Improved Detecton Performance of Cogntve Rado Networks n AWGN and Raylegh Fadng Envronments Yng Loong Lee 1, Wasan Kadhm Saad, Ayman Abd El-Saleh *1,, Mahamod Ismal 1 Faculty of Engneerng Multmeda Unversty 63100 Cyberjaya, Selangor, Malaysa *ayman.elsaleh@mmu.edu.my Department of Electroncs, Electrcal and System Engneerng Faculty of Engneerng and Bult Envronment Unverst Kebangsaan Malaysa 43600 Bang, Selangor, Malaysa ABSTRACT Cogntve rados (CRs) have been recently emergng as prme canddates to enhance spectral effcency by explotng spectrum-aware systems whch can relably montor lcensed users actvtes. CR users montor such actvtes by performng spectrum sensng to detect potental whte spaces. However, ths process of local sensng mght be a challengng task n fadng envronments. The neffcency of spectrum sensng mght cause nterference to lcensees f they are mss-detected by CR users. Thus, cooperatve spectrum sensng s proposed as a means to combat fadng and mprove the detecton performance. However, the detecton performance does not mprove by such cooperaton when low-snr envronment s consdered. In ths paper, cooperatve spectrum sensng wth PSO-based threshold adaptaton s presented to address the aforementoned problem. Smulaton results show that the detecton performance wth PSObased adaptve detecton threshold s mproved, partcularly, n low-snr envronment. Keywords: cogntve rado, cooperatve spectrum sensng, dynamc threshold adaptaton, partcle swarm optmzaton. 1. Introducton As wreless communcaton technology grows rapdly, the demand for spectrum s ncreasng consequently n order to support more wreless servces. However, the lmted rado resources become a great barrer to meet the ncreasng demand for spectrums. A survey s carred out by Federal Communcaton Commsson (FCC) to nvestgate the spectrum usage effcency n a temporal and geographcal area varaton [1]. Ths survey ndcated the that current the lcensed spectrum usage s often mostly under-utlzed. Due to ths fact, cogntve rado (CR) s proposed as one of the most promsng solutons to support the ncreasng need for spectrum by occupyng these under-utlzed lcensed spectrum segments. CR s defned as a rado whch s able to adapt and learn from ts surroundng rado envronment and adjust ts network parameters to optmze the utlzaton of the spectrum whle provdng flexblty n wreless access [1]. In other words, CR s a technology whch s capable of detectng and accessng the under-utlzed spectrums effcently. In order to perform ths capablty, four mportant functons are proposed for CR systems, namely, spectrum sensng, spectrum decson, spectrum sharng and spectrum moblty []. Durng CR operaton, spectrum sensng wll frst be performed to detect all avalable under-utlzed spectrums, also known as spectrum holes or whte spaces [3]. After detectng all the whte spaces, the functon of spectrum decson-makng wll be performed to select the best whte space for nstantaneous transmsson. The spectrum sharng functon n CR provdes coordnaton or schedulng for sharng of spectrum bands wth other secondary users (SUs) and/or CR users. Lastly, the spectrum moblty functon allows SU to smoothly release the spectrum band back to ts owner, also known as prmary user (PU), once detected and move to another avalable whte space. In realty, PU sgnals could be shadowed and faded, hence causng rapd fluctuaton n sgnal Journal of Appled Research and Technology 437
Improved Detecton Performance of Cogntve Rado Networks n AWGN and Raylegh Fadng Envronments, Yng Loong Lee et al. / 437 446 strength, as ndcated n many studes [][4][5]. Local spectrum sensng by a sngle SU mght not be able to determne the presence of PU sgnal accurately due to potental fadng and shadowng effects. To tackle ths ssue, cooperatve spectrum sensng s proposed [][4]. The cooperatve spectrum sensng technques show promsng mprovement n detecton performance [6][7]. However, the cooperatve sensng can hardly provde mprovement to the detecton performance n low-snr envronment as can be observed n [7]. In ths paper, performance nvestgatons for local spectrum sensng and cooperatve spectrum sensng are carred out under dfferent rado envronments,.e., Addtve Whte Gaussan Nose (AWGN) and Raylegh fadng channels. Energy detecton-based spectrum sensng s adopted n ths paper due to ts mplementaton smplcty. Also, an OR-rule hard decson fuson scheme s used to realze cooperatve sensng whle mantanng low communcaton overhead n comparson to soft fuson schemes. Fnally, a cooperatve spectrum sensng wth PSO-based threshold adaptaton s proposed to address the drawback of neffcent performance of cooperatve spectrum sensng n low-snr scenaro. The rest of ths paper s organzed as follows. Secton provdes an overvew on energy detecton-based spectrum sensng technque. In Secton 3, cooperatve spectrum sensng s explaned. Secton 4 presents cooperatve spectrum sensng wth PSO-based threshold adaptaton and ts performance evaluaton. Fnally, several concludng remarks are drawn n Secton 5.. Energy detector-based local spectrum sensng In energy detecton, the PU sgnal s frst receved and sampled. From the sampled receved sgnals by SU, two hypotheses can be deduced as gven below: H 0 : PU does not exst. H 1 : PU does exst Based on the two hypotheses above, the receved sgnal for the -th SU can be expressed as follows [8] y t n ( t) H0 (1) h x ( t ) n ( t ) H1 where y (t) s the sgnal receved by -th SU, x(t) s the PU sgnal, n (t) s the AWGN nose receved by -th SU and h s the channel gan. To determne whether H 0 or H 1 s true, the energy of the receved sgnal, y (t), s estmated from the lcensed channel of nterest wthn an observaton perod or sensng perod, T, and then a test/decson statstc s obtaned. Accordng to [8], the decson statstc, Z, obtaned from the energy detector for the -th SU s gven as Z W TW 1 y, k N W k 1 0 () Where y,k = y (k/w) and N 0 s the one-sded nose power spectral densty. The decson statstc for the -th SU, Z, obtaned from the energy detecton s found to have ch-square dstrbuton [8] and can be characterzed as [9] Z ~ m m H 0 H1 (3) Where m = TW, that s the tme-bandwdth product of the energy detector. For smplcty, m s assumed to be an nteger value. From Equaton 3, m represents a central ch-square dstrbuton wth m degrees of freedom whereas m represents a noncentral ch-square dstrbuton wth m degrees of freedom and a noncentralty parameter of γ for H 1 where γ s the nstantaneous SNR receved at the -th SU. In general, the probablty of false alarm and probablty of detecton for the -th SU are, respectvely, gven as P P Pr( Z H ) (4) f, 0 Pr( Z H ) (5) d, 1 Where λ s the detecton threshold for the -th SU. Hence, from Equatons 3 and 4, the closed-form expresson for probablty of detecton over AWGN channel can be obtaned from [9] as P Q, (6) d, m 438 Vol. 11, June 013
Improved Detecton Performance of Cogntve Rado Networks n AWGN and Raylegh Fadng Envronments, Yng Loong Lee et al. /437 446 where a b Q m, refers to the generalzed Marcum Q-functon defned: Q x a 1, m1 m a (7) b m a b x e I axdx m 1 The probablty of mssed detecton s smply defned as P m, Pd, 1 (8) On the other hand, usng Equatons 3 and 5, the probablty of false alarm over AWGN channel s obtaned as [9]: P f, m, ( m) (9) Where Γ(.) and Γ(.,.) are the complete gamma functon and the upper ncomplete gamma functon, respectvely. It was noted that Equaton 9 s ndependent of γ ; the nstantaneous channel SNR. For fadng channels, the probablty of detecton, P d,, for the -th SU over fadng channels can be, respectvely, gven as [6], f ( x dx Pd, Qm ) (10) Where f (x) s the PDF for γ whch vares wth dfferent fadng models. However, the probablty of false alarm wll reman the same as n Equaton 9 because t s ndependent of γ. When the PU sgnal experences scatterng mposed by the envronment, the PU sgnal undergoes multpath fadng. Due to ths phenomenon, the faded PU sgnal s descrbed by Raylegh dstrbuton. Therefore, γ would have an exponental dstrbuton and thus the probablty of detecton over Raylegh fadng channel can be found as [9] P d, e e 1 m k 0 e 1 k! m k 0 k 1 1 k! 1 m1 k... (11) Fgure 1 shows the complementary ROC curves for local spectrum sensng under AWGN and Raylegh fadng channels for dfferent SNR values. It s observed that the probablty of detecton decreases dmnshngly for a fxed probablty of false alarm under both AWGN channel and Raylegh fadng channel when SNR decreases. 3. Cooperatve spectrum sensng A smple operatng cogntve rado network (CRN) s llustrated n Fgure. Each SU performs spectrum sensng to detect the presence of PU sgnal. When the PU sgnal undergoes deep fadng and shadowng, the sgnal strength vares at dfferent tmes and locatons dependng on the channel condtons represented by the correspondng noses and gans mposed. Fgure shows dfferent sensng scenaros by multple SUs where some SUs may be able to relably detect the PU sgnal such as SU and SU 3 whle others such as SU 1 and SU 4 may not able to detect t due to ntermedate blockng obstacles. Ths observaton leads to the dea of cooperatve sgnal detecton whch nvolves collaboraton among all SUs n a CRN to mprove the detecton performance. In cooperatve spectrum sensng, each SU sends the sensng nformaton to the SU base staton and the base staton makes a global fnal decson. Many decson fuson schemes are proposed n the lterature. One of the well-known decson fuson schemes s the so-called one-out-of-n rule or ORrule, where N s the total number of cooperatng SU. In ths hard decson fuson scheme, all cooperatng SUs send ther local sensng decsons to a common fuson center for fnal decson fuson. A fnal decson corresponds to H 0 s deduced f all N collaboratng SUs ndcated that the PU s absent whereas that corresponds to H 1 s made f there s at least one out of N SUs reports that the PU s present. Assumng that all decsons are ndependent, the probablty of detecton, the probablty of mssed detecton, and the probablty of false alarm of cooperatve spectrum sensng denoted by Q d, Q m and Q f, respectvely, can be defned as [10]: N 1 Q d 1 P d, (1) 1 Journal of Appled Research and Technology 439
Improved Detecton Performance of Cogntve Rado Networks n AWGN and Raylegh Fadng Envronments, Yng Loong Lee et al. / 437 446 N Q m 1 P d, (13) 1 N 1 Q f 1 P f, (14) 1 Fgures 3, 4 and 5 llustrate the detecton performance of cooperatve spectrum sensng usng OR-rule under AWGN and Raylegh fadng channel at dfferent SNR values (.e., 10 db, 0 db, and -10 db). Under both AWGN and Raylegh fadng channels, t can be seen that the detecton performance of cooperatve spectrum sensng was greatly mproved at hgh-snr value (.e., 10 db) where the greater the number of cooperatng SUs, the greater the mprovement observed on the complementary ROC curve. However, Fgure 4, shows that the mprovement of detecton performance dd not mprove sgnfcantly wth the ncreased cooperaton among SUs n medum- SNR scenaro (.e., SNR = 0 db). Even more notceable, n low SNR (.e., SNR = -10 db), the cooperatve spectrum sensng (e.g., N = 5 or N = 10) dd not provde any mprovement n comparson to the detecton performance of local sensng (.e., N = 1), as depcted n Fgure 5. Fgure 1. Complementary ROC curves for local spectrum sensng under AWGN and Raylegh channel wth m = 10. Fgure. Cogntve rado network (CRN). 440 Vol. 11, June 013
Improved Detecton Performance of Cogntve Rado Networks n AWGN and Raylegh Fadng Envronments, Yng Loong Lee et al. /437 446 Fgure 3. Complementary ROC curves for cooperatve spectrum sensng usng OR-rule under AWGN and Raylegh channel wth m = 10 and SNR = 10 db. Fgure 4. Complementary ROC curves for cooperatve spectrum sensng usng OR rule under AWGN and Raylegh channel wth m = 10 and SNR = 0 db. Journal of Appled Research and Technology 441
Improved Detecton Performance of Cogntve Rado Networks n AWGN and Raylegh Fadng Envronments, Yng Loong Lee et al. / 437 446 Fgure 5. Complementary ROC curves for cooperatve spectrum sensng usng OR-rule under AWGN and Raylegh channel wth m = 10 and SNR = -10 db. 4. Cooperatve spectrum sensng wth dynamc threshold adaptaton As mentoned n Secton 3, the cooperatve spectrum sensng dd not provde sgnfcant mprovement to the detecton performance n low- SNR envronment. In order to tackle ths ssue, ths paper presents a cooperatve spectrum sensng technque armed wth PSO-based threshold adaptaton. Fgure 6 llustrates a smple block dagram for the proposed scheme. Assumng that each SU s capable of estmatng SNR at ts recever; each SU wll perform local spectrum sensng and wll calculate ts own decson statstc. Then, every SU wll run ts PSO evolutonary processes to search for an optmal detecton threshold such that hgh probablty of detecton and low probablty of false alarm are jontly attaned. The calculated decson statstc by each SU wll be then compared wth the optmzed threshold and the correspondng decsons made by all SUs wll be sent to an OR-rule common fuson center for developng a fnal global decson on PU avalablty. 4.1 Problem formulaton for threshold adaptaton To optmze the detecton performance n an SNRvaryng envronment, low probablty of mssed detecton and low probablty of false alarm must always be jontly mantaned. Ths s because mnmzng the probablty of mssed detecton makes the PU more protected aganst potental SU transmssons whereas mnmzng the false alarm probablty allows SUs to effcently utlze the unused bands of spectrum. Therefore, the decson threshold has to be adaptvely adjusted to satsfy the aforementoned two conflctng requrements for varous channel condtons. The overall performance objectve of the whole CRN can be put nto a sngle optmzaton problem of mnmzng the total sensng error gven by [11]: P m Pf 1 (15) Where δ s a weghtng constant rangng n (0, 1) for the probablty of mssed detecton relatve to that of false alarm. Because low probablty of false alarm and hgh probablty of detecton are desred, 44 Vol. 11, June 013
Improved Detecton Performance of Cogntve Rado Networks n AWGN and Raylegh Fadng Envronments, Yng Loong Lee et al. /437 446 a constrant has to be mposed to the threshold. In ths paper, probablty of false alarm s lmted to the range of [0.001, 0.1]. It s reasonable to mpose a maxmum lmt to the probablty of false alarm so that low probablty of false alarm can be mantaned. A mnmum lmt s also mposed because a very low probablty of false alarm would mply that that the probablty of detecton s also very low, thus mposng a mnmum lmt to the probablty of false alarm could preserve a reasonable probablty of detecton. Hence, the optmzaton problem becomes argmn (16) s. t. 0.001 0.1 P f 4. Partcle swarm optmzaton Partcle swarm optmzaton (PSO) s a populatonbased random search algorthm developed by Kennedy and Eberhart n 1995 based on swarm behavor of brd flockng and fsh schoolng [13][14]. In PSO, a populaton, also known as swarm, s ntally created based on the search space of a gven optmzaton problem. Each member n the swarm, also referred to as partcle, s dstrbuted randomly wthn the predefned search space. Ths populaton wll randomly fly through the search space to look for the global optmum. The trajectory of each partcle s nfluenced by the best poston personally found so far, whch s called personal best (pbest), and the best poston found by the entre swarm, named as global best (gbest). The velocty-update and poston-update equatons for each partcle are gven as [13][14]: v j t 1 v j t c1r1 y j t xj t c r yˆ t x t t 1 x t v t 1 j j (17) xj j j (18) Where v j (t) denotes the velocty of -th partcle n j- th dmenson at t-th teraton, c 1 and c are referred as acceleraton constants, r 1 and r are unformly dstrbuted random values rangng n [0, 1]. y j (t) s referred as pbest, whch s the best poston found by the -th partcle n j-th dmenson so far by the t- th teraton whereas ŷ j (t) s referred as gbest, whch s the best poston found by the entre swarm Fgure 6. Block dagram for cooperatve spectrum sensng wth PSO-based threshold adaptaton. Journal of Appled Research and Technology 443
Improved Detecton Performance of Cogntve Rado Networks n AWGN and Raylegh Fadng Envronments, Yng Loong Lee et al. / 437 446 n j-th dmenson so far at the t-th teraton. x j (t) denotes the poston of -th partcle n j-th dmenson at t-th teraton. Another parameter, so-called maxmum velocty, V max s mposed to lmt the velocty of each partcle to ensure exploraton wthn the search space. A pseudocode of PSO s shown n Fgure 7 for a gven mnmzaton problem [1]. In ths paper, the performance objectve of CRN s to mnmze the total sensng error. Thus, the ftness functon to be optmzed by PSO s the objectve functon n Equaton 15 and each partcle represents a potental settng of the decson threshold. Fgure 7. Pseudocode for PSO algorthm [1]. 4.3 Results and analyss In ths secton, m s set to 10, δ s set to 0.5 as n [15], and SNR vares from -10 db to 0 db wth a step sze of 1 db. In the PSO algorthm, the search space s set to the threshold values correspond to the constrant of the probablty of false alarm stated n the Equaton 16, V max s set as the dfference between the mnmum and maxmum search space boundares, c 1 and c are both set to as n [13][14], and the number of partcles s set to 10. The PSO evolutonary operaton runs for 100 teratons and to be repeated for 100 tmes to perform averagng (.e., number of realzatons = 100) for each step ncrease n SNR. The convergence performance of PSO for AWGN channel, for nstance, wth SNR = 0 db s depcted n Fgure 8. It can be observed that the PSO algorthm s able to converge wthn the frst 10 teratons, whch s extremely fast. Ths mples that the computatonal tme of the proposed PSO algorthm s suffcently short to meet the real-tme requrements. The smlar convergence performance of the proposed PSO-based cooperatve spectrum sensng scheme can be observed at dfferent SNR values and for channels encounterng Raylegh fadng. Ths convergence performance confrms the effectveness of employng a PSO-based threshold adaptaton at the recever of every SU whch s observed by the fast processng speed of the optmzaton algorthm. Next, we are nterested to show the mprovement on the detecton performance at the common fuson centre of a CRN employng PSO-based threshold adapton at ts CR nodes. Remember that the man objectve of employng PSO for dynamc adapton of CR/SU recever s threshold s to mnmze the overall sensng errors of cooperatve SUs n low-snr envronment. Consder the SNR range of PU sgnal at SU recevers from -10 db to 0 db. Fgure 9 depcts the detecton performance represented by the probablty of sensng error versus SNR for cooperatve network of 5 SUs (.e., N = 5) for AWGN channel. Fgure 10 shows the detecton performance for the same number of SUs but wth assumng Raylegh channels. For the two cases n Fgures 9 and 10, t can be observed that the probablty of sensng error s mnmzed by usng threshold adaptaton at the SUs recevers n comparson to the case of usng statc threshold settng based on a gven probablty of false alarm; P f = 0.1. The statc threshold settng s bascally obtaned from Equaton 9 for AWGN and Raylegh fadng channels, respectvely, gven a fxed probablty of false alarm. Interestngly, the probablty of sensng error s contnually decreasng as the SNR goes lower for both AWGN and Raylegh channels. Ths observaton makes the proposed dea of employng PSO-based threshold adapton a good choce, n partcular, n low-snr scenaros. In addton, snce mnmzng the overall sensng error requres jont mnmzaton of mssed detecton and false alarm probabltes, ths approach provdes a balanced compromse between PU protecton demands and SU wllngness to opportunstcally access avalable unused spectrum bands or whte spaces. 444 Vol. 11, June 013
Improved Detecton Performance of Cogntve Rado Networks n AWGN and Raylegh Fadng Envronments, Yng Loong Lee et al. /437 446 Fgure 8. Convergence performance of PSO. Fgure 9. Sensng error versus SNR for N = 5 under AWGN channel. Fgure 10. Sensng error versus SNR for N = 5 under Raylegh fadng channel. Journal of Appled Research and Technology 445
Improved Detecton Performance of Cogntve Rado Networks n AWGN and Raylegh Fadng Envronments, Yng Loong Lee et al. / 437 446 5. Concluson In ths paper, the detecton performance for local spectrum sensng and cooperatve spectrum sensng usng OR-rule under AWGN and Raylegh fadng channel are evaluated. It was demonstrated that the cooperatve spectrum sensng can hardly mprove the detecton performance n low-snr envronment. Thus, cooperatve spectrum sensng wth PSO-based threshold adaptaton has been proposed to address the aforementoned drawback. The PSO threshold adapton algorthms are mplemented at the recevers of CRN. The am of mprovng the detecton performance of CRN n low-snr envronment was realzed by mnmzng the overall sensng error; that s, by the jont mnmzaton of probablty of false alarm and probablty of mssed detecton at the common fuson centre of CRN. Computer smulatons showed that the performance of the proposed scheme s superor to that wth fxed threshold and provde lower sensng errors n low-snr envronment. It was also observed that the gan of decreasng the sensng error ncreases as the SNR goes lower. Thus, these nterestng fndngs confrmed the effcency of employng PSObased adaptve thresholds to mprove the sensng performance of CRN n deterorated channel condtons. References [1] FCC, ET Docket No 03- Notce of proposed rulemakng and order, 003. [] I. F. Akyldz et al., A Survey on Spectrum Management n Cogntve Rado Networks, IEEE Communcatons Magazne, vol. 46, no.4, pp.40-48, 008. [3] S. Haykn, Cogntve Rado: Bran-Empowered Wreless Communcatons, IEEE Journal on Selected Areas n Communcatons, vol. 3, no., pp. 01-0, 005. [4] D. Cabrc et al., Implementaton Issues n Spectrum Sensng for Cogntve Rados, n Proceedngs of the 38th Aslomar Conference on Sgnal, Systems and Computers, vol. 1, Pacfc Grove, Calforna, USA, 004, pp. 77-776. [5] T. Yucek and H. Arslan, A Survey of Spectrum Sensng Algorthms for Cogntve Rado Applcatons, IEEE Communcatons Surveys & Tutorals, vol. 11, no. 1, pp.116-130, 009. [6] A. Ghasem and E. S. Sousa, Collaboratve Spectrum Sensng for Opportunstc Access n Fadng Envronments, n Proceedngs of the 1st IEEE Symposum on New Fronters n Dynamc Spectrum Access Networks, Baltmore, Maryland, USA, 005, pp. 131-136. [7] J. Duan and Y. L, Performance Analyss of Cooperatve Spectrum Sensng n Dfferent Fadng Channels, n Internatonal Conference on Computer Engneerng and Technology, vol. 3, Chengdu, Chna, 010, pp. 64-68. [8] H. Urkowtz, Energy detecton of unknown determnstc sgnals, Proceedngs of the IEEE, vol. 55, no. 4, pp. 53-531, 1967. [9] F. F. Dgham et al., On the Energy Detecton of Unknown Sgnals over Fadng Channels, n Proceedngs of the IEEE Internatonal Conference on Communcaton, Seattle, Washngton, USA, 003, pp. 3575-3579. [10] Y.-C. Lang et al., Sensng-Throughput Tradeoff for Cogntve Rado Networks, IEEE Transactons on Wreless Communcatons, vol. 7, no. 4, pp. 136-1337, 008. [11] D. R. Josh et al., Gradent-Based Threshold Adaptaton for Energy Detector n Cogntve Rado Systems, IEEE Communcatons Letters, vol. 15, no. 1, pp. 19-1, 011. [1] F. Van den Bergh, An analyss of the partcle swarm optmzer, Ph.D. Thess, Unversty of Pretora, 006. [13] J. Kennedy and R. Eberhart, Partcle Swarm Optmzaton, n Proceedngs of the IEEE Internatonal Jont Conference on Neural Networks, Perth, WA, Australa, 1995, pp. 194-1948. [14] R. Eberhart and J. Kennedy, A new optmzer usng partcle swarm theory, n Proceedngs of 6th Inernatonal Symposum on Mcro Machne and Human Scence, Nagoya, Japan, 1995, pp. 39-43. 446 Vol. 11, June 013