MATCHED FILTER BASED SPECTRUM SENSING FOR COGNITIVE RADIO AT LOW SIGNAL TO NOISE RATIO

Transcription

1 Journal of heoretcal and Appled Informaton echnology th Aprl 4. Vol. 6 No. 5-4 JAI & LLS. All rghts reserved. ISSN: E-ISSN: MACED FILER BASED SPECRUM SENSING FOR COGNIIVE RADIO A LOW SIGNAL O NOISE RAIO R. VADIVELU, K. SANKARANARAYANAN, 3 V. VIJAYAKUMARI Assstant Professor, Department of Electroncs and Communcaton Engneerng, Sr Krshna College of echnology, Combatore, Inda Dean, PG Studes, Department of Computer Scence and Engneerng Sr Ramakrshna Insttute of echnology, Combatore, Inda 3 Assocate Professor, Department of Electroncs and Communcaton Engneerng Sr Krshna College of echnology, Combatore, Inda E-mal: vadveluece@gmal.com, kkdcbesankar@gmal.com, 3 ebnvj@gmal.com ABSRAC Custom usage of a Cogntve Rado s admnstrated by the essental utlzaton aspect of the rado spectrum the natural resource. Cogntve rado tryng to resourcefully share the rado spectrum along wth potental prmary users n the spectrum that must be dentfed n order to evade causng harmful nterference wth other users on the spectrum. he vbrant usage of spectrum belongs to the whte space assessment and how accurately t can be utlzed. In ths paper we put forward an open stuaton of channel estmaton at low sgnal to nose rato. A Matched Flter based system s well-thought-out to make the spectrum sensng resoluton based on the observed sgnal to nose rato from the Cogntve Users. Wth the exstng knowledge of the regulated system parameters, the fuson Centre can make a global sensng decson consstently wthout any addtonal requrements such as channel state nformaton, pror nformaton and pror prospects about the prmary user's sgnal. Numercal results n terms of recever operatng characterstcs show that the sensng performance of the proposed Matched flter based system outperforms the performance of the adaptve akag and Sugeno s fuzzy energy based system model at low Sgnal to Nose Rato and one order Cyclostatonary detecton based on estmated sgnal to nose rato. Keywords: Cogntve Rado, Spectrum ole, Sgnal to Nose Rato, Matched Flter, Spectrum Sensng, Channel State Informaton, Recever Operatng Characterstcs. INRODUCION Current and Innovatve wreless communcaton devces are profcent to offer hgher data rates and nnovatve servces to end users, extensve rado spectrum s avalable for several wreless communcaton servces. On the other hand wth the exponental growth n wreless communcaton devces and ther usage, the rado spectrum s becomng a scarce resource [][]. Cogntve Rado (CR) has been dentfed as a new desgn technque wshes to enrch the profcent utlzaton of electromagnetc rado spectrum by empowerng dynamc spectrum access (DSA) for the current and next-generaton wreless communcaton technology. he stmulus for the desgn of CR communcaton system arses from the fact that the ample portons of the authorzed or lcensed rado spectrum are underutlzed by the prmary users or lcensed users. As per Federal Communcatons Commttee (FCC) about 5-85 % of the spectrum s assessed to be underutlzed [3]. hs lays the strong foundaton for the secondary user or CR user (CU) permtted to access a spectrum band that s unoccuped by the prmary user at a specfc tme and geographc locaton [4]. he spectrum hole or whte space s the frequency band that has been allocated to a PU who s not usng the spectrum at the specfed allotted tme. Opportunstc Spectrum Access (OSA) by the CU depends on how effcently and relably the spectrum s sensed and utlzed by the CU. Moreover, a perodc spectrum sensng s the basc requrement that a CU to transmt data over the spectrum, to overcome the nterference wth the PU [5]. Spectrum can be sensed by varous technques.e., matched flter, energy detecton, Cyclostatonary feature detecton and stochastc process technques to detect the presence of the PU sgnal n the channel or sub channel. Cooperatve spectrum sensng used to estmate the optmal number of CU s nvolved n spectrum sensng. An energy effcent technque for mnmzng the number of CU s subject to the constrants on the probablty of false alarm (PF) and probablty of detecton (PD) are dealt n [6]. In [7] a new cooperatve spectrum sens- 7

2 Journal of heoretcal and Appled Informaton echnology th Aprl 4. Vol. 6 No. 5-4 JAI & LLS. All rghts reserved. ISSN: E-ISSN: ng algorthm was proposed to mprove the bandwdth problem of reportng channels, only the cogntve users wth relable nformaton are allowed to report ther sensng results. Under user less relablty condton, the cogntve user wth hghest reputaton wll report ts sensng result to the fuson center. In Data throughput optmzaton scenaro, the throughput of the CU s network s maxmzed subject to a constrant on the global probablty of detecton (PD) n order to determne the optmal number of CU s dealt n [8]. o avod nterference to the lcensed user by the CU, spectrum sensng process or algorthms must be more accurate and should be hghly relable. he frmness on whether the sgnal s present or absent on the channel can be expedted f we pass the sgnal through a flter that wll accentuate the useful sgnal and suppress the nose sgnal. A Matched Flter wll peak out the sgnal component at some nstant of tme and suppress the nose ampltude at the same tme. If Sgnal s present on the channel, a large peak at ths nstant wll occur and f the sgnal s absent, no such peak wll appear. hs prearrangement wll make t possble to decde whether the sgnal s present or not n the channel. A matched flter detecton technque s the optmal lnear flter used to maxmze the sgnal to nose rato (SNR) n the presence of addtve whte Gaussan nose. Cooperaton among CUs are establshed to estmate the PU s presence or absence, fuson Centre (FC) s used to take the overall resoluton about the PU s. In [9], Quan et al. projected an optmal lnear collaboraton framework for sensng the spectrum n order to exactly detect the weak prmary user sgnal n the spectrum band or sub band. owever, the weakness of algorthm mplementaton n [9] s the overall knowledge of sgnal to nose rato (SNR) and the nose varance of the PU sgnal should be known at the FC durng the spectrum sensng process. In [] a fuzzy nference system was proposed by Keu-Xuan et al., assumng the SNR of the PU s known to the CU whch provdes an advantage of local soft spectrum sensng decson made at CUs termnal. Results n [] shows that the sensng performance of the proposed scheme s comparable wth the sensng performance of the maxmal-rato combnaton (MRC) based scheme whch does not requre SNR of the PU sgnal from CUs to the FC. In real tme t s very challengng for a CU to precsely estmate SNR of the PU n a gven spectrum band or sub band n a non-cooperatve envronment. It s apparent that most of the contemporary cooperatve spectrum sensng schemes makes a speculaton that the SNR of the PU sgnal at the CU s perfectly known. Moreover, the CUs can estmate these parameters well but t s very dffcult to communcate them along wth local observatons to the FC. In [] Keu-Xuan et al. assumed each CU n the CR network estmated the energy of the receved sgnal n the gven band or sub band of nterest and then transmtted the expermental parameters to the FC. Data fuson at the FC s accomplshed by usng an adaptve akag and Sugeno s fuzzy system where fuzzfcaton parameters are adapted from receved data through a Kalman flter. In ths paper we frst estmated the SNR of each CU n the CR network and then t s transmtted along wth the parameters of the receved sgnal n the gven band or sub band of nterest to the FC. Data fuson s performed at the FC by an adaptve Bayesan system where SNR are adapted from receved data through a Kalman flter. It means that the detecton problem and the estmaton problem are solved at the FC concurrently and cooperatvely. herefore, the FC can make a global decson based on local observed SNR n an addtve whte Gaussan nose envronment of PU sgnal at CUs. In [] Waleed Ejaz et.al proposed one order Cyclostatonary detecton wth estmated SNR n terms of reduced detecton tme. he rest of the paper s organzed as: Secton descrbes the system model for adaptve cooperatve spectrum sensng problem. A matched flter detecton overvew s dealt n Secton 3. Secton 4 derves Detecton of prmary user sgnal n addtve whte Gaussan channel. Numercal result analyss s presented n Secton 5. Fnally Secton 6 concludes the paper.. SYSEM MODEL FOR COOPERAIVE SPECRUM SENSING he basc problem of exstence based spectrum sensng s to dfferentate between two postulates.e., based on presence or absence of the prmary user: that the possbly faded prmary user sgnal s present at a suffcently hgh power level or t s absent. he detector that t can relably detect the weakest sgnal present over the channel s ts senstvty. he spectrum sensng problem can be formulated based upon the appearance or nonappearance of PU n the concerned band or sub band based on bnary hypothess testng model [3] as Pr maryuserabsent () Pr maryuser Pr esent Any detecton scheme can be wrtten as a possbly random functon F: RN {, }, where F maps 8

3 Journal of heoretcal and Appled Informaton echnology th Aprl 4. Vol. 6 No. 5-4 JAI & LLS. All rghts reserved. ISSN: E-ISSN: the N dmensonal receved vector y = (y[], y[],y[3],, y[n]) onto the set {, }. ere stands for the decson that the receved sgnal s only nose and stands for the decson that the receved sgnal s sgnal plus nose. Consderng a sngle FC wth N number of CUs scattered across a gven CR network. he receved sgnal at each CU based on the appearance or non-appearance of PU s gven by y (t) = : n (t) () y (t) = : h (t)x (t) + n (t) Where the receved sgnal at the th CU s represented by y (t) and gan of the channel between the PU and the th CU represented by h(t). he sgnal transferred by the PU s represented by x (t) and the addtve whte Gaussan nose (AWGN) at the th CU represented by n (t). In addton to above consderatons we assume that the channels correspondng to dfferent CUs are assumed to be ndependent and dentcally dstrbuted, and the CUs and the PU share a common spectrum of concerned band or sub band. Fgure : Sensng n Cooperatve Envronment Cooperatve spectrum sensng envronment s shown n Fgure. For a gven sequence of sensng, the CU estmates the SNR of ts receved sgnal n the AWGN channel. he SNR observed from the CUs are then communcated to the overall FC through the control channel for fnal decson. Fnally, the FC coordnates wth the observatons of all the CUs and ther observed SNR to make a fnal decson about the appearance or non-appearance of the PU sgnal. Our prmary goal s to detect the PU sgnal n the gven AWGN channel and that s never seamlessly known n the real tme spectrum sensng, hence we assumed that the PU sgnal s present n the AWGN channel and by maxmzng the SNR we detect the presence or absence of PU sgnal usng optmal flter based on fxed threshold [4]. 3. MACED FILER DEECION o enrch the SNR a matched flter s often used at the recever front end. Matched flter coeffcents are bascally gven by the complex conjugated reversed sgnal samples n terms of dscrete sgnals. wo types of coherent or non-coherent recevers are used based on sgnal analyss ether as complex sgnals or noses. If the ampltude and phase of the receved sgnal are known coherent recevers are used results n a perfect match between the matched flter coeffcents and the sgnals. In case of a noncoherent recever, the receved sgnal s modeled as a replca of the orgnal sgnal wth a random phase error. Wth a noncoherent recever the detecton after the matched flter s generally based on the power or magntude of the sgnal snce we need both real and magnary parts to defne the sgnal entrely [5]. Power Spectral Densty (PSD) of the AWGN sgnals s gven N PSD AWGN (f ) = (3) Where N s the nose sgnal and AWGN channel Sgnal to Nose Power measured at the output of the matched flter s gven by S(t) SNR = (4) N(t) he output nose power P n calculated of the n th prmary user s found to be [6] + N Pn = n (f ) df (5) he output sgnal power P s calculated of the n th prmary user s found to be + ejπft Ps n (f )S (f ) df = (6) Output sgnal power P s s decomposed n terms of nput sgnal power S usng Schwartz nequalty + s n s P = (f )df P (7) Now the SNR of the prmary user s smplfed to SNR o Ps SNR o = (8) N he above equaton represents the PU sgnal over the nose. 9

4 Journal of heoretcal and Appled Informaton echnology th Aprl 4. Vol. 6 No. 5-4 JAI & LLS. All rghts reserved. ISSN: E-ISSN: DEECION OF PRIMARY USER SIG- NAL IN ADDIIVE WIE GAUSSIAN NOISE CANNEL he reason behnd ntroducng Matched Flter mplementaton s to model the evoluton of PU sgnal n the spectrum band consdered over tme by measurements usng CR. In ths secton, we consder the basc functonal model of lnear matched flter. Prmarly matched flter mplementaton s best sutable for radar, sonar, wreless communcaton systems, Intellgent Rado Systems and bnary detecton of AWGN channel as shown n Fgure. Fgure : Implementaton of Matched Flter Bnary detecton problem s used to dentfy the state of PU sgnal presence or absence n AWGN channel n the tme nterval of t. he bnary hypothess specfed by and ndcates the presence and absence of PU sgnal [7] n the channel consdered. : Y(t) = N(t) : Y(t) = s(t) + N(t) (9) ere N(t) s the AWGN wth zero mean and covarance of σ δ(t-s), where σ s the power densty of ntensty of the AWGN sgnal. In order to detect the PU sgnal an orthonormal bass functon n sgnal space {Φ (t), ϵ I} of space S [,] ntegrable over the functon [,]. As the frst element of the bass, we select the functon as s(t) Φ (t)= () Where E E = s = s dt () If we dentfed orthonormal bass {Φ (t), ϵ I}, usng Karhunen-Loeve decomposton [8] AWGN nose N(t) decomposed to () N = N(t) Φ (t)dt Smlarly the receved sgnal Y(t) can be decomposed to (3) Y = Y(t) Φ (t)dt Under sgnal space doman, the detecton problem reduces to one dmensonal problem as : Y = N = + : Y E N (4) Now the Bnary ypothess theory n terms of Generalzed Lkelhood Rato est (GLR) [9] statstcs establshed as exp( (y E ) ) L(y ) = σ τ (5) < exp( y ) σ Where τ denotes the threshold and takng logarthms and reorganzng the resultng dentty gves E σ Y Ω + ln( τ) (6) < E ence the optmum flter expressed n terms of suffcent statstcs S as S = Y(t)s(t)dt (7) E Under bnary hypothess t s observed that Y~N(, σ) based on and Y~N(E/, σ) based on. If r denotes the dstance between the two hypotheses measured n terms of nose standard devaton. Dfferent types of responses based on old stmulus to a correct response, called detecton whereas a yes response to a new stmulus s a mss, called a false alarm (FA). A No response gven to a new stmulus s a true response, called a Correct Rejecton whereas a No response to an old stmulus s a false response called a Mss or a false alarm (FA). hese four types of reacton can be organzed as shown n able.

5 Journal of heoretcal and Appled Informaton echnology th Aprl 4. Vol. 6 No. 5-4 JAI & LLS. All rghts reserved. ISSN: E-ISSN: able : he four possble types of reacton Decsons (Reactons) Realty Yes No Sgnal present Detecton Mss Detecton terms of ts ROC curve s evaluated usng Monte- Carlo smulatons. It s assumed that the PU sgnal s lkely-equally Bnary Phase Shft Keyng (BPSK) sgnal [9] wth pror probabltes Pr {} = Pr {} =.5 and the noses at CUs are AWGN wth zero mean and unt varance. Sgnal absent False Alarm Rejecton It s shown that the probablty of detecton PD and Probablty of false alarm PF for the test are gven by r ln( τ) PD = Q( ) (8) r r ln( τ) PF = Q( ) (9) r By usng Neyman-Pearson (NP) test and elmnatng the threshold τ from the above denttes we acheve PD Q(r Q (P F )) = () Usng unnormalsed statcs we obtan the suffcent statc S as E γ + σ τ () < S ln( ) 5. RESUL AND DISCUSSION In order to ncrease the performance of spectrum sensng, we allow varous SU to cooperate by sharng ther nformaton and to reduce the communcaton overheads, users share ther decson statstcs based on the bnary hypothess testng. In sgnal estmaton theory, a recever operatng characterstc (ROC) curve s a graphcal plot whch demonstrates the performance of a bnary classfer system as wth threshold varaton. It s created by plottng the probablty of detecton (PD) vs. Probablty of False alarm (PF), at varous threshold values. In general, f both of the probablty dstrbutons for detecton and false alarm are known, the ROC curve can be generated by plottng the Cumulatve Dstrbuton Functon of the detecton probablty n the y-axs versus the Cumulatve Dstrbuton Functon of the false alarm probablty n x-axs. he sensng performance of the proposed scheme, n Fgure 3: Recever Operatng Characterstcs Curve For 3 Db SNR Under the above assumptons the ROC curve of the proposed scheme s compared wth [], namely the Kalman flter based adaptve fuzzy system scheme and wth [] one order Cyclostatonary detecton based on estmated sgnal to nose rato. A comparson of these results s presented n Fgure 3 and Fgure 4 for dfferent values of lower SNRs. he proposed scheme outperforms the Kalman flter based adaptve fuzzy system scheme and one order Cyclostatonary detecton scheme. o mplement the proposed scheme, the SNR at reduced rate s consdered normally at 3 db and 6 db and also n the proposed scheme each CU must send the SNR of the PU sgnal to the FC.

6 Journal of heoretcal and Appled Informaton echnology th Aprl 4. Vol. 6 No. 5-4 JAI & LLS. All rghts reserved. ISSN: E-ISSN: Fgure 4: Recever Operatng Characterstcs Curve For 6 Db SNR 6. CONCLUSSION In order to sense the spectrum holes consstently and resourcefully, n ths paper we propose a matched flter based cooperatve spectrum sensng n CR networks. he advantage of the proposed scheme comes from the fact that t can work wth very low SNR wth the knowledge of the PU sgnal, the pror probablty of the PU actvty, and SNRs of the PU sgnal at cogntve rado termnals. Monte-Carlo smulaton results based on ROC curve show that the sensng performance of the proposed scheme outperforms the performance of Kalman flter based adaptve fuzzy system scheme and one order Cyclostatonary detecton scheme. he only lmtaton of the proposed scheme s we should have the pror knowledge about the PU sgnal before sensng the channel. he choce of matched flter detecton technque proposed here to estmate the channel n the lower SNR regme. Consequently, fndng lower bound of mn τ (hreshold) and upper bound of max τ s stll an open ssue. Future work s n progress n ths drecton. REFRENCES: [] S. aykn, "Cogntve Rado: Bran-Empowered Wreless Communcatons", IEEE Journal on Selected Areas n Communcatons, vol. 3, 5, pp. -. [] K. Yau, P. Komsarczuk and P.D. eal, "Cogntve Rado-based Wreless Sensor Networks: Conceptual Desgn and Open Issues", IEEE 34th Conference on Local Computer Network, 9, pp [3] Federal Communcatons Commsson (FCC), Spectrum Polcy ask Force, E Docket no.,, pp [4] R.W. Broderson, A. Wolsz, D. Cabrc, S.M. Mshra and D. Wllkomm, "CORVUS: A Cogntve Rado Approach for Usage of Vrtual Unlcensed Spectrum", Whte Paper, Unversty of Calforna, Berkeley, echncal Report, 4. [5] A. Ghasem and E.S. Sousa, "Collaboratve Spectrum Sensng for Opportunstc Access n Fadng Envronments", I st IEEE Symposum on New Fronters n Dynamc Spectrum Access Network, 5, pp [6] F. F. Dgham, M.S. Aloun and M.K. Smon, "On the Energy Detecton of Unknown Sgnals Over Fadng Channels", IEEE Internatonal Conference on Communcatons, vol. 5, 3, pp [7] Le Zhang and Shuquan Xa, "A New Cooperatve Spectrum Sensng Algorthm for Cogntve Rado Networks", Proceedngs of ISECS Internatonal Colloquum on Computng, Communcaton, Control, and Management, vol., 9, pp. 7-. [8] Sna Malek, Sundeep Prabhakar Chepur and Geert Leus, "Energy and hroughput Effcent Strateges for Cooperatve Spectrum Sensng n Cogntve Rados", IEEE th Internatonal Workshop on Sgnal Processng Advances n Wreless Communcatons,, pp [9] Z. Quan, S. Cu and A.. Sayed, "Optmal Lnear Cooperaton for Spectrum Sensng n Cogntve Rado Networks", IEEE Journal of Selected opcs n Sgnal Processng, vol., 8, pp []. Keu-Xuan and I. Koo, "A Cooperatve Spectrum Sensng Usng Fuzzy Logc for Cogntve Rado Networks", KSII ransactons on Internet and Informaton Systems, vol. 4,, pp []. Keu-Xuan and I. Koo, "Cooperatve Spectrum Sensng usng Kalman Flter based Adaptve Fuzzy System for Cogntve Rado Networks", KSII ransactons on Internet and Informaton Systems, vol. 6,, pp [] Waleed Ejaz, Najam Ul asan and yung Seok Km, "SNR Based Adaptve Spectrum Sensng for Cogntve Rado Networks", Internatonal Journal of Innovatve Computng, Informaton and Control, vol. 8,, pp

7 Journal of heoretcal and Appled Informaton echnology th Aprl 4. Vol. 6 No. 5-4 JAI & LLS. All rghts reserved. ISSN: E-ISSN: [3] Rahul andra and Anant Saha, "SNR Walls for Sgnal Detecton", IEEE Journal of Selected opcs n Sgnal Processng, vol., 8, pp [4] Pu Wang, ongbn L and Braham med, "Bayesan Parametrc Approach for Multchannel Adaptve Sgnal Detecton", IEEE Radar Conference, Washngton DC,, pp [5] A. Eftekhar, A. J. Romberg and M.B. Wakn, "Matched Flterng from Lmted Frequency Samples", IEEE ransactons on Informaton heory, vol. 59, 3, pp [6] F. Challan, C. Fraschn and P. Courmontagne, "Stochastc Matched Flterng Method Appled to SAS Imagery", Proceedngs of OCEANS 5, vol., 5, pp [7] Rongfe Fan and a Jang, "Optmal Mult Channel Cooperatve Sensng n Cogntve Rado Networks", IEEE ransactons on Wreless Communcatons, vol. 9,, pp [8] P. A. Regala, "An Adaptve Unt Norm Flter wth Applcatons to Sgnal Analyss and Karhunen-Loeve ransformatons", IEEE ransactons on Crcuts and Systems, vol. 37, 99, pp [9] John Proaks and Masoud Saleh, Dgtal Communcatons, McGraw-ll, 7. [] Peter J. Schreer, Lous L. Scharf and Alfred anssen, "A Generalzed Lkelhood Rato est for Improprety of Complex Sgnals", IEEE Sgnal Processng Letters, vol. 3, 6, pp