M. Hasna, Dr. T. Khattab, Mr. S. Li, Mr. Y. Liu, Dr. P. Samadi, andMr. D. J. Su.Finally, I owe enduring gratitude to my parents and siblings. Sinere gratitudegoes to my wife, Irene, and her family for their love, support, and patiene. Withoutthe sarie and relentless support from my family, I would not be where I am today.xxviiDediationTo my familyxxviiiChapter 1IntrodutionIn a ognitive radio network (CRN), ommuniation devies adaptively hange theirtransmission and reeption harateristis so as to use sare network resoures moreeiently. The adaptation is based on the hanging network environment [1, 2℄, e.g.hannel fading, user behaviors, and network tra. Studies have shown that someradio spetrum bands alloated exlusively to (liensed) primary users (PUs), alsoknown as inumbent users (IUs), are greatly under-utilized [3℄. In a CRN, unliensedseondary users (SUs) an sense and aess liensed spetrum bands as long as theydo not interfere unduly with PUs [2℄. Two CRN paradigms have been proposed:underlay and overlay. The overlay CRN refers to spetrum aess where the SUsutilize unused PU spetrum for transmission opportunistially. In the underlay CRNparadigm, the CRN an oexist with the primary network (PN) on the same spetrumband as long as it does not generate interferene power at the PN whih exeeds apredetermined threshold. A entralized CRN in onjuntion with PU networks (e.g.Publi safety, WiFi, WiMAX, and TV networks) is shown in Fig. 1.1, where theCRN aims to share or utilize the unused PU spetrum. The CRN should utilize thespetrum in a manner that is largely transparent to the PU network.Collaborative spetrum sensing (CSS) has been proposed in whih sensing reportsfrom SUs are sent to one or more deision making entities to produe more reliabledeisions on the state of the spetrum band [47℄. However, in the presene ofmisbehaving SUs (MSUs), the integrity of the reports sent by SUs need to be assessed1Chapter 1. IntrodutionSUBSPublic Safety NetworksUHF/VHF DTVIEEE 802.22 WiMAX NetworksIEEE 802.16WiFi Networks IEEE 802.11MSUHSUWiMAX BSTV Broadcasting  AntennaSU BS PUBroadcasting stationWiFi RouterPUPU (CPE)PUFigure 1.1: A entralized CRN shares the spetrum with TV, Publi Safety, WiFi,and WiMAX networks.to avoid interferene with PUs and improve spetrum utilization [813℄. Trust andreputation management systems (TRMSs) have been proposed to ombat maliiousbehaviors in CRNs [14℄. Our aim is to study eient methods for sensing, reporting,data olletion, and data fusion in CRNs with MSUs.1.1 Bakground and Related WorkWe provide some bakground information and review the existing literature aboutsensing methods, ollaborative spetrum sensing, data fusion tehniques, seuritythreats and mitigation strategies in CRNs, and bandwidth and energy usage in CSS.2Chapter 1. Introdution1.1.1 Sensing Methods and Collaborative Spetrum SensingSpetrum sensing in the ontext of CRN is dened as measuring the energy andobtaining the harateristis of the liensed spetrum as a funtion of time, frequeny,spae, and ode [7℄. Spetrum sensing is one of the most hallenging tasks in CRNs.Although several aspets of spetrum sensing (e.g. hardware requirements [15℄, multi-spetrum band sensing [16℄, sensing duration and frequeny [17℄, spread spetrumsensing user signals [18℄, et.) are being investigated, we briey desribe a few topisrelated to our work. Several sensing methods an be used to detet the state of aspetrum band. The main sensing methods are [7℄: mathed ltering [19℄, ylo-stationary detetion [20, 21℄, and energy detetion [4, 5, 2224℄. The most ommonlyused method is energy detetion as it has the lowest omputational omplexity anddoes not require any a priori knowledge of the PU signal [2224℄. This is an importantfator beause the signaling sheme used by PUs may be unknown to SUs. However,the energy detetion method is also the least aurate [7℄. Sensing reports providedby SUs for a given liensed band may dier due to dierenes in hannel fadinggains, loations of SUs and primary network transmitters, number of signal energyquantization levels used at the sensing SU, and the auray of sensing methods.However, CSS methods are used to improve the overall false alarm (FA) and missdetetion (MD) probabilities. During the CSS proess, sensing reports are gatheredfrom SUs in a entralized or distributed manner using a variety of tehniques [47,2225℄.1.1.2 Data Fusion TehniquesThe nal deision about the state of the spetrum usage an be made at one or morenodes based on the infrastruture of the network. In a entralized CRN, an entity at3Chapter 1. Introdutionthe seondary user base station (SUBS), namely the fusion enter (FC), reeives thesensing reports from SUs and produes a nal deision on the state of the spetrumband availability. In data fusion (DF), also known as deision ombining, tehniquessuh as AND/OR Rule [11, 26℄, Ki Rule [27℄, and Majority Rule [28℄, the FC rulesthat the PU hannel is busy when all/one, i out of K, and at least half of sensingentities report busy hannels, respetively. Other tehniques whih are based on aBayesian riterion and the Neyman-Pearson test [12, 2932℄ require the knowledgeof some a priori probabilities. For example, the a priori probability of busy (oridle) hannel and the a priori probability distribution of the sensed energy given thetype of the SU (e.g. misbehaving SU or honest SU) are required in [29, 31℄ to testthe busy-state and idle-state hypotheses, denoted by H1 and H0, respetively. TheSUs may transmit a binary deision on the state of the band (i.e. hard deision) or areport with the measured energy of the spetrum band (i.e. soft deisions) to the FC.In [27,33℄, the hard deision fusion (HDC) is shown to be omparable to those of softdeision ombining (SDC) with noise unertainty fator. Although SDC an performbetter than HDC [30℄ in some senarios, HDC is less omputationally omplex andmore pratial than SDC.1.1.3 MSUs in Collaborative Spetrum SensingDierenes in the geographial loations of SUs and PU network transmitters, hannelfading gains, and auray of sensing methods are not the only auses of variationsamong the sensing reports. Misbehaving SUs (MSUs) also aim to inuene the naldeision of the FC maliiously or for selsh reasons. A user with maliious behavioraims to degrade the system or disrupt the others in networks with no expliit intentionto maximize its own gain while a user with selsh behavior aims to maximize its gain4Chapter 1. Introdutionat others' expense [14℄.Seurity of the CRN in the presene of threats from MSUs is an ongoing eort. Se-urity threats are divided into two ategories: hard seurity threats and soft seuritythreats. On one hand, hard seurity measures suh as ryptography are proposedin [3436℄ to mitigate against hard seurity threats, i.e. eavesdropping, dynamispetrum aess attaks (DSAA) [810℄, unauthorized users and aess ontrol, PUemulations [36, 37℄, et., in CRNs. On the other hand, soft seurity measures areproposed in [13, 24, 29, 31, 34, 3845℄ to defend against soft seurity threats suh asspetrum sensing data falsiation (SSDF), where MSUs transmit dishonest reportsto degrade (or disrupt) the system or to maximize their own gains in the CRN.The seurity threats an be independent or oordinated. In independent attaks,MSUs try to harm the CRN without onsidering the reports from other SUs [24,29,38,41℄. In oordinated attaks [13,4649℄, MSUs an ollaborate (ollude) and an listento the reports from other SUs (inluding honest SUs) in order to inuene the deisionof the FC more eetively. The impat of MSUs should not be underestimatedbeause they an severely inuene the performane of CSS [8,13,24,29,31,47,48℄ asit has been shown that the CSS in CRNs is inreasingly vulnerable to soft seuritythreats beause it relies heavily on the sensing reports from SUs.1.1.4 Trust and Reputation Management SystemsTrust and reputation management systems have been proposed to ombat maliiousand selsh behaviors in CRNs [13, 24, 29, 31, 4043℄. Reputation refers to publiknowledge and it is established based on the opinion of all the members in a networktowards one member. Trust is the pereption of a member about the behavior ofanother member in the network; however this pereption may not be reiproal [14,5Chapter 1. IntrodutionTrust Management Systems in Centralized CRNsFirst-Hand EvidenceTrust EvaluationInteraction Decision MakingInteraction Outcome EvaluationEvidence SpaceTrust SpaceIndividual-Level Trust Model System-Level Trust ModelTrust-Based Reward/Punishment ModuleFigure 1.2: Trust management system in a entralized CRN50℄. In distributed networks suh as mobile ad Ho networks (MANETs) and wirelesssensor networks (WSNs), both trust and reputation parameters form an opinion ofone member about another member. In entralized networks, only trust is used toform an opinion about eah member that interats with a entral member suh as theFC. The TRMS framework requires interations between the members in the networkand a funtion to evaluate the degrees of positive and negative behavior in order togive rewards and penalties, respetively.We show an example of trust management system (TMS) arhiteture in a en-tralized CRN whih inludes individual and system level trust models in Fig. 1.2.The individual-level trust model onsists of rst hand evidene from the interationbetween an SU and the FC, a trust evaluation (mapping) funtion, a deision based6Chapter 1. Introdutionon the interation with the SUs, and the deision outome based on the interation.For example, if the FC's deision based on the evidene from the SUs with respet tothe interation outome is inaurate, the trust value of the SU will degrade. Thereare several methods whih an be used to ompute trust values: beta distribution (inonjuntion with Laplae smoothing) [5153℄, beta distribution with the onept ofunertainty [25℄, and suspiious value and onsisteny [31℄.For the deision making based on the evidene (sensing reports) from SUs, theFC typially forms a weighted sum of the sensing reports and their auraies,∑Nn Tn,kXn,k, and ompares it against Dth [14,51,52℄, where Xn,k denotes the hard orsoft loal deision (e.g. binary or energy value) reported by the nth SU, denoted bySUn, at kth time slot, Dth denotes the deision threshold at the FC, and Tn,k denotesthe trust value of SUn at the kth time slot. At the system level trust model, the FCan take punitive measures to disourage nodes from behaving selshly or maliiously,e.g. the FC an eliminate misbehaving SUs from partiipating in future CSS or de-prive the misbehaving SUs from aessing the available liensed bandwidth. Otherstrategies suh as Point System [13℄, Abnormality Detetion [11℄, and Multi-StageFiltering [43℄ have also been proposed to defend against MSU attaks.1.1.5 Bandwidth and Energy Eient CSSSeveral studies have examined the energy and bandwidth overhead osts assoiatedwith CSS methods [2226,28,3032,5476℄. In [26,55℄, the authors propose an energydetetion tehnique whih an redue the number of reports transmitted by SUs. Thetehnique uses two energy deision thresholds, denoted by λ1 and λ2, instead of theonventional single energy deision threshold, λ. SUn ompares its deteted energy,denoted by Un, with λ1 and λ2 and proeeds as follows:7Chapter 1. Introdutionˆ If Un ≥ λ2, the SU deides that hannel is busy, i.e. H1ˆ If Un ≤ λ1, the SU deides that hannel is idle, i.e. H0.ˆ If λ1 < Un < λ2, the SU has low ondene in its deision and does not send areport to the FC.This method redues the number of sensing reports sent to the FC. Thus, energy isused more eiently. In [56℄, a brute-fore approah is proposed to nd the optimumnumber of reporting SUs needed when bandwidth (or energy) eieny, global falsealarm probability (denoted by Qf ), and global detetion probability (denoted by Qd)are onsidered. An objetive funtion whih weighs Qf , Qd, and bandwidth (or en-ergy) usage is optimized with respet to the number of reporting SUs. Although moresensing reports from SUs an improve the CSS deision, they also inrease signalingoverhead, energy onsumption, delay before nal deision, and omputational osts.The results show that the number of transmitted reports an potentially be redued.In [57℄, the number of ooperating SUs is optimized given λ, the signal-to-noise ratio(SNR) of the PU signal sensed at the SU, and the deision threshold, Dth, at the FCsubjet to satisfying Qf + Qmd < ǫ, where ǫ is dened as the total error rate limitand Qmd = 1 − Qd denotes the global missed detetion probability. Unfortunately,the onstraint Qf + Qmd < ǫ used in [57℄ does not guarantee that Qf < ǫf andQmd < ǫmd, where ǫf + ǫmd = ǫ, ǫf is the target FA probability, and ǫmd is the targetMD probability. A minimum number of required sensing reports whih an satisfythe Qf < ǫf and Qd > (1 − ǫmd) is numerially evaluated and the energy saving isompared to traditional CSS in [58℄.In [59℄, a method is proposed in whih the sensing frequeny an be adjustedwith respet to the observed ative and idle probability distribution of the PU. Asthe observed busy probability dereases, the sensing frequeny is lowered in order8Chapter 1. Introdutionto redue the energy onsumption of the SUs during sensing. In [60℄, the sensingand transmission periods are optimized, subjet to energy resoure onstraints, tar-get inumbent detetion probability, and the total time (i.e. idling, sensing, andtransmission durations) onstraints.In [61℄, the optimal number of sensing entities whih maximizes a objetive fun-tion is obtained, where the objetive funtion is formulated based on Qmd and thetotal energy onsumption during the CSS. In [62℄, the optimal number of sensingentities whih maximizes the Qd is obtained subjet to a given total available powerfor two hannel models, i.e. additive white Gaussian noise (AWGN) and Rayleighhannel. Eah SU ats as amplify-and-forward relay and forwards what it reeivesfrom PU to the FC. The results show that the maximum detetion probability anbe ahieved by a number less than (or equal to) the total number of SUs.Cluster-based ollaborative spetrum sensing protools are proposed and analyzedin [6365℄, where eah SU transmits its sensing report to the luster-head (CH) (whihis loated loser to the FC) and eah CH forwards the sensing reports, as well as itsown sensing deision, to the FC. The luster-based protools require less energy thanthe traditional CSS protools (where all SUs transmit their report to the FC), beausethe total transmission energy is redued aording to the inverse square-law of powertransmission [77℄.In [66℄, the average energy onsumed at a SU during the spetrum sensing, hannelswithing, and data transmission, are ombined in a utility funtion. The utilityfuntion is then optimized subjet to satisfying multiple onstraints inluding a PUdetetion target probability, an opportunity detetion target probability, and a targetthroughput. The work in [66℄ onsiders pratial onstraints suh as SU delay inswithing between sensing and transmitting whih are not onsidered in [67, 68℄.9Chapter 1. IntrodutionCensoring sensing reports in the CSS is studied in [6972℄, where SUs transmittheir report to the FC only when they are ondent about their report. The ensoringshemes aim to redue the number of inaurate sensing reports during the ollabo-ration proess. In [69℄, sleeping and ensoring behavior at eah SU are jointly studiedas a mehanism to improve the energy onsumption in the CSS while maintaining aglobal MD target probability (i.e. Qf < ǫf) in the CRN. In [70,71℄, Qmd is minimizedsubjet to satisfying an energy onsumption target and a false alarm onstraint (i.e.Qf < ǫf ).Two ollaborative spetrum sensing algorithms based on sequential detetion (SD)and ordered transmission (OT) are proposed in [73℄, where the number of requiredsensing reports satisfying Qf < ǫf and Qmd < ǫmd are dereased and thereby theenergy onsumption in the CSS is redued. In [74℄, a spetrum eieny is formulatedand a sensing framework is proposed to maximize the data transmission time subjetto satisfy the interferene avoidane requirements (i.e. MD target probability). Thesensing framework in onjuntion with spetrum seletion and sheduling poliies inthe presene of multiple liensed spetrum bands and multiple SUs in a CSS are alsoinvestigated in [74℄.1.2 MotivationCSS tehniques an be used in onjuntion with TRMSs to improve the global FA andMD probabilities and the utilization of spetrum opportunities in CRNs. However,the transmission of sensing reports from SUs an represent a signiant energy andspetrum overhead ost, i.e. eah sensing report requires spetrum and energy fortransmission, proessing, and reeption. As the number of SUs inreases, so does theenergy and bandwidth requirements. Dierent aspets of the fundamental trade-o10Chapter 1. Introdutionbetween sensing auray and the overhead osts (i.e. bandwidth and energy) in CSSare studied in [7,2225,2832,5876℄. Dierent aspets of energy and bandwidth on-sumption are studied jointly with target global FA or MD probabilities onstraintsin [26, 5574℄, based on the assumption that there are no misbehaving SUs in theCRNs. More speially, it is assumed that all SUs behave onsistently while ollab-orating on the state of the band. It is expeted that MSUs an inuene Qf and Qmdsigniantly by ompromising the deisions of the FC. In addition, more reports fromSUs may be required to mitigate the eets of misbehaviors in the CRN. With MSUpresent in CRNs [7,2225,2832,75,76℄, traditional CSS (T-CSS) protools are used,where the FC requires at least one sensing report from eah SU that partiipates inCSS.We are interested in strategies whih use bandwidth and energy more eientlythan T-CSS while satisfying the same target FA and MD probabilities as T-CSSprotools in the presene of MSUs. Note that in this thesis, T-CSS is a term used todesribe any CSS sheme whih requires eah partiipating SU in the CSS proess totransmit one report to the FC. Therefore, T-CSS may refer to CSS shemes in whihthe number of partiipating SUs has been minimized to satisfy ertain onstraints(e.g. target FA and MD probabilities) or it may refer to CSS shemes in whih allSUs in the CRN partiipate in the CSS proess.1.3 ContributionsThe main ontributions of this thesis are summarized as follows:1. We propose an energy eient CSS protool, namely EE-CSS, based on a TMS,and derive expressions for the steady-state average trust value and the steady-state average total number of sensing reports transmitted by the SUs in the11Chapter 1. IntrodutionCRN. We formulate energy onsumption models for EE-CSS and T-CSS anduse the models to show the senarios in whih EE-CSS is more energy eientthan T-CSS. A method to evaluate Qf and Qd is proposed and losed-formexpressions for Qf and Qd in the ase with no MSUs are derived. We also ana-lyze the impat of link outages between the FC and the SUs while exhangingsensing reports on the expressions derived. The numerial results show that, forgiven target global FA and MD probabilities, EE-CSS an redue the numberof reports transmitted by HSUs and thus, the energy onsumption (omparedwith a T-CSS tehnique).2. We study the transient behavior of the trust values of SUs as well as the numberof sending reports transmitted by SUs. We derive losed-form expressions forthe transient probability distributions and averages of the trust values for HSUsand MSUs in EE-CSS. We also obtain expressions for the transient averagenumber of sensing reports transmitted by HSUs and MSUs to the FC in eahphase. The results, veried by simulations, show that these expressions anbe used to eiently evaluate the average number of reports transmitted for alarge number of SUs and time slots.3. We propose a trust-based entralized ollusion attak strategy in EE-CSS,where a maliious FC (MFC) obtains information from MSUs and ontrolsthe behaviors of MSUs in eah time slot. The ollusion attak strategy ap-italizes on the fat that the ontent of the ertain types of sensing reports(i.e. disagreement reports) in EE-CSS are impliitly known. Furthermore, wepropose a method in whih a ross-orrelation lter (CCF) method is used toidentify SUs with unusual behavior and to eliminate them from partiipatingin the band state deision proess being made at the FC. We ompare the Qf12Chapter 1. Introdutionand Qmd of the entralized ollusion attak strategy with that of a ommonindependent attak strategy and show that the ollusion attak strategy has amore sever eet impat on Qf and Qmd. The results also show that the CCFmethod is eetive in deteting MSUs in the CRNs.4. In addition to being energy eient, EE-CSS has the potential to be spetrumeient. However, due to the assumed TDMA struture, spetrum eienywas not onsidered. By adapting the EE-CSS in a dynami spetrum aesssheme, we propose a spetrum and energy eient CSS protool (SEE-CSS)whih aims to redue the number of urgent oexistene situation (UCS) noti-ations from the CPEs to the BS while ahieving the sameQf andQmd values asT-CSS in a wireless regional area network (WRAN). We dene throughput andenergy eieny onsumption models to ompare the data throughput and theenergy onsumed in T-CSS and SEE-CSS. We also propose a trust-based en-tralized data falsiation ollusion spetrum sensing attak strategy whih aimsto manipulate the deision of the BS severely. We propose the CCF methodto mitigate the impat of the proposed ollusion attak. The results show thatSEE-CSS is more spetrum and energy eient than T-CSS.Simulation results orroborate the derived analytial results and onrm the en-ergy and bandwidth eienies of EE-CSS and SEE-CSS relative to those of T-CSS.1.4 NotationsFor larity, all vetors are denoted by boldfae letters. Random variables (rvs) andtheir realizations are denoted by upper-ase and lower-ase letters, respetively.13Chapter 1. Introdution1.5 Organization of the ThesisThe struture of the thesis is as followsˆ Chapter 2 is organized as follows. In Setions 2.1 and 2.2, we present a briefintrodution and the system model, respetively. In Setion 2.3, we disussthe omponents of the proposed EE-CSS. In Setion 2.4, we derive expressionsfor the steady-state average trust values, T h and Tm, of honest SUs (HSUs)and MSUs in EE-CSS. Expressions for the steady-state total average number,NH and NM , of transmitted reports from HSUs and MSUs in EE-CSS are alsoderived as well as expressions for Qf and Qd. The energy onsumption in EE-CSS and T-CSS is analyzed. The impat of link outage between the FC andthe SUs on T h, Tm, NH , and NM , in EE-CSS is analyzed. Illustrative resultsand a summary are provided in Setions 2.5 and 2.6, respetively.ˆ Chapter 3 is organized as follows. We present an introdution in Setion 3.1.The system model is desribed in Setion 3.2. In Setion 3.3, we derive losedform expressions for the transient behavior of the average trust values, T h,kand Tm,k, of an HSU and an MSU and the average total number, NH,k andNM,k, of sensing reports transmitted by HSUs and MSUs in EE-CSS. Illustrativeresults are presented in Setion 3.4 and the main ndings are summarized inSetion 3.5.ˆ Chapter 4 is organized as follows. In Setion 4.1, a brief introdution is pre-sented. In Setion 4.2, we present the system model. In Setion 4.3, we proposethe ollusion attak strategy based on integer linear programming optimization(ILPO) at the MFC. A mitigating strategy based on the ross-orrelation ofthe SU report at the FC is proposed in Setion 4.4. Illustrative results and14Chapter 1. Introdutiononlusions are provided in Setions 4.5 and 4.6, respetively.ˆ Chapter 5 is organized as follows. In Setion 5.1, we present a brief introdution.In Setion 5.2, we present the system model inluding the signal propagationmodel, hannel model, and the BS deision method. In Setion 5.3, we pro-pose the SEE-CSS protool and we derive expressions for the average numberof CPEs whih partiipate in T-CSS and SEE-CSS. We dene spetrum andenergy onsumption models for T-CSS and SEE-CSS and we derive two expres-sions to evaluate the bandwidth and energy eieny ratios between the T-CSSand SEE-CSS protools. In Setion 5.4, we disuss the attak strategies andthe mitigating CCF method. Illustrative results and onlusions are providedin Setions 5.5 and 5.6, respetively.ˆ In Chapter 6, we provide a summary of the main ontributions and ndings ofthis thesis and some suggestions for further researh.15Chapter 2Trust-based Centralized EnergyEient CSS (EE-CSS)2.1 IntrodutionIn reent years, CR has attrated onsiderable researh interest due to high demandfor spetrum and low utilization rate of the liensed spetrum bands. CR denesthe standard, hardware, software, and the network speiations with whih its users(SUs) an aess the liensed spetrum bands without degrading or disrupting theliensed users (PUs) in the primary network. Spetrum sensing methods and ol-laborative spetrum sensing in CRNs are proposed to detet the unused liensedspetrum bands (or the PU signal aurately) and improve the auray of the de-tetions, respetively [7℄. In a entralized CRN, the partiipating SUs transmit theirsensing reports to the FC for a nal deision on the state of the spetrum band usage.Although CSS methods improve the auray of detetions, they require energy andspetrum overhead to transmit their reports. CSS methods are studied jointly withenergy onsumption in [26,54,5674℄. Several methods suh as ensoring, lustering,ltering as well as optimization problems with energy onstraints are proposed toredue the number of transmitted reports or the energy onsumption in CSS. How-ever, the majority of the eort is foused on the networks in the absene of MSUs.The performane of eah CSS method is extremely vulnerable to the behavior of the16Chapter 2. Trust-based Centralized EE-CSSsensing entities. We aim to improve the energy onsumption in sensing, reporting,and data olletion in CRNs whih inludes MSUs. In this work, we propose an en-ergy eient CSS protool, namely EE-CSS, based on a TMS, and derive expressionsfor the steady-state average trust value and the steady-state average total numberof sensing reports transmitted by the SUs in the CRN. We formulate energy on-sumption models for EE-CSS and T-CSS and use the models to show the senariosin whih EE-CSS is more energy eient than T-CSS. A method to evaluate Qf andQd is proposed and losed-form expressions for Qf and Qd in the ase with no MSUsare derived. We also analyze the impat of link outages between the FC and the SUswhile exhanging sensing reports on the expressions derived. The numerial resultsshow that, for given target global FA and MD probabilities, detetion EE-CSS anredue the number of reports transmitted by honest SUs (HSUs) and thus, the energyonsumption (ompared with a T-CSS tehnique).The remainder of the hapter is organized as follows. In Setion 2.2, the sys-tem model is presented. In Setion 2.3, the omponents of the proposed EE-CSS isdisussed. In Setion 2.4, expressions for the steady-state average trust values, T hand Tm, of HSUs and MSUs in EE-CSS are derived. We derive expressions for thesteady-state total average number, NH and NM , of transmitted reports from HSUsand MSUs in EE-CSS as well as expressions for Qf and Qd. The energy onsumptionin EE-CSS and T-CSS is analyzed. The impat of link outage between the FC andthe SUs outage on T h, Tm, NH , and NM , in EE-CSS is analyzed. Illustrative resultsand onlusions are provided in Setions 2.5 and 2.6, respetively.17Chapter 2. Trust-based Centralized EE-CSS2.2 System ModelThe system model is shown in Fig. 2.1. We assume that there are H HSUs andM MSUs for a total of N = H +M + 1 sensing entities (inluding the FC) in theCRN. Without loss of generality, we assume that the HSU are numbered from 1 toH and the MSUs are numbered from H + 1 to H + M , i.e. h ∈ {1, 2, . . . , H} andm ∈ {H + 1, H + 2, . . . , H + M}. There is one SUBS and one primary user basestation (PUBS). The PUBS ommuniates with the PUs using the liensed bandand the SUBS ommuniates with SUs via the unliensed and liensed bands. SUsan only ommuniate with the SUBS. We assume that the PUBS and SUBS anommuniate so that the PUBS an send the band-state information (BSI) to theSUBS periodially; this matrix inludes the state of the liensed band for the pastR time slots. The BSI allows the FC to alulate the auray of its own reports inaddition to the auray of the reports from SUs. The FC uses the auray valuealulated from the past reports of an SU to weight future reports from the SU. Inall hapters, we assume that the FC, PUBS, and SUBS annot be ompromised. TheSUs an misbehave independently (Chapters 2,3, and 5) or alloate a entralized nodeto aid with oordinated attaks (Chapters 4 and 5).Note that the information about the state of the spetrum band is not availableinstantly to the FC and annot be used to alloate spetrum holes for SUs in futuretime slots. In addition, this information is transmitted to the FC periodially to easethe burden on the PUBS. We show the impat of BSI delays on the average trustvalues of the SUs, the average number of sensing reports, and the global FA andMD probabilities in Appendix B. The results in Appendix B show that the averagetrust values of the SUs (as well as the average number of sensing reports and theglobal FA and MD probabilities) with short and long delays are similar at instanes18Chapter 2. Trust-based Centralized EE-CSSwhen the BSI is reeived at the SUBS from the PUBS. In addition, the delay inthe BSI transmission does not impat the steady state values of the aforementionedparameters. The existene of suh a link between SUBS and PUBS is motivated bythe fat that the PUBS should be able to ommuniate with the SUBS if SUs interferewith the operation of PUs. The impat of not having the BSI at the FC is exploredin Appendix C. The results from Appendix C show that if the BSI is not available,the loal deision of the FC an be used to evaluate trust values for the SUs. It isshown that the trust values of SUs, Qf , and Qmd (based on a trust-weighted variable)in T-CSS and EE-CSS degrade as the loal deision of the FC worsens. In addition,the results show that while Qf and Qmd are higher when the loal deision of theFC is used, they are equal for both T-CSS and EE-CSS. However, EE-CSS remainsmore energy eient than T-CSS and the steady-state average number of reports inEE-CSS (T-CSS) using the loal deision of the FC is exatly the same as that inEE-CSS (T-CSS) using the BSI. The liensed band is divided into V sub-bands, Bv,v = 1, 2, ..., V , and eah SU an only sense one band at any given time. The FC mayalloate one or more SUs to sense a band in eah time slot in EE-CSS. Hereafter, weassume that the FC requests all SUs to sense one band. EE-CSS an be generalizedto the ase in whih the FC alloates subgroups of SUs to sense dierent sub-bands.2.2.1 Energy Detetion MethodThe signal, yn(t), n = 1, ..., N , reeived by SUn at time t, under the idle and busyhannel hypotheses, denoted by H0 and H1, respetively, are:H0 : yn(t) = wn(t),H1 : yn(t) = hns(t) + wn(t), (2.1)19Chapter 2. Trust-based Centralized EE-CSS SUBS (FC) PUBS  MSU HSU MSU MSU HSU PU2 PU1 PU3 PU3  ( ) Figure 2.1: A entralized overlay CRN with multiple HSUs and MSUs shares thespetrum with PUswhere s(t), hn, and wn(t) denote the transmitted signal from PUBS, the hannelgain from the PUBS to SUn, and additive white Gaussian noise (AWGN) with meanzero and variane σ2n, i.e. wn(t) ∼ N (0, σ2n), respetively. The hannel gains areassumed to be drawn from independent Gaussian random variables with means zeroand variane σ2hn , i.e. hn ∼ N (0, σ2hn) orresponding to a Rayleigh fading hannel.It is assumed that the CSS is performed over one time slot and the hannel gain isonstant during the time slot. As derived in [78℄, the sensed energy (also known as,test statistis) Un,k of the hannel at SUn during the kth time slot, is given byUn,k =2TW−1∑i=0yn,k[i]2, (2.2)where yn,k[i] = yn(t = kT + i2W ), 2W is the sampling frequeny, and T is the obser-vation period. We show the omponents of the energy detetor in Fig. 2.2.Under hypotheses H0 and H1, the distribution of Un,kσ2n follows a entral hi-square(χ2) distribution with 2WT degrees of freedom and a non-entral hi-square distri-20Chapter 2. Trust-based Centralized EE-CSSA/D (.)2 ∑y(t)Noise Pre-Filter Analog-to-digital Converter Squaring Device IntegratorTest  StatisticsFigure 2.2: Components of the energy detetorbution with 2WT degrees of freedom, respetively [78℄, i.e.Un,kσ2n∼χ22WT ,H0χ22WT (ηn) ,H1,(2.3)where ηn is the SNR at SUn, i.e. ηn = |hn|2Esσ2nand Es is the energy of s(t). The loaldeision at SUn for time slot k isDn,k =0 if Un,k < λn1 if Un,k ≥ λn,(2.4)where λn is the energy deision threshold for SUn and Dn,k = 0, 1 orrespond tohypotheses H0 and H1, respetively.2.2.2 Loal Missed Detetion and False Alarm ProbabilitiesThe FA and MD probabilities of SUn for AWGN hannel with no fading, denoted bypf,n and pmd,n,AWGN , are given by [78, 79℄pf,n = Pr{Un,k > λn|H0} =Γ(TW, λn2 )Γ(TW ) , (2.5)andpmd,n,AWGN = 1−QTW (√2 ηnTW ,√λn), (2.6)21Chapter 2. Trust-based Centralized EE-CSSwhere Γ(.) and Γ(., .) denote the omplete and inomplete gamma funtion, respe-tively. Qu(a, b) is the generalized Marum Q-funtion. The expression for pf,n in(2.5) is independent of the hannel fading model between the PUBS and SUs. Thedetetion probability of SUn for AWGN hannel with fading, denoted by pd,n is givenby [79℄pd,n = e−λn2TW−2∑n=01n!(λn2)n +(1 + ηnTWηnTW)TW−1×[e−λn2(1+ ηnTW) − eλn2TW−2∑n=01n!( λn ηnTW2(1 + ηnTW ))2]. (2.7)The loal deisions Dn,k in (2.4) are ommuniated to the FC over the unliensedband.2.3 The Proposed EE-CSS ProtoolThe two main omponents of the proposed EE-CSS are the media aess ontrol(MAC) protool and the trust model. We use a ontention-free MAC protool and atrust-weighted data fusion sheme at the FC.2.3.1 MAC ProtoolEE-CSS attempts to redue the number of transmitted reports from HSUs, basedon the observation that HSUs agree on the spetrum usage more often than theydisagree. EE-CSS uses mini time slots in two phases as shown in Fig. 2.3:1. Phase I: Based on the SUs trust values, the FC hooses a set of SUs whihare hosen to sense the band and transmit their report to the FC in the minitime slots. The FC will broadast a message ontaining the list of hosen SUs.22Chapter 2. Trust-based Centralized EE-CSSThe FC fuses the reports from the hosen SUs with its own loal deision andbroadasts the intermediate deision to all SUs.2. Phase II: If an SU disagrees with the intermediate deision or it does not re-eive the broadast message reliably, it an so indiate via a transmission in itsdesignated mini time slot; otherwise, the SU remains quiet.FC ChoosesSU n*FC Broadcasts IDSU n* Transmits Dn*SU1 ... SU N-2 SU N-1Mini Time SlotsPhase I Phase IITime Slot………...Figure 2.3: The EE-CSS MAC protool: Phases I and IIIt is assumed that the SUs are synhronized to listen to the FC's broadasts in PhaseI. When an SU wishes to go to sleep mode, it noties the FC so that its silene in PhaseII will not automatially be treated as agreement with the FC's intermediate deision.We assume that the FC broadasts speial request messages (in the broadast minitime slots) oasionally asking eah SU to expliitly transmit sensing reports in theiralloated mini time slots. The FC ignores the impliit reports from SUs whih fail toexpliitly report following a request to do so. This prevents SUs from being penalizedor rewarded when they are not within the range of the CRN. Hereafter, we assumethat the FC hooses only one SU in Phase I (denoted by SUn∗) as depited in Fig.2.3. The more general ase will be investigated later.23Chapter 2. Trust-based Centralized EE-CSS2.3.2 Trust ModelThe information ontained in the BSI an be used to evaluate a trust value for eahSU, based on its reporting auray in previous time slots. In [25, 51, 53℄ lassialbinomial beta framework is adopted to estimate reputation and trust values for re-porting nodes in a distributed system. An expression for the expeted trust value ofthe nodes is derived in [53℄. This expression estimates the probability value for thenumber of good behavior minus the number of bad behavior for a seondary user asTn,k =1 +∑ki=1 rn,i1 + k , (2.8)where rn,i denotes the reward assoiated with the deision of SUn at time i; its valueis 1 if a orret deision is made and 0 if the deision is inorret. The `1's in thenumerator and the denominator reet the fat that the trust values are initializedto 1 at k = 0. At i = 1, if a orret deision is made, rn,1 = 1 and Tn,1 = 1+11+1 = 1; ifan inorret deision is made, rn,1 = 0, and Tn,1 = 1+01+1 = 12 .2.3.3 Intermediate DeisionIn Phase I, the most trustworthy SU at time slot k, denoted by n∗k (where n∗k =argmaxnTn,k), is requested by the FC to report on the state of a partiular band.Then, the sensing report reeived from SUn∗ is fused with the FC's own loal deisionin an intermediate fusion proess using the OR rule to form an intermediate deision:IDFC,k =DFC,k, if the report from SUn∗ is not availableDFC,k OR Dn∗,k, otherwise,(2.9)24Chapter 2. Trust-based Centralized EE-CSSwhereDn∗,k ∈ {0, 1} andDFC,k ∈ {0, 1} denote the loal deision by SUn∗ transmittedto the FC and the loal deision of FC for time slot k. The impat of other fusiontehniques during the intermediate fusion proess is urrently being studied. The ORrule an ause high FA probability as well as high detetion probability; therefore,a deision based on the OR rule an protet PUs from interferene situations morethan a deision on other rules. The SUBS broadasts IDFC,k in a mini time slot andwaits for responses from SUs in Phase II. The FA and MD probabilities, pf,ID andPmd,int., of the intermediate deision are:pf,ID =pf,FC , if the report from SUn∗ is not available1− (1− pf,FC)(1− pf,n∗), if SUn∗ is an HSU1− (1− Pp,FC)(1− pf,n∗,FC), if SUn∗ is an MSU(2.10)andpmd,int. =pmd,FC , if the report from SUn∗ is not availablepmd,FCpmd,n∗ , if SUn∗ is an HSUpmd,FCpmd,n∗,FC, if SUn∗ is an MSU(2.11)respetively, where the FA and MD probabilities for the FC are denoted by pf,FCand pmd,FC and the loal FA and MD probabilities of hth HSU are denoted by pf,hand pmd,h. The FA and MD probabilities of mth MSU seen at the FC are denoted bypf,m,FC and pmd,m,FC . Evaluation of pf,m,FC and pmd,m,FC is desribed in onjuntionwith the attak poliy of MSUs in Setion 2.3.5.25Chapter 2. Trust-based Centralized EE-CSS2.3.4 Final DeisionAt the end of Phase II, the FC has N sensing reports for the spetrum band inludingthe loal FC deision, the sensing report from SUn∗ , and N − 2 sensing reports inthe forms of agreements (i.e. silent mini time slots), disagreements (i.e. expliitreports from the disagreeing SUs), and reports from the SUs whih did not reeivethe broadast intermediate deision message reliably. The FC uses the followingtrust-weighted sum funtion to determine the nal deision, Dk, at time slot k,FDFC,k =0, if∑N−1n=1 Tn,kXn,k + TFC,kXFC,k < Dth1, otherwise,(2.12)where TFC,k is the trust value of the FC in time slot k, XFC,k, Xn,k = −1 if DFC,k,Dn,k= 0 and XFC,k, Xn,k = 1 if DFC,k, Dn,k = 1. Using bipolar values for Xn,k(XFC,k), we an implement the trust-weighted majority rule at the FC by settingDth = 0 in (2.12). Note that Dth an be seleted to ahieve the desired overallsystem false alarm [80℄. If the BSI is available at the end of Phase II, the trust valuesfor the SUs are updated using (2.8). A owhart of the deision proess for the FCin EE-CSS is shown in Fig. 2.4. We next desribe the apabilities and the attakpoliy of the MSUs.2.3.5 Maliious Seondary UsersAn MSU has all the apabilities of an HSU, inluding sensing and deteting thestate of a band and reporting its deision about the state to the FC. In addition,an MSU an manipulate its report to benet unfairly from or to disrupt the CRN.In Chapter, we make the simplifying assumption that MSUs at independently (i.e.they do not ooperate or ollude). We will assess the performane of EE-CSS against26Chapter 2. Trust-based Centralized EE-CSSDFC,kXn*,kOR FC Broadcasts IDFC,kAgreements/Disagreements/Explicit MessagesXn,k , nϵ{1,2,…,N-1}\n*(2.12)FDFC,kPhase I Phase II,  Figure 2.4: Two-phase Deision making proess at the FC in EE-CSSan independent attak. Eah MSU an manipulate its report with respet to itsown sensing deision, denoted by Dm,k. Table 2.1 illustrates the MSU attak poliy.By adapting the attak probabilities (pα0,0 ,pα1,0 , pα0,1 , and pα1,1) in Table 2.1, MSUsMSU Reports MSU ReportsIDFC,k Dm,k Dishonestly with Honestly withProbability: Probability:0 0 pα0,0 1− pα0,01 1 pα1,1 1− pα1,10 1 pα0,1 1− pα0,11 0 pα1,0 1− pα1,0Table 2.1: Independent Attak Poliy Ihoose to emphasize either gaining unfair advantage to aess the available spetrumor ausing interferene to the PUs. For example, if the objetive of the attak strategyis to gain unfair aess advantage, eah MSU should attempt to manipulate thedeision of the FC so that it deides that the hannel is busy even though the hannelis atually idle. In this dissertation we only analyze the behavior of the network based27Chapter 2. Trust-based Centralized EE-CSSon this ommonly used independent attak. Therefore, MSUs an use large valuesfor pα0,0 and pα1,0 . Note that if intermediate message is unknown at the mth MSU,it reports dishonestly with pα0 and pα1 for Dm,k = 0 and 1, respetively. The attakpoliy summarized in Table 2.1 is similar to the attak poliies in [8185℄ wherepα0,0 = pα1,0 and pα0,1 = pα1,1 . The FA and MD probabilities of MSUs seen at theFC are pf,m,FC = (1 − pf,m)(pα0,0(1 − pf,ID) + pα1,0pf,ID) + pf,m[(1 − pα1,1)pf,ID +(1 − pα0,1)(1 − pf,ID)]and pmd,m,FC = pmd,m[(1 − pα0,0)pmd,ID + (1 − pα1,0)pd,ID]+(1−pmd,m)(pα1,1pd,ID + pα0,1pmd,ID), where the loal FA and MD probabilities of mthHSU are denoted by pf,m and pmd,m. If intermediate message is unknown at themth MSU, the FA and MD probabilities of MSUs seen at the FC are pBr.f,m,FC =(1− pf,m)(pα0) + pf,m(1− pα1) and pBr.md,m,FC = pmd,m(1− pα0) + (1− pmd,m)pα1 .Note that the SSDF attak poliy in Table 2.1 an impat the deision of the FC.The attak poliy does not guarantee that the SUs will interfere with PUs or thatselsh MSUs will gain unfair aess to the available spetrum, even if the deision ofthe FC is ompromised. When the FC authorizes an SU to transmit on a band thatis oupied by an undeteted PU, the objetive of maliious MSUs may be ahieved.Similarly, eah MSU an obtain its selsh objetive only when it atually uses theunused spetrum that the FC has falsely delared oupied.How MSUs an utilize the spetrum opportunities whih have beome availableby suessfully ausing global false alarms is beyond the sope of this study. Notethat if MSUs onvine the FC that the hannel is busy, given that the hannel is idle,the FC (SUBS) will not alloate that hannel to any entity in a entralized CRN.Therefore, one may suggest MSUs an form a distributed ad ho CRN suh as theones studied in [39, 8688℄ to use the available spetrum.28Chapter 2. Trust-based Centralized EE-CSS2.3.6 Expliit Reports in EE-CSSThe FC an reeive either an expliit sensing report or agreement/disagreement reportfrom eah SU during its mini time slot. The three ases in whih the FC may reeivean expliit sensing report are:ˆ Case 1: SUn∗ sends a sensing report to the FC in Phase I.ˆ Case 2: SU sends a sensing report to the FC in Phase II in response to a speialrequest message from the FC.ˆ Case 3: SU sends a sensing report to the FC in Phase II beause the SU hasnot reeived the broadast sensing report from the FC due to a link outage.Case 1 is desribed in Setion 2.3.1. Motivation for Case 2 is due to potential hit-and-run attaks. An MSU an attak the network for a short duration and an refrain fromattaking by leaving the network for a long duration. MSUs an benet by leaving andnot notifying the FC. Consider the senario in whih orret intermediate deisionsare being made; MSUs whih have left the network without notifying the FC appearto agree with the intermediate deisions about the spetrum usage. Therefore, ifthe intermediate deisions are more often orret than inorret, the trust values ofsuh MSUs will improve while they are absent from the CRN. In order to mitigatethis eet, the FC broadasts speial request messages (in the broadast mini timeslots) oasionally, asking eah SU to expliitly send transmit sensing reports in theiralloated mini time slots. The FC ignores the impliit reports (idle mini time slots)from SUs whih fail to expliitly report following a request to do so. Case 3 ourswhen there is an outage in the link from the FC to an SU. The impat of link outagesis disussed next.29Chapter 2. Trust-based Centralized EE-CSS2.3.7 Link Outages in EE-CSSIn this setion we will investigate the eet of link outages on the performane of EE-CSS. An outage ours when the reeived signal (e.g. for the sensing report or thebroadast intermediate deision) at the FC or SUs does not satisfy a minimum SNRrequirement. The link outage an our due to low hannel gain or interferene fromother users in the network. For our purpose, an outage means that the broadast (orthe sensing) report is not reeived suessfully by an SU (or the FC). The impat oflink outages on Qf and Qd is an important onsideration. Knowledge of hannel stateinformation an help the FC and SUs use appropriate transmission power levels forreliable ommuniation between the FC and SUs. Seure trusted relays an reduethe impat of outages by relaying reports from SUs whih are suering from very poorhannel gains [89,90℄. The CRN in wireless regional area networks (WRANs) utilizesliensed frequeny bands and bak up ontrol hannels to ensure higher reliability inthe link between the SUs and the SUBS. We onsider the following outage senariosfor the analysis of T h, Tm, NH and NM in future setions:ˆ Senario 1: An SU reeives the broadast intermediate deision suessfullyand the FC reeives a disagreement report from the SU suessfully, i.e. thereis no link outage from the FC to the SU and from the SU to the FC.ˆ Senario 2: An SU reeives the broadast intermediate deision suessfullybut the FC does not reeive a report from the SU; therefore the FC interpretsthat the SU is agreeing with the broadast intermediate deision, i.e. there isno link outage from the FC to the SU but there is a link outage from the SUto the FC.ˆ Senario 3: An SU does not reeive the broadast intermediate deision from30Chapter 2. Trust-based Centralized EE-CSSthe FC but the FC reeives a report (informing the FC that the SU did notreeive the broadast intermediate deision) from the SU suessfully, i.e. thereis link outage from the FC to the SU but there is no link outage from the SUto the FC.ˆ Senario 4: An SU does not reeive the broadast intermediate deision andthe FC does not reeive the report from the SU, i.e there is a link outage fromthe FC to the SU and from the SU to the FC.2.4 Analysis of EE-CSSIn this setion, expressions for T h, Tm, NH , and NM in EE-CSS are derived. Theimpat of outages on T h, Tm, NH , and NM in EE-CSS is analyzed. A method toompute Qf and Qd is proposed. We also provide expressions for Qf and Qd forEE-CSS and traditional CSS tehniques, onsidering outages but no MSUs in theCRN.2.4.1 Derivation of T h, Tm, NH, and NMExpressions for T h, Tm, NH , and NM in EE-CSS are derived under steady-stateonsiderations, i.e. these quantities reah a state at whih their value is not varyingwith time slot k. Using (2.8), HSUh an be obtained asT h = 1 − (1− pBr.out,h)(1− pout,h)(pH0pf,h + pH1pmd,h)− (1− pBr.out,h)pout,h(pH0pf,ID + pH1pmd,ID)− pBr.out,h(1− pout,h)(pH0pf,h + pH1pmd,h)− pBr.out,hpout,h(pH0pf,ID + pH1pmd,ID), (2.13)31Chapter 2. Trust-based Centralized EE-CSSwhere pBr.out,h is the outage probability of the link from the FC to the hth HSU andpout,h is the outage probability of the link from the hth HSU to the FC. The term(1−pBr.out,h)(1−pout,h)(pH0pf,h +pH1pmd,h) represents FA and MD probabilities seen atthe FC for Senario 1. Similarly, the terms on the RHS of (2.13) in the following linesrepresent FA and MD probabilities seen at the FC for Senarios 2 to 4. Equation(2.13) an be further simplied to:T h = 1 − pH0(pf,IDpout,h + pf,h(1− pout,h))− pH1(pmd,IDpout,h + pmd,h(1− pout,h)). (2.14)Note that (2.14) is ompletely independent of pBr.out,h. In (2.14), the term pH0pf,IDpout,hrepresents the probability that the intermediate deision is H1 and HSUh experienesan outage, given that the PU hannel is idle. This means that when the intermediatedeision is H1 and HSUh experienes an outage, the FC will onlude that HSUhagrees with the intermediate deision and T h is aeted regardless of the deision ofHSUh. The term pH0pf,h(1−pout,h) represents the probability that HSUh deides H1,and it does not experiene an outage given that the PU hannel is idle. Similarly, thetwo terms in the seond line of (2.14) follow the ase that the PU hannel is busy. Asexpeted, for pout,h = 0 for h = 1, . . . , H , (2.14) redues to T h = 1−pH0pf,h−pH1pmd,h.When MSU does not reeive the broadast, its attak poliy hanges aording thedesription in Setion 2.3.5. We an express and simplify Tm for all the senarios to:Tm = 1 − (1− pBr.out,m)[pH0 (pf,IDpout,m + pf,m,FC(1− pout,m))− pH1 (pmd,IDpout,m + pmd,m,FC(1− pout,m))]− pBr.out,m[pH0(pf,IDpout,m + pBr.f,m,FC(1− pout,m))− pH1(pmd,IDpout,m + pBr.md,m,FC(1− pout,m)) ], (2.15)32Chapter 2. Trust-based Centralized EE-CSSwhere pout,m is the outage probability of the link from the FC to the mth MSU andpBr.out,m is the outage probability of the link from the mth MSU to the FC. The rstterm, (1 − pBr.out,m)pf,IDpout,m, on the RHS of (2.15), represents the probability thatthe FC thinks that MSUm deided H1 due to link outage in Senario 2, given thatthe PU hannel is idle. The seond term, (1 − pBr.out,m)pf,m,FC(1 − pout,m), representsthe probability that the FC thinks that MSUm deided H1 for the link outage inSenario 1, given that the PU hannel is idle. Similarly, the event probabilities fortwo link outage in Senarios 1 and 2, given the PU hannel is busy, are presented bythe third and fourth terms. The fth term, pBr.out,mpBr.f,m,FC(1 − pout,m), represents theprobability that the FC thinks that MSUm deided H1 (i.e. the mth MSU does notreeive the broadast and its expliit report is not reeived at the FC due to outage)for the link outage in Senario 4, given the PU hannel is idle. The sixth term,pBr.out,mpBr.f,m,FC(1 − pout,m), represents the probability that the FC thinks that MSUmdeided H1 for the link outage in Senario 3, given that the PU hannel is idle.Similarly, the event probabilities for two link outage in Senarios 3 and 4, given thePU hannel is busy, are presented by the seventh and eighth terms. As expeted, forpout,m = 0 for m = 1, 2, . . . ,M , (2.15) redues to Tm = 1−pH0pf,m,FC −pH1pmd,m,FC .Assuming T h > Tm (for h = 1, ..., H andm = H+1, ..., H+M), as k inreases, thevarianes of the HSU and MSU trust random variables derease and the probabilityof hoosing an HSU in Phase I inreases beause the FC in EE-CSS hooses the mosttrusted SUs and HSUs tend to have higher trust values than MSUs. In steady-state,the probability, PHI , of hoosing an HSU in Phase I approahes 1. This observation isvalidated by the simulation results. Consequently, the average numbers, NHI = PHIand NMI = 1 − PHI , of transmitted sensing reports in Phase I for HSUs and MSUsapproah 1 and 0, respetively. The steady-state average total number, NH and NM ,33Chapter 2. Trust-based Centralized EE-CSSof transmitted sensing reports by HSUs and MSUs respetively are NH = NHI +NHIIand NM = NMI+NMII , respetively, where NHII and NMII are dened as the steady-state average number of transmitted sensing reports by HSUs and MSUs in Phase II:NHII =H−1∑h[pBr.out,h + (1− pBr.out,h)(NH1hII +NH0hII )](2.16)andNMII =M∑m[pBr.out,m + (1− pBr.out,m)(NH1mII +NH0mII )], (2.17)respetively, where the steady-state average number of reports transmitted by the hthHSUs in Phase II for hypotheses H0 and H1, are NH0hII and NH1hII , respetively. Andthe steady-state average number of reports transmitted by mth MSU in Phase II forH0 and H1 based on the attak poliy of Table 2.1, are NH0mII and NH1mII , respetively.The terms NH0hII , NH1hII , NH0mII and NH1mII are evaluated in Appendix D.2.4.2 Energy Consumption in T-CSS and EE-CSSThe energy onsumption for ollaborative spetrum sensing at a node depends onthe energy onsumed in transmitting, reeiving, and proessing pakets. The energyonsumed for transmission is a dominant fator in the total energy onsumptionin traditional wireless links where the transmission distane is large (greater than100m) [91, 92℄. For example, in WRAN inspired CRNs where the distane from SUsto the SUBS is typially large (e.g. Kilometers), researh eorts should be foused onenergy eient transmissions. However, when a CRN is implemented for short rangewireless appliations suh as wireless sensor networks (WSNs) where the transmissiondistane is usually small (e.g. tens of meters), the energy onsumed for reeiving andproessing pakets an be a signiant fration the total energy onsumption. Forexample, both transmission energy and iruit energy (for proessing information)34Chapter 2. Trust-based Centralized EE-CSSare onsidered in alulating the total energy onsumption in [93℄. For simpliity, weassume that the energy onsumed to proess a paket is inluded in the transmissionand reeiving energy. We also assume that the pakets transmitted from the FC andSUs are of equal length in both EE-CSS and T-CSS. For T-CSS, the average totalenergy, ET , onsumed in the ollaboration proess in one time slot after the FC'sinitial broadast is:ET =H∑h=1Erx,h︸ ︷︷ ︸J1+H∑h=1Etx,h︸ ︷︷ ︸J2+HErx,FC︸ ︷︷ ︸J3, (2.18)where the average energy onsumed for transmitting and reeiving a paket at the hthHSU is denoted by Etx,h and Erx,h, respetively, and the average energy onsumedfor reeiving a paket at the FC is denoted by Erx,FC. For simpliity, we assumedthat the average energy onsumed to transmit (reeive) a paket at any SU is thesame, i.e. Etx = Etx,FC = Etx,h (Erx = Erx,FC = Erx,h). In (2.18), the three termsorrespond to,ˆ J1: H HSUs reeive the announement from the FC.ˆ J2: H HSUs transmit their sensing reports to the FC.ˆ J3: H expliit sensing reports from HSUs are reeived and proessed at theFC.35Chapter 2. Trust-based Centralized EE-CSSFor EE-CSS, the average total energy, EEE, onsumed in the ollaboration proessin one time slot after the FC's initial broadast is:EEE =H∑h=1Erx,h︸ ︷︷ ︸I1+Etx,SUn∗︸ ︷︷ ︸I2+Erx,FC︸ ︷︷ ︸I3+Etx,FC︸ ︷︷ ︸I4+H−1∑h=1Erx,h︸ ︷︷ ︸I5+NHIIEtx︸ ︷︷ ︸I6+NHIIErx,FC︸ ︷︷ ︸I7,(2.19)where the average energy onsumed for transmitting at the FC and SUn∗ are denotedby Etx,FC and Etx,SUn∗ , respetively. In (2.19), the terms orrespond toˆ I1: H HSUs reeive the announement from the FC.ˆ I2: SUn∗ transmits one sensing report to the FC.ˆ I3: The FC reeives one expliit sensing report.ˆ I4: The FC transmits a broadast report.ˆ I5: H − 1 SUs reeive the broadast report.ˆ I6: On average NHII HSUs transmit their sensing reports to the FC.ˆ I7: The FC reeives NHII expliit sensing reports from HSUs.We dene the reeive-to-transmit energy ratio as θ , ErxEtx . We an simplify the totalenergy values to ET = H(2θ + 1)Etx and EEE = [(2H +NHII )θ + (2 +NHII )]Etx.36Chapter 2. Trust-based Centralized EE-CSS2.4.3 Global False Alarm and Global Detetion ProbabilitiesIn this setion we examine Qf and Qd in EE-CSS and T-CSS. In order to omputeQf and Qd, we dene S =∑Hh=1 ThXh +∑M+Hm=H+1 TmXm + TFCXFC for the Nsensing entities, where Xh ∈ {−1, 1} (and Xm ∈ {−1, 1}) for Dh = {0, 1} (andDm = {0, 1}) are the (impliit or expliit) sensing reports transmitted to the FC fromHSUh (and MSUm). For simpliity, we rst assume that all sensing reports from SUsare suessfully reeived by the FC. Using Th, Tm, pf,n, and pd,n, we an ompute theprobability of eah of the 2N possible realizations of S and the orresponding eventprobabilities. For example, suppose M = 1, H = 1 (i.e. N = 3). Then,Pr{S = Th − Tm + TFC|H1} = (1− pmd,h)pmd,m,FC(1− pmd,FC). (2.20)Reall pmd,m,FC denotes the MD probability of MSUm seen by the FC. Therefore,Qd =2N∑i=1Pr{Si > Dth|H1} (2.21)andQf =2N∑i=1Pr{Si > Dth|H0} (2.22)an be omputed. For evaluation of S in (2.20), it is assumed that the reeivedsensing report from eah SU are reeived at the FC suessfully, i.e. there is no linkoutage. Reports gathered at the FC from eah SU are independent deisions based onloal deision of eah SU and therefore, Pr{S} an be presented as a produt of thehypotheses probabilities as shown in (2.20). However, when link outage is onsidered,some reports are not reeived at the FC reliably and therefore, the report from theSU is dependent on the intermediate deision. For this reason (2.20) annot be used37Chapter 2. Trust-based Centralized EE-CSSto evaluate S. The evaluation of S for the ase with link outages from SUs to theFC is urrently being studied.Derivation of Qf and Qd (No MSUs)In this setion, we assess the impat of outage on Qf and Qd in EE-CSS and thetraditional CSS tehnique in [94℄, when there is no MSUs in the CRN. For simpliity,we assume that in omputing the nal deision as in (2.12), the weights are set to 1,i.e.FDFC,k =H0, if∑N−1n=1 Xn,k +XFC,k < DthH1, otherwise,(2.23)where Dth is an integer. Expressions for Qf and Qd for T-CSS, based on (2.23)assuming unit weights and without onsidering outages on the links between the SUsand the FC, are presented in [94℄. To the best of our knowledge, the impat ofoutage has not been onsidered in the derivation of Qf and Qd in T-CSS. The globalmissed detetion and false alarm probability an be evaluated using Poisson binomialdistribution [95℄ asQf = pf,FCN−1∑l=Dth−1Pf,l︸ ︷︷ ︸Kf,1+(1− pf,FC)N−1∑l=DthPf,l︸ ︷︷ ︸Kf,2, (2.24)where, Pf,l =∑A∈Fl∏i∈A βf,i∏j∈Ac(1 − βf,j). Note that βf,n , pf,n(1 − pout,n), Flis the set of all subsets of l integers that an be seleted from {1, 2, . . . , L}, Ac ={1, 2, . . . , L}\A and L , N − 1. In (2.24), the rst term on the RHS represents theprobability that the FC deides H1 and it is informed that at least Dth − 1 of theN − 1 SUs deide H1, given that the PU hannel is idle. The seond term representsthe probability that the FC deides H0 and it is informed that at least Dth of the38Chapter 2. Trust-based Centralized EE-CSSN − 1 SUs deide H1, given that the PU hannel is idle. From (2.23), both of thesenarios will result in the FC deiding H1. Similarly, we haveQd = pd,FCN−1∑l=Dth−1Pd,l︸ ︷︷ ︸Kd,1+(1− pd,FC)N−1∑l=DthPd,l︸ ︷︷ ︸Kd,2, (2.25)where Pd,l =∑A∈Fl∏i∈A βd,i∏j∈Ac(1− βd,j) and βd,n , pd,n(1− pout,n).For EE-CSS, from (2.23), we haveQf = Mf,1N−2∑l=Dth−2Pf,l,1︸ ︷︷ ︸Lf,1+Mf,2N−2∑l=Dth−1Pf,l,2︸ ︷︷ ︸Lf,2+Mf,3N−2∑l=DthPf,l,3︸ ︷︷ ︸Lf,3, (2.26)where Mf,1 , pf,FCpf,n∗(1 − pout,n∗), Mf,2 , pf,FC(1 − pf,n∗) + pf,n∗pf,FCpout,n∗ +pf,n∗(1− pf,FC)(1− pout,n∗), Mf,3 , (1− pf,FC)(pf,n∗pout,n∗ + (1− pf,n∗)), L , N − 2,Pf,l,. =∑A∈Fl∏i∈A γf,i,.∏j∈Ac(1 − γf,j,.), and γf,n,1 , pf,n(1 − pout,n) + pout,n is theprobability that the FC sees the SUn deides H1, given PU hannel is idle. Andγf,n,2 = γf,n,1 and γf,n,3 , pf,n. Note that SUn∗ in Phase I and N − 2 SUs in PhaseII may experiene outages. The rst term, Mf,1Lf,1, on the RHS represents theprobability that the FC deides H1 and it is informed that SUn∗ and at least Dth− 2of N − 2 SUs have deided H1, given that the PU hannel is idle. The seond term,Mf,2Lf,2, represents the probability that either the FC's loal deision is H1 or theFC is informed that SUn∗ has deided H1 and that at least Dth− 1 of the N − 2 SUshave deided H1, given the PU hannel is idle. The third term, Mf,3Lf,3, representsthe probability that the FC deides H0 and it is informed that SUn∗ has deided H0and that at least Dth out of N−2 SUs have deided H1, given the PU hannel is idle.Note that Dint = 0 and if SUn is in outage, the FC sees that the SUn has deided39Chapter 2. Trust-based Centralized EE-CSSH0. Similarly, for Qd we have,Qd = Md,1N−2∑l=Dth−2Pd,l,1︸ ︷︷ ︸Ld,1+Md,2N−2∑l=Dth−1Pd,l,2︸ ︷︷ ︸Ld,2+Md,3N−2∑l=DthPd,l,3︸ ︷︷ ︸Ld,3, (2.27)where Md,1 , pd,FCpd,n∗(1 − pout,n∗), Md,2 , pd,FC(1 − pd,n∗) + pd,n∗(pd,FCpout,n∗ +(1 − pd,FC)(1 − pout,n∗)), Md,3 , (1 − pd,FC)(pd,n∗pout,n∗ + (1 − pd,n∗)), L , N − 2,Pd,l,. =∑A∈Fl∏i∈A γd,i,.∏j∈Ac(1 − γd,j,.), and γd,n,1 , pd,n(1 − pout,n) + pout,n is theprobability that the FC sees the SUn deides H1, given PU hannel is busy. Andγd,n,2 = γd,n,1 and γd,n,3 , pd,n. For βf,n = βf , Qf in (2.24) an be further simpliedusing Bernoulli Binomial distribution by replaing Kf,1 and Kf,2 by B(Dth − 1, N −1, βf) and B(Dth − 1, N − 1, βf), respetively,Qf = pf,FC B(Dth − 1, N − 1, βf) + (1− pf,FC) B(Dth − 1, N − 1, βf),(2.28)where B(a, b, c) = ∑bl=a(bl)(c)l(1− c)b−l. For βd,n = βd, Qd in (2.25) an be obtainedby replaing Kd,1 and Kd,2 with B(Dth − 1, N − 1, βd) and B(Dth − 1, N − 1, βf),respetively. Similarly, for γf,n,1 = γf1 , γf,n,2 = γf2, and γf,n,3 = γf3, we an simplifyQf in (2.26) by replaing Lf,1, Lf,2, Lf,3, by B(Dth − 2, N − 2, γf1), B(Dth − 1, N −2, γf2), and B(Dth, N − 2, γf3), respetively. For γd,n,1 = γd1 , γd,n,2 = γd2 , and γd,n,3 =γd3 , we an simplify Qd in (2.27) by replaing Ld,1, Ld,2, and Ld,3 by B(Dth − 2, N −2, γd1), B(Dth− 1, N − 2, γd2), and B(Dth, N − 2, γd3), respetively. For pout,n = 0, forn = 1, 2, . . . , N , (2.26) and (2.27) redue to (2.24) and (2.25), respetively.40Chapter 2. Trust-based Centralized EE-CSS10 15 20 25 30 35 4005101520253035404550SNR of PU Signal Received at the SU (dB)Number of Reports  T−CSSEE−CSSFigure 2.5: Minimum number of sensing reports (or minimum number of partiipatingSUs, i.e. eah SU transmits one sensing report to the FC) required from all SUs tosatisfy Qmd < ǫmd and Qf < ǫf in the T-CSS versus EE-CSS protools.2.5 Numerial ResultsIn this setion, we ompare the results from the analysis in Setion 2.4 with simu-lation results. First, we ompare NH in EE-CSS and T-CSS without MSUs. Thenwe ompare the energy onsumption of T-CSS and EE-CSS as a funtion of θ. Inaddition, we plot NH , NM , T h, and Tm in the presene of MSUs with and withoutoutages. We show the impat of outages on Qf and Qmd of EE-CSS with MSUs. Weplot and ompare the ROC urves for T-CSS and EE-CSS in the presene of linkoutages. In order to provide a fairer omparison between NH in T-CSS and EE-CSS,we used an exhaustive searh method to nd the minimum number of partiipatingSUs (by adjusting λ, Dth) whih an satisfy Qf < ǫf = 0.05 and Qmd < ǫmd = 0.05for both T-CSS and EE-CSS (ǫf and ǫmd denote the required system FA and MDprobabilities, respetively). If Qf < 0.05 and Qmd < 0.05 annot be met, Qf + Qmdmust be minimized. The resulting minimum number of sensing reports in T-CSS and41Chapter 2. Trust-based Centralized EE-CSSEE-CSS are then ompared. The average steady-state total number, NH , of sensingreports in T-CSS and EE-CSS are shown in Fig. 2.5. In this gure, T-CSS refersto a CSS sheme that the number partiipating SUs is minimized with respet totarget FA and MD probabilities. The at line at low SNR is the result of reahingthe maximum number of available SUs in the CRN. It an be seen that EE-CSSrequires a lower NH than T-CSS at low SNR and a similar number at high SNR. Weexpet the same average number of sensing reports in EE-CSS as that in T-CSS forhigher SNR values. More speially, when the minimum number of sensing reportsrequired to meet Qf < 0.05 and Qmd < 0.05 is 2, the sensing reports from the FCand the hosen SU are suient and Phase II of the protool is not exeuted. Whenthe minimum number of sensing reports required to meet Qf < 0.05 and Qmd < 0.05is 1, the sensing report from the FC is suient. For these two senarios, the numberof sensing reports transmitted in T-CSS and EE-CSS are idential.0 0.2 0.4 0.6 0.8 1 1.2100101102θAverage Total Energy (mJouls)  T−CSS SNR= 20dBEE−CSS SNR= 20dBT−CSS SNR= 25dBEE−CSS SNR= 25dBT−CSS SNR= 30dBEE−CSS SNR= 30dBFigure 2.6: Average total energy onsumed to transmit and to reeive pakets as afuntion of θ for the T-CSS and EE-CSS protools.The average total energy onsumed in T-CSS and EE-CSS as a funtion of θ forthree PU SNR values are shown in Fig. 2.6. For long range wireless ommuniation42Chapter 2. Trust-based Centralized EE-CSS(e.g. 100 meters or more), θ ≃ 0 and for short range wireless ommuniations θ ≃ 1[93℄. The average total energy onsumed for T-CSS and EE-CSS orrespond to thenumber of reports transmitted in Fig. 2.5. When the dierene between the numberof reports in T-CSS and EE-CSS is large in Fig. 2.5, EE-CSS onsumes less energythan T-CSS for 0 < θ < 1.2. Our results onrm that the average total energyonsumed in EE-CSS is lower than that in T-CSS for PU SNR values smaller than20 dB. We have omitted the results for PU SNR values less than 20 dB for brevity.However, when the dierene between the number of reports in T-CSS and EE-CSSis small (i.e. SNR= 30, 35 dB) in Fig. 2.5, T-CSS onsumes less energy than EE-CSSfor 0.5 < θ. This is due to the fat that the energy onsumed in reeiving paketsin EE-CSS osets its eieny in transmission energy. For SNR values whih T-CSSand EE-CSS require only 2 reports (or less), the energy onsumption are the samebeause T-CSS and EE-CSS are idential. To assess the impat of the outage inpα1,0 pα1,1 pH1 pout pBr.outpα0,0 pα0,10.7 0.2 0.5 0.2 0.2Table 2.2: Simulation Parameter Values Ithe presene of MSUs, simulation results were obtained using the parameter valuessummarized in Table 2.2 for H = 4 and M = 4. The results are averaged over100, 000 yles; in eah yle, the values for NH , NM , T h, Tm, Qf , and Qmd aresimulated for 300 time slots. We assume that the PUBS sends the BSM to the FCevery TBSM = 1 time slot period, the average reeived signal SNR at eah SU is thesame (SNRdB = 25), Dth = 0, and λ = 50, i.e. the loal FA probabilities (as well asloal MD probabilities) for all SUs are idential. Based on the parameters values inTable 2.2, T h > Tm for all h = 1, ..., H and m = H + 1, ...,M +H .The probability of hoosing an HSU or MSU in Phase I as a funtion of time slot43Chapter 2. Trust-based Centralized EE-CSS0 50 100 150 200 250 30000.10.20.30.40.50.60.70.80.91Time,  k (Time slot)PHI and PMI  Sim. HSU pout= 0Sim. MSU pout= 0Sim. HSU pout= 0.2Sim. MSU pout= 0.2Figure 2.7: Probability of hoosing an honest or maliious SU in Phase I as a funtionof time, k (in time slots), with pout = pBr.out = 0.2, H = 2, and M = 2k is shown in Fig. 2.7. Note that HSUs and MSUs are equally likely to be hosen atk = 0, e.g. for H = 4 and M = 4, the probability that an HSU or MSU is hosen atk = 0 is 12 . As k inreases, the EE-CSS protool is more likely to hoose an HSU inPhase I; it an be seen that at k = 300, PHI = NHI = 1 − NMI is very lose to 1.Outages an ause the trust values of the HSUs to derease whih in turn result in alonger time for NMI to onverge to its steady-state value, as an be seen in Fig. 2.7.For pout = pBr.out = 0.2 and the given parameter values, the impat of outage on PHI isvery small as shown in Fig. 2.7.The average number of expliit sensing reports transmitted by HSUs and MSUsin Phase II for k = 0 to k = 300 is shown in Fig. 2.8. The results in Figs. 2.7 and 2.8show that on average, HSUs only send a total of approximately NH = NHI +NHII ≃1+1.45 = 2.45 reports per time slot when there are broadast link outages and a totalof approximately 2.05 reports when there are no broadast link outages in Phase I.However, T-CSS requires 1 report from eah HSU, i.e. NH = 4 for H = 4. Outages44Chapter 2. Trust-based Centralized EE-CSS0 50 100 150 200 250 30011.11.21.31.41.51.61.71.81.9Time,  k (Time slots) NHII and  NMII  Sim HSUpout= 0Sim MSU pout= 0Sim. HSU pout= 0.2Sim. MSU pout= 0.2Steady−state values from the analysis Figure 2.8: NMII and NHII as a funtion of time, k (in time slots), with pout = pBr.out =0.2, H = 4, and M = 4. The dashed horizontal lines show the steady state values forthe orresponding urves.an impat the number of suessfully transmitted reports to the FC as shown in Fig.2.8. The number of transmitted reports from MSUs depends on the attaking poliyof MSUs. The steady-state average trust values of HSUs and MSUs, T h and Tm, fork = 0 to 300 are shown in Fig. 2.9. It an be seen that at steady-state, the theoretialand simulation values agree losely. The dashed lines show the steady state valuesof eah urve. Note that outages result in an inrease in Tm. In Fig. 2.9, HSUs aremore likely hosen in Phase I beause they have higher trust values. Link outagesredue the number of suessfully transmitted disagreements from MSUs to the FC.Sine the majority of the disagreements from MSUs at steady-state are inaurate,the number of negative rewards (penalties) in the FC's evaluation of Tm dereases,leading to a higher Tm value.Fig. 2.10 shows a plot of Qf and Qmd as a funtion of time (in time slots). Theimpat of outage on Qf and Qmd in EE-CSS is simulated and shown in Fig. 2.1045Chapter 2. Trust-based Centralized EE-CSS0 50 100 150 200 250 3000.550.60.650.70.750.80.850.90.951Time, k (Time slots)Average Trust Value  Sim. HSU pout= 0Sim. MSU pout= 0Sim. HSU pout= 0.2Sim. MSU pout= 0.2Steady−state values from the analysis The results for 2 cases of HSU are very close.Figure 2.9: The steady-state average trust value as a funtion of time, k (in timeslots), with pout = pBr.out = 0.2, H = 4, and M = 4. The dashed horizontal lines showthe steady state values for the orresponding urves.0 50 100 150 200 250 30000.020.040.060.080.10.120.140.160.180.2Time, k (Time slots)Q f and Q md   Qf (pout= 0)Qmd (pout= 0)Qf (pout= 0.2)Qmd (pout= 0.2)Figure 2.10: The global FA and MD probabilities (Qf and Qmd, respetively) as afuntion of time, k (in time slots), with pout = 0.2, pBr.out = 0.2, H = 4, and M = 4as a funtion of time (in time slots). The results show that Qf and Qmd inreasewhen outages our. However, extensive simulations for dierent attak probability46Chapter 2. Trust-based Centralized EE-CSS0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 100.10.20.30.40.50.60.70.80.91  Qf  Q md  pout = 0 in EE−CSS pout  = 0.5 in EE−CSS pout = 1 in EE−CSS pout = 0 in T−CSS pout  = 0.5 in T−CSS pout  = 1 in T−CSSFigure 2.11: ROC for EE-CSS and T-CSS when H = 3 and M = 1 for three outageprobability values, pout = 0, 0.5, 1.values, pout, pBr.out show this is not always the ase. Sine the number of suessfullytransmitted disagreements derease with pout, the FC interprets that there are moreagreements with the broadast intermediate deision and therefore, the nal deisionis more likely to be the same as the intermediate deision.The simulated impat of outage on Qf and Qmd in both EE-CSS and T-CSS forpout = 0, 0.5, and 1 in steady-state are illustrated in Fig. 2.11. For pout = 0 and1, Qmd and Qf are the same for EE-CSS and T-CSS. In EE-CSS, when pout = 0,expliit and impliit sensing reports are reeived and interpreted orretly at the FC.In T-CSS, when there is no outage, all messages are reeived at the FC suessfully.Therefore, the nal deision of the FC is the same in both T-CSS and EE-CSS.When pout = 1, all the reports transmitted from SUs are aeted by link outages. InT-CSS, the FC uses its own sensing results to make the nal deision. In EE-CSS,the FC interprets that all other nodes have agreed with the deision of the FC andtherefore, the FC make the nal deision based on its own sensing results. When47Chapter 2. Trust-based Centralized EE-CSS0 0.5 1 1.5 2 2.5 300.20.4 Q f and  Q mdH/M  00.51Avg.   NH /HQmd (T−CSS & EE−CSS) Qf (T−CSS & EE−CSS)Avg. NH/H (EE−CSS)Figure 2.12: Left Y axis: Qf and Qmd in EE-CSS and T-CSS when M = 5 as afuntion ofHM . Right Y axis:NHH in EE-CSS M = 5 as a funtion of HM . In this gurepout = 00 < pout < 1 (e.g. pout = 0.5) in EE-CSS, some expliit sensing report do not reahthe destination and the FC interprets them as agreement messages. On the otherhand, the expliit messages (in T-CSS) whih are not transmitted suessfully to theFC due to outages are not used in the deision making proess; the outage an reduethe number of available reports at the FC and therefore impat the Qf and Qmd. Theresults with pout = 0.5 show that for pratial appliations where Qmd should be small(i.e. Qmd < 0.10), EE-CSS has lower FA probability than T-CSS whih is desirable.For Qmd > 0.30 and pout = 0.50, T-CSS has lower FA probability than EE-CSS. Forpout = 0, the results from simulation and theoretial expressions in (2.21) and (2.22)onrm one another.The global FA and MD probabilities, Qf and Qmd (on the left Y axis) are plottedin Fig. 2.12 as a funtion of Honest-to-Maliious SU ratio for M = 5 and pout = 0 insteady-state for EE-CSS. The results from theoretial expressions in (2.21) and (2.22)48Chapter 2. Trust-based Centralized EE-CSSfor Qf and Qmd are onrmed by omputer simulations. Also plotted are NHH (on theright Y axis) as a funtion of Honest-to-Maliious SU ratio for M = 5 and pout = 0in EE-CSS and T-CSS. The results from theoretial expressions in (2.16)-(D.4) areused to evaluateNHH . The results show that as H inreases in the network, Qf (orQmd) dereases and NHH dereases. Note that in EE-CSS,NHH = 1H +NHIIH andNHIIHremains onstant as H inreases. However 1H dereases as H inreases. Therefore,NHH dereases as a funtion of H . The results show that while Qf (or Qmd) is thesame in both EE-CSS and T-CSS in the absene of outage, theNHH is muh lower inEE-CSS ompared to T-CSS. In EE-CSSNMM = 0.5 for all values of HM . In T-CSSNMM =NHH = 1 for all values of HM . For larity, we have omitted these results fromFig. 2.12.2.6 SummaryA CSS protool is proposed whih aims to improve the energy eieny of T-CSSprotools by reduing the number of sensing reports from SUs to the FC. Expres-sions for the steady-state SU average trust values and the steady-state average totalnumber of sensing reports transmitted for eah band state evaluation were derived.Expressions for Qf and Qmd for a ommonly used deision fusion tehnique were alsoobtained. Outages on the FC and SU links aet the reeption of sensing reportsat the FC. The eet of suh outage on steady-state SU average trust values andaverage number of reports were analyzed. The eet of outage on Qf and Qmd forsenarios with no MSUs in the network were also analyzed. For a given Qf and Qmdtarget values, it is found that EE-CSS requires a smaller steady-state average totalnumber of sensing reports. It is shown that EE-CSS an greatly redue the energyonsumption in EE-CSS ompared to that in T-CSS for long range ommuniations49Chapter 2. Trust-based Centralized EE-CSSwhere the transmission energy is dominant.50Chapter 3Transient Analysis in EE-CSS3.1 IntrodutionA entralized two-phase trust-based energy eient ollaborative spetrum sensing(EE-CSS) protool was proposed in Chapter 2. EE-CSS an redue the number ofsensing reports transmitted to the FC in the presene of MSUs. The steady-stateaverage trust values and average number of sensing reports transmitted in the twophases were analyzed for EE-CSS. To obtain pratial design insights into the impatof dierent parameters on the onvergene of the trust and the number of reports totheir steady-state values in EE-CSS, we study the transient behavior of EE-CSS inthis hapter. In CRNs with TRMSs, the global FA and MD probabilities are mostlyvulnerable during the initialization period, where all SUs are assumed to be HSUs.It is shown in [24,31,4043,81℄ as the trust or reputation values of the SUs omputedat the FC reah their steady-state values, the FC is more eetive in deteting theMSUs and eliminating them from the nal deision proess.The losed-form expressions for the transient probability distributions and av-erages, T h,k and Tm,k, of the trust values for HSUs and MSUs are derived. Exatlosed-form expressions for the average total number, NH,k and NM,k, of sensing re-ports transmitted by HSUs and MSUs are derived. In ontrast to the highly omplexmathematial method we proposed in [96℄, the derived losed-form expressions in thishapter an be used to easily obtain the numerial results for large values of H , M ,51Chapter 3. Transient Analysis in EE-CSSand k, (e.g. H = 5 , M = 10, and k = 30), as illustrated in Setion 3.4.The remainder of this hapter is organized as follows. The system model is de-sribed in Setion 3.2. In Setion 3.3, we derive losed form expressions for 1) T h,k,Tm,k, 2) the average number, NHI ,k, and NMI ,k, of reports transmitted by HSUs andMSUs in Phase I, respetively and 3) the average number, NHII ,k, and NMII ,k, ofreports transmitted by HSUs and MSUs in Phase II, respetively, in EE-CSS. Illus-trative results are presented in Setion 3.4 and the main ndings are summarized inSetion 3.5.3.2 System ModelThe system model in this hapter is similar to that in Chapter 2 and the sameassumptions are used. For onveniene, we present a short summary of the systemmodel and the important equations from Chapter 2. For more details, you may referto Setions 2.2 and 2.3. We onsider a CRN with H HSUs, M MSUs and one FC fora total of N = H +M + 1 sensing entities, where we assume that the hannel gainsare identially and independently distributed (i.i.d.) aording to Rayleigh fading asdisussed in Setion 2.2. An overlay CRN is studied with energy detetors employedat eah sensing entity. The MSUs attak aording to the independent attak poliydisussed in Table 2.1. The FA and MD probabilities of the MSUs seen at the FCbased on the EE-CSS protool and the attak probabilities are :pf,m,FC = (1− pf,m)(pα0,0(1− pf,ID) + pα1,0pf,ID)+ pf,m[(1− pα1,1)pf,ID + (1− pα0,1)(1− pf,ID)], (3.1)52Chapter 3. Transient Analysis in EE-CSSandpmd,m,FC = pmd,m[(1− pα0,0)pmd,ID + (1− pα1,0)pd,ID]+ (1− pmd,m)(pα1,1pd,ID + pα0,1pmd,ID), (3.2)respetively. Note that Pf,ID = 1− (1− Pf,FC)(1− Pf,n∗), Pmd,ID = (1− Pd,FC)(1−Pd,n∗), where n∗ is the most trusted sensing entity hosen by the FC. The FC isloated at the SUBS and the SUBS ommuniates with SUs using both liensed andunliensed bands. The PUBS ommuniates with PUs on the liensed band. The FCommuniates periodially with the PUBS to learn the state of the liensed band inthe previous time slots, i.e. BSI. The BSI allows the FC to ompute trust values forthe nth SU at time slot k asTn,k =1 +∑ki=1 rn,i1 + k , (3.3)where rn,i denotes the reward assoiated with the deision of SUn at time i; its valueis 1 if a orret deision is made and 0 if the deision is inorret. The FC employs atrust-weighted sum deision tehnique to make its nal deision, FDFC,k, about thestate of the hannel at time kFDFC,k =H0, if∑N−1n=1 Tn,kXn,k + TFC,kXFC,k < 0H1, otherwise,(3.4)where Xn,k = −1 for H0 and Xn,k = +1 for H1. We assume that there are no linkoutages (i.e. pout = pBr.out = 0) in this hapter.53Chapter 3. Transient Analysis in EE-CSS3.3 Derivation of Transient Average Trust Valuesand Number of Transmitted Sensing ReportsIn this setion, we derive exat expressions for the transient average trust values ofHSUs and MSUs as well as the transient average total number of sensing reportstransmitted by HSUs and MSUs.3.3.1 Evaluation of T h,k and Tm,kThe probability that the FC reeives a orret report from HSU h isph = 1− pH0pf,h − pH1pmd,h, (3.5)where pH0 and pH1 denote the probability that the hannel is idle and busy, re-spetively. Assuming that an MSU transmits a dishonest report to the FC withprobabilities pα0,0 , pα1,0 , pα0,1 , and pα1,1 dened in Table 2.1, the probability that theFC reeives a orret report from MSU m ispm = 1− pH0(pf,m,FC)− pH1(pmd,m,FC). (3.6)For onveniene, let the HSUs be numbered from 1 to H and the MSUs be numberedfrom H + 1 to H +M , i.e. h ∈ {1, 2, . . . , H} and m ∈ {H + 1, H + 2, . . . , H +M}.We an obtain the transient distribution of the trust value, Th,k, for an HSU and thetrust value, Tm,k, for an MSU at time k as follows. From (5.9), it an be seen thatTh,k or Tm,k is a disrete rv whih takes on value i1+k , i = 1, 2, . . . , k + 1, if and onlyif exatly (i− 1) of the k sensing reports from time slot 1 to time slot k are deemedto be orret by the FC using the BSI. Sine the probability that the FC reeives a54Chapter 3. Transient Analysis in EE-CSSorret report from an HSU (MSU) is ph (pm), the probability mass funtion (pmf)of the trust value for an HSU and MSU an be written aspTl,k( i1 + k)=( ki− 1)(1− pl)k−i+1pi−1l , i = 1, 2, . . . , k + 1, (3.7)where l = h for an HSU and l = m for an MSU. From (3.7), it follows that theaverage trust value at time k for an HSU isT h,k =1 + kph1 + k (3.8)and for an MSU isTm,k =1 + kpm1 + k . (3.9)3.3.2 Evaluation of NHI ,k and NMI ,kWe next onsider the average number of sensing reports transmitted by HSUs andMSUs. To derive the transient average number of reports transmitted by HSUs andMSUs in Phase I, during whih the most trustworthy SU is requested to transmit itssensing report to the FC, we use (3.7) to determine the probability that the mosttrusted SU is an HSU or an MSU. Let Xk = maxh {Th,k} and Yk = maxm {Tm,k}. Thenthe average number, NHI ,k, of sensing reports transmitted by HSUs in Phase I intime slot k is simply the probability that an HSU is hosen, i.e.NHI ,k = Pr{Xk > Yk}+ Pr{Xk = Yk,HSU is hosen}, (3.10)where Pr{Xk = Yk,HSU is hosen} denotes the probability that at least one HSU55Chapter 3. Transient Analysis in EE-CSSand at least one MSUs have the highest trust value in Phase I at time slot k andHSU is hosen. Similarly, the average number, NMI ,k, of sensing reports transmittedby MSUs isNMI ,k = Pr{Xk > Yk}+ Pr{Xk = Yk,MSU is hosen} (3.11)= 1−NHI ,k, (3.12)where Pr{Xk = Yk,MSU is hosen} denotes the probability that at least one HSUand at least one MSUs have the highest trust value in Phase I at time slot k and MSUis hosen. In Appendix E, the losed form expressions for the terms Pr{Xk > Yk},Pr{Xk < Yk}, Pr{Xk = Yk,HSU is hosen}, and Pr{Xk = Yk,MSU is hosen} arederived and their omputational omplexities are analyzed.3.3.3 Evaluation of NHII ,k and NMII ,kWe next derive the average number of reports transmitted by HSUs and MSUs inPhase II in time slot k, denoted by NHII ,k and NMII ,k, respetively. For onveniene,let E1(E2) denote the event that an HSU (MSU) is hosen by the FC to transmitits sensing report in Phase I. Also, let E3(E4) denote the event that in Phase II, thedeision of an HSU (MSU) other than SUn∗ diers from the ID. Then, we an writeNHII ,k = Pr {E1} · (H − 1) · Pr {E3|E1}+ Pr {E2} ·H · Pr {E3|E2} . (3.13)56Chapter 3. Transient Analysis in EE-CSSpH1 M H λ TW SNRdB0.5 10 5 50 20 25Table 3.1: Simulation Parameter Values IIIn (3.13), Pr {E1} is given by the sum of (E.3) and (E.6) whereas Pr {E2} is givenby the sum of (E.4) and (E.8). Similarly, we an writeNMII ,k = Pr {E1} ·M · Pr {E4|E1}+ Pr {E2} · (M − 1) · Pr {E4|E2} . (3.14)In Appendix F, the losed form expressions for the terms Pr {E3|E1}, Pr {E3|E2},Pr {E4|E1}, and Pr {E4|E2} are derived.3.4 Numerial ResultsIn this setion, numerial results are presented to ompare the analytial results forT h,k, Tm,k, NHI ,k, NMI ,k, NHII ,k, and NMII ,k with simulation results. The followingparameter values are used in obtaining the simulation results in Figs. 3.1, 3.2, and3.3. In addition, pα0,0 = pα1,0 = pα0,1 = pα1,1 = 0.5. Fig. 3.1 shows T h,k and Tm,k asa funtion of time slot k. It an be seen that the theoretial results, obtained using(3.8) and (3.9), agree losely with the simulation results. The transient average HSUand MSU trust values approah their orresponding steady state values quite rapidlyas k inreases.In Fig. 3.2, we plot NHI ,k and NMI ,k as a funtion of time slot k. As k inreases,NHI ,k inreases to 1 whereas NMI ,k dereases to 0, orresponding to the fat thatan HSU is inreasingly likely to be hosen by the FC in Phase I. The upper boundand the lower bound for NHI ,k are Pr{Xk > Yk}+ Pr{Xk = Yk} and Pr{Xk > Yk},57Chapter 3. Transient Analysis in EE-CSS0 5 10 15 20 25 300.50.550.60.650.70.750.80.850.90.951Time,  k (Time slots)Average Tt,k  HSU TheoryHSU  SimulationMSU TheoryMSU SimulationFigure 3.1: T h,k and Tm,k as a funtion of time slot k. Theoretial urves obtainedusing (3.8) and (3.9).respetively. Similarly, the upper bound and the lower bound for NMI ,k are Pr{Xk <Yk} + Pr{Xk = Yk} and Pr{X < Y }, respetively. The theoretial and simulationresults for NHI ,k and NMI ,k math losely. As k inreases, Pr{Xk = Yk} → 0 andthe upper and lower bounds onverge to the same values, as an be seen in Fig. 3.2.Moreover, as k inreases Pr{Xk < Yk} → 0 due to the lower trust values of MSUsand hene the probability that the FC would hoose an MSU in Phase I approahes0, i.e. NMI ,k = 0.In Fig. 3.3, NHII and NMII are plotted as a funtion of time slot k. The upperand lower bound values for NHII ,k are obtained by using the upper and lower boundvalues of NHI ,k from Fig. 3.2 in (3.13). Similarly, those of NMII ,k are obtained byusing the upper and lower bound values of NMI ,k from Fig. 3.2 in (3.14). Fig. 3.3shows that in steady-state, the majority of HSUs do not send expliit sensing reportsto the FC.58Chapter 3. Transient Analysis in EE-CSS0 5 10 15 20 25 3000.10.20.30.40.50.60.70.80.91Time,  k (Time slots)Average NHI , k and  NMI , k  HSU ExactHSU SimulationHSU Upper BoundHSU Lower BoundMSU ExactMSU SimulationMSU Upper BoundMSU Lower BoundFigure 3.2: NHI ,k and NMI ,k as a funtion of time slot k. Theoretial urves obtainedusing (E.9) and (E.10).0 5 10 15 20 25 3011.522.533.544.555.5Time,  k (Time slots)Average NHII , k and  NMII , k  HSU ExactHSU SimulationHSU Upper BoundHSU Lower BoundMSU ExactMSU SimulationMSU Upper BoundMSU Lower BoundFigure 3.3: NHII ,k andNMII ,k as a funtion of time slot k. Theoretial urves obtainedusing (3.13) and (3.14).59Chapter 3. Transient Analysis in EE-CSS3.5 SummaryExpressions for the transient probability distributions and averages of the trust val-ues for honest and maliious seondary users (HSUs and MSUs) in EE-CSS werederived. Closed form expressions for the transient average number of sensing reportstransmitted by HSUs and MSUs to the FC in eah phase were also obtained. Thelosed form expressions from Setion 3.3 are less omputationally omplex (and moretime eient) than the method we proposed in [96℄ for the evaluation of the averagenumber of transmitted reports in EE-CSS. The results show that these expressionsan be used to eiently evaluate the average number of reports transmitted for alarge number of SUs (M +H = 15) and time slots (k = 30).60Chapter 4A Collusion Attak in EE-CSS4.1 IntrodutionIn a CRN, the SUs an use a variety of sensing and ollaboration tehniques inorder to redue MDs and FAs. While the objetive of the CRN is to improve theutilization of the spetrum and minimizing the interferene with the PUs, MSUs anmanipulate their sensing reports in CSS to ompromise the deision of the FC inorder to gain unfair advantage like aess to the spetrum or ausing interfereneswith the PUs [14, 29, 43, 4749, 8185, 97100℄. Trust and reputation systems havebeen proposed to mitigate the eets of maliious and dishonest behavior in CRNs[14, 24, 47, 49, 81, 84, 97, 98℄.The impat of SSDF ollusion attaks on Qf and Qmd are studied in [47,81,83,99℄.In [47℄, 2 MSUs ollude in an eort to manipulate the nal deisions of the FC inthe absene of TRMSs, where the nal deision at the FC is based on the majorityrule. The eet of the ollusion attak is then ompared with that of a stationaryindependent attak poliy (i.e. MSUs attak poliy does not hange in time) similarto that in [8185℄ and summarized in Table 2.1, where the MSUs report dishonestlywith ertain probabilities for H0 and H1 hypotheses. The results shows that theollusion attak an impat both Qf and Qmd more severely than the independentattak. In [83, 99℄, a ollusion attak strategy similar to that in [47℄ is studied ina CRN with more than 2 MSUs, where a maliious agent ollets sensing reports61Chapter 4. A Collusion Attak in EE-CSSfrom MSUs and ditates what MSUs report to the FC. The results show that as thenumber of olluding MSUs inreases, the data fusion tehniques without TRMS annot obtain pratial values for Qf and Qmd. In [81℄, a reputation-based method isstudied to mitigate the ollusion attaks of [47, 83, 99℄. The results show that themethod an derease Qf and Qmd. The FC lters the MSUs and applies a majorityrule to the reports it reeives in [81℄. The MSUs in [47,81,83,99℄ do not take advantageof other information in the CRN.In Chapter 2, an independent attak poliy was proposed for EE-CSS and trust-weighted deision was proposed to mitigate the eets of the attak. In the inde-pendent attak, it is assumed that eah MSU attaks the CRN based on its ownobservation (i.e. do not ollude with other MSUs).In this hapter, we propose a entralized trust-based ollusion attak strategy forEE-CSS, where a misbehaving FC (MFC) explores the fat that the ontent of thesensing reports after the broadast an be obtained from MSUs, i.e. an MSU adjaentto an HSU an monitor whether the HSU is transmitting or not in its alloated minitime slot and an potentially obtain information about the ontent of the sensingreports. The MFC ollets information from MSUs and ditates the behavior ofMSUs in eah time slot. Furthermore, we propose a method with a CCF to identifySUs with unusual behavior and to eliminate them from partiipating in the bandstate deision proess being made at the FC.The remainder of this hapter is organized as follows. In Setion 4.2, we presentthe system model. In Setion 4.3, we propose the ollusion attak strategy at theMFC. A mitigating strategy at the FC is proposed in Setion 4.4. Illustrative resultsand onlusions are provided in Setions 4.5 and 4.6, respetively.62Chapter 4. A Collusion Attak in EE-CSS SUBS (FC) PUBS  MSU HSU MSU MSU HSU PU2 PU1 PU3 PU3 FC MFC Figure 4.1: A entralized overlay CRN with H HSUs, M MSUs, an FC, and an MFC,the MFC and MSUs ollaborate in a ollusion attak strategy.4.2 System ModelThe system model in this hapter is similar to those in Chapters 2 and 3 in someaspets. We onsider N−1 SUs (H HSU andM MSUs), an FC at SUBS and a PUBSin the CRN. In addition, we onsider an MFC whih an ollet available informationin the network and an ommuniate with the MSUs. We also assume that all MSUsollude with the MFC. The BS has no knowledge about MCPEs. An overlay CRN isonsidered with energy detetors employed at eah sensing entity. We assume thatthe PUBS sends periodi BSI to the FC so that it an ompute the auray of thesensing reports transmitted from the SUs to the FC during past time slots. Thesystem model is illustrated in Fig. 4.1.The FC ollets N − 1 sensing reports using the EE-CSS protool from the SUsin addition to its own sensing report to make a nal deision on the state of the bandbased on the trust-weighted deision of (2.12). For onveniene, we briey summarize63Chapter 4. A Collusion Attak in EE-CSSthe EE-CSS protool. In EE-CSS, the reports are olleted in two phases. In PhaseI, SUn∗, is hosen to send its report to the FC. The FC applies an OR rule to itsown report and the report from the SUn∗ in order to make an intermediate deision.The intermediate deision and a transmission shedule for N − 2 remaining SUs arethen broadasted to all SUs. In Phase II, eah SU disagreeing with the broadasttransmits its sensing reports to the FC in its alloated mini time slots. We referto these reports as expliit sensing reports hereafter. The FC interprets the idlemini time slot orresponding to a partiular SU as agreeing with the broadastedintermediate deision. Next we propose an attak strategy whih apitalizes on thisproperty of EE-CSS to impat FDFC,k eetively.4.3 Centralized Trust-based Collusion Attak inEE-CSSWe propose a entralized trust-based ollusion attak strategy in whih the MFCditates MSUs what to transmit to the FC. In this strategy, the MFC obtains infor-mation from the network in two phases:ˆ Phase I: the MFC obtains the sensing information from MSUs.ˆ Phase II: the MFC ollets the observational information of MSUs about theiradjaent HSUs.It is assumed that the MSUs report honestly to the MFC. In Phase I, the MFCapplies a trust-weighted deision to MSUs sensing report in order to make a deision,64Chapter 4. A Collusion Attak in EE-CSSdenoted by IDMFC,k, about the state of the band at time slot k,IDMFC,k =0, if ∑Mm=1 T̂m,kYm,k + T̂MFC,kYMFC,k < 01, otherwise,(4.1)where Ym,k ∈ {−1, 1} denotes sensing report of MSUm to the MFC and YMFC,k ∈{−1, 1} denotes the MFC loal deision, i.e. Ym,k,YMFC,k =−1 forH0 and Ym,k, YMFC,k =+1 for H1. Also note that IDMFC,k = 0 and IDMFC,k = 1 are the intermediate de-ision of the MFC for H0 and H1 hypotheses, respetively. In addition, T̂m,k andT̂MFC,k denote the trust values of MSUm and the auray of the MFC estimated atthe MFC based on the FDMFC,k, where FDMFC,k is the nal deision of the MFCand will be disussed shortly. The observational information about HSUs is availableat the MFC in Phase II and therefore, an not be used in (4.1). The MFC appliesthe following ILPO problem for IDMFC,k = 0,min.M∑m=1Zm,ks.t.M∑m=1T̂m,kZm,k +H∑h=1T̂h,kÊ{Xh|H0}+ T̂FC,kÊ{XFC|H0} > 0T̂m,k ≥ τth,MBS for m = 1, ...,M,Zm,k ∈ {−1, 1} for m = 1, ...,M, (4.2)where T̂h,k and T̂FC,k are the trust values of the HSUh and the FC estimated at theMFC based on the information obtained by the MFC from MSUs during Phase IIin the past time slots. Note that the solution to the ILPO is subjet to satisfying athreshold trust value, denoted by τth,MBS. The expeted value of the reports fromHSUh estimated by MSU given H0 is denoted by Ê{Xh|H0}, i.e. (−1)(1 − p̂f,h) +65Chapter 4. A Collusion Attak in EE-CSS(1)p̂f,h, where P̂f,h is the estimated FA probability of HSUh based on the informationobtained by the MFC from MSUs during Phase II from the past time slots. Forexample, if IDMFC,k = 1 and MSU observes that the hth HSU is disagreeing withIDFC,k = 1, the MFC deides that the hth HSU made an inorret deision. The esti-mated value of sensing deisions at the FC, denoted by Ê{XFC |H0}, an be obtainedfrom p̂f,h and the observed p̂f,ID, respetively, i.e. if p̂f,ID and p̂f,h are auratelyestimated, p̂f,FC an be obtained from (2.10). If there are solutions for the ILPOin (4.2), the MFC hooses one solution randomly and ditates MSUs to transmitsensing reports aording to the solution of the ILPO, e.g. for a CRN with M = 3, ifZ1,k = 1, Z2,k = −1, Z3,k = −1, then MSUs report to the FC as X1,k = 1, X2,k = −1,X3,k = −1. If there is no solution to ILPO above, MSUs will report IDMFC,k to theFC instead of their own deisions, i.e. X1,k = X2,k = X3,k = IDMFC,k. Similarly, forIDMFC,k = 0 we have,max.M∑m=1Zm,ks.t.M∑m=1T̂m,kZm,k +H∑h=1T̂h,kÊ{Xh|H1}+ T̂FC,kÊ{XFC|H1} < 0T̂m,k ≥ τth,MBS for m = 1, ...,M,Zm,k ∈ {−1, 1} for m = 1, ...,M, (4.3)where the expeted value of the reports from HSUh and the FC observed by MSUgiven H1 are denoted by Ê{Xh|H1} and Ê{XFC|H1}, respetively. The expetedvalue of the sensing reports from HSUh estimated by MSU given H1 is denotedby Ê{Xh|H1} and an be obtained from (−1)(1 − p̂f,h) + (1)p̂f,h. Note that theexpeted value of sensing deisions at the FC given H1, denoted by Ê{XFC|H1}, an66Chapter 4. A Collusion Attak in EE-CSSbe estimated from p̂d,ID and p̂d,h. If τth,MBS is large (e.g. τth,MBS = Th,k), (4.2) and(4.3) are less likely to have solutions and as a result, MSUs are less likely to impatFA and MD probabilities drastially. However, the MFC aims to maintain the trustvalues of MSUs evaluated at the FC as high as τth,MBS. The hoie of τth,MBS involvesstudying the trade o between the high trust value against the impat on FA andMD probabilities.The optimization problems in (4.2) and (4.3) aim to use the minimum numberof dishonest reports whih is expeted to impat the TW deision of the FC. Thisoordinated attak inreases the hane of suess while reduing the impat on thetrust values of MSUs omputed at the FC. In Phase II, the MFC uses the solution toan integer linear programming optimization problem to ditate MSUs what to reportto the FC.In Phase II, MSUs will report their observation about their adjaent HSUs tothe MFC. This observation an reveal the ontent of reports transmitted from theHSUs to the FC. Reall when the intermediate deision is broadasted, only SUswhih disagree with the broadast will report to the FC and hene, MSUs adjaentto an HSU an detet whether the HSU is transmitting (disagreeing) or remainingsilent (agreeing). The MFC uses MSUs' sensing reports from Phase I and MSUs'observational information from Phase II in trust-weighted deision to make the naldeision,FDMFC,k =0, if ∑Mm=1 T̂m,kYm,k + T̂MFC,kYMFC,k +∑Hh=1 T̂h,kXh,k < 01, otherwise,(4.4)where FDMFC,k denotes the nal deision of the MFC in Phase II at time slot k.Note that the results obtained from IDMFC,k are only used for the integer linear67Chapter 4. A Collusion Attak in EE-CSSprogramming optimization problem (ILPO) and that from FDMFC,k is only used toestimate to the trust values, FA probabilities, and MD probabilities of HSUs, MSUs,MFC, and the FC. For larity, we present the steps whih takes plae at the FC andthe MFC during Phases I and II in a hart in Fig. 4.2.• The FC chooses SUn* and request for its report• The FC awaits transmission from SUn*• The FC broadcasts IDFC,k as well as mini time slot schedules for each SU • The FC listens to sensing reports from HSUs and MSUs; implicit and explicit sensing reports.• The FC makes FDFC,k• The FC computes Tn,k based on BSM• The MFC collects Ym,k• The MFC uses (4.1) to get IDMFC,k • The MFC uses (4.2) and (4.3) to solve for optimal Zm,k• The MFC and MSUs listen to IDFC,k• The MFC and MSUs adjacent to HSUs observe whether HSUs are transmitting in their allocated mini time slots.• The MFC collects observational information from  MSUs • The MFC estimates T n,kbased on FDMFC,kPhase IPhase IIThe FC The MFCFigure 4.2: The operations of the FC and MFC during the two-phase entralizedollusion attak in EE-CSS68Chapter 4. A Collusion Attak in EE-CSS4.4 Cross-orrelation Filter Method in EE-CSSWe propose a method whih uses ross-orrelation value of eah pair of sensing entities(inluding the FC) in order to identify misbehaving SUs. This method aims to identifyabnormal behavior and to eliminate MSUs from the nal deision proess at the FCin (2.12). The FC updates the trust value and ross orrelation for all the SUs inevery time slot and at every Tcc, respetively. The misbehaving SUs an improvetheir trust values by reporting honestly and an be identied as behaving SU. Theross orrelation of the reports for SUi and SUj from time slot 1 to time slot k anbe obtained fromRk,ij =k∑l=1Xi,kX∗j,k, (4.5)where i 6= j, i, j ∈ {1, 2, . . . , N}, the orrelation lag is assumed to be equal to zero,and X∗i,k = Xi,k sine the reports are real values (i.e. Xi,k, Xj,k ∈ {−1, 1}). The FComputes the ross-orrelation between the reports up to kth time slot for two SUsusing (4.5). The FC obtains a set of Rk at time slot k whih inludes all possibleross-orrelation values for eah pair of sensing entities, i.e. Rk = {Rk,ij|i 6= j, i, j ∈{1, 2, . . . , N}, where |Rk| =(N2). In addition, the FC obtains the expeted value andthe variane of the ross-orrelation of HSUs fromE{Rk,ij} = k(aij + bij) (4.6)andσ2Rk,ij = k[(aij + bij)− (aij − bij)2], (4.7)69Chapter 4. A Collusion Attak in EE-CSSwhere ai,j orresponds to two senarios in whih the reports from the two SUs aresimilar, i.e.ai,j = pH0(pf,ipf,j + (1− pf,i)(1− pf,j))+ pH1(pd,ipd,j + (1− pd,i)(1− pd,j)), (4.8)and bi,j orresponds to two senarios in whih the reports from two SUs are dierent,i.e.bi,j = pH0(pf,i(1− pf,j) + (1− pf,i)pf,j)+ pH1(pd,i(1− pd,j) + (1− pd,i)pd,j), (4.9)where the term b orresponds to two senarios in whih the reports from two SUsare not the same. Assuming all loal FA probabilities, as well as MD probabilities,are the same pf,i = pf,j = pf and pd,i = pd,j = pd for i, j = 1, 2, . . . , H + M + 1,aij = pH0(p2f +(1−pf)2)+pH1(p2d+(1−pd)2) and bij = 2pH0pf(1−pf )+2pH1pd(1−pd).The higher and lower (than mean) ross-orrelation value orresponding to SUi andSUj an indiate misbehavior of at least one of the two SUs, e.g. the two SUs maybeolluding and reporting dishonestly at the same time. It an also indiate that thetwo SUs are suering from link outage and their silene is being interpreted as anagreement at the FC. While the SU suers from outage and transmits inauratereports to the FC, it will be ategorized as misbehaving SU.The FC allows pairs of SUs with ross orrelation values within a ondeneinterval, ψth,k, of the expeted ross orrelation value at time slot k to partiipate in(2.12). The ondene interval an be obtained from70Chapter 4. A Collusion Attak in EE-CSSif k = Tcc thenCompute expeted and variane ross-orrelation value of the reports from twoHSU, i.e. E{Rk,ij} and σ2Rk,ij .Compute the ondene interval, i.e. ψth,k.Evaluate observed Rk,ij for(N2)possible ross-orrelation values, i.e. Rk ={Rk,i,j|i 6= j, i, j = 1, 2, . . . , N} and |Rk| = N(N−1)2 .Categorize all SUs as an MSU.for eah member of Rk doif |Rk,i,j − E{Rk,ij}| ≤ ψth thenCategorize i and j as HSU.end ifend forend ifFigure 4.3: Pseudoode for the ross-orrelation lter Methodψth,k =√σ2Rk,ijǫup, (4.10)where ǫup is the upper bound error in the Chebyshev's inequalityPr(|Rk,i,j − E{Rk,ij}| ≥ ψth) ≤ ǫup, (4.11)where Rk,ij is obtained from (4.5).As the number of observation inreases, the variane of the observed ross-orrelationvalues for HSUs dereases. And the CCF method is more suessful to eliminate mis-behaving SUs. The Pseudoode for the ross orrelation lter method is summarizedin Fig. 4.3. The SUs ategorized as MSUs are not onsidered as SUn∗ in Phase I andare not used in (2.12).71Chapter 4. A Collusion Attak in EE-CSS4.5 Numerial ResultsIn this setion, the impat of two attak poliies (the proposed ollusion attak andthe independent attak) on Qf and Qmd are simulated and ompared. In addi-tion, omputer simulations are used to show the eetiveness of the CCF method ineliminating the misbehaving SUs from the deision making proess at the FC. Thesimulation results presented are averaged over 10,000 yles. In addition, MAT-LAB Smooth funtion in R2011b edition with a window equal to 11 is used tosmooth the urves generated in Figs. 4.4 and 4.6. The parameters used to generatethe simulation results in Figs. 4.4-4.7 are summarized in Table 4.1. In addition,pα0,0 = pα1,0 = pα0,1 = pα1,1 = 0.5 and τth,MBS is hosen to be τth = 0.90T h.pH1 PUSNR M H TW λ τth,MBS Tcc50% 25 (dBm) 4 3 20 40 0.9T h 100Table 4.1: Simulation Parameter Values IIIFig. 4.4 shows the average global FA and MD for two CRNs, one with independentattak poliy and the other with the proposed ollusion attak poliy, as a funtionof time slot k. The ollusion attak impats the FA and MD probabilities worsethan the independent attak. The steady-state value of Qf has inreased with theollusion attak poliy when ompared to the independent attak. Similarly, theollusion attak has aused the steady-state value of Qmd to inrease. Therefore,the ollusion attak poliy at the MFC an utilize the available information in theognitive network to ompromise the deision of the FC more eetively than theindependent attak poliy.More investigations on the values of Qf and Qmd in CRN, with the ollusionattak poliy indiate that they are highly dependent on the value of λ. For ex-ample, when λ = 50, the steady state values of Qf and Qmd are approximately72Chapter 4. A Collusion Attak in EE-CSS0 100 200 300 400 500 600 700 800 900 100000.10.20.30.40.50.60.70.80.91Time,  k (Time slots)Q f  and  Qmd   Qmd  Independent AttackQmd   Collusion AttackQf  Independent AttackQf  Collusion AttackFigure 4.4: The global FA and MD probabilities (Qf and Qmd, respetively) as afuntion of time slot k for the CRN with independent and ollusion attaks2% and 75%, respetively. This an be explained as follows. When λ = 50, theloal FA probabilities, pf,FC and pf,h, of FC and HSUs are very low ompared tothe loal MD probabilities, pmd,FC and pmd,h, at eah SU and therefore, the value of∑Hh=1 T̂h,kE{Xh|H0} + T̂FC,kE{XFC |H0} in the rst onstraint of (4.2) is expetedto be small. Hene, (4.2) is less likely to have any solutions. If there is no solutionfor (4.2), MSUs report IDMFC,k to the FC whih is a trust-weighted deision of theMFC and is a more aurate deision than the individual MSU deisions. When thehannel is busy,∑Hh=1 T̂h,kE{Xh|H1} + T̂FC,kE{XFC|H1} in (4.3) is more likely tobe satised due to large values of pmd,FC and pmd,h, i.e. lower number of MSUs arerequired to report dishonestly in order to satisfy the rst onstraint in (4.3). As aresult, the ollusion attak poliy is more likely to manipulate the deision of the FCfor IDMFC,k = 1 and hene, Qmd is large.Further investigations have shown that when λ is hosen to be 30, the ollusion73Chapter 4. A Collusion Attak in EE-CSS0 100 200 300 400 500 600 700 800 900101102103Time,  k (Time slots) RXk,ij  HSU−HSUMSU−MSUHSU−MSUFigure 4.5: The average ross-orrelation of SU pairs as a funtion of time slot k.attak poliy auses the Qf to inrease and Qmd to derease drastially.Fig. 4.5 shows the ross-orrelation between sensing reports for three pairs,namely, MSU-MSU, HSU-MSU, HSU-HSU, as a funtion of time. The ross-orrelationis alulated at periods equal to Tcc at the FC. The results illustrate that the ross-orrelation values of HSU-HSU pairs remain below those of MSU-MSU pairs andabove those of HSU-MSU pairs.Fig. 4.6 shows Qf and Qmd in CRNs with the ollusion attak poliy, with andwithout the CCF method at the FC, as a funtion of time. The FC identies the SUswith the observed ross-orrelation value larger than a ondene interval omputedfrom (4.11) with ǫ = 0.1%. Furthermore, the FC eliminates the MSU reports frominuening FDFC,k as k inreases. The ltering proess allows a redution in Qmdand Qf . Given (4.6) and the limits obtained from Chebyshev's inequality, the FCan eliminate the SUs whih tend to have lower or higher ross-orrelation valuesthan the expeted value. Fig. 4.7 shows the average suess per attak ratio (SAR)74Chapter 4. A Collusion Attak in EE-CSS0 100 200 300 400 500 600 700 800 90000.10.20.30.40.50.60.70.80.91Time,  k (Time slots)Q f  and  Qmd   Qf Collusion Attack with CCFQf  Collusion AttackQmd  Collusion Attack with CCFQmd  Collusion AttackFigure 4.6: The global FA and MD probabilities (Qf and Qmd, respetively) as afuntion of time slot k for CRN with ollusion attak and CCF method0 100 200 300 400 500 600 700 800 9000.20.30.40.50.60.70.80.91Time,  k (Time slots)Success per Attack Ratio (SAR)   Collusion AttackCollusion Attack with CCFIndependent AttackFigure 4.7: The average suess-per-attak ratio (SAR) as a funtion of time slot k.for CRNs with independent attak, the proposed ollusion attak, and the ross-orrelation lter at the FC as a funtion of time slot k. The suess per attak75Chapter 4. A Collusion Attak in EE-CSSorresponds to senarios in whih at least one MSU is reporting dishonestly and theFC makes an inorret deision. Note that SAR also inludes the impat of FAs andMDs from HSUs. We use SAR to get an insight into the suess ratio of attaks orounter-attak strategies. In EE-CSS with independent attak, only 25% of attakingsenarios ause the FC to make an inorret deision. However, in EE-CSS with theproposed ollusion attak, 50% of the attaking senarios ause the FC to make aninorret deisions. The eetiveness of the ross-orrelation lter an be observedas k inreases. Note that 30% SAR in ollusion with CCF signies that with theremoval of MSUs, there are only 3 HSUs whih ollaborate with the FC to make anal deision and they ontribute to 30% ombined FA and MD probabilities.4.6 SummaryA entralized trust-based ollusion attak strategy was proposed whih explores thefat that the ontent of the sensing reports transmitted from SUs to the FC in thePhase II of the EE-CSS an potentially be obtained. The proposed ollusion attakstrategy was shown to impat Qf and Qmd more severely than the independent attakstrategy. A CCF method was proposed to identify and to eliminate the misbehavingSUs from ontributing to the deision making proess at the FC in the ognitivenetwork. The proposed CCF method was shown to be eetive in eliminating themisbehaving SUs and in improving Qf and Qmd.76Chapter 5Trust-based Centralized Spetrumand Energy Eient CSS (SEE-CSS)5.1 IntrodutionThe growing demand for wireless servies has shown a dramati inrease in reentyears due to ontent sharing (i.e. videos, pitures, et.), web browsing, and emails onmobile devies. Numerous studies have shown that several spetral bands, partiu-larly the TV broadast bands remain under utilized [101℄. The TV broadast bandsare desirable for long range ommuniations due to their low free spae propagationloss and building penetration [102℄. The IEEE 802.22 working group has proposedthe rst entralized CRN known as WRAN, whih aims to utilize the TV broadastbands to provide broadband servies to diverse or hard-to-reah geographial areas aslong as it does not interfere unduly with the IUs [103,104℄. In the ontext of WRAN,SUs are known as CPEs.In the IEEE 802.22 standard WRAN, numerous tools are available at the basestation (BS) and CPE to protet the IUs [105℄, e.g. quiet period synhronization,sensing methods, and CSS. Synhronization mehanisms and quiet periods are usedby the BS and CPEs to sense the spetrum in a oordinated manner [103℄. The CPEsommuniate the spetrum sensing reports to the BS, on demand or periodially, inorder to improve the IU detetion and spetrum opportunity detetion [44℄. The77Chapter 5. Trust-based Centralized SEE-CSSBS reates a list of available hannels and bak up hannels based on the onsensusand propagates the list among CPEs periodially. The sensing reports from CPEsat the BS may dier due to sensing errors, shadow fading, misbehavior, et. Par-tiularly, misbehaving CPEs (MCPEs) may share dishonest sensing reports with theBS in an attempt to ompromise the nal deision regarding the state of a spe-trum band. MCPEs may be motivated to disrupt the operation of the network orto gain unfair advantages to the available spetrum. Trust and reputation systemshave been proposed to mitigate the eet of misbehaving and dishonest behavior inCRNs [14, 29, 43℄.The energy and bandwidth onsumed during CSS in CRNs have been studiedin [26,54,56,57,59,60,106109℄. The number of sensing reports are optimized subjetto satisfying target inumbent detetion and spetrum opportunity detetion on aliensed hannel in [56,57℄. In [106℄, a sequential FFT based pilot sensing method isproposed to detet pilots in Advaned Television Systems Committee (ATSC) signalswhere the CPE varies the number of dwells (i.e. number of times required to observea hannel in order to obtain target false alarm and miss detetion probabilities)dynamially aording to the SNR of the reeived ATSC signals. As a result, CPEsan redue the number of dwells when the SNR is high and the sensing entity isondent in its deision. In [59℄, a sensing method is proposed in whih the numberof sensing frequenies (i.e. number of times to observe a hannel in a period) variesaording to the observed ative and idle probability distribution of the IU, therebyimproving the energy onsumption in the spetrum sensing when the hannel statetransition probabilities are small. In [107℄, a robust sensing method is proposedfor a type of primary user signal, namely PCP-OFDM, at low SNR values. Thedetetion sheme an redue the sensing time and inrease the data transmission78Chapter 5. Trust-based Centralized SEE-CSSand thereby, inrease the spetrum eieny. The tradeo between sensing aurayand utilizing spetrum opportunities is studied in [60, 108℄. Objetive funtion withsensing duration and data transmission duration is dened in [60℄. The objetivefuntion is optimized subjet to satisfying energy resoure and target inumbentdetetion probability onstraints. A novel approah, namely EE-CSS, based on a trustmanagement system was shown to improve the energy eieny of the CSS in thepresene of MSUs in a two-phase ontention-free MAC protool in [109℄. The resultsshow that the number of reports transmitted by the HSUs are redued signiantly.Although the proposed protool is energy eient, it is not spetrum eient dueto the TDMA struture. Here, we adapt EE-CSS to a dynami spetrum multipleaess (DSMA) framework, inspired by the IEEE 802.22 standard, where we aim toimprove the energy onsumption as well as the bandwidth usage in the CSS.In the IEEE 802.22 standard, eah WRAN ell an servie several tens of CPEs.As the number of CPEs in WRAN inreases, the energy and spetrum overheadosts assoiated with IU detetion and spetrum opportunity detetion also inreasein T-CSS, where eah sensing entity sends sensing a report to the deision mak-ing entity. In the presene of MCPEs, the HCPEs have the burden to oset thenegative impat of MCPEs in the CSS. In [26, 54, 56, 57, 59, 60, 106108℄, energy orbandwidth onsumption have been onsidered in the absene of MSUs or MCPEs.In this hapter, we propose a spetrum and energy eient CSS protool (SEE-CSS)whih aims to redue the number of urgent oexistene situation (UCS) notiationsfrom the CPEs to the BS while meeting the same Qf and Qmd as T-CSS protool inWRAN. We dene throughput and energy eieny onsumption models to omparethe data throughput and the energy onsumed in the T-CSS and SEE-CSS protoolsin WRAN. We propose a trust-based entralized data falsiation ollusion spetrum79Chapter 5. Trust-based Centralized SEE-CSSsensing attak strategy whih aims to manipulate the deision of the BS severely.We propose the CCF method to mitigate the impat of the proposed ollusion at-tak. The ollusion attak strategy and the CCF methods in this hapter are similarto those in Chapter 4. We show that the SEE-CSS protool is more spetrum andenergy eient than the T-CSS protool.The remainder of the paper is organized as follows. In Setion 5.2, we presentthe system model inluding the signal propagation model, hannel model, and theBS deision method. In Setion 5.3, we propose the SEE-CSS protool and we deriveexpressions for the average number of CPEs whih partiipate in the T-CSS and SEE-CSS protools. We dene a spetrum and energy onsumption of T-CSS and SEE-CSS and we derive two expressions to evaluate the bandwidth and energy eienyratios between the T-CSS and SEE-CSS protools. In Setion 5.4, we disuss theattak strategy and the mitigating CCF method. Illustrative results and onlusionsare provided in Setions 5.5 and 5.6, respetively.5.2 System ModelIn this setion we will desribe some aspets of the IEEE 802.22 standard whihare relevant to this hapter, e.g. the signal propagation and noise models, sensingmethod, and the deision making method used at the BS.5.2.1 Network ModelWe onsider a WRAN ell and a total of N − 1 CPEs and one BS (i.e. a total of Nsensing entities) in our network model, where the CPEs and BS are stationary. TheWRAN ell is outside the proteted ontour (keep-out-radius∼ 150.3 km) of the TVbroadast area as shown in Fig. 5.1. There are H HCPEs and M MCPEs, where80Chapter 5. Trust-based Centralized SEE-CSSBSHCPE TV Antenna(PUBS)IU Distance > 150kmMCPE Figure 5.1: Network model in a WRANH +M = N − 1. For onveniene, let the HCPEs be numbered from 1 to H and theMCPEs be numbered from H + 1 to H + M . We assume that all CPEs an sensethe hannel and send bak the reports to the BS. The system model onsists of aprimary user base station (PUBS) where the TV broadast antenna is loated. TheTV antenna operates on a UHF band transmitting at an eetive radiated power(ERP) of 1 MW (90 dBm) [103℄.5.2.2 Signal Propagation Model, Noise Model, and SensingMethodThe TV signal power in dB at the jth sensor entity, denoted by Pj,dB, an be expressedas [110℄:Pj,dB = PTV,dB + 10αpl log10(d0dj) + SdB, (5.1)where PTV,dB, αpl, d0, dj, and SdB are the TV signal transmit power, path lossexponent, referene distane, distane from the TV broadast antenna to the jth81Chapter 5. Trust-based Centralized SEE-CSSCPE, and shadow fading power gain, respetively. The (log-normal) shadow fadingis desribed by a Gaussian rv with mean zero and variane σ2dB, i.e. SdB ∼ N (0, σ2dB).The average noise power at sensing nodes is [105℄Pn0,dBm = −174 + 10log10(B) +NF, (5.2)where the −174 (dBm/Hz), B (6 MHz), and NF (∼ 11 dBm) denote the noiseonstant per 1 Hz bandwidth, hannel bandwidth, and noise gure, respetively.Noise temperature is obtained KBTc = 1.38 × 10−23 × 300 J, where KB is theBoltzmann onstant in (J/K) and Tc is the noise temperature in (K). It is om-mon to refer to the noise power in logarithmi sale with respet to 1mWatt, i.e.10log10(1.38×10−23×300×6×1061mWatt ) ≃ −95 dBm. The noise gure is the ratio of the SNRof the PU signal at the input to output of the iruit at the CPE reeiver. Pra-tial values for the average reeived power of the TV signal and the orrespondingshadow fading standard deviation at eah sensing CPE (based on its geoloationand terrain with respet to the TV antenna) an be obtained from the InternationalTeleommuniation Union Radioommuniation (ITU-R) douments [111℄.The SNR in dBm at the jth sensing node, denoted by γj,dBm, an be obtained asγj,dBm = Pj,dBm − Pn0,dBm and its distribution is given by [105℄fΓj(γj,dBm) =12πσ2dBme(−(γj,dBm−µj,dBm)2σ2dBm)(5.3)where µj,dBm = PTV,dBm + αpl10 log10(d0dj ). Energy detetion method is used at eahsensing node due to its simpliity [7℄. The test statistis of the signal reeived at the82Chapter 5. Trust-based Centralized SEE-CSSjth sensor an be approximated for low SNR region by [105℄Uj ∼N (Pn0,dBm,P 2n0,dBmNs ) ,H0N (Pj,dBm + Pn0,dBm,(Pj,dBm+Pn0,dBm)2Ns ) ,H1,(5.4)where Ns is the number of samples at eah sensor and H0 and H1 denote idle andbusy inumbent hannel hypotheses, respetively. Note that Ns is a funtion ofsampling frequeny, fs, and observation period, Ts, (i.e. Ns = Tsfs) and is hosenat a suiently large value so that the distribution of rv Uj onverges to a Gaussianrv in (5.4). Given a FA probability, pf,j , at the jth CPE, the energy threshold isobtained by [105℄λj = Pn0(1 +Q−1(pf,j)√Ns), (5.5)where Q−1 is the inverse Q funtion, Pn0 is the noise power in Watt, and λj is theenergy threshold at the jth CPE. The orresponding detetion probability, pd,j , atthe jth CPE an be obtained by [105℄pd,j = 1−Q( NsPj + Pn0[(Pj + Pn0)− λj])1−Q( Ns(γj + 1)Pn0[((γj + 1)Pn0)− λj]), (5.6)where Pj is the reeived signal power at the jth CPE in Watt and γj = PjPn0 . Basedon the SNR distribution, we an obtain the average detetion probability aspd,j =∫ ∞0pd,j |γjfΓj (γj)dγj. (5.7)If Uj ≥ λj , the jth CPE deides the hannel is busy, i.e. Dj = 1, where Dj denotes83Chapter 5. Trust-based Centralized SEE-CSSthe deision of the jth CPE; if Uj < λj , the jth CPE deides the hannel is idle, i.e.Dj = 0.5.2.3 The T-CSS Protool in the IEEE 802.22 WRANIn the IEEE 802.22 standard, the BS ollets sensing reports for in-band (i.e. theoperating hannel and the immediately adjaent hannels) and out-of-band hannels(i.e. any hannel that is not in-band) hannels from the CPEs based on demand fromthe BS or periodially [103℄. In this setion, we analyze the energy and spetrumoverhead osts in CSS for the in-band hannels.During the T-CSS protool in the WRAN, the CPEs use UCS notiations toinform the BS about the presene of IUs on the in-band hannels. The notiationsan be sent to the BS in a MAC frame after a quiet period [103℄ in one of twodierent periods in a frame: dediated ontention-based UCS period or ontention-free upstream (US) period. If a CPE with US resoure detets an IU signal on thein-band hannels, it sets the UCS ag bit in the generi MAC header (GMH) andsends UCS notiations to the BS during the ontention-free period. If the CPEdoes not have US resoures, it uses the UCS ontention-based period to transmit theGMH (32 bits). If the CPE is unable to transmit its UCS notiation in the urrentframe (i.e. due to ollisions), it requests US resoures for future frames or attemptsto transmit the UCS notiation in future UCS ontention-based periods.Although the T-CSS protool desribed in [103℄ may take more than two framesto transmit all the UCS notiations to the BS, for the sake of omparing the T-CSS and SEE-CSS protools, we assume that the BS ollets sensing reports in twoonseutive frames (i.e. Frames i and i+ 1) in CSS period (CSSP) k and i = 2k − 1and k = ⌊ i+12 ⌋ for k ∈ N and eah frame is 10 ms long.84Chapter 5. Trust-based Centralized SEE-CSS5.2.4 Deision Fusion and Trust ModelThe BS uses a trust-weighted (TW) variable to make a nal deision, FDBS,k, at theend of Frame i+ 1 in CSSP k asFDBS,k =0, if∑N−1j=1 Tj,kXj,k + TBS,kXBS,k < 01, otherwise,(5.8)where FDBS,k ∈ {0, 1} for H0 and H1), respetively. The sensing report reeived atthe BS from the jth CPE during CSSP k is denoted by Xj,k ∈ {−1, 1}, where Xj,k =+1 if a UCS notiation is reeived from the jth CPE and Xj,k = −1 otherwise.Similarly, the loal sensing deision of the BS during CSSP k is denoted by XBS,k ∈{−1, 1} and we assume that the BS annot be ompromised. Note that −1 and +1orrespond to H0 and H1 hypotheses, respetively. The trust value, Tj,k, of the jthCPE during CSSP k and the auray of the BS during CSSP k is denoted by TBS,k.The value of Tj,k is obtained by usingTj,k =1 +∑kl=1 rj,l1 + k , (5.9)where∑kl=1 rj,l denotes the reward assoiated with the deision of the jth CPE inCSSP k. Note that rj,l = 0 or 1, if an inorret or orret deision is made, respe-tively. The `1's in the numerator and the denominator reet the fat that the trustvalues are initialized to 1 at k = 0. The trust values are evaluated at the end of CSSP,if spetrum BSI for the previous periods is available. This an be obtained from thePUBS periodially. The average transient and steady-state trust values of HCPEsare derived in [96℄. Based on FDBS,k, the BS deides whether CPEs should vaateor remain on the in-band hannels. If the BS deides that the operating hannel85Chapter 5. Trust-based Centralized SEE-CSSshould be vaated, it will exeute a spetrum management ommand, in whih theBS requests all CPEs to swith hannel (i.e. CHS-REQ) and the BS will transmit theappropriate hannel parameters in the next frame ontrol header (FCH) in the down-stream (DS). The IEEE 802.22 standard mandates the BS and CPEs to vaate thein-band hannels in 2 s, if an IU signal is deteted in the operating in-band hannel.5.3 The Proposed SEE-CSS ProtoolThe SEE-CSS protool aims to derease the number of UCS notiations from theCPEs to the BS while meeting the same global FA and global MD probabilities asthose ahieved in T-CSS and thereby, to derease the overhead spetrum and energyonsumption. The SEE-CSS protool is based on the observation that HCPEs aremore likely to agree than to disagree on the state of a spetrum band. An exampleof the SEE-CSS protool is presented in two onseutive frames, Frames i and i+ 1,as shown in Fig. 5.2. Eah frame onsists of a DS and an US. For simpliity, we haveomitted time intervals suh as time buers between the US and DS and ontention-based periods (i.e. ranging, BW request, self oexistene window, et.) in Fig. 5.2beause these periods are ommon in both the SEE-CSS and T-CSS protools.The T-CSS and the SEE-CSS protools are similar in Frame i. We next disussthe details of the SEE-CSS and T-CSS protools next.5.3.1 Intermediate Deision and BroadastThe BS reeives UCS notiations from a subset of CPEs in Frame i, denoted byIi, i.e. the CPEs whih have deteted an IU signal on the in-band hannel withUS resoures and the ones without US resoures whih suessfully transmitted theUCS notiations. At the end of Frame i during CSSP k, the BS uses its own loal86Chapter 5. Trust-based Centralized SEE-CSSBSCPE 1CPE 3CPE 2Contention-based UCS PeriodContention-free US PeriodQuiet PeriodFrame i =2k-1 Frame i+1=2kUCS+ DataDTV BusyCPE nUSDSDS*DS USBLM-REPBLM-REPDSUCS-DUCSUCS-DCSSP kContention-free US PeriodContention-based UCS PeriodFigure 5.2: The SEE-CSS protool in CSSP k (k = ⌊ i+12 ⌋)deision, DBS,k, and the UCS notiations to make an intermediate deision, IDBS,k,asIDBS,k =DBS,k, if Tj∗,k < τth,BS1, otherwise ,(5.10)where IDBS,k, DBS,k ∈ {0, 1} (for H0 and H1), Tj∗,k = maxj∈Ii {Tj,k}, and τth,BS is athreshold trust value used by the BS to minimize inaurate UCS notiations fromimpating IDBS,k. The value of τth,BS is determined by the BS based on the observedtrust values of the most aurate CPEs. The rule used in (5.10) is a modied versionof the OR rule mandated by the IEEE 802.22 standard for ertain regulatory domainssuh as the US for the protetion of the IUs. Denoting the set of CPEs by I∗i suh87Chapter 5. Trust-based Centralized SEE-CSSthat Tj,k ≥ τth,BS for j ∈ I∗i and I∗i ⊆ Ii, the false alarm and detetion probabilitiesof IDBS,k during CSSP k, denoted by pf,ID and pd,ID, an be obtained aspf,ID =1−∏j∈I∗i(1− pf,j)+ pf,BS∏j∈I∗i(1− pf,j) (5.11)andpd,ID =1−∏j∈I∗i(1− pd,j) + pd,BS∏j∈I∗i(1− pd,j), (5.12)respetively, where pf,BS and pd,BS are the false alarm and detetion probabilities ofthe BS. We denote the probability that at least one UCS report is reeived at the BSfrom a trusted CPE (with Tj∗,k ≥ τth,BS) by pCPE∗.At the beginning of Frame i + 1, the BS broadasts the intermediate deisionduring the DS as shown by DS∗in Fig 5.2. From the subset of CPEs whih were unableto send UCS notiations in Frame i (due to ollisions, lak of US resoures), theCPEs whih disagree with the broadasted IDBS,k will transmit UCS disagreement(UCS-D) notiations in Frame i+1 in the subsequent UCS ontention-based period,as shown in Fig. 5.2. Note that if the BS reeives a GMH with a UCS ag bit setto 1 in Frame i + 1 (subsequent to the broadast of IDBS,k), the BS interpret thatas a disagreement with IDBS,k. At the end of Frame i + 1, the BS uses the UCSnotiations in Frame i and the UCS-D notiations in Frame i + 1 to makes thenal deision based on (5.8).5.3.2 Partiipating CPEs in SEE-CSS and T-CSSWe assume that eah CPE is ative and remains ative (i.e. CPE partiipates in theCSS) for the duration of CSSP k with the probability pact,k. The subset of CPEswhih are ative in CSSP k are denoted by Ak. We denote the set of ative HCPEs88Chapter 5. Trust-based Centralized SEE-CSSand MCPEs by Ak,H and Ak,M , where Ak,H,Ak,M ⊆ Ak (Ak,H ∩ Ak,M = φ andAk,H ∪ Ak,M = Ak).In the urrent frame, the BS an alloate US resoures to CPEs whih haverequested US resoures in the past. Let pus,H,i and pus,M,i denote the probability thatthe BS alloates US resoures to HCPEs and MCPEs in Frame i, respetively. Asthe BS obtains more insight into the behaviors and auray of the CPEs in CSSP,it an ategorize them into honest and misbehaving CPEs and alloate US resouresaordingly, i.e. the honest behavior is enouraged by an inrease in the probabilityof US resoure alloation and the misbehavior is disouraged by a derease in theprobability of US resoure alloation. We dene Ui,H and Ui,M , as the subset of ativeHCPEs and MCPEs with US resoures in Frame i, respetively, i.e. Ui,H ⊆ Ak,H andUi,M ⊆ Ak,M .Let Bk,H|H0 and Bk,H|H1 denote the subset of ative HCPEs whih have deidedhannel is busy (H1) during CSSP k, given that the hannel is idle and busy, re-spetively (i.e. Bk,H|H0,Bk,H|H1 ⊆ Ak,H). We assume that the hannel state remainsonstant during CSSP k and therefore we use the index k instead of i. Subsets ofative HCPEs with US resoures whih have deided H1, given that the hannel isidle and busy, areGi,H|H0 , {Ui,H ∩ Bk,H|H0} (5.13)andGi,H|H1 , {Ui,H ∩ Bk,H|H1}, (5.14)respetively. Similarly, let Bk,M |H0 and Bk,M |H1 denote the subset of ative MCPEswhih have deided H1 during CSSP k, given that the hannel is idle and busy,respetively (i.e. Bk,M |H0,Bk,M |H1 ⊆ Ak,M). Let the subsets of ative MCPEs withUS resoures whih have deided H1, given that the hannel is idle and busy, be89Chapter 5. Trust-based Centralized SEE-CSSdenoted byGi,M |H0 , {Ui,M ∩ Bk,M |H0} (5.15)andGi,M |H1 , {Ui,M ∩ Bk,M |H1}, (5.16)respetively. The subsets of HCPEs without US resoures whih an suessfullytransmit their UCS notiations to the BS in Frame i, given that the hannel is idleand busy, be denoted bySi,H|H0 , {PS i,H ∩ Bk,H|H0} (5.17)andSi,H|H1 , {PS i,H ∩ Bk,H|H1}, (5.18)respetively, where PS i,H ⊆ U ci,H and it is the set of HCPEs whih an transmit UCSnotiations suessfully with a probability psucc,i in Frame i, if IU signal is deteted.Similarly, the subsets of MCPEs without US resoures whih an suessfully transmittheir UCS notiations to the BS in Frame i, given that the hannel is idle and busy,are denoted bySi,M |H0 , {PS i,M ∩ Bk,M |H0} (5.19)andSi,M |H1 , {PS i,M ∩ Bk,M |H1}, (5.20)respetively, where PS i,M ⊆ U ci,M and is the set of MCPEs whih an transmit UCSnotiations suessfully with a probability psucc,i in Frame i, if IU signal is deteted.Let Ii|H0 and Ii|H1 denote the subset of CPEs whih have suessfully transmittedUCS notiations to the BS in Frame i given that the hannel is idle and busy,90Chapter 5. Trust-based Centralized SEE-CSSrespetively, i.e.Ii|H0 = {Si,H|H0 ∪ Si,M |H0 ∪ Gi,H|H0 ∪ Gi,M |H0} (5.21)andIi|H1 = {Si,H|H1 ∪ Si,M |H1 ∪ Gi,H|H1 ∪ Gi,M |H1}. (5.22)In the T-CSS and SEE-CSS protools, the numbers of UCS notiations sent fromCPEs to the BS in Frame i are equal to |Ii|H0| and |Ii|H1| for the idle and busyhannel state hypotheses, respetively. If a ollision ours, the CPEs with UCSnotiations are eligible to transmit in Frame i+ 1 and their bak-o window is setfor an opportunity in Frame i+ 1.In SEE-CSS, let the subsets of HCPEs (MCPEs) whih disagree with the IDBS,kin Frame i+ 1, given that the hannel is idle and busy, be denoted by Di+1,H|H0 andDi+1,H|H1 (Di+1,M |H0 and Di+1,M |H1), whereDi+1,H|H0 ,Di+1,H|H1 ⊆ PSci,H (5.23)andDi+1,M |H0,Di+1,M |H1 ⊆ PSci,M . (5.24)In T-CSS, the subsets of HCPEs (MCPEs) whih an suessfully transmit theirUCS notiations to the BS in Frame i+ 1 for idle and busy hannel are denoted by91Chapter 5. Trust-based Centralized SEE-CSSSi+1,H|H0 and Si+1,H|H1 (Si+1,M |H0 and Si+1,M |H1), respetively, whereSi+1,H|H0 ,Si+1,H|H1 ⊆ PSci,H (5.25)andSi+1,M |H0,Si+1,M |H1 ⊆ PSci,M . (5.26)In the T-CSS and SEE-CSS protools, the set of CPEs from {PSci,H ∪ PSci,M}transmit the UCS notiations to the BS in Frame i+1. The IEEE 802.22 standardmandates that if the UCS-ontention period in Frame i is insuient for the numberof UCS notiations, the BS should alloate longer UCS notiation ontention-based periods in the following frames to failitate the transmission of more UCSnotiations. In addition, the BS an request the subset of CPEs from the sets Ii|H0and Ii|H1 to send their bulk measurement (BLM) reports, denoted by BLM-REPin [103℄, to the BS. We derive expressions for the average number of UCS reports inAppendix G. The minimum size of a BLM report is 32 bits and based on the type ofthe report requested by the BS, its size an inrease up to several tens of bytes [103℄.The number of UCS notiations as well as the number of BLM reports from CPEsinrease with the number of CPEs in the network. Next we study the energy andbandwidth overhead osts in SEE-CSS and T-CSS. The average ardinality of thesets disussed here (e.g. |Ak,H|, |Gi,H|H0|, |Gi,H|H1|, et.) are derived in Appendix G.The results from Appendix G are used in future setions.92Chapter 5. Trust-based Centralized SEE-CSS5.3.3 Spetrum and Energy Overheads in SEE-CSS andT-CSSSpetrum and energy eienies of CSS protools have been studied in [44, 112℄.Studies have shown that when more honest CPEs exhange sensing information withthe BS, more aurate deision on the state of the spetrum bands an be obtaineddue to spae diversity gain. However, as the number of sensing reports (i.e. UCSnotiations in WRAN) inreases, CPE data throughput in WRAN dereases as moreresoures are alloated to the CSS protool.Spetrum OverheadWe obtain the number of bits exhanged in the T-CSS and SEE-CSS protools andwe dene a throughput eieny ratio to ompare the spetrum onsumption ineah protool. The number of bits exhanged during Frame i of CSSP k betweenthe BS and the CPEs is the same for both the T-CSS and SEE-CSS protools,i.e. bucs|Si|H0 | + bdata∗ |Gi|H0| and bucs|Si|H1 | + bdata∗ |Gi|H1|, for idle and busy hannelhypotheses, where bdata∗ is the number of bits in a MAC data paket with UCSnotiation ag bit set to 1 for transmission in US, bucs is the number of bits in a GMHin UCS notiation during the UCS ontention-based period, Si|H0 , Si,H|H0∪Si,M |H0,Si|H1 , Si,H|H1 ∪ Si,M |H1, Gi|H0 , Gi,H|H0 ∪ Gi,M |H0, and Gi|H1 , Gi,H|H1 ∪ Gi,M |H1. Forsimpliity, we have assumed that all CPEs have equal length BLM reports as well asequal user data pakets.In the T-CSS protool, the number of bits transmitted from CPEs to the BSduring Frame i+ 1 when an IU is deteted, given that the hannel is idle and busy,93Chapter 5. Trust-based Centralized SEE-CSSdenoted by bT,i+1|H0 and bT,i+1|H1, an be obtained frombT,i+1|H0 = bucs|Si+1|H0|+ bblm|Ii|H0 | (5.27)andbT,i+1|H1 = bucs|Si+1|H1|+ bblm|Ii|H1 | (5.28)respetively, where Si+1|H0 , {Si+1,H|H0 ∪ Si+1,M |H0} and Si+1|H1 , {Si+1,H|H1 ∪Si+1,M |H1}. The number of bits in the BLM report paket is denoted by bblm. Inthe SEE-CSS protool, the number of bits transmitted from the CPEs to the BS inFrame i+ 1, given that the hannel is idle and busy, are denoted by bSEE,i+1|H0 andbSEE,i+1|H1, an be obtained bybSEE,i+1|H0 = bucs|Di+1|H0|+ bblm|Ii|H0| (5.29)andbSEE,i+1|H1 = bucs|Di+1|H1|+ bblm|Ii|H1 |, (5.30)respetively, where Di+1|H0 , {Di+1,H|H0 ∪ Di+1,M |H0} and Di+1|H1 , {Di+1,H|H1 ∪Di+1,M |H1}. Note that the number of BLM reports requested by the BS may besmaller than |Ii|H0 | (|Ii|H1 |) for idle (busy) hannel states, i.e. the BS may requestonly a few trusted CPEs to send more detailed sensing reports. The throughput ofthe T-CSS and SEE-CSS protools in Frame i+ 1, denoted by CT,i+1 and CSEE,i+1,94Chapter 5. Trust-based Centralized SEE-CSSan be obtained from:CT,i+1 = R[Tf − Tc −1R (bT,i+1)](5.31)andCSEE,i+1 = R[Tf − Tc −1R (bSEE,i+1)], (5.32)where R is the net data rate between the relay and the BS, Tf is the frame period.Note thatbT,i+1 ,(pH0bT,i+1|H0 + pH1bT,i+1|H1)(5.33)andbSEE,i+1 ,(pH0bSEE,i+1|H0 + pH1bSEE,i+1|H1), (5.34)where pH0 is the probability that in-band hannels are idle, and pH0 is the probabilitythat the in-band hannels are busy. Also note Tc , Tds+Tusr+Tscw+Tttg+Trtg+2Tbuff ,where Tds, Tusr, Tscw, Tttg , Trtg, and Tbuff are the DS period, US request period, self-oexistene window (SCW), transmit-to-reeive transition gap, reeive-to-transmittransition gap, buer before/after the SCW, respetively. We dene the throughputeieny ratio by adapting the energy eieny ratio denition in [58℄ asηi+1 =CT,i+1 − CSEE,i+1CT,i+1, (5.35)where the eieny inreases as the number of overhead bits in SEE-CSS protool95Chapter 5. Trust-based Centralized SEE-CSSwith respet to that in the T-CSS protool dereases.Energy ConsumptionThe energy onsumption for CSS at a node depends on the energy onsumed in trans-mitting, reeiving, and proessing pakets. The energy onsumed for transmission isa dominant fator in the total energy onsumption in WRAN links where the trans-mission distane is large (greater than 100m) [91,92℄. The average total energy, ET,k,onsumed in T-CSS during Frames i and i+ 1 an be obtained fromET,k = (pH0 |Si|H0|+ pH1 |Si|H1|)Eucs+ (pH0 |Si+1|H0 |+ pH1 |Si+1|H1 |)Eucs+ (pH0 |Ii|H0|+ pH1 |Ii|H1|)Eblm, (5.36)where Eucs and Eblm denote the average energy used for the transmission of the UCSnotiations and BLM reports at eah CPE, respetively. For simpliity, we haveassumed that the average energy onsumed to transmit a paket at eah CPE is thesame, i.e. Eucs,j = Eucs and Eblm,j = Eblm for j = 1, 2, . . . , H + M . Similarly, theaverage total energy, ESEE,k, onsumed in SEE-CSS isESEE,k = (pH0 |Si|H0|+ pH1 |Si|H1|)Eucs+ (pH0 |Di+1|H0|+ pH1 |Di+1|H1|)Eucs+ (pH0 |Ii|H0|+ pH1 |Ii|H1|)Eblm. (5.37)We omit the energy onsumed at the BS during the broadast of IDBS,k in (5.36) and(5.37). This is due to the fat that the BS is required to transmit FCH to all CPEsinforming them about transmission/reeption parameters of the US and DS during96Chapter 5. Trust-based Centralized SEE-CSSthe DS in eah frame. Therefore, the energy onsumption of an extra bit (IDBS,k)in the FCH is negligible ompared to the total energy onsumed to broadast FCH.Note that energy onsumption in T-CSS and SEE-CSS are dierent only in Framei+ 1 in (5.36) and (5.37). The energy eieny is dened as [58℄:ηE,k =ET,k − ESEE,kET,k. (5.38)5.4 Attak Strategies and a Mitigating MethodDierent attak strategies are proposed to degrade or disrupt the CSS in the CRNs[35,113116℄, e.g. denial of servie (DoS), Jamming, PU emulator, and sensing spe-trum data falsiation (SSDF). Mitigating methods against suh attak strategiesare also studied in [35, 113116℄. In this setion we fous on the SSDF attaks, i.e.the MCPEs aim to manipulate the nal deision of the BS by falsifying their sens-ing reports. We study independent and oordinated SSDF attaks for SEE-CSS. Weassume an MCPE has all the apabilities of an HCPE, inluding sensing in-bandhannels and reporting to the BS. In addition, an MCPE an manipulate its reportto benet unfairly from or to disrupt the operation of WRAN. We investigate theimpat of an independent attak strategy, the entralized ollusion attak strategy,on Qf and Qmd, and we propose a CCF method to mitigate against the proposedollusion attak.5.4.1 Independent AttakA ommon independent attak poliy is studied in [8184℄, where eah MCPE ma-nipulates its own sensing deision independently from other CPEs with probabilitiesshown in Table 5.1.97Chapter 5. Trust-based Centralized SEE-CSSDj,k MCPE Reports MCPE ReportsDishonestly with Honestly withprobability: probability:0 pα0 1− pα01 pα1 1− pα1Table 5.1: Independent Attak Poliy IIIf the jth MCPE deides H0 (H1) during CSSP k, it manipulates its loal deisionwith probability pα0 (pα1), i.e. if Dj,k = 0 (Dj,k = 1), the MCPE transmits Xj,k = 1(Xj,k = −1) to the BS with pα0 (pα1). This attak strategy is a simplied versionof the one studied in Chapter 2 and does not onsider the IDBS,k. By adaptingthe attak probabilities (pα0 , pα1) in Table 5.1, MCPEs hoose to emphasize eithergaining unfair advantage to aess the available spetrum or ausing interferene tothe PUs. For example, if the objetive of the attak strategy is to gain unfair aess,eah MCPE attempts to manipulate the deision of the BS so that it deides that thehannel is busy, when the hannel is atually idle, i.e. MCPEs an use large valuesfor Pα0 .5.4.2 Centralized Collusion Attak in IEEE 802.22 WRANWe propose a entralized trust-based ollusion attak strategy in whih the misbe-having BS (MBS) ollets information from the network, as well as via the MCPEs,and mandates the MCPEs on how to report to the BS in order to manipulate thedeision of the BS more eetively. The MBS aims to nd the minimum numberof dishonest reports whih is expeted to manipulate the deision of the BS in aneort to redue the negative impat of dishonest reporting on the trust values of theMCPEs evaluated at the BS. The maliious entities take advantage from the fat thatthe ontents of the UCS notiations during the UCS ontention-based period an be98Chapter 5. Trust-based Centralized SEE-CSSobtained by MCPEs. In other words, MBS monitors the CPEs whih transmit duringthe UCS notiation during the ontention-based period and estimates FA and MDprobabilities of CPEs in the network. This information from the observation enablesthe MBS to attak the BS more eetively.After the quiet period and before the UCS ontention-based period in Frame i,the MBS ollets sensing reports from MCPEs. The MBS uses the reports to makean intermediate deision in CSSP k, denoted by IDMBS,k, using a TW deision, i.e.IDMBS,k =0, if T̂MBS,kYMBS,k +∑j∈Ak,m T̂j,kYj,k < 01, otherwise,(5.39)where Yj,k ∈ {−1, 1} denotes sensing reports of jth MCPE reeived at the MBSduring CSSP k and YMBS,k ∈ {−1, 1} denotes the loal deision of the MBS in CSSPk. The average trust values of the MBS and the jth MCPE during CSSP k, denotedby T̂MBS,k and T̂j,k, are obtained using (5.9). However, the MBS uses IDMBS,k asa referene to measure the auray of the CPEs and to estimate their trust valuesbeause BSI is not available to MBS.If IDMBS,k = 0, the MBS applies the following integer linear programming opti-mization (ILPO) problem,min.∑j∈Ak,MZj,ks.t.∑j∈Ak,MT̂j,kZj,k +∑j∈Ak,HT̂j,kE{Xj,k|H0}+ T̂MBS,kYMBS,k ≥ 0T̂j,k ≥ τth,MBS,Zj,k ∈ {−1, 1} for j ∈ Ak,M , (5.40)99Chapter 5. Trust-based Centralized SEE-CSSwhere Zj,k is the solution of the ILPO and is a sensing report from the jth MCPEto the BS. The expeted value of the jth HCPE deision given that the hannelis idle, denoted by E{Xj,k|H0}, an be obtained by observing and traking theUCS notiation during the UCS ontention-base period in previous CSSPs, i.e.E{Xj,k|H0} = (−1)(1− p̂f,j) + (1)(p̂f,j), where j ∈ Ak,H and p̂f,j is the estimated FAprobability of jth CPE based on observing the UCS notiations of HCPEs duringthe past CSSPs. The estimated trust value of the jth CPE during the CSSP is de-noted by T̂j,k, i.e. 1 − pH0 p̂f,j − pH1 p̂md,j . The rst onstraint ensures that MCPEsreport dishonestly, only when the minimum number of dishonest report is expetedto manipulate the deision of the BS. Note that the MBS inludes its report in theTW deision in the rst onstraint in (5.40) to ompensate for the impat of the BSontribution in the nal deision. Note that the solution to the ILPO is subjet tosatisfying a threshold trust value, denoted by τth,MBS. The MFC aims to maintainthe trust values of MSUs evaluated at the BS as high as τth using this onstraint.If there is a solution for the ILPO in (5.40), the MBS ditates MCPEs to transmitsensing reports aording to the solution of the ILPO, i.e. Xj,k = Zj,k. If there isno solution to the ILPO in (5.40), the jth MCPE reports Xj,k = 2(IDMBS,k)− 1 forj ∈ Ak,M to the BS instead of its own deision. This will improve the trust valueof the MCPEs at the BS, beause the FA and MD probabilities of the IDMBS,k are100Chapter 5. Trust-based Centralized SEE-CSSlower than those of the individual MCPEs. Similarly, for IDMBS,k = 1, we havemax.∑j∈Ak,MZj,ks.t.∑j∈Ak,MT̂j,kZj,k +∑j∈Ak,HT̂j,kE{Xj,k|H1}+ T̂MBS,kYMBS,k < 0T̂j,k ≥ τth,MBS,Zj,k ∈ {−1, 1} for j ∈ Ak,M , (5.41)where E{Xj,k|H1} is the expeted value of reports from HCPEs to the BS given thatthe hannel is busy, based on the observing the UCS notiation during the UCSontention-base period, i.e. E{Xj,k|H1} = (−1)(1− p̂d,j) + (1)(p̂d,j).5.4.3 Cross-orrelation Filter MethodWe use the CCF proposed in Chapter 4 to detet abnormalities in the behaviors ofCPEs and to eliminate the MCPEs reports from impating the nal deision at theBS in (5.8). If an MCPE stops misbehaving, it an improve its trust values as kinreases and redeem itself from being ategorized as MCPE. The ross-orrelationof the reports for 2 CPEs in CSSP k an be obtained from (4.5). In addition, the BSobtains the expeted value and the variane of the ross-orrelation of HCPEs from(4.6) and (4.7), respetively.The ross-orrelation value whih diers largely from an expeted trust value oftwo HCPE indiates abnormalities in the behavior of at least one of the CPE. Forexample, higher than mean ross orrelation of CPE 1 and CPE 2 orresponds totwo CPEs reporting dishonestly in a oordinated manner (e.g. ollusion) or non-oordinated (e.g. both CPEs suer from shadowing).101Chapter 5. Trust-based Centralized SEE-CSSThe BS ategorizes pairs of CPEs with ross-orrelation values within a ondeneinterval, ψth,k, of the expeted ross-orrelation value in CSSP k as HCPEs and allowsthem to partiipate in (5.8). The ondene interval and the upper bound error inChebyshev's inequality an be obtained from (4.10) and (4.11).As the number of observation inreases, the variane of the observed ross-orrelationvalues for HCPEs dereases. And the CCF method is more suessful in detetingabnormal behavior and eliminating the misbehaving CPEs. The Pseudoode for theCCF method is similar to that presented in Fig. 4.3 in Chapter 4.5.5 Numerial ResultsIn this setion, the transient impat of these two attak strategies on Qf and Qd inWRAN with TW and OR [11℄ deisions as well as the impat of the proposed CCFmethod on Qf and Qd are studied using omputer simulations. The steady-stateaverage number of UCS and UCS-D notiations, energy eieny ratio, throughputeieny ratio, pCPE∗ as a funtion of various parameters (i.e. pus,i, psucc,i, Ns, et.)are numerially evaluated via simulations and analysis. For the results in this setion,the energy threshold is omputed by assuming that the loal FA probability for allsensing entities is equal to 0.1, (pf,j = pf,BS = pf,MBS = pf). The simulation resultspresented are averaged over 10,000 yles. In Figs. 5.3 and 5.4, eah yle begins withtrust values initialized to 1 and ends at k = 1000, where steady-state is reahed. ForFigs. 5.5-5.9, the trust values at the steady-state are used and the simulation resultsare averaged over 10,000 yles. The simulation parameter values are summarized inTable 5.2.102Chapter 5. Trust-based Centralized SEE-CSSParameters Values (unit) Commentspus,i, psucc,i 0.3 pus,i, psucc,i = 0.1 in Figs. 5.5 and 5.6pact,k 1pf , pf,BS, pf,MBS 0.1pH0 , pH1 0.5pα0 , pα1 0.5Tusr, TSCW 1/5 ∗ Tf (ms)Trtg, Tttg 0.13 (ms) Propagation delay [103℄Tbuff 0.1 (ms)Tds 2/5 ∗ Tf (ms)Tf 10 (ms) [103℄τth,BS, τth,MBS 0.7bbc, bucs 32 (bit) [103℄bblm 64 (bit) [103℄Ebc, Eucs 1 (unit energy)Eblm 2 (unit energy)BW 6 (MHz) [103℄R 10.21 (Mbps) Raw data rate [103, 117℄Ns 600 Ns = 103 in Figs. 5.3 and 5.4µj,dB -105 (dBm) Average reeived power [105℄µBS,dB, µMBS,dB -103 (dBm)Pn0,dB -95.2 (dBm) Average noise power [105℄Table 5.2: Simulation Parameter Values IVFig. 5.3 shows the impat of the ollusion attak (denoted by CA on the legend)with and without the ross-orrelation lter (denoted by CCF on the legend) and theindependent attaks (denoted by IA on the legend) on Qf as a funtion of k withH = 5 and M = 10 for TW and OR deision methods at the BS. The omparisonwith OR is provided beause in ertain regulatory domains suh as the U.S.A., ORRule based CSS is required [103℄. The global FA probabilities in T-CSS and SEE-CSSin eah ase (i.e. 'no attak', CA, IA, CA with CCF) are the same; hene, we presentonly one urve for eah ase. First we observe the impat of the attak strategiesand the CCF method on the OR deision. With no attaks (all nodes are HCPEs,i.e. H = 15 and M = 0), we expet to see Qf at approximately 0.81 omputed from103Chapter 5. Trust-based Centralized SEE-CSS0 100 200 300 400 500 600 700 800 900 100000.10.20.30.40.50.60.70.80.91CSSP ( k )Global FA Probability (Qf )  TW Rule − No AttackOR Rule − No AttackTW Rule − Collusion AttackOR Rule − Collusion AttackTW Rule − Collusion Attack w/ CCFOR Rule − Collusion Attack w/ CCFTW Rule − Independent AttackOR Rule − Independent AttackOR RuleTW RuleFigure 5.3: The global false alarm probability in a WRAN as a funtion of k withollusion attak (with and without CCF) and independent attak using OR and TWdeisions at the BS with M = 10, H = 5, Ns = 1000∑H+1i=1(H+1i)pif(1− pf)H+1−i, where the fator 1 in H +1 orresponds to pf,BS. Withindependent attaks, the probability that at least one CPE reports dishonestly withprobabilities given in Table 5.2 inreases and so does Qf . The impat of ollusionattak on Qf is similar to that of the independent attak in steady-state (k > 800).Note that the ollusion attak was optimized in ILPO to impat a TW deision atthe BS and it was not optimized for the OR deision. However, beause OR deisionis highly vulnerable to FAs, Qf approahes 1 at steady-state. The derease in Qfwith ollusion attak for k < 180 is aused beause p̂f,j , p̂md,j , and T̂i,j values of allHCPEs estimated by the MBS are inaurate for low k values. For this reason, theMBS either does not attak (i.e. MCPEs transmit IDMBS,k to the BS) or it attemptsto to ause miss detetions and therefore, FA probability improves. Soon after, p̂f,j,p̂md,j , and T̂i,j beome more aurate, the MBS realizes it is more likely to impatglobal FA than the global MD due to larger loal FA probabilities. The BS an104Chapter 5. Trust-based Centralized SEE-CSS0 100 200 300 400 500 600 700 800 900 1000100CSSP ( k )Global Detection Probability (Qd )  TW Rule − No AttackOR Rule − No AttackTW Rule − Collusion AttackOR Rule − Collusion AttackTW Rule − Collusion Attack w/ CCFOR Rule − Collusion Attack w/ CCFTW Rule − Independent AttackOR Rule − Independent AttackFigure 5.4: The global detetion probability in a WRAN as a funtion of k withollusion attak (with and without CCF) and independent attak using OR and TWdeisions at the BS with M = 10 and H = 5detet the MCPEs olluding with CCF method and eliminate them from attakingin steady-state. For OR deision, Qf in ollusion attak with CCF improves to∑H+1i=1(H+1i)pif(1 − pf )H+1−i for H=5 in steady-state, i.e. approximately 0.47. Notethat Qf based on OR deision is worse than that in TW deision in eah ase. ForTW deision, we observe that Qf in the steady-state degrades 10 folds (i.e. from 0.01to 0.1) from independent to ollusion attak. However, when CCF in onjuntionwith TW deision is used at the BS, the impat of the ollusion attak dereases to0.01 beause the BS an detet abnormal behaviors from the olluding MCPEs. Notethat TW with CCF (H = 5) annot ahieve Qf as low as that in the ase with noattak (H = 15) beause more HCPEs are ollaborating in the ase with no attak.Fig. 5.4 shows the impat of the ollusion attak with and without the ross-orrelation lter and the independent attaks on Qd as a funtion of k with H = 5and M = 10. For larity, we have shown the Y-Axis in logarithmi sale. The results105Chapter 5. Trust-based Centralized SEE-CSSshow that Qd with OR deision is approximately 1 for all 4 ases (i.e. 'No Attak',CA, IA, and CA with CCF) beause it deides that the hannel is busy if at leastone CPE deides that the hannel is busy. We also observe the TW deision withollusion attak with and without CCF auses Qd to degrade from 1 to 0.87 at lowk values and to improve to approximately 1 for k > 300. The ollusion attak poliylearns that it an manipulate the deision of the BS when the hannel is idle withless penalties than when the hannel is busy as k inreases. In other words, it learnsthat lower number of misbehaving CPEs are required to manipulate the deision ofthe BS when the hannel is idle as opposed to when the hannel is busy. This is dueto the fat that the loal FA probabilities are higher than MD probabilities for eahCPEs (i.e. pf,j >> pmd,j for j = 1, 2, . . . , H +M) and therefore, smaller number ofmisbehaving CPEs are required to send dishonest reports when the hannel is idlethan when it is busy. More investigations on the values of Qf and Qd with ollusionattak indiates that they are highly dependent to the value of λj. For example, whenλj is low (i.e. pf,j << pmd,j for j = 1, 2, . . . , H + M) the ollusion attak severelyimpats Qd. We observe that Qd at low k values for TW deision with CCF is higherthan that without the CCF. However the improvement is marginal sine the ollusionattak strategy refrains from attaking when the hannel is busy.Fig. 5.5 shows the steady-state average normalized number of UCS and UCS-Dnotiations (i.e. |Si+1,H |/H and |Di+1,H |/H) in the T-CSS and SEE-CSS protoolstransmitted in Frame i + 1 by HCPEs and MCPEs (normalized by H and M , re-spetively) as a funtion of Ns for various values of H and M . In steady-state, theolluding MCPE an be eliminated and therefore, we assume that MCPE an only at-tak independently. The average normalized number of UCS notiations for HCPEs(denoted by |Si+1,H |/H) in T-CSS inreases and reahes an asymptoti value as Ns106Chapter 5. Trust-based Centralized SEE-CSS400 600 800 1000 1200 1400 1600 1800 2000 2200 24000.20.250.30.350.40.450.5NsNormalized Number of UCS & UCS−D Notifications  |Si+1,H | /H  (H=25,250)|Si+1,M | /M, Di+1,M /M (M=25,250)|Di+1,H | /H (H=250)|Di+1,H | /H (H=25)Figure 5.5: The normalized number of UCS and UCS-D notiations in the T-CSSand SEE-CSS protools, respetively, by HCPEs and MCPEs in Frame i + 1 as afuntion of Ns for H = 25, 250 (M = 25, 250)inreases from 500 to 2400. As Ns inreases, the detetion probability inreases toapproximately 1 and the average normalized number of UCS notiations inreasesand onverges to an asymptoti value whih is obtained from (G.13) and (G.14) aswell as from the simulations. Note that the average normalized values of |Si+1,H |/Hfor H = 25, 250 are the same. Also note that for (5.4) to be valid, the samples in thequiet period must be independent and hene, Ns an not be hosen as an arbitrarylarge number. The average normalized number of UCS-D notiations for HCPEs inFrame i + 1 in SEE-CSS (denoted by |Di+1,H |/H) dereases and reahes an asymp-toti values as Ns inreases. This is due to the fat that the number of disagreementswith the IDBS,k dereases and reahes an asymptoti value as the loal detetionprobabilities improve to approximately 1 as Ns inreases. For Ns < 620, H = 250,and M = 250, the average number of UCS notiation reports in T-CSS is less thanthat of UCS-D in SEE-CSS (i.e. |Di+1,H|/H > |Si+1,H |/H), beause pf,j and on-107Chapter 5. Trust-based Centralized SEE-CSS400 600 800 1000 1200 1400 1600 1800 2000 2200 2400−0.0500.050.10.150.20.250.3Throughput Efficiency Ratio  −0.0500.050.10.150.20.25Energy Efficiency RatioNs ηC,i+1 ( H=25, M=25)ηC,i+1  ( H=250, M=250)ηE,k  ( H=25, M=25)ηE,k ( H=250, M=250)Figure 5.6: Left Y axis: the throughput eieny ratio as a funtion of Ns. Right Yaxis: the energy eieny ratio as a funtion of Nssequently pf,ID (whih results from at least one CPE reporting UCS) are large, theprobability of disagreement with IDFC,k is also large. As pf,j dereases, so does pf,IDand the number of disagreements with IDFC,k. The average number of UCS andUCS-D notiations for MCPEs is dependent on Pα0 and Pα1 as well as the loal FAand MD probabilities of eah MCPEs. The number of UCS notiations in T-CSSand EE-CSS are the same beause the attak strategy of MCPEs is independent ofIDBS,k. Note that |Di+1,H|/H for H = 250 is higher than that for H = 25. This anbe explained as follows. pf,ID for H = 250 is higher than that for H = 25, therefore,the number of disagreements with IDFC,k is also higher for H = 250.Fig. 5.6 shows the throughput eieny ratio (left Y axis) and the energy e-ieny ratio (right Y axis) as a funtion of number of samples (Ns) for various H andM values. It is shown that both the throughput eieny and the energy eienyratios improve as Ns inreases beause less number of CPEs disagree with the IDBS,kdeision as Ns inreases (i.e. due to lower loal FA and MD probabilities). The energy108Chapter 5. Trust-based Centralized SEE-CSS0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 100.10.20.30.40.50.60.70.80.91pus,i  and psuc,ip CPE* and  pBS  pCPE* (H=250)pCPE* (H=25)pBS (H=250)pBS (H=25)Figure 5.7: The probabilities pCPE∗ and pBS (pBS , (1 − pCPE∗)) as a funtion ofpus,i and psucc,i for H = 25, 250 (M = 25, 250) and Ns = 2000improves up to approximately 23% and the throughput improves up to approximately25% in SEE-CSS protool ompared to T-CSS protool for large Ns (e.g. 2400). Weobserve that the energy and the throughput eienies for large number of CPEs(H +M = 500) are less than that for small number of CPEs (H +M = 50) beausethe average numbers of UCS and UCS-D notiations are more for (H +M = 500)as shown in Fig.5.5. The energy and spetrum eieny ratios have negative valuesfor Ns < 600 and H = 250 (M = 250). This is due to the fat that the number UCSand UCS-D notiations during Frame i+ 1 in T-CSS is less than that of SEE-CSSas disussed for Fig. 5.5.Fig. 5.7 shows the probability that at least one UCS notiation from a trustedCPE (i.e. Tj∗ > τth,BS) is reeived in Frame i at the BS (denoted by pCPE∗) and pBS(, 1−pCPE∗) as a funtion of pus,i and psucc,i for two values of H (H = 25, 250). ThepCPE∗ and pBS have impats on the FA and MD probabilities of the intermediatedeision at the BS and their study is important in the desription of the next gure.109Chapter 5. Trust-based Centralized SEE-CSS0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 100.10.20.30.40.50.60.7pus,i and  psucc,iNormalized Number of UCS & UCS−D Notifications  |Gi,H | /H, H=25,250|Si+1,H | /H, H=25,250|Di+1,H | /H, H=250|Di+1,H | /H, H=25Figure 5.8: The normalized number of UCS and UCS-D notiation reports in T-CSSand SEE-CSS by HCPEs in Frames i and i + 1 as a funtion of pus,i and psucc,i forH = 25, 250 (M = 25, 250) and Ns = 2000As pus,i and psucc,i inrease, more CPEs beome available for CSS partiipation inFrame i and more likely that a trusted UCS is transmitted from a CPE to the BSand therefore, pCPE∗ inreases and pBS dereases. When H is large (H = 250), pCPE∗an reah the asymptoti value at low pus,i (psucc,i) values as shown in Fig. 5.7, e.g.pCPE∗ = 1 for pus,i = psucc,i > 0.1.Fig. 5.8 shows the steady-state average normalized number of UCS and UCS-Dnotiation reports in T-CSS and SEE-CSS by HCPEs in Frames i and i + 1 as afuntion of pus,i and psucc,i for H = 25, 250 (M = 25, 250). As pus,i and psucc,i inrease,the average normalized number of UCS notiations transmitted in Frame i for T-CSSand EE-CSS (|Gi,H |/H) inreases beause the number of CPEs whih an suessfullytransmit UCS in Frame i inreases. In T-CSS, the average normalized number of UCSnotiations in Frame i+ 1 (|Si+1,H |/H) from CPEs to the BS dereases as pus,i andpsucc,i inreases, i.e. less number of CPEs are available to transmit in Frame i + 1110Chapter 5. Trust-based Centralized SEE-CSSas psucc,i and pus,i inrease. Note that |Si+1,H |/H is the same for H = 25 and 250beause in T-CSS, eah HCPE transmits a UCS notiation report with the sameprobability. In EE-CSS, the average normalized number of UCS-D notiation reportsin Frame i+ 1 (|Di+1,H|/H) from CPEs to the BS rst inreases and then dereasesas pus,i and psucc,i inreases. At pus,i = psucc,i = 0, pCPE∗ = 0 (from Fig. 5.7) andIDBS,k is solely based one the loal deision of the BS (DFC,k). As pus,i and psucc,iinrease, the number of UCS notiations whih are transmitted to the BS inreasesand pf,ID inreases. As a result, more CPEs disagree with IDBS,k in Frame i + 1when the hannel is idle and |Di+1,H |/H inreases. Note that the rate of inreasefor H = 250 is muh larger than that for H = 25. This is due to the rapid inreasein pf,ID for H = 250. As pus,i and psucc,i inrease, the number of available CPEs inFrame i + 1 dereases and hene the number of |Di+1,H|/H begin to derease. Atpus,i = psucc,i = 1, |Si+1,H |/H and |Di+1,H |/H are zero beause all CPEs an sendUCS in US resoures and in the ontention base period of Frame i. The results showthat the average number of UCS and UCS-D in SEE-CSS is less than that in T-CSS.Fig. 5.9 shows the throughput eieny ratio (left Y axis) and the energy e-ieny ratio (right Y axis) as a funtion pus,i and psucc,i forH = 25, 250 (M = 25, 250).The throughput and energy eieny ratios derease as pus,i and psucc,i inrease. Atpus,i = psucc,i = 0, we expet to see the maximum throughput and energy eieniesas all CPEs only transmit UCS and UCS-D notiation reports in Frame i + 1. Atpus,i = psucc,i = 1, the throughput and energy eieny ratios are zero as |Si+1,H |/Hand |Di+1,H |/H are both equal to zero. Note that the throughout eieny forH = 25 is higher and lower than that for H = 250 for pus,i = psucc,i ≤ 0.5 andpus,i = psucc,i > 0.5, respetively. Similarly, the energy eieny for H = 25 is higher111Chapter 5. Trust-based Centralized SEE-CSS0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 100.050.10.150.20.250.30.350.40.45Throughput Efficiency Ratio  00.050.10.150.20.250.30.350.40.45Energy  Efficiency Ratiopus,i and psucc,i ηE,k  (H=250, M=250)ηE,k  (H=25, M=25)ηC,i+1  (H=250, M=250)ηC,i+1  (H=25, M=25)Figure 5.9: Left Y axis: the throughput eieny ratio as a funtion of pus,i and psucc,ifor H = 25, 250 (M = 25, 250) and Ns = 2000. Right Y axis: the energy eieny(right Y-axis) ratio as a funtion of pus,i and psucc,i for H = 25, 250 (M = 25, 250)and Ns = 2000and lower than that for H = 250 for pus,i = psucc,i ≤ 0.6 and pus,i = psucc,i > 0.6,respetively. Extensive numerial results show that bT,i+1 − bS,i+1 for H = 250 de-reases at a faster rate than that in H = 25 for pus,i = psucc,i ≤ 0.5 and bT,i+1 − bS,i+1for H = 25 dereases at a faster rate than that in H = 250 for pus,i = psucc,i ≤ 0.5.5.6 SummaryA trust-based spetrum and energy eient CSS protool, namely SEE-CSS, fora WRAN was proposed. Expressions for the average number of UCS notiationstransmitted from CPEs to the BS were derived for SEE-CSS and the T-CSS, whereeah CPE transmits a UCS notiation, if the CPE detets IU signals. Spetrumand energy eieny models for the T-CSS and SEE-CSS protools were dened andompared. The numerial results show that signiant improvements in the spetrum112Chapter 5. Trust-based Centralized SEE-CSSand energy onsumption an be obtained when using SEE-CSS ompared to T-CSS.Furthermore, a entralized trust-based ollusion attak strategy was proposed basedon the fat that the ontents of the UCS notiations in WRAN is impliitly known.The attak strategy aims to manipulate the deision of the BS and to impat theglobal false alarm and miss detetion probabilities more severely by oordinatingattaks from MCPEs. A ross-orrelation lter method was proposed to detet theCPEs with abnormal behaviors. The numerial results show that the CCF methodenables the BS to detet and to eliminate olluding MCPEs so as to redue the impaton the deision of the BS.113Chapter 6Conlusions and Topis for FutureResearhIn this hapter, we summarize our nding and ontributions in Setion 6.1. Sometopis for future related researh are outlined in Setion 6.2.6.1 ConlusionsIn this thesis, we foused on the design and analysis of a CSS protool whih ismore spetrum and energy eient in the presene of MSUs than T-CSS protools,in whih eah sensor node transmits a report to the FC. The main results from eahhapter are reviewed.ˆ In Chapter 2, we proposed a two-phase trust-based energy eient ollaborativespetrum sensing (EE-CSS) protool. Closed-form expressions for the averagesteady-state trust values of the HSUs and MSUs as well as those for the totalsteady-state average number of reports transmitted from the HSUs and MSUsto the FC were derived. Energy onsumption models for the EE-CSS and T-CSSprotools were dened based on the energy onsumed to transmit, reeive, andproess sensing reports. The results from the energy onsumption of EE-CSSand T-CSS show that when the number of reports required to satisfy the targetglobal FA and MD probabilities is large (i.e. SNR of the PU signal reeived at114Chapter 6. Conlusions and Topis for Future ResearhSUs is low and therefore, large number of SUs are required to partiipate in theCSS), EE-CSS is more energy eient than T-CSS. In addition, in traditionalommuniation systems where the energy onsumed in transmitting pakets ismuh larger than those onsumed in reeiving and proessing pakets, EE-CSSis more energy eient than T-CSS. We derived expressions for steady-stateQf and Qmd for the EE-CSS and T-CSS protools. The numerial resultsonrmed that EE-CSS is more energy eient than T-CSS for the same Qfand Qmd values. Furthermore, we studied the impat of link outages on thesteady-state average trust values and the total steady-state average number ofreports transmitted from the HSUs and MSUs to the FC in the EE-CSS andT-CSS protool. The results show that the impat of link outages on Qf andQmd in EE-CSS is less that than in T-CSS.ˆ In Chapter 3, we analyzed the transient behavior of trust values and numberof sensing reports from HSUs and MSUs to the FC in EE-CSS to gain insightinto when the EE-CSS protool reahes the steady-state. In partiular, wederived expressions for the transient probability distribution and the averagesof the trust values for HSUs and MSUs in EE-CSS. We also derived losed-formexpressions for the transient average number of sensing reports transmitted bythe HSUs and MSUs to the FC.ˆ In Chapter 4, we proposed a trust-based entralized ollusion attak strategy forthe EE-CSS protool. In the proposed attak strategy, the MFC and the MSUsobserve the disagreement reports transmitted from the HSUs to the FC and usethem to estimate the loal FA and MD probabilities as well as the trust valuesof HSUs. The MFC uses the information obtained from observing HSUs as wellas the loal deisions of MSUs in an integer linear programming optimization115Chapter 6. Conlusions and Topis for Future Researhproblem to nd the minimum number of dishonest reports likely to hange thedeision of the FC. The MFC aims to inrease Qf and Qmd while trying toredue the negative impat on the trusts value of the MSUs as evaluated at theFC. The MFC requests the MSUs to transmit reports aording to the solutionof the ILPO. The results show that the proposed ollusion attak impat Qfand Qmd more severely than a ommonly used independent attak model. Inaddition, we proposed a CCF method to nd SUs with abnormal behaviors. Theresults show that CCF an eetively detet misbehaving SUs in the CRN.ˆ In Chapter 5, we proposed a trust-based spetrum and energy eient CSSprotool (SEE-CSS) in a WRAN. We derived expressions for the average num-ber of UCS notiations transmitted from CPEs to the BS were derived forSEE-CSS and T-CSS (i.e. in T-CSS, eah CPE transmits a UCS notiation,if the CPE detets IU signals). We dened spetrum and energy eienymodels for T-CSS and SEE-CSS. The numerial results show that signiantimprovements in the spetrum and energy onsumption an be obtained whenusing SEE-CSS ompared to T-CSS. Furthermore, we proposed a entralizedtrust-based ollusion attak strategy (similar to that in Chapter 4) for SEE-CSS based on the fat that the ontents of the UCS notiations in WRAN isimpliitly known. The attak strategy aims to hange the deision of the BSand to impat Qf and Qmd more severely by oordinating attaks from MCPEs.The CCF method is used to detet the CPEs with abnormal behaviors. Thenumerial results show that the CCF method enables the BS to detet and toeliminate olluding MCPEs so as to redue the impat on the deision of theBS.116Chapter 6. Conlusions and Topis for Future ResearhWe an make the following observations about our proposed protools and theattak strategies:ˆ The EE-CSS and SEE-CSS protools proposed in this thesis an replae the T-CSS protools where the sensing reports, in the form of binary hypotheses (i.e.busy and idle), are olleted for hard deision ombining at the FC (or BS).The proposed protools broadast an intermediate deision whih a sensing nodean disagree with (by transmitting an expliit reports to the FC) or agree with(by remaining silent after the broadast). In CSS protools with soft deisionombining, where the sensing entities report the deteted energy about the stateof the band, the proposed are not diretly appliable.ˆ The results from Chapters 2-5 as well as the results in [14,24,47,49,81,84,97,98℄show that trust and reputation management systems an signiantly reduethe impat of node misbehaviors in CSS in CRNs. In addition, we observethat the ollusion attaks suh as the one studied in Chapters 4 and 5 as wellas the ones in [47, 83, 99℄ an degrade or disrupt the CSS more severely thanindependent attak poliies suh as the one studied in Chapter 2 as well as theones in [8185℄. It is important to note that the majority of mitigation methodsagainst ollusion attaks use a form of ross-orrelation of reports from the SUsto detet abnormal behaviors and to eliminate the misbehaving SUs from theCSS proess. Therefore, it is natural to assume that misbehaving SUs an useross-orrelation of reports in ollusion attak strategies to remain undetetedby the FC.117Chapter 6. Conlusions and Topis for Future Researh6.2 Future WorkWe now outline some topis for possible future researh:1. EE-CSS and SEE-CSS in multi-band CSS: Multi-band detetion shemesfor wide-band liensed spetrum sensing senarios are proposed in [118120℄.In [119,120℄, a parallel fusion arhiteture is proposed in whih SUs are grouped,with members of eah group ollaborating on the state of one spetrum band;eah SU sends its hard loal deision to its orresponding aess point, whihthen applies the fusion deision to make a nal deision about the state ofthe spetrum band. The EE-CSS and SEE-CSS an be utilized in multi-banddetetion shemes in eah parallel CSS proess to save energy and improvespetrum usage. For example, optimization methods proposed in [56, 57℄ anbe used to nd the minimum number of SUs required to ensure Qf < ǫf andQmd < ǫmd for eah spetrum band. In addition, the two-phased EE-CSS orSEE-CSS an be used to make a deision about the state of eah band and toredue the number of expliit sensing reports exhanged between the SUs andthe FC.2. EE-CSS and SEE-CSS in entralized luster-based CSS: The energy andspetrum onsumption are studied for luster-based CSS protools in [6365℄,where eah luster head (whih is loser to the FC than other luster member)ollets hard (or soft) loal deisions from its own luster member (CM) andapplies a hard (or soft) deision ombining shemes to the sensing reports. Eahluster head (CH) then transmits its nal deision to the FC, where anotherhard (or soft) deision ombining sheme is used to make a nal deision aboutthe state of the band. The luster-based CSS is more energy eient than118Chapter 6. Conlusions and Topis for Future Researhother CSS methods beause the total transmission energy is redued aord-ing to the distane-power transmission relations (i.e. square-law of distaneand power transmission [65, 77℄, path-loss exponent law of distane and powertransmission Chapter 5, et.). We propose using the two-phase EE-CSS andSEE-CSS between the CMs and CHs as well as between the CMs and the FCjointly with hard deision ombining at the FC. While the transmission energyan be improved in luster-based CSS, the total number of reports transmittedbetween the CMs and CHs as well as between CHs and the FC an also beredued using EE-CSS or SEE-CSS.3. EE-CSS and SEE-CSS in distributed CRNs: The SUs in ad ho CRNsollaborate with one another in a distributed manner [39℄. Eah SU rst mea-sures the energy of the spetrum band and then either sends its loal deisionor its measured energy to the neighboring SUs. The overhead spetrum andenergy osts assoiated with distributed CSS an be very large. Methods suhas lustering have been proposed to redue the ost assoiated with sensingreports during the CSS. It would be interesting to onsider using EE-CSS orSEE-CSS in distributed CSS jointly with hard deision ombining at the FC.4. Evaluation of Trust: In Appendix C, we explored the impat when thePUBS does not send BSI to the FC. We showed that the FC may use its ownloal deision to evaluate trust for SUs. We propose using the nal deision ofthe FC in onjuntion with the loal deision of the FC for trust evaluation.The trust-weighted nal deision in (2.12) is shown to outperform individualdeisions or the traditional deision ombining shemes (the OR, AND, andKi rules) in the presene of MSUs. The reason for using the nal deisionjointly with the loal deision of the FC is to prevent MSUs from manipulating119Chapter 6. Conlusions and Topis for Future Researhthe FC in a oordinated attak in the initial time slots. If MSUs outnumberHSUs, then in initial time slots when the trust values of all SUs are the same,the trust-weighted nal deision ould be manipulated by MSUs. As a result,rewards and penalties will be given to MSUs and HSUs, respetively, and thetrust values of MSUs (HSUs) will improve (degrade). One approah may be togive more weight to the loal deision of the FC than to the nal deision ofFC in the initial time slots.120Bibliography[1℄ J. Mitola and G. Maguire, Cognitive radio: making software radios more per-sonal, IEEE Pers. Commun., vol. 6, no. 4, pp. 1318, Aug. 1999.[2℄ S. Haykin, Cognitive radio: brain-empowered wireless ommuniations, IEEEJ. Sel. Areas Commun., vol. 23, no. 2, pp. 201  220, Feb. 2005.[3℄ G. Staple and K. Werbah, The end of spetrum sarity, IEEE Spetrum,vol. 41, no. 3, pp. 48  52, Mar. 2004.[4℄ G. Ganesan and Y. Li, Agility improvement through ooperative diversity inognitive radio, in Pro. IEEE GLOBECOMM, 2005, pp. 25052509.[5℄ , Cooperative spetrum sensing in ognitive radio networks, in Pro.IEEE DySPAN, 2005, pp. 137143.[6℄ D. Cabri, A. Tkahenko, and R. Brodersen, Spetrum sensing measurementsof pilot, energy, and ollaborative detetion, in Pro. IEEE MILCOM, 2006,pp. 17.[7℄ T. Yuek and H. Arslan, A survey of spetrum sensing algorithms for ognitiveradio appliations, IEEE Commun. Surveys Tuts., vol. 11, no. 1, pp. 116 130,First Quarter 2009.[8℄ W. Wang, Y. Sun, H. Li, and Z. Han, Cross-layer attak and defense in ogni-tive radio networks, in Pro. IEEE GLOBECOM, 2010, pp. 1 6.[9℄ O. Leon, J. Hernandez-Serrano, and M. Soriano, A new ross-layer attak toTCP in ognitive radio networks, in Pro. IEEE IWCLD, 2009, pp. 1 5.[10℄ A. Toledo and X. Wang, Robust detetion of selsh misbehavior in wirelessnetworks, IEEE J. Sel. Areas Commun., vol. 25, no. 6, pp. 1124 1134, Aug.2007.[11℄ H. Li and Z. Han, Cath me if you an: An abnormality detetion approahfor ollaborative spetrum sensing in ognitive radio networks, IEEE Trans.Wireless Commun., vol. 9, no. 11, pp. 3554 3565, Nov. 2010.[12℄ R. Chen, J.-M. Park, and K. Bian, Robust distributed spetrum sensing inognitive radio networks, in Pro. IEEE INFOCOM, 2008, pp. 1876 1884.121Bibliography[13℄ E. Noon and H. Li, Defending against hit-and-run attakers in ollaborativespetrum sensing of ognitive radio networks: A point system, in Pro. IEEEVTC Spring, 2010, pp. 1 5.[14℄ H. Yu, Z. Shen, C. Miao, C. Leung, and D. Niyato, A survey of trust and rep-utation management systems in wireless ommuniations, Pro. of the IEEE,vol. 98, no. 10, pp. 1755 1772, Ot. 2010.[15℄ T. Yuek and H. Arslan, MMSE noise plus interferene power estimation inadaptive OFDM systems, IEEE Trans. Veh. Tehnol., vol. 56, no. 6, pp. 38573863, Nov. 2007.[16℄ S. Geirhofer, L. Tong, and B. Sadler, Dynami spetrum aess in the timedomain: Modeling and exploiting white spae, IEEE Commun. Mag., vol. 45,no. 5, pp. 6672, May 2007.[17℄ C. Cordeiro, K. Challapali, , and D. Birru, IEEE 802.22: An introdution tothe rst wireless standard based on ognitive radios, JCM J. Commun., vol. 1,no. 1, pp. 3847, Apr. 2006.[18℄ T. Yuek and H. Arslan, Spetrum haraterization for opportunisti ognitiveradio systems, in Pro. IEEE MILCOM, 2006, pp. 16.[19℄ J. G. Proakis, Digital Communiations. New York, NY: MGraw-Hill, 2001.[20℄ N. Shankar, C. Cordeiro, and K. Challapali, Spetrum agile radios: utilizationand sensing arhitetures, in Pro. IEEE DySPAN, 2005, pp. 160169.[21℄ H. Sadeghi and P. Azmi, Cylostationarity-based ooperative spetrum sensingfor ognitive radio networks, in Pro. IEEE IST, 2008, pp. 429434.[22℄ A. Ghasemi and E. Sousa, Collaborative spetrum sensing for opportunistiaess in fading environments, in Pro. IEEE DySPAN, 2005, pp. 131136.[23℄ , Opportunisti spetrum aess in fading hannels through ollaborativesensing, J. of Commun., vol. 2, no. 2, pp. 71  82, Mar. 2007.[24℄ H. Chen, X. Jin, and L. Xie, Reputation-based ollaborative spetrum sensingalgorithm in ognitive radio networks, in Pro. IEEE PIMRC, 2009, pp. 582587.[25℄ H. Chen, H. Wu, X. Zhou, and C. Gao, Agent-based trust model in wirelesssensor networks, in Pro. ACIS SNPD, 2007, pp. 119 124.[26℄ L. Hong, J. Ma, F. Xu, S. Li, and Z. Zhou, Optimization of ollaborativespetrum sensing for ognitive radio, in Pro. IEEE ICNSC, 2008, pp. 17301733.122Bibliography[27℄ S. Mishra, A. Sahai, and R. Brodersen, Cooperative sensing among ognitiveradios, in Pro. IEEE ICC, 2006, pp. 16581663.[28℄ Z. Quan, S. Cui, and A. Sayed, Optimal linear ooperation for spetrum sens-ing in ognitive radio networks, IEEE J. Sel. Topis Signal Proess., vol. 2,no. 1, pp. 28 40, Feb. 2008.[29℄ W. Wang, H. Li, Y. Sun, and Z. Han, Cathit: Detet maliious nodes inollaborative spetrum sensing, in Pro. IEEE GLOBECOM, 2009, pp. 1 6.[30℄ E. Visotsky, S. Kuner, and R. Peterson, On ollaborative detetion of TVtransmissions in support of dynami spetrum sharing, in Pro. IEEE DyS-PAN, 2005, pp. 338345.[31℄ W. Wang, H. Li, Y. Sun, and Z. Han, Attak-proof ollaborative spetrumsensing in ognitive radio networks, in Pro. IEEE CISS, 2009, pp. 130 134.[32℄ J. Hillenbrand, T. Weiss, and F. Jondral, Calulation of detetion and falsealarm probabilities in spetrum pooling systems, IEEE Commun. Lett., vol. 9,no. 4, pp. 349  351, Apr. 2005.[33℄ R. Tandra and A. Sahai, Fundamental limits on detetion in low SNR undernoise unertainty, in Pro. IEEE IWCMC, 2005, pp. 464469.[34℄ Z. Gao, H. Zhu, S. Li, S. Du, and X. Li, Seurity and privay of ollaborativespetrum sensing in ognitive radio networks, IEEE Trans. Wireless Commun.,vol. 19, no. 6, pp. 106112, De. 2012.[35℄ T. Jiang, H. Zeng, Q. Yan, W. Lou, and Y. Hou, On the limitation of embed-ding ryptographi signature for primary transmitter authentiation, IEEEWirelese Commun. Lett., vol. 1, no. 4, pp. 324327, Aug. 2012.[36℄ A. Alahmadi, M. Abdelhakim, J. Ren, and T. Li, Defense against primaryuser emulation attaks in ognitive radio networks using advaned enryptionstandard, IEEE Trans. Inf. Forensis Seurity, vol. 9, no. 5, pp. 772781, May2014.[37℄ R. Chen and J.-M. Park, Ensuring trustworthy spetrum sensing in ognitiveradio networks, in Pro. IEEE SDR, 2006, pp. 110119.[38℄ X. He, H. Dai, and P. Ning, HMM-based maliious user detetion for robustollaborative spetrum sensing, IEEE J. Sel. Areas Commun., vol. 31, no. 11,pp. 21962208, Nov. 2013.[39℄ H. Tang, F. Yu, M. Huang, and Z. Li, Distributed onsensus-based seuritymehanisms in ognitive radio mobile ad ho networks, IET Communiations,vol. 6, no. 8, pp. 974983, May 2012.123Bibliography[40℄ T. Qin, H. Yu, C. Leung, Z. Shen, and C. Miao, Towards a trust aware ognitiveradio arhiteture, ACM SIGMOBILE, vol. 13, pp. 8695, Sep. 2009.[41℄ S. Parvin, S. Han, L. Gao, F. Hussain, and E. Chang, Towards trust establish-ment for spetrum seletion in ognitive radio networks, in Pro. IEEE AINA,2010, pp. 579 583.[42℄ S. Parvin, S. Han, B. Tian, and F. Hussain, Trust-based authentiation forseure ommuniation in ognitive radio networks, in Pro. IEEE/IFIP EUC,2010, pp. 589 596.[43℄ M. Akbari and A. Falahati, SSDF protetion in ooperative spetrum sensingemploying a omputational trust evaluation algorithm, in Pro. IEEE IST,2010, pp. 23 28.[44℄ S. Althunibat, V. Suasas, H. Marques, J. Rodriguez, R. Tafazolli, andF. Granelli, On the trade-o between seurity and energy eieny in o-operative spetrum sensing for ognitive radio, IEEE Commun. Lett., vol. 17,no. 8, pp. 15641567, Aug. 2013.[45℄ W. Wang, L. Chen, K. Shin, and L. Duan, Seure ooperative spetrum sensingand aess against intelligent maliious behaviors, in Pro. IEEE INFOCOM,2014, pp. 12671275.[46℄ H. Li and Z. Han, Collaborative spetrum sensing with a stranger: Trust, ornot to trust? in Pro. IEEE WCNC, 2010, pp. 1 6.[47℄ C.-K. Yu, M. Laghate, A. Sayed, and D. Cabri, On the eets of olludedstatistial attaks in ooperative spetrum sensing, in Pro. IEEE SPAWC,2013, pp. 275279.[48℄ W. Wang and Q. Zhang, Privay-preserving ollaborative spetrum sensingwith multipleservie providers, IEEE Trans. Wireless Commun., vol. PP,no. 99, pp. 11, 2014.[49℄ Q. Yan, M. Li, T. Jiang, W. Lou, and Y. Hou, Vulnerability and protetionfor distributed onsensus-based spetrum sensing in ognitive radio networks,in Pro. IEEE INFOCOM, 2012, pp. 900908.[50℄ J. Esh, Prolog to a survey of trust and reputation management systems inwireless ommuniations, Pro. IEEE, vol. 98, no. 10, pp. 17521754, Ot.2010.[51℄ S. Ganeriwal, L. K. Balzano, and M. B. Srivastava, Reputation-based frame-work for high integrity sensor networks, Pro. ACM SASN, vol. 4, no. 3, pp.137, Jun. 2008.124Bibliography[52℄ Y. Wang and M. P. Singh, Formal trust model for multiagent systems, inPro. ACM IJCAI, 2007, pp. 15511556.[53℄ S. Buhegger and J.-Y. Le Boude, The eet of rumor spreading in reputationsystems for mobile ad-ho networks, in Pro. WiOpt, 2003.[54℄ J. Zhu, Z. Xu, F. Wang, B. Huang, and B. Zhang, Double threshold energydetetion of ooperative spetrum sensing in ognitive radio, in Pro. EAICrownCom, 2008, pp. 1 5.[55℄ L. Duan, L. Zhang, Y. Chu, and S. Liu, Cooperative spetrum sensing withdouble threshold detetion based on reputation in ognitive radio, in Pro.IEEE WiCom, 2009, pp. 1 4.[56℄ Y. Chen, Optimum number of seondary users in ollaborative spetrum sens-ing onsidering resoures usage eieny, IEEE Commun. Lett., vol. 12, no. 12,pp. 877 879, De. 2008.[57℄ W. Zhang, R. Mallik, and K. Letaief, Optimization of ooperative spetrumsensing with energy detetion in ognitive radio networks, IEEE Trans. Wire-less Commun., vol. 8, no. 12, pp. 5761 5766, De. 2009.[58℄ S. Althunibat, R. Palaios, and F. Granelli, Energy-eient spetrum sensingin ognitive radio networks by oordinated redution of the sensing users, inPro. IEEE ICC, 2012, pp. 13991404.[59℄ H. Su and X. Zhang, Energy-eient spetrum sensing for ognitive radionetworks, in Pro. IEEE ICC, 2010, pp. 15.[60℄ Y. Wu and D. Tsang, Energy-eient spetrum sensing and transmission forognitive radio system, IEEE Commun. Lett., vol. 15, no. 5, pp. 545547, May2011.[61℄ C. An, H. Ji, and Y. Li, Energy-eient ollaborative sheme for ompressedsensing-based spetrum detetion in ognitive radio networks, in Pro. IEEEWCNC, 2012, pp. 13601364.[62℄ D. Godarzi, K. Arshad, Y. Ko, and K. Moessner, Seleting users in energy-eient ollaborative spetrum sensing, in Pro. IEEE WCNC, 2012, pp.10291033.[63℄ C. Sun, W. Zhang, and K. Ben, Cluster-based ooperative spetrum sensingin ognitive radio systems, in Pro. IEEE ICC, Jun. 2007, pp. 25112515.[64℄ C. Sun and K. Letaief, User ooperation in heterogeneous ognitive radio net-works with interferene redution, in Pro. IEEE ICC, May 2008, pp. 31933197.125Bibliography[65℄ J. Wei and X. Zhang, Energy-eient distributed spetrum sensing for wirelessognitive radio networks, in Pro. IEEE INFOCOM, 2010, pp. 16.[66℄ S. Wang, Y. Wang, J. Coon, and A. Doufexi, Energy-eient spetrum sensingand aess for ognitive radio networks, IEEE Trans. Veh. Tehnol., vol. 61,no. 2, pp. 906912, Feb. 2012.[67℄ A. T. Hoang, Y.-C. Liang, D. Wong, Y. Zeng, and R. Zhang, Opportunistispetrum aess for energy-onstrained ognitive radios, IEEE Trans WirelessCommun., vol. 8, no. 3, pp. 12061211, Mar. 2009.[68℄ Y. Chen, Q. Zhao, and A. Swami, Distributed spetrum sensing and aess inognitive radio networks with energy onstraint, IEEE Trans. Signal Proess.,vol. 57, no. 2, pp. 783797, Feb. 2009.[69℄ S. Maleki, A. Pandharipande, and G. Leus, Energy-eient distributed spe-trum sensing for ognitive sensor networks, IEEE Sensors J., vol. 11, no. 3,pp. 565573, Marh 2011.[70℄ S. Appadwedula, V. Veeravalli, and D. Jones, Deentralized detetion withensoring sensors, IEEE Trans. Signal Proess., vol. 56, no. 4, pp. 13621373,Apr. 2008.[71℄ , Energy-eient detetion in sensor networks, IEEE J. Sel. Areas Com-mun., vol. 23, no. 4, pp. 693702, Apr. 2005.[72℄ C. Sun, W. Zhang, and K. Letaief, Cooperative spetrum sensing for ognitiveradios under bandwidth onstraints, in Pro. IEEE WCNC, 2007, pp. 15.[73℄ H. Yang, Z. Zhao, and H. Zhang, Hard ombining based energy eient spe-trum sensing in ognitive radio network, in Pro. IEEE GLOBECOM, 2013,pp. 10381043.[74℄ W.-Y. Lee and I. Akyildiz, Optimal spetrum sensing framework for ognitiveradio networks, IEEE Trans. on Wireless Commun., vol. 7, no. 10, pp. 38453857, Ot. 2008.[75℄ L. Lu et al., Tehnology proposal lariations for ieee 802.22 wran systems,in Pro. IEEE P802.22 Wireless RANs, Mar. 2006.[76℄ T. Clany and N. Goergen, Seurity in ognitive radio networks: Threats andmitigation, in Pro. EAI CrownCom, 2008, pp. 1 8.[77℄ P. Tipler and G. Mosa, Physis for Sentists and Engineers: Magnetism, Light,and Elementary Modern Physis. W. H. Freeman & Co, 2004.126Bibliography[78℄ H. Urkowitz, Energy detetion of unknown deterministi signals, Pro. IEEE,vol. 55, no. 4, pp. 523  531, Apr. 1967.[79℄ F. F. Digham, M.-S. Alouini, and M. K. Simon, On the energy detetion ofunknown signals over fading hannels, IEEE Trans. Commun., vol. 55, no. 1,pp. 21 24, Jan. 2007.[80℄ N. Katenka, E. Levina, and G. Mihailidis, Loal vote deision fusion for targetdetetion in wireless sensor networks, IEEE Trans. Signal Proess., vol. 56,no. 1, pp. 329338, Jan. 2008.[81℄ C. Hyder, B. Grebur, L. Xiao, and M. Ellison, ARC: Adaptive reputation basedlustering against spetrum sensing data falsiation attaks, IEEE Trans.Mobile Comput., vol. 13, no. 8, pp. 17071719, Aug. 2014.[82℄ Y. Cai, Y. Mo, K. Ota, C. Luo, M. Dong, and L. Yang, Optimal data fusionof ollaborative spetrum sensing under attak in ognitive radio networks,IEEE Netw., vol. 28, no. 1, pp. 1723, Jan. 2014.[83℄ A. Rawat, P. Anand, H. Chen, and P. Varshney, Collaborative spetrum sens-ing in the presene of Byzantine attaks in ognitive radio networks, IEEETrans. Signal Proess., vol. 59, no. 2, pp. 774786, Feb. 2011.[84℄ Y. Cai, L. Cui, K. Pelehrinis, P. Krishnamurthy, M. Weiss, and Y. Mo, De-oupling trust and wireless hannel indued eets on ollaborative sensingattaks, in Pro IEEE DySPAN, 2014, pp. 224235.[85℄ B. Liu, M. Wang, C. Zhang, and C. W. Chen, Robust sensor seletion forollaborative spetrum sensing with attaks under orrelated shadowing, inPro. IEEE ICC, 2014, pp. 279284.[86℄ P. Velloso, R. Laufer, D. de O Cunha, O. Duarte, and G. Pujolle, Trust man-agement in mobile ad ho networks using a salable maturity-based model,IEEE Trans. Network and Servie Management, vol. 7, no. 3, pp. 172185,Sep. 2010.[87℄ S. Li, H. Zhu, Z. Gao, X. Guan, and K. Xing, Yousense: Mitigating entropyselshness in distributed ollaborative spetrum sensing, in Pro. IEEE IN-FOCOM, 2013, pp. 25352543.[88℄ S. Li, H. Zhu, B. Yang, C. Chen, and X. Guan, Believe yourself: A user-entrimisbehavior detetion sheme for seure ollaborative spetrum sensing, inPro. IEEE ICC, 2011, pp. 15.127Bibliography[89℄ G. Ganesan and Y. Li, Cooperative spetrum sensing in ognitive radio, partI: Two user networks, IEEE Trans. Wireless Commun., vol. 6, no. 6, pp. 22042213, Jun. 2007.[90℄ , Cooperative spetrum sensing in ognitive radio, part II: Multiuser net-works, IEEE Trans. Wireless Commun., vol. 6, no. 6, pp. 2214 2222, Jun.2007.[91℄ S. Cui, A. J. Goldsmith, and A. Bahai, Energy-onstrained modulation opti-mization, IEEE Trans. Wireless Commun., vol. 4, no. 5, pp. 23492360, 2005.[92℄ A. El Gamal, C. Nair, B. Prabhakar, E. Uysal-Biyikoglu, and S. Zahedi,Energy-eient sheduling of paket transmissions over wireless networks,in Pro. IEEE INFOCOM, 2002, pp. 17731782.[93℄ A. Wang, S. Cho, C. Sodini, and A. Chandrakasan, Energy eient modulationand MAC for asymmetri rf mirosensor systems, in Pro. IEEE ISLPED,2001, pp. 106111.[94℄ P. K. Varshney, Distributed Detetion and Data Fusion. NY USA: Springer-Verlag, 1997.[95℄ Y. Wang, On the number of suesses in independent trials, Statistia Sinia,vol. 3, pp. 295312, 1993.[96℄ S. Mousavifar and C. Leung, Transient analysis in ognitive radio ollaborativespetrum sensing, in Pro. IEEE WCNC, 2014, pp. 954959.[97℄ K. Guan, L. Gharai, S. Dehnie, R. Ghanadan, and S. Kumar, Trust manage-ment for distributed deision fusion in sensor networks, in Pro. IEEE Fusion,2009, pp. 119124.[98℄ K. Zeng, P. Pawezak, and D. Cˇabric´, Reputation-based ooperative spetrumsensing with trusted nodes assistane, IEEE Commun. Lett., vol. 14, no. 3,pp. 226228, Mar. 2010.[99℄ M. Wang, B. Liu, and C. Zhang, Detetion of ollaborative SSDF attaksusing abnormality detetion algorithm in ognitive radio networks, in ProIEEE ICC, 2013, pp. 342346.[100℄ H. Li and Z. Han, Cathing attaker(s) for ollaborative spetrum sensing inognitive radio systems: An abnormality detetion approah, in Pro. DyS-PAN, 2010, pp. 112.[101℄ D. Makris, G. Gardikis, and A. Kourtis, Quantifying TV white spae apa-ity: A geoloation-based approah, IEEE Communiations Magazine, vol. 50,no. 9, p. 145, 2012.128Bibliography[102℄ T. Baykas, M. Kasslin, M. Cummings, H. Kang, J. Kwak, R. Paine, A. Reznik,R. Saeed, and S. Shellhammer, Developing a standard for TV white spae o-existene: tehnial hallenges and solution approahes, IEEE Wireless Com-mun., vol. 19, no. 1, pp. 1022, Feb. 2012.[103℄ IEEE Standard 802.22a-2011, IEEE Standard for Information TehnologyTeleommuniations and information exhange between systems Wireless Re-gional Area Networks (WRAN)Spei requirements - Part 22: CognitiveWireless RAN Medium Aess Control (MAC) and Physial Layer (PHY) Spe-iations: Poliies and Proedures for Operation in the TV Bands Amendment1: Management and Control Plane Interfaes and Proedures and Enhanementto the Management Information Base (MIB), 2014.[104℄ IEEE Standard 802.22.2-2012, IEEE Reommended Pratie for InformationTehnology - Teleommuniations and information exhange between systemsWireless Regional Area Networks (WRAN) - Spei requirements - Part 22.2:Installation and Deployment of IEEE 802.22 Systems, 2012.[105℄ S. Mathur, R. Tandra, S. Shellhammer, and M. Ghosh, Initial signal proessingof aptured DTV signals for evaluation of detetion algorithms, IEEE 802.22-06/0158r4, Sep. 2006.[106℄ N. Kundargi and A. Tewk, Sequential pilot sensing of ATSC signals in IEEE802.22 ognitive radio networks, in Pro. IEEE ICASSP, 2008, pp. 27892792.[107℄ X. Wang, H.-W. Chen, Y. Wu, J.-Y. Chouinard, and C.-L. Wang, Identiationof PCP-OFDM signals at very low SNR for spetrum eient ommuniations,in Pro. IEEE VTC Spring, 2009, pp. 15.[108℄ T. Zhang and D. Tsang, Cooperative sensing sheduling for energy-eientognitive radio networks, IEEE Trans. Veh. Teh., vol. PP, no. 99, pp. 11,2014.[109℄ S. Mousavifar and C. Leung, Energy eient ollaborative spetrum sensingbased on trust management in ognitive radio networks, IEEE Trans. on Wire-less Commun., vol. PP, no. 99, pp. 113, 2014.[110℄ A. Goldsmith, Wireless ommun. Cambridge university press, 2005.[111℄ I.-R. P.1546-1, Method for point-to-area predition for terresterial servies inthe frequeny range 30 MHz to 3000 MHz. Aademi Press, 6th Edition, 2005.[112℄ G. Deepak and K. Navaie, On the sensing time and ahievable throughput insensor-enabled ognitive radio networks, in Pro. IEEE ISWCS, 2013, pp. 15.129Bibliography[113℄ Z. Luo, C. Lou, S. Chen, S. Zheng, and S. Li, Spei primary user sensingfor wireless seurity in IEEE 802.22network, in Pro. IEEE ISCIT, 2011, pp.1822.[114℄ R. Chen, J.-M. Park, and J. Reed, Defense against primary user emulationattaks in ognitive radio networks, IEEE J. Sel. Areas Commun., vol. 26,no. 1, pp. 25 37, Jan. 2008.[115℄ S. Prasad and D. Thuente, Jamming attaks in 802.11g -A ognitive radiobased approah, in Pro. IEEE MILCOM, Nov., pp. 12191224.[116℄ K. Bian and J.-M. J. Park, Seurity vulnerabilities in IEEE 802.22, in Pro.EAI WICON, 2008, pp. 19.[117℄ V. Gardellin, S. Das, and L. Lenzini, Self-oexistene in ellular ognitive radionetworks based on the ieee 802.22 standard, IEEE Wireless Commun., vol. 20,no. 2, pp. 5259, Apr. 2013.[118℄ Z. Quan, S. Cui, A. Sayed, and H. Poor, Optimal multiband joint detetionfor spetrum sensing in ognitive radio networks, IEEE Trans. Signal Proess.,vol. 57, no. 3, pp. 11281140, Mar. 2009.[119℄ P. Kaligineedi and V. Bhargava, Sensor alloation and quantization shemes formulti-band ognitive radio ooperative sensing system, IEEE Trans. WirelessCommun., vol. 10, no. 1, pp. 284293, Jan. 2011.[120℄ X. Sun and D. Tsang, Energy-eient ooperative sensing sheduling formulti-band ognitive radio networks, IEEE Trans. Wireless Commun., vol. 12,no. 10, pp. 49434955, Ot. 2013.[121℄ IEEE P802.22aD1, IEEE Draft Standard for Information TehnologyTeleommuniations and information exhange between systems Wireless Re-gional Area Networks (WRAN)Spei requirements - Part 22: CognitiveWireless RAN Medium Aess Control (MAC) and Physial Layer (PHY) Spe-iations: Poliies and Proedures for Operation in the TV Bands Amendment1: Management and Control Plane Interfaes and Proedures and Enhanementto the Management Information Base (MIB), 2013.130Appendix AList of Other PubliationsOther ontributions not presented in this dissertation:Some researh works ompleted during my Ph.D. program at UBC but not diretlyrelated to this dissertation have been published or are under review as follows.ˆ S. Li, S. A. Mousavifar, C. Leung, Lifetime Distribution of an Amplify-and-Forward Wireless Relay Network, submitted De. 2015.ˆ S. A. Mousavifar, Y. Liu, Y. Deng, C. Leung, M. Elkashlan, Wireless EnergyHarvesting in a Cognitive Relay Network, submitted Jan. 2015.ˆ S. A. Mousavifar, Y. Liu, T. Q. Duong, M. Elkashlan, and C. Leung, WirelessEnergy Harvesting and Spetrum Sharing in Cognitive Radio, in Pro. IEEEVTC Fall 2014, Vanouver, Canada, Sep. 2014, pp. 1-5.ˆ S. A. Mousavifar, and C. Leung, Lifetime Analysis of a Two-Hop Amplify-and-Forward Opportunisti Wireless Relay Network, IEEE Trans. WirelessCommun., vol. 12, no. 3, pp. 1186-1195, Mar. 2013.ˆ S. A. Mousavifar, T. Khattab, and M. Hasna, Sequential Random Sele-tion Relaying for Energy Eient Wireless Sensor Networks, in Pro. IEEEGLOBECOM, Miami, FL, U.S.A., Deember 2010, pp. 1-6.131Appendix BEet of Delay in Communiating theBSI from PUBS to SUBSIn this Appendix, we show the impat of the BSI delay on the average steady-stateand transient trust, number of sensing reports and global FA and MD probabili-ties using omputer simulations. Let's assume that the PUBS delays Nd number oftime slots to send the BSI to the SUBS. Then, we show the impat of the delay onthe aforementioned parameters using omputer simulations with parameter valuessummarized in Table C.1.pα pH1 M H λ TW SNRdB Nd0.4 0.5 10 5 50 20 25 1,10,50Table B.1: Simulation Parameter Values VIFig. B.1 shows the average trust value of HSUs and MSUs as a funtion oftime slot k for Nd = 1, 10 and 50. The average trust values of HSU or MSU forNd = 1, 10, 50 onverge to the steady-state average trust value (i.e. illustrated byat dotted lines for HSUs and MSUs) as k inreases. It is shown that the averagetrust values for Nd = 10 and 150 at the time of update (i.e. k = 10, 20, . . . , 150 andk = 50, 100, 150, respetively) are relatively lose that for Nd = 1. For example, thePUBS sends the BSI from k = 1, 2, . . . , 10 at k = 10 for Nd = 10. The results alsoshow that Nd an impat the evaluation of the trust values of the SUs as a funtionof time slot k.132Appendix B. Eet of Delay in Communiating the BSI from PUBS to SUBS0 50 100 1500.550.60.650.70.750.80.850.90.951Time,  k (Time slots)Average Trust Value   Nd = 50HSUsMSUs Nd = 10 Nd = 1Figure B.1: The average trust value of HSUs and MSUs as a funtion of time slot kfor several Nd values. The dotted at line shows the steady-state average trust valueof HSUs and MSUs.0 50 100 15000.10.20.30.40.50.60.70.80.91Time,  k (Time slots)NHI and  NMI    Nd = 10 Nd = 1HSUsMSUs Nd = 50Figure B.2: The probability of HSU or MSU hosen in Phase I as a funtion of timeslot k for several Nd values133Appendix B. Eet of Delay in Communiating the BSI from PUBS to SUBSFigs. B.2 and B.2 show the probability of HSU or MSU hosen in Phase I and theaverage number of reports transmitted by HSUs (MSUs) in Phase II, respetively, as afuntion of time slot k for Nd = 1, 10 and 50. The probability of HSU or MSU hosenin Phase I (i.e. 1 and 0, respetively) and the average number of reports transmittedby HSUs and MSUs in Phase II reah their steady-state values (i.e. illustrated by atdotted lines for HSUs and MSUs) for Nd = 1, 10 and 50 as k inreases. Note that theprobability of an HSU hosen in Phase I in Nd = 10 (during the delay) is lower thanthat in Nd = 1 for k < 10. This is due to the fat that the trust values are assumeto be 1 and all SUs are equally likely to be hosen. Between BSI updates in Nd = 10and Nd = 50, the trust value of HSUs remains higher than that in Nd = 1 and hene,HSUs are more likely to be hosen in Nd = 10 and Nd = 50 than in Nd = 1. As aresult, MSUs are less likely to be hosen in Phase I in Nd = 10 and Nd = 50 than inNd = 1 as shown in Fig. B.2.0 50 100 15011.522.533.544.55Time,  k (Time slots)NHII and  NMII   HSUsMSUs Nd = 10 Nd = 1  Nd = 50Figure B.3: The average number of reports transmitted in Phase II as a funtion oftime slot k for several Nd values134Appendix B. Eet of Delay in Communiating the BSI from PUBS to SUBS0 50 100 1500.010.020.030.040.050.060.070.080.09Time,  k (Time slots)Q fa and Q md   Qf Nd = 10 Nd = 1  Nd = 50QmdFigure B.4: The global FA and MD probabilities as a funtion of time slot k forseveral Nd valuesFig. B.4 shows the global FA and MD probabilities as a funtion of time slot kfor Nd = 1, 10 and 50. The global FA and MD probabilities onverge to their steady-state values as k inreases. The results show that when Nd is large, it takes longerfor the global Fa and MD probabilities to reah their stead-state values.135Appendix CEvaluating Trust Based on the LoalDeision of the FCIn this Appendix, we show the impat of using the loal deision of the FC in plaeof the BSI to evaluate SU trust values. The advantage of this approah is that thereis no burden plaed on the PUBS to transmit the BSI and the FC an evaluate SUtrust more frequently. The disadvantage of this approah is that the trust values ofHSUs (Th) are degraded more than those of MSUs (Tm). For onveniene, we willdenote the trust value of HSUs and MSUs when the loal deision of the FC is usedfor trust evaluation by T FCh and T FCm , respetively. The orresponding global FA andMD probabilities are denoted by QFCf and QFCmd , respetively.We rst onsider the senario in whih the SNR of the PU signal reeived at theFC is similar to those at SUs (i.e. 25 dB). Next, we also onsider a pratial senarioin whih the FC has a higher SNR value of the PU signal [121℄. For a fair omparison,we assume that T-CSS and EE-CSS both use the same trust evaluation proess, i.e.if available, they both use BSI; otherwise they both use the loal deision of the FCto evaluate trust values.For the senario in whih the trust is evaluated using the loal deision of the FC,the reward is given to eah SU whih agrees with the loal deision of the FC at timeslot k in (2.8), i.e. if Xn,k = −1 and DFC,k = 0 or Xn,k = 1 and DFC,k = 1, then136Appendix C. Evaluating Trust Based on the Loal Deision of the FCrk = 1. The average steady-state trust values of the hth HSUs and the mth MSUsan be evaluated asTh,k = 1− pH0((pf,h(1− pf,FC) + pf,FC(1− pf,h))− pH1((pmd,h(1− pmd,FC) + pmd,FC(1− pmd,h)) (C.1)andTm,k = 1− pH0((1− pα)(pf,h(1− pf,FC) + pf,FC(1− pf,h))+ pα(pf,hpf,FC + pf,FCpf,h))− pH1((1− pα)(pmd,h(1− pmd,FC) + pmd,FC(1− pmd,h))+ pα(pmd,hpmd,FC + pmd,FCpmd,h)). (C.2)For simpliity, we have assumed that pα = pα0,0 = pα1,0 = pα0,1 = pα1,1 in (C.2). Thesteady-state analysis obtained in Chapter 2 for NHI , NHII , NMI , NMII , Qf , and Qmdremains the same. The parameter values are summarized in Table C.1.pα pH1 M H λ TW SNRdB0.4 0.5 10 5 50 20 25Table C.1: Simulation Parameter Values VFig. C.1 shows Th, Tm, T FCh , T FCm , T FC 35dBh , and T FC 35dBm , where T FC 35dBh andT FC 35dBm denote the average steady-state trust values of HSUs and MSUs when theSNR of the PU signal reeived at the FC is 10 dB higher than at SUs (i.e. SNR =35 dB and λ = 55). This assumption is reasonable sine the BS may have largerantenna gain and lower RF loss in the iruitry than the SUs. The results show thatthe trust values of HSUs and MSUs degrade when the trust is evaluated using theloal deision of the FC. There are senarios in whih the HSU has made a orret137Appendix C. Evaluating Trust Based on the Loal Deision of the FCdeision and it would be penalized due to the inorret loal deision at the FC in(C.1), i.e. pH0(1− pf,h)pf,FC and pH1(1− pmd,h)pmd,FC . These two terms do not havean impat on the trust value of HSUs when the trust is evaluated using BSI.0 10 20 30 40 50 60 70 80 90 1000.50.60.70.80.91Time,  k (Time slots)Average Trust Value   Tm FC Th FC Tm Th  Tm FC 35dB Th FC 35dBFigure C.1: The steady-state average trust value of HSUs and MSUs as a funtion oftime slot k evaluated using BSI or the loal deision of the FCFig. C.2 shows the probability of HSU (MSU) hosen in Phase I as a funtionof time slot k, denoted by NHI (NMI ), NFCHI (NFCMI ), NFC 35dBHI (NFC 35dBMI ), for threetrust evaluation senarios: using BSI, the FC loal deision with SNR = 25 dB and35 dB, respetively. The results show that the steady-state probability that an HSUis hosen in Phase I approahes 1 for all trust evaluation senarios. However, forthe senarios in whih the trust is evaluated using the loal deision of the FC, thesteady-state probability approahes 1 at larger k values than the senarios in whihthe trust is evaluated using BSI.Fig. C.3 shows the average number of reports transmitted by HSUs (MSUs) inPhase II as a funtion of time slot k, denoted by NHII (NMII ), NFCHII (NFCMII ), NFC 35dBHII138Appendix C. Evaluating Trust Based on the Loal Deision of the FC0 10 20 30 40 50 60 70 80 90 10000.10.20.30.40.50.60.70.80.91Time,  k (Time slots)NHI and  NMI    NMIFC,  Pα = 0.4 NHIFC,  Pα = 0.4 NMI,  Pα = 0.4 NHI,  Pα = 0.4 NMIFC 35dB,  Pα = 0.4 NHIFC 35 dB,  Pα = 0.4Figure C.2: The probability of HSU or MSU hosen in Phase I as a funtion of timeslot k for trust evaluation using BSI and the FC loal deision.(NFC 35dBMII ), for three trust evaluation senarios: using BSI, the FC loal deision withSNR = 25 dB and 35 dB, respetively. The results show thatNFCHII andNFCMII onvergeto NHII and NMII , respetively. Note that NFC 35dBHII and NFC 35dBMII are smaller thanNHII and NMII , i.e. the number of disagreements with the intermediate deision ofthe FC is the lowest when the loal deision at the FC is based on a PU signal withSNR = 35 dB is used. At high SNR values, the loal FA and MD probability of theFC improve signiantly. As a result the FA and MD probabilities of the intermediatedeision also improve. Consequently, NFC 35dBHII and NFC 35dBMII , whih are the numberof disagreements with the intermediate deisions, derease.Fig. C.4 shows the global FA (global MD) probabilities, denoted by Qf (Qmd),QFCf (QFCmd ), and QFC 35dBf (QFC 35dBmd ) as a funtion of time slot k for three trustevaluation senarios: using BSI, the FC loal deision with SNR = 25 dB and35 dB, respetively. The results show that both QFCf and QFCmd are slightly higher139Appendix C. Evaluating Trust Based on the Loal Deision of the FC0 10 20 30 40 50 60 70 80 90 10011.522.533.544.55Time,  k (Time slots) NHII and  NMII   NMIIFC,  Pα = 0.4 NHIIFC,  Pα = 0.4 NMII,  Pα = 0.4 NHII,  Pα = 0.4 NMIIFC 35dB,  Pα = 0.4 NHIIFC 35dB,  Pα = 0.4Figure C.3: The average number of reports transmitted in Phase II as a funtion oftime slot k for trust evaluation using BSI and the FC loal deision.than Qf and Qmd, respetively. This is due to the fat that the trust values of theHSUs degrades more than those of MSUs in the absene of BSI, and therefore thetrust-weighted deision in (2.12) is more likely to make an inorret deision in thepresene of MSUs. Note that QFC 35dBf < Qf and QFC 35dBmd < Qmd. This an beexplained as follows. When the SNR of the SU signal at the FC is 35 dB, not only isthe deision of the FC more aurate with higher weight in (2.12), but also the trustvalues of HSUs beome more aurate with higher weight. In this setion, we showedthat while QFCf and QFCmd are higher than Qf and Qmd, respetively, QFCf and QFCmdare the same for both T-CSS and EE-CSS and the number of reports transmitted inEE-CSS is less than that in T-CSS.140Appendix C. Evaluating Trust Based on the Loal Deision of the FC0 10 20 30 40 50 60 70 80 90 1000.020.030.040.050.060.070.080.09Time, k (Time slots)Q f  and   Qmd   Qf FC,  Pα = 0.4Qmd FC,  Pα = 0.4Qf ,  Pα = 0.4Qmd,   Pα = 0.4QfaFC 35dB,  Pα = 0.4QmdFC 35dB,  Pα = 0.4Figure C.4: The global FA and MD probabilities (Qf and Qmd, respetively) as afuntion of time slot k for trust evaluation using BSI and the FC loal deision.141Appendix DEvaluating NH0hII , NH1hII , NH0mII , andNH1mIIThe steady-state average number, NH1hII and NH0hII , of reports transmitted by the hthHSUs in Phase II for hypotheses H1 and H0, an be given by:NH1hII = pH1 ((1− pmd,ID)pmd,h + pmd,ID(1− pmd,h)) (D.1)andNH0hII = pH0 ((1− pf,ID)(pf,h) + pf,ID(1− pf,h)) , (D.2)respetively. Similarly, the steady-state average number, NH1mII and NH0mII , of reportstransmitted by an MSU in Phase II for H1 and H0 based on the attak poliy ofTable 2.1, areNH1mII = pH1[(1− pmd,m)((1− pmd,ID)pα1,1 + pmd,ID(1− pα0,1))+ pmd,m((1− pmd,ID)(1− pα1,0) + pmd,IDpα0,0) ](D.3)and142Appendix D. Evaluating NH0hII , NH1hII , NH0mII , and NH1mIINH0mII = pH0[(1− pf,m)((1− pf,ID)pα0,0 + pf,ID(1− pα1,0))+ pf,m((1− pf,ID)(1− pα0,1) + pf,IDpα1,1) ], (D.4)respetively. Assuming NHI = 1 and NMI = 0 are true in steady-state, the termpBr.out,h in (2.16) represents the senario in whih there may be a link outage from theFC to hth HSU and the hth HSU transmit expliit sensing reports to the FC inPhase II. The same senario between the FC and the mth MSU an be explained forpBr.out in (2.17). In (D.1), the rst term, (1− pmd,ID)pmd,h, on the RHS represents theprobability that the intermediate deision is H1 and the deision of HSUh is H0, giventhe PU hannel is busy. The seond term, pmd,ID(1− pmd,h), on the RHS representsthe probability that the intermediate deision is H0 and the deision of HSUh is H1,given the PU hannel is busy. HSUh will transmit a message in eah of these twosenarios. Similarly, (D.2) desribes the ase in whih the PU hannel is idle. In(D.3) and (D.4), eah term represents the probability of a senario in whih the mthMSU disagrees with the FC's intermediate message, e.g. (1− pmd,m)(1− pmd,ID)pα1,1orresponds to the senario in whih both the intermediate and the mth MSU deidethat the hannel is busy but the MSU is dishonest with the FC given that the hannelis busy.143Appendix EEvaluating NHI ,k and NMI ,kIn order to nd the losed form expressions for the terms Pr{Xk > Yk}, Pr{Xk <Yk}, Pr{Xk = Yk,HSU is hosen}, and Pr{Xk = Yk,MSU is hosen}, we rst denethe PMF of Xk and Yk. The pX( i1+k ) an be obtained as FXk( i1+k ) − FXk( i−11+k ) fori = 1, . . . , k + 1, where FXk( i1+k ) = Pr{T1,k ≤ i1+k , T2,k ≤ i1+k , . . . , TH,k ≤ i1+k}denotes the umulative distribution funtion (CDF) of the maximum trust value ofthe HSUs at time slot k. Note thatFXk( i1 + k)=[ i∑j=1pTh,k( j1 + k)]H, (E.1)sine {Th,k, h = 1, 2, . . . , H} are idential and independently distributed (i.i.d.) rv's.Similarly, pYk( i1+k ) an be obtained as FYk( i1+k )− FYk( i−11+k), whereFYk( i1 + k)=[ i∑j=1pTm,k( j1 + k)]M. (E.2)Note that FXk( i1+k ) = 0 and FYk( i1+k) = 0 at i = 0. We an obtain Pr{Xk > Yk} attime k as144Appendix E. Evaluating NHI ,k and NMI ,kPr{Xk > Yk} =k+1∑i=1pYk( i1 + k)Pr{Xk > Yk∣∣Yk =i1 + k}=k+1∑i=1pYk( i1 + k)Pr{Xk >i1 + k}, (E.3)sine Xk and Yk are independent. Similarly, Pr{Xk < Yk} and Pr{Xk = Yk} an beobtained asPr{Xk < Yk} =k+1∑i=1pYk( i1 + k)Pr{Xk <i1 + k}(E.4)andPr{Xk = Yk} =k+1∑i=1pYk( i1 + k)Pr{Xk =i1 + k}. (E.5)The probability that at least one HSU (and no MSU) has the highest trust valuein Phase I at time slot k an be evaluated using (E.3). Similarly, the probability thatat least one MSU (and no HSUs) has the highest trust value in Phase I at time slot kan be evaluated using (E.4). The probability that at least one HSU and at least oneMSUs have the highest trust value in Phase I at time slot k is given by (E.5). In theevent that several SUs have the maximum value, one of them is hosen at randomby the FC to report its sensing deision in Phase I. Note that when Xk = Yk, theprobability that an HSU is hosen may not be equal to the probability that an MSUis hosen by the FC for reporting in Phase I. The joint probability that at time k anHSU is hosen and Xk = Yk an be expressed as145Appendix E. Evaluating NHI ,k and NMI ,kPr{Xk = Yk,HSU is hosen} =k+1∑i=1H∑h=1M∑m=1hh+m(Hh)[pTh,k( i1 + k)]h [ i−1∑j=1pTh,k( j1 + k)]H−h×(Mm)[pTm,k( i1 + k)]m [ i−1∑j=1pTm,k( j1 + k)]M−m . (E.6)The right hand side (RHS) of (E.6) sums the probabilities of all events in whihat least one HSU and at least one MSU share the same highest trust value and anHSU is hosen. Similarly, the joint probability that at time k an MSU is hosen andXk = Yk an be expressed asPr{Xk = Yk,MSU is hosen} =k+1∑i=1H∑h=1M∑m=1mh+m(Hh)[pTh,k( i1 + k)]h [ i−1∑j=1pTh,k( j1 + k)]H−h×(Mm)[pTm,k( i1 + k)]m [ i−1∑j=1pTm,k( j1 + k)]M−m , (E.7)or, simply we an writePr{Xk = Yk,MSU is hosen} = Pr{Xk = Yk} − Pr{Xk = Yk,HSU is hosen}.(E.8)The omputational omplexity of (E.6) and (E.7) is (K+1)HM operations whihis less than that in the method we studied in [96℄ (i.e. exponential omputationomplexity). Note that the average number, NHI ,k, of sensing reports transmitted by146Appendix E. Evaluating NHI ,k and NMI ,kHSUs in Phase I in time slot k is simply the probability that an HSU is hosen, i.e.NHI ,k = Pr{Xk > Yk}+ Pr{Xk = Yk,HSU is hosen} (E.9)whereas the average number, NMI ,k, of sensing reports transmitted by MSUs isNMI ,k = 1−NHI ,k. (E.10)147Appendix FEvaluating NHII ,k and NMII ,kLet E5(Ec5) denote the event that the ID is orret (inorret), E6(Ec6) denote the eventthat an HSU deision is orret (inorret) and E7(Ec7) denote the event that the deisionan MSU sends to the FC in Phase II is orret (inorret).The terms Pr {E3|E1} and Pr {E3|E2} in (3.13) an be obtained asPr {E3|E1} = pH0 [Pr {E5, Ec6|H0, E1}+ Pr {Ec5, E6|H0, E1}]+ pH1 [Pr {E5, Ec6|H1, E1}+ Pr {Ec5, E6|H1, E1}] (F.1)andPr {E3|E2} = pH0 [Pr {E5, Ec6|H0, E2}+ Pr {Ec5, E6|H0, E2}]+ pH1 [Pr {E5, Ec6|H1, E2}+ Pr {Ec5, E6|H1, E2}] . (F.2)Similarly, the terms Pr {E4|E1} and Pr {E4|E2} in (3.14) an be obtained asPr {E4|E1} = pH0 [Pr {E5, Ec7|H0, E1}+ Pr {Ec5, E7|H0, E1}]+ pH1 [Pr {E5, Ec7|H1, E1}+ Pr {Ec5, E7|H1, E1}] (F.3)andPr {E4|E2} = pH0 [Pr {E5, Ec7|H0, E2}+ Pr {Ec5, E7|H0, E2}]+ pH1 [Pr {E5, Ec7|H1, E2}+ Pr {Ec5, E7|H1, E2}] . (F.4)148Appendix F. Evaluating NHII ,k and NMII ,kThe terms on the RHS of (F.1)-(F.4) an be expressed asPr {E5, Eca|Hb, Ec} = Pr {E5|Hb, Ec}Pr {Eca|Hb, Ec}Pr {Ec5, Ea|Hb, Ec} = Pr {Ec5|Hb, Ec}Pr {Ea|Hb, Ec} (F.5)where a ∈ {6, 7}, b ∈ {0, 1} and c ∈ {1, 2}. Finally, the terms on the RHS of (F.5)an be evaluated usingPr {E5|H0, E1} = (1− pf,FC)(1− pf,h)Pr {E6|H0, E1} = Pr {E6|H0, E2} = 1− pf,hPr {E5|H1, E1} = (1− pmd,FCpmd,h)Pr {E6|H1, E1} = Pr {E6|H1, E2} = 1− pmd,hPr {E5|H0, E2} = (1− pf,FC)pf,m,FCPr {E5|H1, E2} = (1− pmd,FC)pmd,m,FCPr {E7|H0, E1} = Pr {E7|H0, E2}= pf,m,FCPr {E7|H1, E1} = Pr {E7|H1, E2}= pmd,m,FC .(F.6)149Appendix GThe Average Cardinality of CPE SetsIn this appendix we derive the expeted value of the sets dened in Setion 5.3.2.The average number of ative CPEs, HCPEs, and MCPEs in CSSP k an be obtainedfrom |Ak| = (N − 1)(pact,k), |Ak,H| = H(pact,k), and |Ak,M | = M(pact,k), respetively.The average number of Ui,H and Ui,M an be obtained from |Ui,H | = |Ai,H |(pus,H,i) and|Ui,M | = |Ai,M |(pus,M,i), respetively. The average number of UCS notiations trans-mitted using the US resoures, given the hannel is idle and busy, an be obtainedfrom:|Gi,H|H0| =∑j∈Ui,Hpf,j (G.1)and|Gi,H|H1| =∑j∈Ui,Hpd,j, (G.2)respetively. Assuming pf,j = pf and pd,j = pd for j ∈ Ai,H,|Gi,H|H0| = H(pact,k)(pus,H,i)pf (G.3)and|Gi,H|H1| = H(pact,k)(pus,H,i)pd. (G.4)150Appendix G. The Average Cardinality of CPE SetsSimilarly for MCPEs we have,|Gi,M |H0| =∑j∈Ui,M(pf,j(1− pα1) + (1− pf,j)pα0)(G.5)and|Gi,M |H1| =∑j∈Ui,M(pd,j(1− pα1) + (1− pd,j)pα0), (G.6)respetively. Assuming pf,j = pf and pd,j = pd for j ∈ Ai,M ,|Gi,M |H0| = M(pact,k)(pus,M,i)(pf(1− pα1) + (1− pf)pα0)(G.7)and|Gi,M |H1| = M(pact,k)(pus,M,i)(pd(1− pα1) + (1− pd)pα0). (G.8)The average number of UCS notiations transmitted by HCPEs using UCS ontention-based period in Frame i, given the hannel is idle and busy, an be obtained from|Si,H|H0| = H(pact,k)(1− pus,H,i)(psucc,i)pf (G.9)and|Si,H|H1| = H(pact,k)(1− pus,H,i)(psucc,i)pd. (G.10)Similarly for MCPEs we have|Si,M |H0| = Mpact,k(1− pus,M,i)psucc,i× [pf (1− pα1) + (1− pf)pα0 ] (G.11)151Appendix G. The Average Cardinality of CPE Setsand|Si,M |H1| = Mpact,k(1− pus,M,i)psucc,i× [pd(1− pα1) + (1− pd)pα0 ] . (G.12)The average number of UCS notiations transmitted by HCPEs using UCSontention-based period in Frame i + 1, given the hannel is idle and busy, anbe obtained from|Si+1,H|H0| = Hpact,k(1− pus,H,i)(1− psucc,i)pf (G.13)and|Si+1,H|H1 | = Hpact,k(1− pus,H,i)(1− psucc,i)pd. (G.14)Similarly for MCPEs we have|Si+1,M |H0| = Mpact,k(1− pus,M,i)(1− psucc,i)× [pf(1− pα1) + (1− pf )pα0 ] (G.15)and|Si+1,M |H1| = Mpact,k(1− pus,M,i)(1− psucc,i)× [pd(1− pα1) + (1− pd)pα0 ] . (G.16)In SEE-CSS, the average number of UCS-D notiations transmitted by HCPEs usingUCS ontention-based period in Frame i+ 1, given the hannel is idle and busy, an152Appendix G. The Average Cardinality of CPE Setsbe obtained from|Di+1,H|H0| = Hpact,k(1− pus,H,i)(1− psucc,i)× (pf,ID(1− pf) + (1− pf,ID)pf) (G.17)and|Di+1,H|H1| = Hpact,k(1− pus,H,i)(1− psucc,i)× (pd,ID(1− pd) + (1− pd,ID)pd). (G.18)Similarly for MCPEs we have|Di+1,M |H0| = Mpact,k(1− pus,M,i)(1− psucc,i)×[pf,ID(1− pf)(1− pα0) + pf,IDpfpα1+ (1− pf,ID)pf (1− pα1)+ (1− pf,ID)(1− pf)pα0](G.19)and|Di+1,M |H1| = Mpact,k(1− pus,M,i)(1− psucc,i)×[pd,ID(1− pd)(1− pα0) + pd,IDpdpα1+ (1− pd,ID)pd(1− pα1)+ (1− pd,ID)(1− pd)pα0]. (G.20)153

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0166185/manifest

Comment

Related Items