COGNITIVE RADIO ENGINE MODEL UTILIZING SOFT FUSION BASED GENETIC ALGORITHM FOR COOPERATIVE SPECTRUM OPTIMIZATION

Transcription

1 Internatonal Journal of Computer Networks & Communcatons (IJCNC Vol.5, No., arch 3 COGNIIVE RADIO ENGINE ODEL UILIZING SOF FUSION BASED GENEIC ALGORI FOR COOERAIVE SECRU OIIZAION ABSRAC d. Kamal ossan, Ayman Abd El-Saleh Faculty of Engneerng, ultmeda Unversty, Cyberjaya Campus, alaysa md.kamal.hossan@hotmal.com Cogntve rado (CR s to detect the presence of prmary users (Us relably n order to reduce the nterference to lcensed communcatons. Genetc algorthms (GAs are well suted for CR optmzaton problems to ncrease effcency of bandwdth utlzaton by manpulatng ts unused portons of the apparent spectrum. In ths paper, a bnary genetc algorthm (BGA -based soft fuson (SF scheme for cooperatve spectrum sensng n cogntve rado network s proposed to mprove detecton performance and bandwdth utlzaton. he BGA-based optmzaton method s mplemented at the fuson centre of a lnear SF scheme to optmze the weghtng coeffcents vector to maxmze global probablty of detecton performance. Smulaton results and analyses confrm that the proposed scheme meets real tme requrements of cogntve rado spectrum sensng and t outperforms conventonal natural deflecton coeffcent- (NDC-, modfed deflecton coeffcent- (DC-, maxmal rato combnng- (RC- and equal gan combnng- (EGC- based SDF schemes as well as the OR-rule based hard decson fuson (DF. he propose BGA scheme also converges fast and acheves the optmum performance, whch means that BGAbased method s effcent and qute stable also. KEYWORDS Genetc Algorthm, Cogntve rado, Cooperatve spectrum sensng, Soft fuson.. INRODUCION In wreless communcaton system, electromagnetc spectrum s n scarcty consderng the avalablty of resources. Wth the burgeonng of technology, the demand for the spectrum s also ncreasng nsstently whch results scarcty n spectrum-avalablty. revous assumpton about the crss of spectrum avalablty resulted msconcepton. It s resolved by the Federal Communcatons Commsson (FCC [] that underutlzaton of the lcensed spectrum bands ether temporally or spatally s the prncpal reason for spectrum scarcty. In order to fully utlze the spectrum resources, CR [],.e. rado systems wth adaptve ntellgence, s fascnatng researchers and developers to overtake spectrum congeston bottlenecks. Cogntve rado takes advantage of the rapdly ncreasng complexty of rado equpment. It has been consdered as the key enablng technology for current spectrum scarcty problem arsng because of the necessty for secure and robust communcatons and s becomng more apparent every day. Wreless servces are becomng largely ubqutous, despte ts expensve mplementaton. he CR users are consdered as secondary users (SUs n Cogntve Rado Network (CRN. DOI :.5/jcnc.3.5 3

2 Internatonal Journal of Computer Networks & Communcatons (IJCNC Vol.5, No., arch 3 he man dea of CR s to perodcally montor the rado spectrum, detect the occupancy and then opportunstcally use spectrum holes wth mnmal nterference wth prmary users (Us. In order to detect the U sgnal wth unknown locaton, structure and strength, energy detecton exhbts smplcty and serves as the optmal spectrum sensng scheme. owever, owng to hdden termnal problem and shadowng effect, the U sgnal packets mght not be receved by CR recever wthn the sharp sensng nterval and thus the sensng performance wll be rendered fragle at a partcular geographcal locaton [3]. hs amplfes potental nterference caused to the U. In other words, energy detecton s extremely vulnerable to the channel effects such as multpath fadng and nose-power fluctuatons that supposed to be avoded. In [4], [5], cooperatve spectrum sensng was used to overcome these drawbacks and vacate the band mmedately f Us presence has been detected. he decson on the presence of U s acheved by combnng all ndvdual decsons of local SUs at a central Fuson Centre (FC usng varous fuson schemes [6], [7]. hese schemes can be classfed as hard decson fuson (DF [4], [8], [9], soft decson fuson (SDF [], [] or softened hard decson fuson (SDF [], []. In DF, the local sensors, or SUs, make ther own judgments on the presence of a U and ther correspondng resultant -bt decsons are sent to the FC for fuson. Apparently the SDF schemes want the local sensors to report ther measurements as raw data to the FC where data wll be fused to construct a fnal decson on the presence of Us. In [], [3] SDF-based schemes have shown better detecton performance than DF schemes. In [], cooperatve spectrum sensng was proposed and NDC, DC, RC, EGC, OR-Rule based methods were used to fnd the optmal weghtng vector for all possble cogntve rados. In ths paper, a BGA-based soft decson fuson (SDF scheme for cooperatve spectrum sensng n cogntve rado network s proposed to enhance the detecton performance. A genetc algorthm (GA s a search heurstc that mmcs the process of natural evoluton. GAs are well equpped wth many tools to reduce computatonal complexty and produce a dverse set of solutons whch can be mplemented on semconductor devces and enable rapd ntegraton wth wreless technologes. Genetc algorthm s a knd of self-adaptve global searchng optmzaton algorthm. It s populaton-based n whch each ndvdual s evolved n parallel and the optmal ndvdual s preserved and obtanable from the last populaton. he contrbutons of ths paper are lsted as follows: (a In order to mprove detecton performance of cooperatve spectrum sensng n Cogntve Rado Network (CRN, U detecton problem has been reformulated usng BGA. (b hen, the proposed method wll be compared wth conventonal NDC, DC, RC and EGC based SDF schemes as well as the OR-rule based hard decson fuson (DF to verfy the supremacy of the proposed method. he rest of ths paper s structured as follows: Secton presents some related research background. Secton 3 brefly explans system model, whle Secton 4 shows comparsons and results of performance of Genetc Algorthm over other methods. Fnally, Secton 5 concludes the paper and provdes future works.. RELAED WORKS CR echnology has been evolved as key enablng technology to the problem of spectrum underutlzaton. CR desgned to allow unlcensed or SUs to access spectrum bands whch has been allocated to Us when nterference to U remans below a gven threshold. he man nspraton behnd ths CR technology s the new spectrum lcense ntated by the FCC, whch wll be more flexble to allow unlcensed or SUs to access the spectrum as long as the lcensed or 4

3 Internatonal Journal of Computer Networks & Communcatons (IJCNC Vol.5, No., arch 3 Us are not nterfered. hat s why there s an ncreased nterest of researchers n ths technology n academa, ndustry and engneers n the wreless ndustry and also from spectrum polcy makers. In [5], [], the authors proposed an optmal weghtng scheme for cooperatve spectrum sensng n cogntve rado networks, under the constrant of equal probabltes of false alarm and mss detecton. ultple cooperatve SUs smply serve as relay nodes n the network to provde space dversty for spectrum sensng. An optmal soft fuson scheme and a double-threshold strategy were proposed to nvestgate the overall spectrum sensng performance n soft fuson and hard fuson, respectvely. From ther observatons and analyss proposed herarchcal cooperatve spectrum sensng schemes can acheve sgnfcant mprovements n spectrum sensng performance. In [3], GA-based weghted collaboratve spectrum sensng strategy was proposed to reduce the effects of channel and enhance spectrum sensng performance. Authors proposed optmum spectrum sensng framework s based on a model that s realstc and also takes nto account both channels, that s, channel between U and SUs as well as the reportng channels. It was shown n ths paper that mperfect reportng channel and dfferent SU SNR values have drect mpact on the performance of CSS. SUs transmt ther soft decsons to the fuson centre and a global decson s made at the fuson centre whch s based on a weghted combnaton of the local test statstcs from ndvdual SUs. he weght of each SU s ndcatve of ts contrbuton to the fnal decson makng. Fnally n [4], the authors addressed SDF-based scheme to explot the advantage of optmum detecton performance of SDF. he SDF schemes had been mplemented usng weghtng coeffcents vector based on NDC, DC, RC and EGC based SDF schemes as well as the ORrule based hard decson fuson (DF. hese SDF schemes were mplemented wthn the cluster. he -bt U-avalablty decsons of several cluster then forwarded to a common recever (BS at whch an OR-Rule DF scheme used to come out wth a global sngle decson on the presence of a U. Authors analyss concerned SDF performs better than DF scheme. In our approach, we are gong to mplement BGA-based soft decson fuson (SDF scheme for cooperatve spectrum sensng n cogntve rado network. 3. SYSE ODEL In Cogntve Rado Networks (CRNs, the detecton performance mght be vulnerable when the sensng decsons forwarded to a fuson centre through fadng channels. A SDF-based cooperatve spectrum sensng s used to mprove detecton relablty. Fgure shows a deployment of CRN usng cooperatve spectrum sensng. Fgure : SDF-base Cooperatve spectrum sensng n a CRN 5

4 Internatonal Journal of Computer Networks & Communcatons (IJCNC Vol.5, No., arch 3 As seen n Fgure, SUs are relayng ther ndvdual statstcal measurements of U avalablty to a common FC. he FC functons as a decson centre whch manages the CR network as well all assocated SUs. he use of a weghtng vector n the lnear soft fuson helps to elmnate the need for fndng optmal thresholds for ndvdual SU. Fgure also presents two progressve lnks, namely, prmary user-secondary user (U-SU lnk and secondary user-fuson centre (SU-FC lnk. he man operatons carred out on these two lnks are spectrum sensng and SDF, respectvely. 3. Characterzaton of rmary User-Secondary User (U-SU lnk SUs n the network are grouped nto multple clusters by some upper layer dstrbuted clusterng algorthms [5]. he SU performs local spectrum sensng ndvdually to detect U s presence. he sensng technque s formulated as bnary hypothess test. When U s absent o : X [n] W [n] ( When U s present : X [n] g S[n] +W [n] ( where X [n] s the receved sampled sgnal at the th SU recever, n,,, K, where K s the number of samples of the receved sgnal and t s defned as K s B where s s the sensng tme,,,,, where s the number of cooperatve SUs, g s the sensng channel gan between the U and the th SU whch accommodates for any channel effects such as multpath fadng, shadowng, and propagaton path loss, S[n] s the U transmtted sgnal whch s assumed to be ndependent and dentcally dstrbuted (..d. Gaussan random process wth zero mean and varance S,.e., S[n] ~N(, S, and W [n] s the th sensng channel nose whch s assumed to be addtve whte Gaussan wth zero mean and varance experencng..d. fadng effects,.e., W [n] ~ N(, W. All these varances are collected nto the vector W [ W, W,..., W ] and the sampled sgnals receved at the SUs are collected nto the vector X X, X,..., ]. he channel gans of the U-SU and SU-FC lnks, g and h, [ X W Fgure.Detaled system model of SDF -based cooperatve spectrum sensng 6

5 Internatonal Journal of Computer Networks & Communcatons (IJCNC Vol.5, No., arch 3 respectvely, are assumed to be constant over each sensng perod; ths can be justfed by the slow-fadng nature over these lnks where the delay requrement s short compared to the channel coherence tme consdered as quas-statc scenaro [6]. A detaled system model smplfed from [4] for the SDF-based cooperatve spectrum sensng s shown n Fgure. 3.. Characterzaton of Secondary User- Fuson Centre (SU-FC lnk In [4], the SDF process s ntalzed by notfyng the SUs to relay ther ndvdual measurements of Us sgnal, X, FC through a dedcated control channel n an orthogonal manner. In the paper the justfcaton of usng amplfy-and-forward (AAF nstead of the less complexty decode-and-forward (DAF scheme s mentoned to ts ablty to ncrease the detecton performance by employng some sgnal processng technques at the FC. he channel noses {N } of the SU-FC lnks are expected to be zero mean and spatally uncorrelated addtve whte Gaussan wth varances { } whch has been composed nto the vector [,,..., ] m. hen, the sgnal receved by correspondng FC from the th SU wll be Y [n] R h X [n] + N [n] (3 where R the transmt power of the th relay h s the ampltude channel gan of the SU- FC lnk. he use of AWGN model here s justfed by the slow-changng nature of the channels between the SUs and ther correspondng FC. Now, by consderng the two hypotheses n ( and (, the receved sgnal at FC can be expressed as Y [n o ] R h W [n]+ N [n] u [n] (4 Y [n ] R h g S[n] + R h W [n] + N [n] R h g S[n] + u [n] (5 whose statstcal propertes are Y [n o ] ~ N(, ~ N(, N(, ~ N(,, R g h S, +., R h W + and Y [n ] ~ In a atrx form, the receved sgnals at the FC through the control channel under and, respectvely, can be wrtten as R h Y[ n ] R gh Y[ n ] R R g h h W[ n] N[ n] + W [ n] N [ n] R h S[ n] u [ n] + S [ n] u [ ] n R g h (6 (7 7

6 Internatonal Journal of Computer Networks & Communcatons (IJCNC Vol.5, No., arch 3 At the correspondng FC of SUs, each receved sequence Y [n] wll be ndvdually averaged and squared usng a separate energy detector to estmate ts own energy as shown n Fgure. hus, the estmated energy collected by the th SU all the way to the FC can be expressed as K Z Y [ ] ;,,.., (8 n n By denotng {Z o, } {Z } and {Z, } {Z }, the two sets of test statstcs can be wrtten as Z Z, Z... Z ] and Z Z, Z... Z ]. ere we have consdered that for a large [,,, [,,, number of samples, the central lmt theorem (CL approxmates each test statstc nto the vectors Z and Z. he normally dstrbuted wth mean and varance gven can be expressed by E ( Z K, K ( R h W +, var( Z (9 4 K, K ( R h W + ( E ( Z K, K( R g h S +,, var( Z ( 4 K, K ( R g h S +, ( where [,,..., ],,, and [,,..., ],,,. Let us assume here that K R g h S and [,,..., ], then conclude that,, + or +. Next, from Fgure 4. that all the ndvdual test statstcs {Z } are multpled by weghtng coeffcent vector and used to lnearly formulate the resultant test statstc of the FC, Z, whch can be expressed as Z Z Z (3 where the weghtng coeffcents vector [,,..., ] ; satsfyng the condton; whch s used to optmze the detecton performance. Dfferent weght settngs wll be addressed later. Snce {Z } are all normal random varables, ther lnear combnaton, whch represent FC test statstc Z, wll also be normally dstrbuted wth statstcs gven by E( Z + K ( R h W K, + K ( R g h S K,,, E( Z var( Z var( Z K( R h W + K( R g h S +,, (4 (5 (6 (7 8

7 Internatonal Journal of Computer Networks & Communcatons (IJCNC Vol.5, No., arch 3 9 where the covarance matrces are 4, K and +, ( S R h g K. Consderng that the global threshold at the FC β, the lkelhood rato s Z β. As such, the overall probablty of detecton, d, and probablty of false alarm, f, for the SUs of FC can be wrtten as > var( ( ( f Z E Z Z (8 > var( ( ( d Z E Z Z (9 In CRNs, the probabltes of false alarm and detecton have unque ndcatons. Specfcally, (- d measures the probablty of nterference from SUs on the Us. On the other hand, f determnes an upper bound on the spectrum effcency, where a large f usually results n low spectrum utlzaton. hs s because the SU s allowed to perform transmssons f and only f the U s undetected under ether or. If the U s undetected under ether or, axmzng d by controllng the weghtng vector whle meetng a certan requrement on the f and vce versa. hen, for a gven f, d can be wrtten as ( f d ( Smlarly, for a gven d, f can be expressed as + ( d f ( 3.3. Conventonal SDF-based cooperatve spectrum sensng schemes ere some conventonal SDF optmzaton schemes for weghtng vector settng at Cs are dscussed as n [4]. hese are NDC, DC, RC and EGC. he EGC scheme s an exstng weghtng scheme that s smlar to the one used n systems wth multple receve antennas.. he ndvdual weghts assgned to the SUs sgnals at the FC n (8 and (9 are all equal and expressed by ( In RC, the weght coeffcent assgned for a partcular SU sgnal at a FC represents ts contrbuton to the overall decson made. hus, f a SU has a hgh U sgnal-to-nose rato (SNR at ts recever that may lead to a correct detecton on ts own, t should be assgned a larger weghtng coeffcent. For those SUs experencng deep fadng or shadowng, ther weghts are

8 Internatonal Journal of Computer Networks & Communcatons (IJCNC Vol.5, No., arch 3 decreased n order to reduce ther negatve contrbuton to the fnal decson. By mantanng ω, ndvdual weght for the th SU s measurement for RC as follows SNR (3 SNR where,,, N and SNR s the sgnal-to-nose rato at the C recever estmated for the th SU. he deflecton coeffcent (DC s a measure of the detecton performance as t s formulated based on the dstance between the centers of and. he DC based weght settng scheme can be realzed by maxmzng the normal DC or the modfed DC as shown below. NDC provdes a good measure of the detecton performance because the covarance matrx under hypothess s used to characterze the varance-normalzed dstance between the centers of the two condtonal DFs of Z j under and. o ensure and normalzng each weghtng co-effcent, we obtan the optmal weghtng vector * opt, NDC / (4 opt, NDC opt, NDC he maxmzaton of DC n order to fnd the optmal weghts settng for the SDF at the C. he DC can be defned to ensure and normalzng each weghtng co-effcent, we obtan the optmal weghtng vector 4. ROOSED SOLUION * opt, DC / (5 opt, DC opt, DC 4. roposed BGA based cooperatve spectrum sensng Genetc Algorthm (GA s classfed as an evolutonary algorthm that s a stochastc search method mmcs natural evoluton. GA s a knd of self-adaptve global searchng optmzaton algorthm. It has been used to solve dffcult problems lke, Non determnstc problems and machne learnng as well as also for smple programs lke evoluton of pctures and musc. he man advantage of GAs over the other methods s ther parallelsm. GAs travel n a search space that uses more ndvduals for the decson-makng and hence are less lkely to get stuck n a local extreme lke the other avalable decson-makng technques. It s a populaton-based n whch each ndvdual s evolved n parallel and the optmal ndvdual s preserved and obtanable from the last set of populaton. In general the genetc algorthm (GA mechansm starts wth randomly generatng a set of chromosomes. hese chromosomes consttute a populaton ( pops. As the genetc algorthm models natural processes such as selecton, crossover and mutaton, t performs on a populaton of ndvduals nstead of a sngle ndvdual. A random populaton of chromosomes wll be ntalzed and then wll be evaluated by ftness functons of a partcular problem. It wll then check for optmzaton crteron defned by the engneer and wll generate a new populaton from the prevous populatons f termnaton crteron s not met. hese new ndvduals are selected accordng to ther ftness values [7]. he chromosomes whch are consdered ft wll be selected 3

9 Internatonal Journal of Computer Networks & Communcatons (IJCNC Vol.5, No., arch 3 as parents and wll undergo matng wth crossover and mutaton for better producton. hese new offsprng wll then be evaluated and becomes parents to the new generaton f termnaton crteron s not satsfed. And ths cycle wll be looped untl a certan crteron s met, where each teraton s called a generaton. In our proposed BGA method, an ntal populaton of pops possble solutons s generated randomly and each ndvdual s normalzed to satsfy the constrants [3]. Our goal s to fnd the optmal set of weghtng vector values to maxmze detecton performance. When t reaches the predefned maxmum generaton, BGA s termnated and the weghted vector values that makes hghest ftness s consdered as the best soluton. Let us assume that we have SUs and Z, Z Z are the soft decsons of SU, SU.SU on the presence of Us, and j s the weghtng vector of the j th ndvdual that conssts of,.. hus, the ftness value for the j th ndvdual s defned as f j ( d j where j (6 he man operatons of the proposed BGA are selecton, crossover, and mutaton. For selecton, the dea s to choose the best chromosomes for reproducton through crossover and mutaton. Larger the ftness value better the soluton obtaned. In ths paper we use Roulette Wheel selecton method. he probablty of selectng the j th ndvdual or chromosome, p j, can be wrtten as f j p j (7 pops f j j he chromosomes wth maxmum probablty value wll be transferred to next generaton through eltsm operaton. After selecton process s done, the next step s crossover. he crossover starts wth parng to produce new offsprng. We use a unform random number generator to select the row numbers of chromosomes as mother ( ma or father ( pa whch are generated as macel(n*rand(, N/and pacel(n*rand(, N/, where cel rounds the value to the next hghest nteger.e. pops4, a random number generator generates followng two pars of random numbers: (.67,.85 and (.793,.3 4. hen the followng chromosomes are randomly selected for matng: ma [ 3] and pa [3 ]. In ths paper, for BGA we use double pont crossover, everythng between these two ponts s swapped between the parent chromosomes [7]. he BGA crossover s shown n Fgure 3. Fgure 3. BGA crossover operaton 3

10 Internatonal Journal of Computer Networks & Communcatons (IJCNC Vol.5, No., arch 3 Fgure 4. BGA utaton operaton Next step s mutaton operaton. For BGA the total no of varables that can be mutated are (mutaton rate* populaton* sze no of bts per varable. he row and column number s nomnated randomly and then the nomnated bt flps to correspondng bnary dgt wth sngle chromosome at a tme. Fgure 4 shows BGA mutaton operaton. he mutaton operaton actually helps to provde new search space. In concluson the proposed GA based optmzaton algorthm for cooperatve spectrum sensng proceeds as follows: Step : Set t and randomly generate a populaton of pops chromosomes each of whch s ( *nbts bts long, where s the number of secondary users n the network and nbts s the number of bts represent each chromosomes. Step : Decode each chromosome n the random populaton nto ts correspondng weghtng coeffcents vector Where the weghtng coeffcents vector [,,..., ] ; satsfyng the condton; whch s used to optmze the detecton performance. Step 3: Normalze the weghtng coeffcent vector dvdng [,,..., ] by ts -norm such that * (, the constrant * has to be satsfed. * Step 4: Cmpute the ftness value of every normalzed decoded weghtng vector, rank ther correspondng chromosomes accordng to ther ftness value and dentfy the best chromosomes pops*elte, where the elte s a parameter determnes a fracton of pops.e elte [, and. denotes floor operaton Step 5: Update tt+ and reproduce pops *( elte new chromosomes (canddate. solutons usng genetc algorthm operatons: selecton, crossover and mutaton where denotes celng operaton. Step 6: Construct a new set of populaton pops by concatenatng the newly pops* ( elte reproduced chromosomes wth the best pops*elte found n (t-. Step 7: Decode and normalze the chromosomes of the new populaton pops as n Step and Step 3 respectvely Step 8: Evaluate the ftness value of each chromosome as n Step 4 Step 9: If t s equal to predefned number of generaton(teratons ngener, stop. Otherwse go to Step 5 3

11 Internatonal Journal of Computer Networks & Communcatons (IJCNC Vol.5, No., arch 3 5. RESULS AND DISCUSSION 5. estng GA for Optmal Set of arameters he GA algorthm has been smulated wth dfferent values of same parameter to fnd out the optmal set of parameters. For smulaton the dfferent parameters has been used s mentoned n able. able : Dfferent GA parameters used for testng arameter name Rate used Bts per varable (bts [,4,6,8,] opulaton sze (pops [,, 3,4,5] crossover rate (c [.5,.65,.75,.85,.95] mutaton rate (p m [.,.,.5,.,.3,.6,.9] populaton for reproducton (rep [.5,.6,.7,.8,.9] And total number of realzatons averaged for tmes. And from smulaton results we found the optmal parameters for our CR problem are n able. able 3: Optmal set of GA parameters arameter name Optmal parameter value Bts per varable (bts opulaton sze (pops 5 Crossover rate (c.95 utaton rate (p m. opulaton for reproducton rate(rep.9 In ths paper, a gven probablty of false alarm f.5 has been used for further calculaton. 5. erformance of GA method In ths secton, we smulate the proposed GA method wth optmal set of parameters whch have obtaned from the prevous secton. Accordng to our paper n Fgure 3 ftness value defnes the probablty of detecton. It can be seen that proposed GA soluton converges and obtan maxmum achevable soluton that s. he parameters s used for BGA are no of generatons (gener, populaton sze (pops 5, secondary users ( 8, crossover rate ( c.95, mutaton rate pm., percentage of populaton for reproducton rep.9 and probablty of false alarm f.5. 33

12 Internatonal Journal of Computer Networks & Communcatons (IJCNC Vol.5, No., arch 3 Ftness rogress.95.9 Ftness axmum ftness Bnary GA ean ftness Bnary GA Number of generatons Fgure 5: erformance of GA method 5.3 erformance Comparson of roposed BGA ethod wth other conventonal method he proposed GA s compared wth conventonal NDC, DC, RC and EGC based SDF schemes as well as the OR-rule based hard decson fuson (DF. he default sensng tme and sensed bandwdth are set as s 5 us and B 6 z, respectvely. he relay transmt power s set to dbm and the channel gans of the U-SU and SU-FC are {g } and {h}, s normally dstrbuted but reman constant wthn each sensng nterval s, as s s suffcently small. {g } and {h }, s randomly-generated so that a low SNR envronment at SU and FC s realzed (SNR < - db. he smulaton results are obtaned from 5 realzatons of channel gans and nose varances. Fgure 4 shows the performance comparson of the proposed BGA-based method versus conventonal NDC, DC, RC and EGC based SDF schemes as well as the OR-rule based hard decson fuson (DF. he detecton performance s characterzed by the ROC curve whch s obtaned by plottng s the best detecton performance comparng to NDC, DC, RC and EGC based SDF schemes as well as the OR-rule based DF. 34

13 Internatonal Journal of Computer Networks & Communcatons (IJCNC Vol.5, No., arch 3.9 robablty of Detecton, d GA-asssted.3 SDF(NDC. SDF(DC SDF(RC. SDF(EGC DF(OR-Rule robablty of False Alarm, f Fgure 6: ROC performance comparson of cooperatve spectrum sensng usng SDF and DF schemes he OR-rule scheme, as expected, s nferor to all other methods as t suffers from a sgnfcant loss of nformaton content beng a DF process. he EGC SDF scheme shows better performance than the OR-rule DF scheme but t s nferor to all other SDF schemes due to ts fxed and equal weghtng coeffcents assgned to the energy measurements of the SUs at the correspondng FC. he RC-based scheme shows better performance than the EGC one due to ts adaptablty. he RC scheme assgns larger weghts for the SUs wth hgh SNRs and smaller weghts for those wth low SNRs and therefore, t controls the contrbutons of each SU n the overall decson taken at the FC stage. he NDC scheme outperforms the DC one wth nontrval dfference. he detecton performance of NDC s slghtly better than that of DC. It verfes that the computaton complexty of the proposed method meets real tme requrements of cogntve rado spectrum sensng. he proposed BGA soluton converges fast and acheves the optmum performance, whch satsfes that BGA-based method s qute stable. 6. CONCLUSION AND FUURE WORKS A BGA-based method s proposed to optmze the optmal weghtng vector requred for lnear SDF-based at a common fuson centre. he BGA control parameters were frst tested and nvestgated to fnd the best set sutable for the gven CR problem. In ths paper, proposed BGA demonstrate as fast and effcent resource allocaton algorthms to enable SUs to adapt CRN parameters n the rapdly changng envronment. It also verfes that the computaton complexty of the proposed method meets real tme requrements of cogntve rado spectrum optmzaton. and t outperforms conventonal SDF schemes. In ths paper, Neyman-earson Crteron s consdered where d s maxmzed for a gven f, and the optmal set of BGA parameters have been found usng set-and test approach. In future we wll 35

14 Internatonal Journal of Computer Networks & Communcatons (IJCNC Vol.5, No., arch 3 consder contnuous genetc algorthm (CGA and nmax crteron, where d and f are jontly optmzed, wll be taken nto account. We wll compare BGA and CGA for CRN envronment and best algorthm wll be used for further research. And then Double-GA engne wll be developed where the st GA wll be used to dynamcally fnd the optmal parameters whereas the nd GA wll use these optmal parameters to optmze the problem at hand. REFERENCES [] Federal Communcatons Commsson, Spectrum polcy task force report, FCC -55, Nov.. [] S. aykn, Cogntve rado: Bran-empowered wreless communcatons, IEEE Journal on Selected Areas n Communcatons, vo. 3, ssue, pp., 5. [3] Ghasem, A., and Sousa, E.S.:, Collaboratve spectrum sensng for opportunstc access n fadng envronments, n roc. Of IEEE DySAN 5, Baltmore, D, USA, 5, pp [4] Ganesan and Y. G. L, Cooperatve Spectrum Sensng n Cogntve Rado, art I: wo User Networks IEEE rans. on Wreless Communcatons, vo. 6, no. 6, 7. [5] Chunme, G.., Wang, J., Shaoqan, L., Weghted-clusterng Cooperatve Spectrum Sensng n Cogntve Rado Context, n roc. of Int. Conf. on Communcatons and oble Computng, vo., pp. -6, 9. [6] Z. Char,. K.Varshney, Optmal data fuson n multple sensor detecton systems, IEEE rans. Aerospace Electron. Syst. (January ( [7]. K. Varshney, Dstrbuted Detecton and Data Fuson, Sprnger, Sprnger, 997. [8] W. Zhang, R. K. allk, and K. B. Letaef, Cooperatve spectrum sensng optmzaton n cogntve rado networks, n roc. of IEEE Int. Conf. Communcatons, pp , 8. [9] Y. C. Lang, Y. Zeng, E. C. Y. eh, and A.. oang, Sensng throughput tradeoff for cogntve rado networks, IEEE rans. on Wreless Communcatons, vo. 7, pp , 8. [] Z. uan, Shuguang Cu, and Al. Sayed, Optmal Lnear Cooperaton for Spectrum Sensng n Cogntve Rado Networks, IEEE Journal of Selected opcs n Sgnal rocessng, vo., no., 8. [] B. Shen and K. S. Kwak, Soft Combnaton Schemes for Cooperatve Spectrum Sensng n Cogntve Rado Networks, ERI Journal, vo. 3, no. 3, 9 [] Jun a and Ye (Geoffrey L, Soft Combnaton and Detecton for Cooperatve Spectrum Sensng n Cogntve Rado Networks, n roc. of IEEE Global Communcatons Conference, pp , 7. [3] Kamran Arshad, uhammad Al Imran, and Klausoessner Collaboratve SpectrumSensng Optmsaton Algorthms for Cogntve Rado Networks, Centre for Communcaton Systems Research, Unversty of Surrey, Guldford GU 7X, UK. [4] Ayman A. El-Saleh, ahamod Ismal, ohd Alaudn ohd Al, and Israna. Arka, ybrd SDF-DF Cluster- Based Fuson Scheme for Cooperatve Spectrum Sensng n Cogntve Rado Networks, KSII RANSACIONS ON INERNE AND INFORAION SYSES VOL. 3, NO., December [5] O. Youns and S. Fahmy, Dstrbuted clusterng n ad hoc sensor networks: a hybrd energy-effcent approach, roc. IEEE INFOCO 4, pp , ong Kong, Chna, ar. 4. [6] Ekram ossan,vjay Bhargava, Cogntve Wreless Communcaton Networks, Sprnger, 7. [7] Randy L. aupt and Sue Ellen aupt, ractcal Genetc Algorthms, New Jersey: Wley, 4. Authors d. Kamal ossan completed hs B.Eng. degree n Electroncs majorng n elecommuncaton from ultmeda Unversty, alaysa,. Currently he s workng as Research offcer at ultmeda Unversty under elecom alaysa R&D grant towards hs masters degree. s current research nterests nclude general communcaton theores, cooperatve communcaton, moble & satellte communcatons, cogntve rados and wreless mesh networks. Ayman A. El-Saleh receved hs B.Eng. degree n Communcatons from Omar El-ukhtar Unversty (OU, Lbya, n 999, hs.sc. n croelectroncs Engneerng from Unverst Kebangsaan alaysa (UK, n 6, and hs hd n Cogntve Wrekess Communcatons from UK as well, n. In October 6, he joned the Faculty of Engneerng, ultmeda Unversty (U, at whch he s currently a Senor Lecturer teachng several telecommuncatons and electroncs courses. e s also a member of ICICE and IACSI nternatonal bodes. s research nterests nclude wreless communcatons, spectrum sensng technques, cogntve rado, soft computng usng genetc algorthm and partcle swarm optmzaton and FGA-based dgtal system desgn. 36