Cognitive Radio Techniques for GSM Band

Transcription

1 Cognitive Radio Techniques for GSM Band Baiju Alexander, R. David Koilpillai Department of Electrical Engineering Indian Institute of Technology Madras Abstract Cognitive Radio has generated a lot of interest as a novel approach for improving the utilization of spectrum resource in recent years. Cognitive radios have to reliably sense the RF environment to co-exist with legacy wireless networks. The critical design problem is to reliably detect the presence of primary users. In this paper we compare the performance of different schemes for signal presence detection as applied to the GSM band. We also propose a hybrid scheme to detect the presence of GSM signals. I. ITRODUCTIO Wireless systems today are characterized by static spectrum allocations which are under-utilized [1][2]. GSM systems use MHz [3] band for uplink and MHz [3] band for downlink. However, in many places some of these frequencies may not be used or some of the signal frequencies may not be present in a particular area due to RF propagation effects. It can also happen that there is no signal present due to lack of GSM deployment, like in remote villages. These unused frequencies can then be used by cognitive radios [4] for their operations. Cognitive radio can be used only if permitted by the regulatory authority. A possible scenario is that the spectrum owner may deploy equipment which use cognitive radio techniques to do self configuration and autonomous frequency planning. Cognitive radios are considered lower priority or secondary users of spectrum allocated to a primary user. Their fundamental requirement is to avoid interference to the primary users in their vicinity. Spectrum sensing has been identified as a key enabling functionality to ensure that cognitive radios do not interfere with primary users, by reliably detecting primary users signal. The first application of spectrum sensing is studied under IEEE standard group [5] in order to enable secondary use of UHF spectrum for fixed wireless access. In addition, there are a number of indoor and rural applications where spectrum sensing would increase spectrum efficiency and utilization. Spectrum sensing is often considered as a detection problem [6], the key challenge of spectrum sensing is the detection of weak signals in noise with a very small probability of missed detection, which requires better understanding of very low SR regimes [7]. Our goal is to study different detection schemes applicable for GSM band and to compare these methods based on probability of missed-detection. Some of the detection schemes used in cognitive radios are matched filter detectors [7], energy detectors [8] and cyclostationary feature detectors [10][11]. The following are the topics discussed in this paper. Study of energy detection method for GSM signals. Limitations of the energy detector performance due to presence of noise level uncertainty and fading. Study of cross-correlation-based detector. A hybrid detection scheme for GSM signals. The paper is organized as follows: Section II reviews the energy detector model, and derives its performance and addresses the limitations. In Section III, we study the performance of cross-correlation-based detector. In Section IV, we discuss a hybrid detection scheme for primary signal in GSM band. The summary and conclusions are presented in Section V. II. EERGY DETECTOR CHARACTERIZATIO We consider the detection of a weak GSM signal in additive noise, and the effect of multipath fading. The signal power is confined within an apriori known bandwidth B (200 khz for GSM) [3], around the carrier frequency f c. We assume that activity outside of this band is unknown. A sub-optimal energy detector is adopted, which can be applied to any signal type. An energy detector consists of a low pass filter to reject out of band noise and adjacent signals, A/D converter, squarelaw device and integrator. Without loss of generality, we can consider a complex baseband equivalent of the energy detector. The spectrum sensing mechanism is attempting to classify the given GSM channel as either occupied by a GSM signal or as vacant. This is a binary hypothesis testing problem, the two hypotheses are summarized below. H 0 : y[n] = w[n] signal absent H 1 : y[n] = x[n] + w[n] signal present where n = 0, 1,, 1 ( - observation interval). x[n] = received signal samples w[n] = noise samples y[n] = received Samples The noise is assumed to be additive, white and Gaussian (AWG) with zero mean and variance σ 2 w. The signal samples can also be modeled as Gaussian random process with variance σ 2 x. A decision statistic for energy detector is: = 1 1 n=0 y(n)y (n)

2 A. Performance Performance of energy detector is measured by a resulting pair of probability of detection (P d ) and probability of false alarm (P fa ). Each pair is associated with the particular threshold (γ) that tests the decision statistic: γ decide signal present < γ decide signal absent When the signal is absent, the decision statistic has a central chi-square distribution with degrees of freedom [8]. When the signal is present, the decision statistic has a non-central chi-square distribution with the same number of degrees of freedom [8]. If is large we can use the central limit theorem to approximate the test statistic as ormal(m, σ 2 ) that is Gaussian with mean m and variance σ 2. ormal (σ 2w, (σ2 w) 2 ) under H 0 ormal (σ 2x + σ 2w, (σ2 x + σw) 2 2 ) under H 1 Fig. 1. Theoretical and simulated performance of energy detector in AWG, number of samples = 568 where = umber of samples Then probability of missed detection (P md ) can be evaluated as [9]: ( ) P md = Q σx 2 + σw 2 [(σx 2 + σw) 2 γ] (1) where threshold γ is: γ = σ 2 w ( ) 1 + Q 1 (P fa ) (2) Fig. 2. Probability of false alarm in percentage for energy detector where the function Q(.) is defined as: Q(α) = 1 2π α e x2 2 dx (3) The performance of the energy detector was evaluated using Monte Carlo simulations. The number of samples used were = 568, probability of false alarm was fixed at 10 %, noise power in 200 khz bandwidth was assumed to be 116 dbm. Detection threshold was calculated using eqn.(2) and simulation was run for different received signal power. Fig. 1 shows theoretical curve for probability of missed detection at different SR. Theoretical curve was plotted using eqn.(1). Fig. 1 also shows the simulated performance of energy detector at different SR. Fig. 2 shows the simulated probability of false alarm for the energy detector. Fig. 1 also validates the Gaussian approximation used to derive theoretical performance of energy detector as the theoretical missed detection curve closely matches with the simulated one for low SRs. B. Limitations In simulating the above results we assumed that the additive noise is white, and Gaussian, with zero mean and with known variance. However, the noise term is an aggregation of various sources including, not only thermal noise at the receiver and underlying circuits, but also interference due to nearby unintended emissions, weak signals from far away transmitters etc. Second, we assumed that noise variance is precisely known to the receiver, so that the threshold can be set accordingly. However, this is not possible as noise could vary over time due to temperature change, ambient interference, filtering, etc. Even if the receiver estimates it, there is a resulting estimation error due to limited amount of time. Due to this noise uncertainty, there is a minimum SR below which signal cannot be detected, this minimum SR level is referred to SR wall [12]. Performance of energy detector with noise uncertainity in AWG channel was simulated for = 142 samples, Fig. 3 shows the simulated missed detection curve of

3 Fig. 3. Performance of energy detector with noise uncertainity, number of samples = 142 Fig. 4. Comparison of synchronization burst and normal burst training sequence based cross-correlation detectors in AWG energy detector with noise uncertainity of 1 db in comparison with simulated missed detection curve of energy detector with noise uncertainity of 0 db. We can see from Fig. 3 that noise uncertainity limits the performance of energy detector and detection of primary signal is not possible below 2 db SR. Up to this point, we have considered spectrum sensing performed in AWG-like channels. In fading channels, however, sensing requirements are set by the worst case channel conditions introduced by multipath, shadowing and local interference. These conditions could easily result in SR regimes below the SR wall, where the detection will not be possible. Fig. 9 shows the probability of missed detection at different SRs for energy detector with noise uncertainity of 1 db in multipath fading environment. For the simulation the hilly terrain model [13] was used and speed was assumed to be less than 2 Kmph, the number of samples used for simulation was = 64. We consider the hilly terrain model because it has the longest delay spread when compared to other GSM multipath channel models. We can see from Fig. 9 that even at high SRs in range of 8 to 10 db, the probability of missed detection is more than Performance limitations of simple energy detector necessitate enhanced detection schemes, which are discussed in the next sections. III. CROSS CORRELATIO BASED DETECTORS Every GSM timeslot has a training sequence embedded in it [14]. These training sequence are known sequence used for channel estimation. By cross-correlating received samples with training sequence we can detect the presence of training sequence even in very low SR. Cross-correlation-based detector is a coherent detector, coherent detectors have been studied earlier for cognitive radio [15]. Cross-correlation of two wide sense stationary sequence s 1,s 2 is given by R(m) = E [s 1 (n)s 2(n m)] Where, E[ ] is expectation and m denotes the number of samples by which s 2 (n) is delayed. In practice, an estimator of the cross-correlation that is used is R(m) = 1 1 n=0 s 1 (n)s 2(n m) 0 m M where, M is the maximum lag for the cross-correlation. Different types of GSM bursts (normal burst and synchronization bursts) have different length of training sequence [14]. The normal burst has 26-bit training sequence and synchronization burst has 64-bit training sequence [16]. To evaluate the performance of cross-correlation-based detector, computer simulations were carried out. Detection threshold (γ) was fixed such that probability of false alarm (P fa ) was 10 %. Fig. 5 shows the simulated probability of false alarm for the cross-correlation-based detector. The detector performance was evaluated at different SRs for AWG and fading channels. Fig. 4 compares the probability of missed detection at different SRs for cross-correlation-based detector which uses two different training sequences (of length 26, and 64 bits); normal burst training sequence and synchronization burst training sequence. It is observed from Fig. 4 that longer the training sequence more reliably we can detect the signal in low SR. For all subsequent simulations the synchronization burst training sequence based cross-correlation detector is used. When compared with energy detector, the cross-correlationbased detector detects the primary signal more reliably in AWG channel as shown in Fig. 6. The cross-correlationbased detector also perform better than energy detector in multipath fading environment, Fig. 9 shows the probability of missed detection for the cross-correlation-based detector at different SRs with multipath fading. For simulating the probability of missed detection curves for energy detector in Fig. 6 and 9, we used = 64. IV. HYBRID DETECTIO SCHEME From the previous section we see that cross-correlationbased detector performs better than the energy detector; it can

4 Fig. 7. Block diagram of the hybrid detector Fig. 5. detector Probability of false alarm in percentage for cross-correlation-based Fig. 8. Probability of false alarm in percentage for hybrid detector Fig. 6. Comparison of energy detector and cross-correlation-based detector using synchronization burst training sequence, number of samples = 64 detect weak signals more reliably and is not affected by noise uncertainty. The main disadvantage of cross-correlation-based detector is that it has to scan for longer time to detect the presence of primary signal. The duration of scanning depends on the training sequence used and type of burst used. Different type of bursts have different repeat rate, ormal burst occurs at least once in 8 timeslots of broadcast carrier and synchronization burst occurs once in 80 timeslots. In GSM each timeslot (also called burst) has duration of 577 µsec. Cross-correlationbased detection scheme gets further complicated because the detector is not time synchronized with primary user signal, and hence, the exact location of training sequence is not known to the detector. To take advantage of both the detection schemes we propose a hybrid scheme, which uses the energy detector for initial detection, and uses cross-correlation method for confirmation. The key steps are enumerated below. 1) Energy detector is used in first stage of hybrid detector. 2) The probability of false alarm energy detector is fixed at 10 %. 3) The cross-correlation method is used in the second stage. 4) The probability of false alarm cross-correlation detector is fixed at 10 %. 5) Threshold for each detector is fixed based on probability of false alarm. Block diagram of the hybrid detector is shown in Fig. 7. All the received samples first pass through the energy detector, if the energy detector fails to detect primary signal then the received samples are analyzed using cross-correlation detector. Hybrid detector can make faster decision than cross-correlation-based detector because it uses cross-correlation based detector only when energy detector fails to detect the primary signal. Fig. 9 shows performance of the hybrid detector in multipath fading channel. The overall probabilty of false alarm for hybrid detector was found to be 20 % and the probability of missed detection of hybrid detector for different SRs is similar to cross-correlation based detector. For cognitive radios the missed detection probability is more crucial than false alarm probability so higher false alarm probabilty of the hybrid detector does not degrade the performance. Fig. 8 shows the simulated probability of false alarm for the hybrid detector. V. COCLUSIO In this paper we investigated the performance of the energy detector and the cross-correlation based detector to detect the

5 Fig. 9. Comparison of energy detector with 1 db noise uncertainity, crosscorrelation based detector and hybrid detector in fading channel, number of samples = 64, Hilly terrain multipath fading model used [11] W. A. Gardner, Exploitation of Spectral Redundancy in Cyclostationary Signals, IEEE Signal Processing Magazine, Vol. 8, o. 2, pp , April [12] R. Tandra, A. Sahai. Fundamental limits on detection in low SR under noise uncertainty, 2005 International conference on wireless networks, communications and mobile computing, Vol. 1, pp , June [13] GSM 03.30, Radio network planning aspects, ETSI, Jan 1994 [14] 3rd Generation Partnership Project (3GPP), TSG GSM/EDGE Radio Access etwork, 45 series Radio aspects, [15] A. Sahai, R. Tandra, Shridhar Mubaraq Mishra,. Hoven. Fundamental design tradeoffs in cognitive radio systems, Proc. of the first international workshop on Technology and policy for accessing spectrum, [16] GSM 05.01, Physical layer on the radio path, ETSI, Sept [17] IEEE Working group on wireless regional area networks,project homepage, [18] GSM 05.02, Multiplexing and multiple access on radio path, ETSI, Sept [19] Mouly M., Pautet M. B., The GSM system for mobile communications, [20] R. Etkin, A. Parekh, D. Tse, Spectrum sharing for unlicensed bands, Proc. of IEEE DySPA 2005, ovember 2005, pp presence of the primary user s signals. We also proposed a hybrid detector and evaluated its performance. We simulated the achievable probability of missed detections and probability of false alarm, and the minimum detectable signal levels in AWG and fading channels for different detectors. For energy detector, it was found that the presence of RF propagation channel uncertainties sets practical limits on minimum detectable signal levels, which cannot be further improved by signal processing. For correlation-based detector we found that it can reliably detect low SR signals but require longer sensing time. It is demonstrated that the hybrid detector requires lesser sensing time and can also reliably detect low SR signals. REFERECES [1] FCC, ET Docket o otice of proposed rule making and order, December [2] S. Haykin, Cognitive radio: brain-empowered wireless communications, IEEE Journal on Selected Areas in Communications, vol. 23, no. 2, Feb 2005, pp [3] GSM 05.05, Radio transmission and reception, ETSI, [4] J. Mitola III, Cognitive radio: an integrated agent architecture for software defined radio, Ph.D Thesis, KTH Royal Institute of Technology, [5] C. Cordeiro, K. Challapali, K. Birru, S. Shankar, IEEE : the first worldwide wireless standard based on cognitive radios, Proc. of DySPA05, ovember 2005, pp [6] H. Urkowitz, Energy Detection of Unknown Deterministic Signal, Proc. of the IEEE, April 1967, pp [7] A. Sahai,. Hoven, R. Tandra. Some Fundamental Limits on Cognitive Radio, In Allerton Conference on Communications, Control and Computing, October [8] Danijela Cabric, Artem Tkachenko, Robert W. Brodersen. Experimental Study of Spectrum Sensing based on Energy Detection and etwork Cooperation, TAPAS06 First International Workshop on Technology and Policy for Accessing Spectrum, August 5, 2006, Boston, MA, United States. [9] Steve Shellhammer, Performance of the Power Detector, IEEE /0075r0, May [10] D. Cabric, S.M. Mishra, R.W. Brodersen, Implementation issues in spectrum sensing for cognitive radios, Proc. of 38th Asilomar Conference on Signals, Systems and Computers 2004, ovember 2004, pp