Full Paper Proc. of Int. Con! on Advances in Recent Technologies in Communication and Computing 2011

Transcription

1 Proc. of nt. Con! on Advances in Recent Technologies in Communication and Computing 2011 FPGA REALZATON OF SPECTRUM SENSNG BASED ON BAYESAN FRAMEWORK FOR COGNTVE RADO NETWORK S. Srinu School of Physics, University of Hyderabad, Hyderabad , ndia Samrat L. Sabat School of Physics, University of Hyderabad, Hyderabad , ndia Abstract- n this paper, spectrum sensing is based on Energy detection in the frequency domain with Bayesian criterion. The sensing performance is analyzed using Monte-Carlo methods. n practice, the perfo=ance of the single node degrades due to noise effects in the channel such as fading, shadowing, and the receiver uncertainty. So, we have mitigated these effects using cooperative sensing and compare the results with single node. The simulation result discloses that the proposed detection algorithm can detect noisy signals of signal to noise ratio (SNR) up to -21dB using single node, -26dB using five nodes in cooperation at required probability of detection and false alarm probability (Pd ;::: 0.9 and P :::; 0.1). The single node sensing algorithm is also implemented in Viliex-4 (XC4VSX35-FFG668-10) Field Programmable Gate Array (FPGA). Keywords- Cognitive Radio Network, Cooperative spectrum sensing, Energy detection, Bayesian criterion, Signal to noise ratio, FPGA. NTRODUCTON The Radio spectrum is an indispensable natural resource for evolution of future generation wireless systems. However, due to rigid licensing policies radio spectrum has becoming scarcity. On the other side, the recent statistical studies on radio spectrum usage have shown that pre-allocation of spectrum bands to specific wireless communication applications leads to poor utilization of those allocated bands in terms of different dimensions such as frequency, time and geographical space [1]. Hence, research into novel techniques for 15 efficient spectrum utilization is being aggressively engaged. Cognitive radio (CR) has emerged as a promising technology to resolve the impending spectral scarcity and under utilization of allocated spectrum [2]. The spectrum sensing is the primary and essential component in the CR functions. Several signal processing techniques are being used for signal detection through spectrum sensing. Popular methods among them are Matched filtering-can perform coherence detection, Energy detection-non coherent detection, cyclostationary feature detection based on the characteristics of the signals [3]. n practice, sensing performance (i.e., detection probability (Pd), false alarm probability (Pt)) of single CR is often compromised with multi path fading and shadowing issues in the channel. To mitigate the impact of these issues, cooperative sensing has been shown to be an effective method to enhance the detection performance by exploiting spatial diversity [4]. n this work, spectrum sensing is based on Energy detection in frequency domain with Bayesian framework. n the Bayesian framework, the prior probability P(H]), P(Ho} are assumed as known. Where P(HJ is the probability for which the PU is active and P(Ho} as the probability for which the PU is inactive. Here, centralized cooperative spectrum sensing with a fusion center (FC) is used to improve the performance of the system, which uses energy detection in each CR locally. Digital Video Boardcasting-terristrial (DVB-T), Qudrature phase shift keying (QPSK) are considered as PU signals under A WGN, Rayleigh fading and shadowing environment according to the standards specified in EEE [5]. n present, FPGAs are being popularly used for signal processing application [7]. Spectrum sensing implementation in Xilinx Virtex-

2 Proc. of nt. Con! on Advances in Recent Technologies in Communication and Computing (XC4VSX35-FFG668-10) Field Programmable Gate Array (FPGA) is proposed. We organize the rest of the paper as follows. Section presents spectrum sensing algorithm for single node and multi node. Simulation results described in section. FPGA implementation of sensing algorithm details in section V. Finally, our conclusions are presented in section V. SPECTRUM SENSNG ALGORTHM CR or secondary user (SU) must sense the PUs band to know whether the channel is being used by primary user or not. Fig.1. describes the block diagram of energy detection in frequency domain. assumed as known, Cpq is the probability of each outcome. Here, we assume Cpp = Cgq and is equal to 90%. The R.H.S term of above equation gives us the threshold (.!c) value. Then eqn. (1) becomes, P(ro,lj,r ,rN_ Ho) H A (2) <, [p(ro,lj,r2,,rn_... _ H)) > HO We make the following assumptions, the noise w(,y and s(n) are Gaussian, independent and identically distributed (iid) random process with mean zero and variance (O', ).and variance 0': respectively. Thus, the likelihood ratio (LR) can be approximated as,... - l T(y )., ::" 'L R(k)."R(k) 'L W (k)."w(k) >- ( 3 ) Figure 1: Block diagram of energy detection A. Single node sensing algorithm The fundamental problem of spectrum sensing in CR is to discriminate the following two hypotheses [7]. Ho: PU signal is absent. H: PU signal is present H HQ Where R(k) are the fast Fourier transform (FFT) coefficients, and W(k) are the FFT coefficients of scaled noise according to the received signal variance. The optimum decision test is to minimize the Bayes risk, i.e., to minimize the probability of misdetection and probability of false alarm. Since we have a binary test, Bayes risk for the detection process will be [6],! C",, = L CJ(Hi')P Hi'HJ (4) -[.,.,:)=0 Ho : r(n) = l1{n), : r(n) =s(n)+ l1{n),n =0,1,2,..,N-l Where r[n}=[(r(o), r(j),...,r(n-1)] is the received signal sequence, s(n) is licensed user transmitted signal, w(n) is the A WGN in the channel. Here, decision test is based on Bayesian criterion, which can be expressed concisely as [6] B. Multinode sensing algorithm Presently, cooperative is being used to mtigate the impact of shadowing and fading issues. Fig. 2 represents centralized cooperative sensing model. Assume that there are M nodes in the cooperation and the received signals of all nodes are independent, then the hypothesis test takes the following form, Where P(r H1), P(r HO) are the probability density functions of H and Ho respectively. P(H)' P(Ho) are the prior probability of each hypothesis H and Ho and 16 F4 : r (n)= l ;,,{n),m= 0, 1, H! : r;;,(n) = s3n) + l ;.{n),n=o, 1,...V-1 Cooperative sensing is classified mainly into soft and hard decision techniques.

3 Proc. of nt. Con! on Advances in Recent Technologies in Communication and Computing PU. 1\ Thus, the Cd-MOST is the average of all individual detection probability. SMULATON RESULTS '"!...,,,' R Nt:rwQrk..." Utlc: am Rce-elVer lddu tennlll:ftl Figure 2: Centralized cooperative sensing model 1) Soft decision fusion: n Soft decision techniques such as weighted gain combining (WGC), equal gain combining (EGC) are popularly being used [7]. WGC assigns weight to each node according to the SNR of the received signal. The global decision for WGC is where Ym is the soft decision from the m'" node, W m is the weight of the mth node. n general, the FC does not know the prior information about SNR. Hence, the FC gives equal weight to all nodes and combines their measurements to generate a global decision in EGC. 2) Hard decision fusion: n Hard decision techniques OR, AND, and MOST logics are popularly being used. n case of OR logic, the FC decides H when any one of the CRs in the cooperative network reports signal detection. The detection probability of OR fusion is, (5) n order to illustrate the perfolmance of the spectrum sensing algorithm, both DVB-T and QPSK signals are considered under A WGN and Rayleigh fading channel environment at required P d and Pi' The simulation parameters are listed in Table.. The observation time is based on time bandwidth product. Detector evaluates the ratio of Power spectral density (PSD), a measurement of the energy at various frequencies for the received signal and the scaled A WGN noise with signal to noise ratio of received signal in a desired band is considered as the test statistic. Since, the closed form solution for Pfand Pd does not exist, the performance of the detection is analyzed using Monte Carlo methods of iterations in simulation. We have also considered the consequences of shadowing and channel fading in our simulation. We simulated AND, OR, and MOST rules of hard decision and WGC, EGC of soft decision. Fig. 3 represents SNR Vs Pd for DVB-T and QPSK signal with (PrO., ) with sample size of 64. Results reveal that the P d increases as the SNR of the received signal increases. Fig. 4 ill ustrates the complementary receiver operating characteristics (CROC) for (p(n) =0.4, 0.5, 0.6) at one particular SNR. From the figure it is seen that the misdetection probability increases as the prior probability of signal presence in the channel reduces. Fig. 5 illustrates the ROC curves for hard and soft fusion logics with no. of nodes M= 3. From the figure 5, M Cd-DR =1-T(1-Pd",J (6) nfl Where Pd,m is the detection probability of the m'h CR user (or mth node). n case of AND logic, The detection probability of AND fusion is, PU signal DVE-T, QPSK Band width (W), Observed time 6MHz,5.4fJSec P, Pd values 0.1,0.9 Prior probabilities of H and Ho 0.5,0.5 Table. SPECFCATONS FOR SMULA TON M C d- A.'D = T Pd,., m=l Finally, MOST logic decides H based on voting rule. (7) soft decision logic (EGC, WGC) is optimal on account of reliability. However, WGC emulates the EGC with small reduction of detection accuracy. n conclusion that soft decision techniques are more reliable. 17

4 Proc. of nt. Con! on Advances in Recent Technologies in Communication and Computing " /".'.. O'Ya.T..F1.:::(l1 crve-t,pi..o -OvH"" 1n OPSK..P':lJ. ORSK.Pt::::OO(i -CP.iK. ::OO' - -Q -! :sr--1i11ll91 Figure 3: SNR vs P d for DVB-T and QPSK signal. Figure 5: ROC curves for hard and soft fusion logic with no. of nodes M=3... 'i 10,"t "=2,,,,= "- ".ll 'O' CJ.l: -...J..L..L..L,u,O.,b ;;; =i&t"f Problbllity r ra t.. alarm Probability of false alarm Figure 4: CROC for DVB- signal with different prior probabilities at fixed SNR, N=64. Figure 6: Bayes risk vs false alarm probability with single node (M=) and Multi node (M=3). Fig. 6 depicts the Bayes risk of all data fusion rules. The optimum detection test is to minimize the Bayes risk. From the figure, it is seen that the Bayes risk reduces for all fusion logics even though Pf value rises, except for AND logic. Moreover, the risk is less for cooperative sensing compared to single node sensing. Fig. 7 shows the relation between SNR vs P d with different fixed false alarm probability (PrO., 0.01) and number of nodes (M) in cooperation. n conclusion, cooperative sensing enhance the performance of the system compared to single node (M=). We observed that the least SNR required to achieve the desired Pd and Pf values (Pd ;::: 0.9 and Pf ::;; 0.05) using EGC fusion is -26dB for M= 5. Table. presents the least SNR required for DVB-T signal with different fixed false alarm probability (Pf=O.1 and 0.01) using EGC fusion. From the table it is obvious that the detection probability increases with more number of users. 18 HARDWARE MPLEMENTATON The spectrum sensing algorithm using Energy measurement in the desired frequency band is implemented in Xilinx Virtex-4 (XC4VSX35-FFG668-10) FPGA. Fig. 8 represents the spectrum sensing architecture along with hardware in loop (HL) simulation based on Energy of the received PU's signal. The architecture consists of an N point FFT unit, Energy unit, Test statistic, Threshold unit, and a decision unit. DVB-T signals of different SNR under A WGN and Rayleigh fading environment are generated in MATLAB and given to detector which calculates the energy of signal and scaled noise in the desired band using N-FFT unit. Test statistic computes the ratio of energy. Based on the test statistic and threshold value,

5 Proc. of nt. Con! on Advances in Recent Technologies in Communication and Computing 2011 Figure 7: SNR vs Pd for EGC fusion with (PrO.,0.01, N=64, M=,3, S,7). EGC, Pd=0.9 M= M=3 M=S M=7 Pr=O.1-19dB -24dB -26dB -27dB Pr=O.OS -14dB -19dB -21dB -23dB P=O.O -lldb -16dB -18dB -19dB Table 3. The least SNR required to achieve required (Pd) Vs (Pt) as a function of (Al) with EGC fusion. Furthermore, misdetection probability as well as Bayes risk reduces even in shadowing and fading environment with cooperation. We can detect the low SNR signals up to -26dB using EGC fusion based on five nodes in cooperation at required Pd and Pr. But the performance depends on statistical properties of the channel. FPGA implementation of sensing architecture can reliably detect the PU's activity in the desired frequency band. REFERENCES [1] FCC, "ET Docket No.02-13S, Spectrum policy task force (SPTF) Rep0l1", "Federal Communications Commission", Nov.S [2] S.Haykin, "Cognitive radio: Brain-empowered wireless communications", EEE Journal on Selected Areas in Communications, vol. 23, pp , Feb- 200S. Figure 8: FPGA implementation of spectrum sensing architecture. decision unit decides the status of PU signal in the band. The decision" 1" and "0" represents signal presence and absence in the desired frequency band respectively. Resource utilization for the Bayesian detector using sample size of 64 requires 82S slices, 978 flip flops, 1361 LUT's, and 103 lob's. The maximum frequency of operation for the detector is MHz with sample size of 64. CONCLUSONS n this paper, we have detected the radio signals based on energy detection in the frequency domain. Simulation result shows that our detection algorithm can detect the DYB-T signal with 90% Pd and 10% P, even at low SNR up to -21dB with single node. 19 [3] Tevtic Yucek and Huseyin Arslan, "A Survey of Spectrum Sensing Algorithms for CR Applications", EEE Communications surveys and tutorials, vol:ll, pp: , First quarter [4] an F. Akyildiz and Brandon F. Lo and Ravikumar Balakrishnan, "Cooperative spectrum sensing cognitive radio networks: A survey ", Physical Communication, vol. 4, pp , July [5] K.C. Carlos and D. Birru., " EEE : An introduction to the first wireless standard based on cognitive ardio", EEE Journal of communications, vol.1, no., pp: 38-47, April [6] S.M. Kay, "Fundamentals of statistical Signal processing and estimation theory," Prentice Hall [7] Srinu, S. and Sabat, S.L. and Udgata, S.K, ""FPGA implementation of cooperative spectrum sensing for Cognitive Radio networks", EEE proceedings, (UKWCWS), pp. -S, Dec-20lO.