Cooperative Sensing Decision Rules over Imperfect Reporting Channels Nian Xia1, a, Chu-Sing Yang1, b

Transcription

1 2nd Internatonal Conerence on Advances n Mechancal Engneerng Industral Inormatcs (AMEII 206) Cooperatve Sensng Decson Rules over Imperect Reportng Channels an Xa, a, Chu-Sng Yang, b Insttute o Computer Communcaton Engneerng, Department o Electrcal Engneerng, atonal Cheng Kung Unversty, Tanan, 700, Tawan a emal: brastme@gmal.com, bemal: csyang@ee.ncku.edu.tw Keywords: cooperatve spectrum sensng; hard decsons; Selecton-based MRC; mperect reportng channels Abstract. Cogntve Rado (CR) s proposed to solve the spectrum scarcty problem caused by the explosve growth o wreless devces the xed spectrum allocaton polcy. Spectrum Sensng s a undamental method n CR. Cooperatve Spectrum Sensng (CSS) can eectvely mprove detecton accuracy. umerous works consderng perect reportng channels exst. However, mperect reportng channels have not ganed enough attenton untl now. In ths paper, we nvestgate mperect lat Raylegh adng lstenng reportng channels. We use channel state normaton to assst Fuson Center (FC) to make nal decsons, propose several channel-normaton-based MRC decson rules. Fnally, tradtonal hard decson rules are compared wth our proposed channel-normaton-based MRC decson rules. Our Selecton-based MRC decson rule can acheve the lowest total error rate. Introducton CR was ntally proposed by Mtola [] n 999. Smon Haykn [2] dened CR as: an ntellgent wreless communcaton system that s aware o ts envronment uses the methodology understng-by-buldng to learn rom the envronment adapt to statstcal varatons n the stmul. Dynamc Spectrum Access (DSA) allows secondary users (SUs) to use lcensed spectrum resources when prmary users (PUs) are dle or the ntererence to PUs s below a constrant [3]. Spectrum Sensng (SS) s used to detect the presence o PU. There are three major spectrum sensng detectors [4-6]. Energy detector (ED) s a non-coherent detector whch detects PU based on the sensed energy. ED s wdely used due to ts smplcty no need pror knowledge o PU. However, ED does not perorm very well n low SR regons. Matched lter detector s a lnear detector whch can maxmze the output sgnal-to-nose rato (SR), but pror knowledge o PU s needed. Cyclostatonary detector uses the perodcty n the receved sgnal to determne PU s state, t s robust to nose uncertanty. Cooperatve Spectrum Sensng (CSS) s proved to be able to eectvely mprove the spectrum sensng detecton perormance by usng spatal dversty. Derent SUs ndependently sense lcensed channels orward ther decsons to FC va reportng channels. Derent decson rules are employed to decde whether PU s dle or not [7, 8]. Decson rules n the recever have ganed enough attentons recently. E. Axell et al. nvestgate hard decson rules whch use one bt to represent PU s actvty. They compare AD, OR Votng rules under derent lstenng channel condtons. The Votng rule s regarded to perorm best among the three hard decson rules [9]. Lteratures above all assume that reportng channel s error-ree. Data transmssons rom SUs to FC are not error ree due to multpath, nose, shadowng so on. Sachn Chaudhar et al. [0, ] prove that Bt Error Probablty (BEP) wall exsts over mperect reportng channels n derent combng rules. Sot decson rules gan lower BEP n the cost o channel bwdth to transmt more than one bt normaton. However, they assume that reportng channel s Bnary Symmetrc Channel (BSC) whch s comparatvely not realstc. In ther recent works, they generalze reportng channels prove that BEP wall stll exsts over Raylegh adng reportng channels [2]. To smply mathematcal dervatons, they assume that lstenng channel ollows Gaussan dstrbuton. Furthermore, they do not use channel normaton whch can 206. The authors - Publshed by Atlants Press 20

2 be easly obtaned by usng estmaton theory [3]to assst FC to make nal decsons. In ths paper, we consder Cogntve Rado etworks (CRs) wth one FC. We assume that both lstenng channels reportng channels ollow Raylegh dstrbutons. Common Control Channels (CCC) are used to exchange normaton between SUs FC. We compare several channel-normaton-based MRC (Maxmze Raton Combng) decson rules, propose a novel Selecton-based MRC decson rule. Ths paper s contrbutons are summarzed as ollows: Investgate mperect Raylegh adng reportng channel. Propose compare several channel-normaton-based MRC decson rules. Propose a novel Selecton-based MRC decson rule. Basc Prncples Cooperatve sensng system model We consder a cooperatve sensng system contanng a PU transmtter, SUs a FC n Fg.. Both lstenng channels reportng channels are ndependent dentcally dstrbuted (..d.). ED s employed n each SU to detect PUs. One bt hard decson s then orwarded by SUs to FC. FC then uses derent decson rules to make a nal decson. We dene I as an ndcator uncton. SUs compare receved sgnal s energy E wth predetermned energy detector threshold γ. ndcates that PU s usng the channel, 0 ndcates that PU s absent. We express I as ollows:, E γ () I = 0, E < γ s(t) denotes that SU transmts equvalent low-pass bnary decson. h θ represent the attenuaton actor phase sht or the channel respectvely. n(t) denotes Addton whte Gaussan nose (AWG). The receved low-pass sgnal can be wrtten as ollows: jθ r( t) = he s( t) + n( t) 0 t T. (2) Lstenng channel R eportng channel transm tter SU Fuson center SU 2 SU - SU Fg.. The wreless communcaton system where lstenng channel reportng channel both ollow Raylegh dstrbuton. False alarm probablty mss detecton probablty H 0 Hdenote that PU s dle busy respectvely. P = Pr( I = H 0), also known as alse alarm probablty, denotes that we decde PU s present whle PU s actually absent. Smlarly, P m = Pr( I = 0 H), termed as mss detecton probablty, denotes that we decde PU s absent whle PU s actually occupyng the channel. We denote the -th SU s detecton probablty as ollows: Pd, = Pm,. (3) In (3), P, denotes the -th SU s mss detecton probablty. We assume that every SU has the m 2

3 same energy detector threshold γ, so every SU has the same detecton probablty: Pd = Pd,, =,2,...,. (4) The -th SU s total error rate can be shown as ollows: P total, = P, + Pm,. (5) Sutable energy detector threshold γ s obtaned to decrease FC s total error rate. Smlarly, we express FC s total error rate as ollows: rule rule rule P = P + P. (6) In (6), rule total m P, rule P rule respectvely denote alse alarm probablty, mss detecton probablty derent decson rules. Decson Rules m Exstng decson rules In perect reportng channels, several hard decson rules ncludng AD rule, OR rule Votng rule have been dscussed. Votng rule can acheve the lowest total error rate. We brely ntroduce the basc schemes o the above three hard decson rules n turn. AD rule: PU s sad to be present only when all SUs decde that PU s present. The detecton probablty, alse alarm probablty, mss detecton probablty n the FC are wrtten as ollows: P d = P,. (7) d,2,..., P = P. (8) m,,2,..., P = P. (9) d,,2,..., OR rule: FC decdes that PU s present when at least one SU declares that PU s present. Smlarly, the probabltes or the OR rule are shown below: or P = ( P ). (0) d,2,..., d, or P = ( P ). () or,2,..., P m = ( P,2,..., (2) Votng rule: FC decdes that PU s present more than hal SUs decde that PU s present. The probabltes or the votng rule n the FC are shown as ollows: d, )., major k k P d = Pd ( Pd ). k = / 2 k major k k P = P ( P ). k = k / 2 major k k Pm = Pd ( Pd ). k = k / 2 (5) EGC-based decson rule MRC-based decson rule Consderng mperect reportng channels, FC can combne derent SUs transmtted sgnals wth correspondng channel state normaton (CSI) by usng derent combng rules to make nal decsons. Assume derent reportng channels wth equal nose power, we can express the FC combner output as ollows: d c = j r w e (3) (4) θ (6) 22

4 In (6), w denotes the weght o -th SU s decson w =, the complex exponental s used or equalzng the phase o each reportng channel, r denotes the receved sgnal rom -th SU s local decson as dened n (2). EGC combng rule: Every SU has the same weght, then w equals to. Below we ntroduce three channel-normaton-based MRC decson rules. Frst, we dene some necessary notatons. h, h, denote adng ampltudes or the -th SU s lstenng channel reportng ls re channel respectvely. We denote the multplcaton o lstenng channel reportng channel o the -th SU as ollows: h = h h. (7) prod, ls, * re, In the Lsten-based MRC, Report-based MRC Channel-based MRC, FC uses lstenng channel state normaton, reportng channel normaton the multplcaton o lstenng reportng channels to do MRC respectvely. The -th SU s weght w mod, s wrtten as ollows: w h mod, mod, = h mod, Where h mod, s set to h ls,, h re, h respectvely n derent MRC rules. prod, Selecton-based MRC In our assumptons, the channel lnk gan between FC SUs s comparatvely hgher than the channel lnk gan between PU transmtter SUs. Thereore, lstenng channel s more mportant than reportng channel. To reduce SU s energy consumpton decrease system s total error rate, we choose SUs wth better lstenng channel qualtes to coordnate.. (8) hλs, w =. (9) h s. t. h λs, λ λs, λ denotes the lstenng channel qualty threshold. We wll dscuss more about the threshold λ n the next secton. Smulaton Results Expermental envronments In our experment, lstenng channels reportng channels ollow Raylegh dstrbuton. ose s AWG wth mean zero varance one. We employ the BPSK modulaton n both the PU transmtter SUs. Carrer Frequency s set to be 800MHz bwdth s 2MHz. Reportng channel SR s 25dB. We smulate more than 0000 tmes. Smulaton results Frstly, we nvestgate the relatonshp between EGC, Channel-normaton-based MRC decson rules lstenng channel threshold λ system s total error rate. We assume that the lstenng channel SR s 0dB the reportng channel SR s 25dB. We vary the lstenng channel threshold λ rom 0 to 3 observe the correspondng total error rate. In Fg. 2., among compared ve decson rules, only Selecton-based MRC wll choose SU s wth better lstenng channel condton to coordnate, so only ts total error rate vares wth derent lstenng threshold λ. Suboptmal lstenng channel threshold λ exsts wth the varyng lstenng λ channel values. Wth the ncreasng lstenng channel threshold, the sensng accuracy wll ncrease; however, the number o selected SUs wll decrease. System s total error rate wll ncrease because we can t select enough SUs to coordnate. There s a tradeo between threshold λ the number o selected SUs. Lsten-based MRC acheves the second lowest total error rate due to the act that lstenng channel has a hgher prorty than reportng channel. Smlarly, we can explan why 23

5 Report-based MRC has the hghest total error rate. In our proposed Selecton-based MRC, SUs wth lstenng channel s parameter equal or larger than threshold λ are selected to assst FC makng nal decsons. By choosng sutable threshold λ, Selecton-based MRC can acheve the lowest total error rate. Fg. 2. The relatonshp o lstenng channel thresholds wth total error rate. Fg. 3. Compare derent decson rules n derent lstenng channel SRs. In Fg. 3., our proposed Selecton-based MRC has the lowest total error rate among derent decson rules n derent SR stuatons. However, n low SR regons, due to the lmted number o cooperatve SUs, we may not nd sutable SUs to coordnate. So t s reasonable that our proposed Selecton-based MRC does not always acheve the lowest total error rate n low SR regons. Meanwhle, our proposed channel-normaton-based MRC decson rules perorm better than tradtonal hard decson rules. Concluson In ths paper, we nvestgate mperect Raylegh adng reportng channels propose several channel-normaton based MRC decson rules: Lsten-based MRC, Report-based MRC, Channel-based MRC decson rules. Furthermore, we propose a novel Selecton-based MRC decson rule whch chooses SUs wth better lstenng channel qualty to coordnate. Fnally, we compare our proposed scheme wth tradtonal hard decson rules. Our Selecton-based MRC decson rule can acheve the lowest total error rate. In our uture work, we wll consder SUs wth derent spatal locatons, SUs can adapt ther transmsson power to save energy. In addton, ths proposed scheme needs to be comprehensvely evaluated when the channel estmaton error s consdered. Theoretc analyss s requred to prove our smulaton results. Furthermore, we wll nvestgate the decson rule perormance when malcous SUs exst. 24

6 Acknowledgement Ths work was supported n part by Tawan Mnstry o Scence Technology (MOST) under Grant MOST E Reerences [] J. Mtola III G. Q. Magure, "Cogntve rado: makng sotware rados more personal," IEEE Personal Communcatons, vol. 6, no. 4, pp. 3-8, Aug [2] S. Haykn, "Cogntve rado: bran-empowered wreless communcatons," IEEE Journal on Selected Areas n Communcatons, vol. 23, no. 2, pp , Feb [3] Z. Qng B. M. Sadler, "A Survey o Dynamc Spectrum Access," IEEE Sgnal Processng Magazne, vol. 24, no.3, pp , May [4] P. D. Sutton, J. Lotze, K. E. olan, L. E. Doyle, "Cyclostatonary Sgnature Detecton n Multpath Raylegh Fadng Envronments," 2nd Internatonal conerence on Cogntve Rado Orented Wreless etworks Communcatons, [5] T. Yucek H. Arslan, "A survey o spectrum sensng algorthms or cogntve rado applcatons," IEEE Communcatons Surveys Tutorals, vol., no., pp. 6-30, [6] A. Saha,. Hoven, R. Tra, "Some Fundamental Lmts on Cogntve Rado," 42sd Allerton Conerence on Communcaton, Control Computng, [7] I. F. Akyldz, B. F. Lo, R. Balakrshnan, "Cooperatve spectrum sensng n cogntve rado networks: A survey," Physcal Communcatons, vol. 4, no., pp , 20. [8] S. allagonda, S. D. Roy, S. Kundu, "Cooperatve spectrum sensng wth censorng o cogntve rados n Raylegh adng channel," 202 atonal Conerence on Communcatons (CC), 202. [9] E. Axell, G. Leus, E. G. Larsson, "Overvew o spectrum sensng or cogntve rado," 2nd Internatonal Workshop on Cogntve Inormaton Processng (CIP), 200. [0] S. Chaudhar, J. Lunden, V. Kovunen, H. V. Poor, "Cooperatve Sensng Wth Imperect Reportng Channels: Hard Decsons or Sot Decsons?," IEEE Transacton on Sgnal Processng, vol. 60, no. pp. 8-28, Jan [] S. Chaudhar, J. Lunden, V. Kovunen, "BEP walls or collaboratve spectrum sensng," 36th IEEE Internatonal Conerence on Acoustcs, Speech Sgnal Processng, 20. [2] S. Chaudhar, J. Lundén, V. Kovunen, H. V. Poor, "BEP walls or cooperatve sensng n cogntve rados usng K-out-o- uson rules," Sgnal Processng, vol. 93, pp , 203. [3] S. M. Kay, "Fundamentals o Statstcal Sgnal Processng: Estmaton Theory," Prentce Hall,