Two Stage spectrum Sensing for Cognitive Radio using CMME

Transcription

1 Two Stage spetrum Sensing or Cognitive Radio using CMME Naresh Gunihetty, S M Hiremath, Member, IEEE, S K Patra, Senior Member, IEEE Abstrat Spetrum sensing is undamental untionality in Cognitive Radio CR to identiy spetral white spaes or opportunisti ommuniation. Around last deade thorough study o spetrum sensing suggests that traditional energy detetion ED perormane is better and ast at high SNR and worst in low SNR, whereas ylo-stationary and Eigenvalue based methods perorm better at low SNR, but implementation omplexity is high. As suggested in IEEE standard or CR, we propose in this paper, novel two stage spetrum sensing based on energy detetion as oarse sensing irst stage and ombination o maximum-minimum eigen value based detetion tehnique CMME as ine sensing seond stage. The dual stage threshold parameter is designed to maximize the probability o detetion or given onstraints on the probability o alse alarm. Comparative analysis o single stage ED, Cylostationary detetion CSD, Combination o maximum-minimum eigen value detetion CMME, Combination o Cylostationary detetion and ED detetion and Combination o CMME and ED detetion is perormed using probability detetion versus SNR urve. Keywords- Cognitive radio, CMME, energy detetion, probability o detetion, Two-stage spetrum sensing I. INTRODUCTION The ineiient usage o the eletromagneti spetrum and the ever inreasing demand or more and more spetrum or wireless ommuniation has led to an extreme sarity o available spetrum [1]. Reent measurement analysis o spetrum oupany by various regulatory organisations in dierent ountries suggests that liensed spetrum is underutilized in dierent time and geographial loations [2]. To overome the problem o low spetrum utilization, Cognitive Radio CR onept given by Dr. Mitola has emerged as a good solution. CR by sensing and adapting to environmental onditions is able to detet or sni the ree spetrum availability in liensed band o primary user PU and intelligently alloate unliensed or seondary userssu without interering to PUs [1]. The ritial untionality o CR is to sense the wide spetral bandwidth ontinuously. Muh researh is arried on spetrum G.Naresh is with the Department o Eletronis and Communiation gnaresh446@gmail.om S M Hiremath is with the Department o Eletronis and Communiation hiremaths@nitrkl.a.in S K Patra is with the Department o Eletronis and Communiation skpatra@nitrkl.a.in sensing tehniques, sine various wireless tehnologies use dierent modulation tehniques, data rate and bandwidth [3], so diiult to detet. Many spetrum sensing tehniques have been proposed in literature, that are inluded in emerging IEEE Wireless Regional Area Network standard or CR are energy detetioned, ovariane based detetion as blind tehnique, ylostationary detetion tehnique CSD as semi-blind tehnique and Mathed ilter as non-blind tehnique [4] [5] [6]. ED is simple and ast tehnique, whih do not require prior inormation about PU and works better in high SNR. But ED is not robust at low SNR and unable to dierentiate between noise and signal. CSD is robust to low SNR, but requires that signal have statistial ylostationary property that are inherent in all modulated signals and its mathematial omplexity does not support hardware implementation [4]. Also sensing time that is ritial or CR is high in CSD. Covariane based tehniques depend on the presene o orrelation between deteted signals and these are blind tehniques among them max-min eigenvalue based detetion is more explored [7]. CMME requires large matrix manipulation, so sensing time required is bit high ompared to ED. So or eiient sensing IEEE standard preers two stage sensing that is oarse sensing whih overs large bandwidth and sensing time is low and ine sensing that onentrates on lower bandwidth and using very robust sensing tehniques like CSD or Eigenvalue based tehniques [6] [8]. In [9] disuss the two stage spetrum sensing in whih, both stages uses the ED and divides bandwidth into oarse resolution and ine resolution that depends on mean sensing time. But inreasing sensing time in ED does not have any eet under low SNR onditions. And [10] uses two stage sensing tehnique that uses ED in oarse sensing and CSD in ine sensing stage. In this paper, we propose a two-stage sensing approah based on energy detetion and eigen value based detetion. For the given hannel; energy detetion is perormed in the irst stage. I the energy is greater than the threshold λ, the hannel is delared as oupied. Otherwise, in the seond stage CMME detetion is perormed. I the deision metri exeeds the threshold γ 2 in the seond stage, the hannel is delared as oupied. Otherwise, it is delared as empty and available or seondary use. We are omparing with [10] two-stage spetrum sensing using ED and ylostationay approah based on the probability o detetion versus SNR. Also optimal thresholds λ and γ 2 or two stage plot based on mathematial analysis shown and analysed. The rest o the paper is organized as ollows. Following this introdution the remaining part o the paper is organized as ollows. Setion II desribes the two stage spetrum sensing by explaining energy detetion as oarse sensing and CMME as

2 ine sensing. Setion III presents the mathematial analysis or two stage sensing. Simulation results are presented in Setion IV. The observation rom Setion IV is shown in Setion V. Finally setion VI onludes this paper. II. TWO STAGE SPECTRUM SENSING SCHEME Analysis or two stage sensing taken quite similar to [10] with assuming that, there are L hannels to be sensed. In this sheme, the irst stage i.e., oarse sensing stage the hannel is tested by using energy detetion tehnique. I the deision in oarse sensing D is greater than the threshold λ, then the hannel is delared as oupied. Else the reeived signal is sensed by using the seond stage i.e., Fine sensing stage by using CMME tehnique. I the deision in ine sensing D is greater than the threshold γ 2, then the hannel is onsidered as oupied else it is empty. Proposed two stage sensing shown in Fig1: Fig. 1: Two-stage spetrum sensing sheme ED and CMME detetion tehnique A. Coarse Sensing In oarse sensing stage, the signal is sensed by using energy detetion tehnique [10]. The energy detetor aumulates the energy 2 o M samples and ompares it with the threshold λ to deide the presene or absene o primary user. Probability o presene o a primary user is denoted by H 1 and absene is denoted by H 0. The reeived signal in the irst stage is given as, x k where k = 1, 2,..., M, the primary users signal s k and reeiver noise n k. The noise is assumed as an i.i.d. Random Gaussian proess with variane σ 2 n and mean zero, while the signal is assumed to be an i.i.d. Random proess with variane and mean zero. OFDM is a key tehnology in most o the emerging tehnologies like WRAN, WiMAX, Long Term Evolution LTE, Digital Video Broadasting DVB et. [11]. So in our analysis we use DVB-T 2K mode OFDM signal as primary signal [11]. xk = { nk under H0 sk + nk under H 1 1 The deision rule or the energy detetor is given as, D = xk 2 > < M k=1 H 1 H 0 λ 2 The test statisti D or large M an be modeled by a Gaussian distribution as ollows [12] [10], { NM D σ 2 = n, σn 4 under H 0 3 NM σn 2 + σs, 2 σn 2 + σs 2 2 under H 1 The probability o alse alarm, P, and probability o detetion, Pd, or the energy detetor irst stage are: P λ M σn 2 = Q 4 σn 4 λ M σn 2 + σ 2 = Q s 5 σn 2 + σs 2 2 P d where Q a is the Q-untion. B. Fine Sensing I the ondition D > λ ails in the irst stage i.e., ED stage then the signal is again sensed by using CMME detetion tehnique. Sine as mentioned previously due to SNR limiting ondition, ED perormane is very poor irrespetive o inreasing sensing time as preerred in IEEE WRAN standard [6] [8]. Whereas CMME works better in low SNR with orrelated signals without any prior inormation about primary signal and hannel noise [7] [13] [14]. Assume that there are K 1 seondary users. Then the reeived signal at the i th seondary user is denoted by x i ki = 1, 2,..., K. Then the statistial matrix an be deined as: xk = [x 1 k, x 2 k,..., x K k] T sk = [s 1 k, s 2 k,..., s K k] T 6 nk = [n 1 k, n 2 k,..., n K k] T Where the reeived signal is given by x k, k = 1, 2,..., N where N is the number o samples in CMME tehnique. s k is the transmitted signal passed through a wireless hannel and n k is the additive white Gaussian noise AWGN with mean zero and variane σ 2 n. Aording to the above deinitions, 1 an be written as, x = s + n 7 Considering the statistial ovariane o the reeived signal, transmitted signal and noise signal as, R x = E xx T R s = E ss T R n = E nn T 8 Let us onsider N onseutive samples, then the statistial ovariane matries o the reeived signal, transmitted signal and noise signal beomes, R x N = 1 N xxt R s N = 1 N sst 9 R n N = 1 N nnt

3 Assuming that the noise is real. Let N N 2, AN = σ R n 2 n N, µ = 1 + K N 1/3. = 1 + K 1 N K Assume that K λmaxan µ lim N N = α 0 < α < 1. Then onverges with probability one to the Tray-Widom distribution o order 1 as mentioned in [7] [13] [15]. Assuming that the noise N 2, is omplex. Let AN = N σ R n 2 n N, µ = + K N 1/3. = + K 1 N + 1 K Assume that K lim N N λmaxan µ = α 0 < α < 1. Then onverges with probability one to the Tray-Widom distribution o order 2 [13]. When the parameter N is large then, µ and µ, and are nearly same, but their limit distribution is dierent [7]. From [14] whih provides details about the tables or Tray- Widom distribution untion that are alulated by numerial omputation. For example F = 0.45, F = 0.98, F = 2.02 λ max and λ min are the maximum and minimum eigen values o the reeived statistial ovariane matrix R x N. Algorithm [13] [6] P λ = P max > γ λ max λ min From above assumptions, we an get: P σ 2 = P N λ maxan > γ λ max λ min P λ 2 max AN > γ N K Let γ N K µ = 1 F 1 13 P d = Q γ 2 4N N 2 + K γ = N K λ max λ max λ min λ max λ max λ min 14 N 2/3 + K NK 1/6 F 1 1 The threshold 2 is: γ 2 = γ +1 γ = γ. 1 P 15 Thereore, the judgment rule or CMME detetion tehnique is: III. H 1 D λ max > = λ max λ min < γ 2 16 H 0 MATHEMATICAL ANALYSIS The overall probabilities o alse alarm and detetion or a single hannel are given by [10], P = P 1 P P 17 P d = P d 1 P d P d 18 The objetive is to design a deision strategy determination o λ and γ 2 in order to maximize the probability o detetion o eah hannel subjet to a alse alarm rate onstraint. Thereore the orresponding problem is given by, max P dλ, γ 2 19 λ,γ 2 s.t.p β The inequality onstraint in the problem 19 an be redued to an equality onstraint by using the theorem in [10]. Theorem: The optimal value o the probability o detetion in 19 is attained by P = β. Proo or the ollowing is analysed similar to the [10] sine P d onsists o Tray-widom distribution untion whih is similar to Gaussian untion dierentiable and Q-untion dierentiable. Hene P d is dierentiable and dereasing untion o the thresholds λ and γ 2. Thus, the irst derivative o P d w.r.t λ and γ 2 is negative. Hene, the maximum P d is attained or the lowest possible λ and γ 2. Same analysis holds or P. Assuming λ, γ2 to be the optimal solution o 19 orresponding to P < β. With keeping either o them λ or γ2 onstant and varying one o them produe the better solution than as assumed. Thereore, λ, γ2 annot be the optimal solution o the problem. Thus, the optimal P d is attained by P = β. Hene, it is rewritten as similar in [10], max P dλ, γ 2 20 λ,γ 2 Solving 17, β = P s.t.p = β Substituting 4 and 13 in 21, we get, β 1 F 1 λ = γ 2 = Q F 1 sine γ = 1 γ P P 21 γ N K 2 µ γ N K 2 µ σ 4 n + M σ 2 n 22

4 Thereore, the problem 20 an be simpliied to an unonstrained problem as ollows : max γ 2 P d γ 2, γ 2 23 The optimal γ 2 and λ = γ 2 an be obtained rom 23 and 22. So the problem is unimodal in γ 2 and the optimal value o γ 2 is alulated at eah SNR whih gives the maximum probability o detetion P d. The inal optimal equation P d whih is the untion o γ 2 an be evaluated in MATLAB. IV. SIMULATION RESULTS In these simulations, OFDM signal 2k mode o DVB-T standard is taken or the analysis. The transmitted OFDM signal is organised in rames. Eah rame onsists o 68 OFDM symbols with 1705 sub-arriers in 2k mode and transmitted with a symbol duration o T s =244 µse [16]. Fig. 3: Probability o Detetion versus SNR Fig. 2: Probability o Detetion versus γ 2 From Fig. 2 we observe that probability o detetion variation w.r.t γ 2 or dierent values o β or two-stage sensing. It is lear rom the Fig. 2 that the maximum probability o detetion is attained when the probability o alse alarm satisies onstraint 19 with equality as shown in Theorem. Fig. 3 shows the perormane o probability o detetionp d vs SNR o all the three spetrum sensing tehniques. From Fig. 3, we observe that or the SNR less than -24dB the two-stage spetrum sensing sheme perorms better than the Energy detetion tehnique or CMME detetion tehnique. From Fig.4, we an observe the probability o detetion P d vs SNR o the spetrum sensing tehniques Energy Detetion ED, Cylostationary based detetion tehnique, CMME detetion, Two-stage spetrum sensing using ED and Cylostationary based detetion tehnique [10], Two-stage spetrum sensing using ED and CMME. As we an see that the probability o detetion P d o Two-stage spetrum sensing using ED and CMME giving better perormane ompared Fig. 4: Probability o Detetion versus SNR to Two-stage spetrum sensing using ED and Cylostationary based detetion tehnique. On all the three sensing shemes it is assumed that the Probability o alse alarm onstraint, β and SNR is same or all the hannels. V. OBSERVATION From TABLE I we an observe that the probability o detetion P d is better or Two-stage spetrum sensing shemes ompared to individual detetion tehniques in both the ases. The probability o detetion P d perormane o two-stage spetrum sensing using ED and CMME is better ompared to two-stage spetrum sensing using ED and CSD [10].

5 SNR db ED P d CSD P d CMME P d Two-stage spe.sen. ED-Cylo P d Two-stage spe.sen. ED- CMME P d TABLE I: Comparison o Prob. o detetion P d o spe.sen. teh s with SNR VI. CONCLUSION Two-stage sensing shemes are analysed in terms o its detetion perormane. In partiular, we designed optimum thresholds or the energy detetion, ylostationary based detetion and ombination o maximum-minimum eigen value detetion CMME stages so as to maximize the probability o detetion with onstraints on the probability o alse alarm. We onsidered DVB-T 2k mode OFDM signal or our analysis and ompared our two-stage sensing shemes and observed the probability o detetion perormane P d. We observed that at low SNR, where the energy detetor s perormane is poor, the two-stage sensing shemes provides better detetion. Perormane analysis o two stage spetrum sensing tehnique based on timing need to be arried out and urther CMME an be used to estimate the noise variane and ed it bak to ED to enhane the perormane o oarse sensing. Indiretly making the dual stage as ull blind and sel-adaptive. Also, detetion perormane needs to be heked using real-time measured data. REFERENCES [1] H. Arslan, Cognitive radio, sotware deined radio, and adaptive wireless systems. Springer London, Limited, [2] C. Cordeiro, K. Challapali, D. Birru, S. Shankar N et al., Ieee : An introdution to the irst wireless standard based on ognitive radios, Journal o ommuniations, vol. 1, no. 1, pp , [3] E. Ghasemi, A.and Sousa, Spetrum sensing in ognitive radio networks: Requirements, hallenges and design trade-os, IEEE Communiations Magazine, vol. 46, no. 4, pp , [4] E. Axell, G. Leus, E. Larsson, and H. Poor, Spetrum sensing or ognitive radio: State-o-the-art and reent advanes, Signal Proessing Magazine, IEEE, vol. 29, no. 3, pp , [5] H. ArslanEd., Ed., Cognitive Radio, Sotware Deined Radio, and Adaptive wireless Systems. Springer, [6] IEEE standard or inormation tehnology teleommuniations and inormation exhange between systems wireless regional area networkswran speii requirements part 22: Cognitive wireless ran medium aess ontrol ma and physial layer phy speiiations: Poliies and proedures or operation in the tv bands, IEEE Std , pp , [7] F. Liu, H. Chen, L. Xie, and K. Wang, Maximum-minimum eigenvalue detetion-based method to mitigate the eet o the puea in ognitive radio networks, in Wireless Communiations and Signal Proessing WCSP, 2011 International Conerene on. IEEE, 2011, pp [8] A. Mody and G. Chouinard, Ieee wireless regional area networks: Enabling rural broadband wireless aess using ognitive radio tehnology do.: Ieee /0073r03, Teh. Rep., June 2010.[Online]. Available: ieee802. org/22, Teh. Rep. [9] L. Luo, N. M. Neihart, S. Roy, and D. J. Allstot, A two-stage sensing tehnique or dynami spetrum aess, Wireless Communiations, IEEE Transations on, vol. 8, no. 6, pp , [10] S. Maleki, A. Pandharipande, and G. Leus, Two-stage spetrum sensing or ognitive radios, in Aoustis Speeh and Signal Proessing ICASSP, 2010 IEEE International Conerene on. IEEE, 2010, pp [11] S. Chaudhari, J. Lundén, and V. Koivunen, Collaborative autoorrelation-based spetrum sensing o odm signals in ognitive radios, in Inormation Sienes and Systems, CISS nd Annual Conerene on. IEEE, 2008, pp [12] D. Cabri, A. Tkahenko, and R. W. Brodersen, Spetrum sensing measurements o pilot, energy, and ollaborative detetion, in Military Communiations Conerene, MILCOM IEEE. IEEE, 2006, pp [13] H. Liu and W. Chen, A robust detetion algorithm based on maximumminimum eigenvalue or ognitive radio, in Wireless Communiations, Networking and Mobile Computing WiCOM, th International Conerene on. IEEE, 2012, pp [14] Y. Zeng, C. L. Koh, and Y.-C. Liang, Maximum eigenvalue detetion: theory and appliation, in Communiations, ICC 08. IEEE International Conerene on. IEEE, 2008, pp [15] M. Yang, J. An, X. Bu, and L. Sun, An improved eigenvalue-based algorithm or ooperative spetrum sensing, in Wireless Communiations Networking and Mobile Computing WiCOM, th International Conerene on. IEEE, 2010, pp [16] M.-G. Di Benedetto and F. Bader, Cognitive Communiation and Cooperative HetNet Coexistene. Springer, 2013.