Available online at ScienceDirect. Procedia Computer Science 37 (2014 )

Transcription

1 Available online at ScienceDirect Procedia Computer Science 37 (204 ) The 5th International Conference on Emerging Ubiquitous Systems and Pervasive Networks (EUSPN-204) A Log-Likelihood Based Cooperative Spectrum Sensing Scheme for Cognitive Radio Networks Sheeraz Akhtar Alvi a,, Muhammad Shahzad Younis b, Muhammad Imran c, Fazal-e-Amin c a Al-Khawarizmi Institute of Computer Science, University of Engineering and Technology, Lahore 54000, Pakistan b School of Electrical Engineering and Computer Science, National University of Sciences and Technology, Islamabad 44000, Pakistan c College of Computer and Information Sciences, King Saud University, Riyadh 2372, Saudi Arabia Abstract In cognitive radio networks, cooperative spectrum sensing is used to exploit spatial diversity of secondary users, in order to reliably detect an unoccupied licensed spectrum. Since each secondary user experience different channel conditions, therefore observations of secondary users are weighted according to the reliability factor (weight) of individual secondary user. However, the weight estimation is done in each sensing interval resulting in high cooperation overhead in terms of time, processing and reporting channel bandwidth. In this paper, we propose a novel cooperative spectrum sensing scheme which is based on the Log-Likelihood ratio (LLR). The secondary users employ two threshold level to minimize false-alarms and miss-detection of the primary user s signal. In addition, unknown parameters for weight computation are estimated/updated after certain number of sensing intervals instead of estimating them in every sensing interval. In this way, the detection performance is increased while the cooperation overhead reduces. To evaluate the detection performance of the proposed scheme we do a simulation study and compare it with the popular optimal detection schemes. c 204 The Authors. Published by Elsevier B.V. by B.V. This is an open access article under the CC BY-NC-ND license ( Selection and peer-review under responsibility of Peer-review under responsibility of the Program Chairs of EUSPN-204 and ICTH 204. Keywords: Cognitive radio; data fusion; energy combining; spectrum sensing; cooperation overhead. INTRODUCTION The immense adaptation of wireless communication services and devices has increased the demand of spectrum bandwidth. However, fixed spectrum allocation policy restricts spectrum utilization only to licensed users, resulting in poor spectrum efficiency. Since, some spectrum bands remain unoccupied more often than others, thus a Dynamic Spectrum Access (DSA) between licensed primary users (PUs) and unlicensed secondary users (SUs) has been shown to significantly improve the spectrum efficiency 2,3. In cognitive radio DSA, secondary users (SUs) temporarily access the unoccupied frequency bands also known as spectrum holes or white space to improve spectrum efficiency. SUs Corresponding author: Sheeraz Akhtar Alvi address: sheeraz.akhtar@kics.edu.pk The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license ( Peer-review under responsibility of the Program Chairs of EUSPN-204 and ICTH 204. doi:0.06/j.procs

2 Sheeraz Akhtar Alvi et al. / Procedia Computer Science 37 ( 204 ) are supposed to reliably detect the presence of licensed users or primary users (PUs) and if the spectrum is available (PU is absent). This process of sensing the RF environment is known as Spectrum Sensing (SS) 2,3,4. In practical cognitive radio ad hoc networks (CRAHNs), the SUs experience different environmental conditions. Thereby, if sensing data from a SU having weak PU signal power is combined with SUs observing stronger signal power then the combined sensing data can significantly improve the detection performance 5. Obstacles in communication environment such as houses, building, among others cause the incoming signal to reflect/deflect resulting in multipath fading in addition to the path loss experienced by the signals. Obstacles may also completely block the desired signal causing shadowing effect. Similarly interference between SU and PU due to miss-detection of presence of PU also affects the DSA performance. Interference can also take place when a SU is out of range of PU but within the range of a PU receiver. These issues make the performance of detection unreliable if individual spectrum sensing is done. As discussed earlier that not all SUs experience same channel conditions, so the sensing data of every SU is not reliable. Thus for efficient cooperative sensing, the sensing data should be combined based on the reliability factor of a SU referred as weight to achieve an effective cooperative gain 6,7. Received signal-to-noise ratio (SNR) strictly depends upon the wireless channel conditions (issues discussed above), so SNR can be an excellent candidate to evaluate the reliability of a SU s sensing data (in other words its weight). However, the weight (in our case SNR) estimation carries costly processing overhead as well frequent updates since channel conditions are time-varying. Consequently for simplicity the channel conditions are considered same at all SUs and sensing data is combined using equal weights 8.In 9 we proposed another cooperative spectrum sending scheme. The two thresholds were calculated based on the probability of false alarm only. However, in this paper we compute the two threshold levels using the log-likelihood technique which gives an optimal estimation/detection. In this paper, a cooperative spectrum sensing scheme is proposed in which the SUs send their observed sensed data to the fusion center (FC) where the sensing data is combined in a beneficial approach to take the decision. The cooperative overhead due to the weight estimation process is decreased and the presence/absence of PU is detected/identified without prior knowledge of the received signal statistics. To decrease weight estimation overhead the correlation among the SNR estimation samples is exploited such that the detection performance remains satisfactory. SNR estimation is not done in every sensing interval instead it is done after a certain number of sensing intervals if PU is detected. When SNR estimation is not required SUs only report observed energy measurements to the FC and weighted sensing data combining is done using latest SNR estimates. 2. SPECTRUM SENSING TECHNIQUES To detect/identify the presence of the PU signal, a binary decision is taken on the basis of a statistical hypothesis test. The binary hypothesis tests and H representing the absence and presence of the PU signal, respectively. In our proposed system model, the received signal by the mth SU at the output of its low pass filter can be given as: : r m [k] = n m [k] () H : r m [k] = α m s[k] + n m [k] (2) Where m =, 2,..., M; and k =, 2,..., K such that, M is the number of SUs and K is the number of observed samples. {n m [k]} K represents the additive noise samples, that are independent and identically distributed (i.i.d.) Gaussian random variables with zero mean and variance σ 2 n (unknown). {s[k]} K represents the PU signal samples that are deterministic and unknown. The combined effect of fading, path loss and carrier phase on the signal received at mth radio is represented by complex variable α m that is unknown. Let the probability distribution of the received signal for and H be P 0,m (r m [k]) and P,m (r m [k]), respectively. Using the central limit theorem 5, it is reported in 0 that P 0,m (r m [k]) and P,m (r m [k]) can be approximated with a Gaussian distribution. : E m P 0,m (r m [k]) E m N(Kσ 2 n, Kσ 4 n) (3)

3 98 Sheeraz Akhtar Alvi et al. / Procedia Computer Science 37 ( 204 ) H : E m P,m (r m [k]) E m N(K ( + γ m ) σ 2 n, K ( + 2γ m ) σ 4 n) (4) Where the notation N ( μ, σ 2) represents a Gaussian random variable with mean value μ and variance σ. The mean and variance values in the expressions given above depend on the modulation scheme used by the PU 0. We assume that the PU transmits a phase-shift keying (PSK) modulated signal with a modulation order of 2. Let E s represents the symbol energy, then the SNR value measured at the mth radio is given as: γ m = E s α m 2. σ 2 n (5) The probability of false alarm (P fa ) and the probability of miss-detection (P md ) are the vital metrics based on which the detection performance of a scheme is evaluated. Let d represents the decision based on a specific threshold value η, then the P fa and the P md will be P fa = P {decision = H } = P {d η } (6) P md = P {decision = H } = P {d η H } (7) The energy detection method comes with drawbacks like poor performance in low SNR, sensitivity towards noise power 7, inability to distinguish signals of primary user and secondary user 8, among others. Sensing techniques such as matched filter and cyclostationary feature detection can perform relatively better than energy detection yet energy detection is preferred due to its low complexity and non-coherent signal detection (see 6,7,8,9,20,2 ). In centralized collaborative spectrum sensing, local sensing data is shared with fusion center where it is combined for hypothesis testing, this process is called data fusion. The sensing data is reported to fusion center using a control channel which usually has a small bandwidth. The two main combining methods used at the fusion center are soft combining and hard combining. However, we restrict our discussion on the soft energy combining schemes, since their detection performance outperforms hard combining schemes. There are three main soft energy combining methods 20 ; Equal Gain Combining (EGC), Weighted Linear Combining (WLC), and Optimal Combining (OC). In EGC, the SUs measure channel energy and send the actual energy samples to the FC. The FC combines the energy samples and compares the result with a predefined threshold. In EGC, the reliability of sensing data is not considered (i.e., the PU signal strength at a SU), thus performance of EGC reduces when the SUs experience different from the channel conditions. Nonetheless, the EGC method is widely used ( 8 ) because of its non-coherent feature. R EGC = m= E m H η EGC To introduce reliability in the data fusion process, in WLC SNR dependent weights are used that are computed using estimated unknown parameters. Thereby, the SUs experiencing better SNR contribute more towards the decision and vice versa. In 9, a heuristic methodology is used to find near-optimal weights w m = R WLC = m= w m E m H η WLC γm +2γ m. OC is based on the likelihood ratio test in which for a given hypothesis H i sensing data of a SU is assumed to be independent of other SUs. WLC and OC are optimal energy combing methods, since reliability of SUs is exploited in the decision. However, instantaneous parameter estimation is required to compute weights which results in a large processing overhead. R OC = M m= / M P (E m /H ) P (E m / ) H η OC (0) m= (8) (9)

4 Sheeraz Akhtar Alvi et al. / Procedia Computer Science 37 ( 204 ) As discussed in hard combining the control channel overhead is small, but the loss of information results in limited detection capability. On the contrary the detection performance of soft combining is better, though cooperation overhead is much higher. Another aspect about the SU cooperation is that in most of the cooperative sensing techniques the fused data is compared with a single threshold. This comparison lacks confidence measure and results in false alarm or miss detection when resultant is closer to the threshold. This scenario is highly likely when SNR levels are low or noise variance and uncertainty is elevated. 3. PROPOSED SCHEME In realistic cognitive radio networks, wireless channel are time varying. Path loss, shadowing, mobility, multipath propagation and fading cause irregular variations in signal power at the SU radio frequency front end. In realistic scenario fading is frequency selective but it is correlated and approximated to be flat for a small duration known as coherence time. Similarly during a short time period when SU changes its position yet the relative distance of PU from SU s current and previous position is not much different. In this case it is highly likely that SU experiences same path loss and shadowing i.e. they are also correlated. WLC and OC are optimal soft energy combining detection methods but they require instantaneous SNR estimates of each SU. This results in high control channel overhead which makes the WLC and the OC impractical in real systems. Another aspect which is insignificantly addressed in previous studies, is the processing overhead at the FC. In weighted combing methods, the FC is responsible for the parameter estimation for each individual SU, weights computation, data fusion and reporting of the final decision. Thus, increased precessing overhead at the FC results in a higher cooperation overhead. It is worth noting that in most cognitive radio networks, the FC is a wireless station which occupies a limited energy. That is why high processing overhead is infeasible in real cognitive radio networks. In hard combing, although overhead is small nevertheless reliability measures are not usually considered resulting in low PU signal detection capability. Furthermore most of the energy combining schemes uses a single threshold level for hypothesis testing which lacks the certainty measure. For example, if the resultant is just above or just below the threshold level then the uncertainty is high and the decision will be unreliable, consequently probabilities P fa and P md tend to increase respectively. Considering this problem, we introduce a non-deterministic region containing the probabilities of uncertain decisions. When a local test result lies in the non-deterministic region then SU does not report the statistics, i.e., SU does not contribute in final decision making at the FC. Those uncertain local sensing results that do not contribute effectively to the cooperative SS mechanism are suppressed, thus improving the detection performance and reducing the reporting channel overhead. To detect the presence of PU signal, initially every SU performs a local sensing test (L m ) based on the LLR. Following the expression that is used to compute LLR at the mth SU LLR m = K k= m= ln 2πσ n 2 exp ( 2πσ n 2 exp ( rm[k] αms[k] 2 2σ 2 n rm[k] 2 2σ 2 n ) ) The local sensing test is performed after each SI, in which the LLR m value is compared with two threshold levels η md and η fa. If the LLR m value lies between the two thresholds levels, then neither the local test result (L m ) nor the estimated unknown parameter values are transmitted to the FC. However, the received PU signal samples are saved for consequent SNR parameter estimation. If a SU makes a certain local decision then the local test result L m is reported as -, +, 0 if the PU signal presence is false, true, uncertain. Threshold levels η md and η fa can be tuned to get the desired detection performance, i.e., depending on the tolerance level of false alarm and miss-detection. It is shown in 6, if the probability of the false alarm and the probability of the miss-detection are sufficiently small, then for LLR based test expressions of P fa and P md can be given as ( ) Pmd η fa = ln (2) P fa ()

5 200 Sheeraz Akhtar Alvi et al. / Procedia Computer Science 37 ( 204 ) ( η md = ln P md P fa ). (3) After each SI, FC performs a binary hypothesis test by combining the local results (L m ) reported by the SUs using respective weights. The global test result is compared with a threshold value of η SCH (we set it equal to 0). If the result is lower (higher) than η SCH than (H ) is reported back to all SUs. Current set of weights is sustained by the FC till new SNR values are sent by the SUs. At the FC, the following decision rule is used to make the final decision R SCH = M m= w m L m H η SCH. In the WLC and the OC, the SNR parameter estimation is done every SIwhich results in high processing overhead. On the contrary, in our proposed scheme, the weight estimation overhead is alleviated by exploiting the correlation in channel conditions. Depending upon the final decision of the FC (H X, fc ) and the last decision of the local test result (H X,lm ), each SU selects a certain number of SIs for which it does not estimate SNR and till then it uses the last estimated values. For the network simulation model presented in Section 4, we set the SNR estimation interval to 5, 25, 5, 5 sensing intervals. Thus, instead of estimating the SNR every SI, it is estimated (less than 20% of the time) only when its SI x has expired. The actual signal samples required in the WLC and the OC, have significant overhead and dependency on the number of sensing samples. Generally, censoring techniques are used to save energy that is utilized in reporting data. However, energy is still utilized in SS. In our proposed scheme, unreliable sensing data is censored to save the control channel bandwidth and received signal samples are used (later) in the SNR estimation process. Note that when the result lies between two thresholds, the SU doesn t have to access the control channel which is usually modeled by a random medium access (e.g., CSMA/CA) that consumes a significant amount of time and energy. (4) 4. SIMULATION RESULTS To evaluate the performance of the proposed scheme, we consider a CRAHN containing a stationary PU and M mobile SUs among which one is acting as the FC. Local test results and estimated parameter values are transmitted to the FC using a small bandwidth reporting channel. We assume that the reporting channel errors (if any) are removed at the FC using forward error correction. A random walk process based mobility model 23 is used for the movement of SUs within an area of m, in which SUs choose new direction (0-2π) and speed (0-2 m/sec) after each SI=00 msec. Due to the Rayleigh fading and path loss, the received SNR of a SU varies between 0 to 0 db, depending on its distance from the PU. We assume that both the fading and path loss are time-varying, however the SNR value of a SU remains constant for a SI. The PU transmit a BPSK modulated signal at 500 kb/sec. The number of samples used for parameter estimation and for LLR based test LLR m are 500 and 0 samples, respectively. Unless specified the values of E s,m,p fa, P md, and σ 2 n are, 6, 0.0, 0.0, and, respectively. The detection performance of the proposed scheme (SCH), OC, and WLC is shown in Fig.. As expected the detection performance increases with the increase in the P fa, also the performance of the proposed scheme closely follows the performance of optimal OC and WLC schemes. If too small values of P fa and P md are selected then the non-deterministic region gets larger that may result in suppressing the majority of the local test results that is why the detection performance is slightly lower at small P fa. However, in our proposed scheme the cooperation overhead is significantly lower as compared to OC and WLC. The performance of any spectrum sensing technique significantly depends upon the SNR values experienced by the SUs. To evaluate the performance of the proposed scheme, we vary the average value of SNR experience by a SU keeping E s and σ 2 n constants. It can be seen in Fig. 2 that the performance of the proposed scheme quickly gets to the performance level of the optimal WLC and OC for the SNR as low as 6 db. For any cooperative SS scheme, the number of cooperating users is very important. As shown in Fig. 3, when the number of participating SUs is increased the detection performance of all schemes increases. However, the performance of the proposed scheme for very small number of SUs is lower due to the possibility of suppression of all local test results.

6 Sheeraz Akhtar Alvi et al. / Procedia Computer Science 37 ( 204 ) OC WLC SCH Pd P fa Fig.. Receiver Operating Characteristic (ROC) Curve OC WLC SCH Pd Avg. SNR (db) Fig. 2. Avg. SNR = M m SNR m M of SUs OC WLC SCH Pd SUs Fig. 3. Effect of Number of Cooperative SUs.

7 202 Sheeraz Akhtar Alvi et al. / Procedia Computer Science 37 ( 204 ) CONCLUSION Cooperative SS comes with cooperation overhead in terms of sensing time, delay, processing overhead, among others. Therefore, a LLR based cooperative spectrum sensing scheme is proposed which decreases the cooperation overhead, by suppressing ineffective local sensing results using two threshold levels. Cooperation overhead for the proposed scheme is similar to hard energy combining schemes. However, the simulation results show that the proposed scheme s detection performance is near-optimal when compared to the optimal OC and WLC detection schemes. However, cooperation overhead of the proposed scheme is significantly lower as compared to the optimal weighted energy combining schemes. 6. ACKNOWLEDMENT This research work is supported by the Research Center of College of Computer and Information Sciences at King Saud University through project No. RC3034. References. Force, S. P. T. Spectrum policy task force report. Federal Communications Commission ET Docket 02 35, I.F. Akyildiz, W.-Y. Lee, and K.R. Chowdhury, CRAHNs: cognitive radio ad hoc networks, Ad Hoc Networks, vol. 7, no. 5, pp , July I.F. Akyildiz, W.-Y. Lee, M.C. Vuran, and S. Mohanty, NeXt generation/ dynamic spectrum access/cognitive radio wireless networks: a survey, Computer Networks, vol. 50, no. 3, pp , September I.F. Akyildiz, B. F. Lo, and R. Balakrishnan, Cooperative spectrum sensing in cognitive radio networks: A survey, Physical Communication, vol. 4, pp , December D. Cabric, S. Mishra, and R. Brodersen, Implementation issues in spectrum sensing for cognitive radios, in Proceedings of Asilomar Conference on Signals, Systems, and Computers, vol., pp , November F. E. Visser, G. J. M. Janssen, and P. Pawelczak, Multinode spectrum sensing based on energy detection for dynamic spectrum access, in Proceedings of IEEE Vehicular Technology Conference (VTC-spring), Singapore, pp , May S. S. Jeong,W. S. Jeon, and D. G. Jeong, Collaborative spectrum sensing for multiuser cognitive radio systems, IEEE Transaction on Vehicular Technology (TVT), vol. 58, no. 5, pp , June Y. Chen, Analytical performance of collaborative spectrum sensing using censored energy detection, IEEE Transaction on Wireless Communications (TWC), vol. 9, no. 2, pp , December Alvi, S. A., Younis, M. S., & Imran, M. (204). A Weighted Linear Combining Scheme for Cooperative Spectrum Sensing. Procedia Computer Science, vol. 32, pp , June Y. C. Liang, Y. Zeng, E. C. Y. Peh, and A. T. Hoang, Sensing-throughput tradeoff for cognitive radio networks, IEEE Transaction on Wireless Communications (TWC), vol. 7, no. 4, pp , April F. Hameed, O. A. Dobre, and Dimitrie C. Popescu, On the likelihood-based approach to modulation classification, IEEE Transaction on Wireless Communications (TWC), vol. 8, no. 2, pp , December C. Y. Huang and A. Polydoros, Likelihood methods for MPSK modulation classification, IEEE Transaction on Communications (TCOM), vol. 43, pp , March P. Panagiotou, A. Anastasopoulos, and A. Polydoros, Likelihood ratio tests for modulation classification, in Proceedings of IEEE 2st Century Military Communications Conference (MILCOM), Los Angeles, CA, pp , October V. G. Chavali and C. R. C. M. da Silva, Maximum-likelihood classification of digital amplitude-phase modulated signals in flat fading non- Gaussian channels, IEEE Transaction on Communications (TCOM), vol. 59, no. 8, pp , Angust A. Papoulis and S. U. Pillai, Probability, Random Variables and Stochastic Processes, New York: McGraw-Hill, Q. Zou, S. Zheng, and A. H. Sayed, Cooperative Sensing via Sequential Detection, IEEE Transaction on Signal Processing (TSP), vol. 58, no. 2, pp , December A. Sahai, N. Hoven, and R. Tandra, Some fundamental limits on cognitive radio, in Proceedings of Allerton Conference, Monticello, October S. Shankar N, C. Cordeiro, and K. Challapali, Spectrum agile radios: Utilization and sensing architectures, in Proceedings of IEEE Symposium New Frontiers Dynamic Spectrum Access Network (DySPAN), Baltimore, MD, pp , November Z. Quan, S. Cui, and A. H. Sayed, Optimal linear cooperation for spectrum sensing in cognitive radio networks, IEEE Journal on Selected Topics in Signal Processing (JSAC), vol. 2, no., pp , Febuary J. Ma, G. Zhao, and Y. Li, Soft combination and detection for cooperative spectrum sensing in cognitive radio networks, IEEE Transaction on Wireless Communications (TWC), vol. 7, no., pp , November T. Do and B. L.Mark, Joint spacial-temporal spectrum sensing for cognitive radio networks, IEEE Transaction on Vehicular Technology (TVT), vol. 59, no. 7, pp , September V. G. Chavali and C. R. C. M. da Silva, Collaborative Spectrum Sensing Based on a New SNR Estimation and Energy Combining Method, IEEE Transaction on Vehicular Technology (TVT), vol. 60, no. 8, pp , October T. Camp, J. Boleng, and V. Davies, A survey of mobility models for ad hoc network research, Wireless Communications and Mobile Computing (WCMC), vol. 2, no. 5, pp , August 2002.