Research On Spectrum Sensing In Non-gaussion Noise For Cognitive Radio System

With the rapid development of wireless communication applications and services, the limitedfrequency spectrum is becoming increasingly precious and rare. With the traditional fixed spectrumallocation method, most of the licensed radio spectral bands are under-utilized, leading to a lowutilization efficiency of the radio resource. How to improve the spectrum utilization has been themost important problem in the next generation wireless communication to be solved. Under thisbackground, cognitive radio (CR) has emerged as a key technology of improving the spectrumutilization efficiency through opportunistic access, sharing and use of the radio resources to solvethe spectrum shortage problem. The CR system needs to detect the presence/absence of primaryusers by spectrum sensing in order to avoid causing interference to the primary users. When thespectrum is free, the second users can use the spectrum. While the second users are using thelicensed spectrum, they should also be able to detect the primary users when they become activeand vacate the spectrum within a certain amount of time. As a result, spectrum sensing is one of themost important technologies of the CR system to realizing spectrum management and sharing andwireless communication. Researches on spectrum sensing become extremely significant. There areseveral traditional signal detection methods, which can be used to spectrum sensing. The problem iseven more complicated due to the fact that the propagation channels undergo non-Gaussian noise,and the performance of the traditional spectrum sensing detectors optimized against Gaussian noisemay degrade drastically. Futhermore，the sensing performance of single user is very low because ofthe low signal-noise-ratio of primary user with the fading and shadow sensing channel. In thisdissertation, spectrum sensing technologies in non-Gaussian noise are studied thoroughly with theaids of statistical signal estimation and detection theory and fractional lower statistical theory andmulti-users cooperation.The main work and contributions of this dissertation can be summarized as follows:(1) A Rao-based detector with linear model in the background noise modeled by the Gaussianmixture distribution (GMD) is researched. The generalized likelihood ratio test (GLRT) is proposedfor spectrum sensing with unknown parameters. However, the maximum likelihood estimation(MLE) of unknown parameters under two hypotheses is very difficult, making the GLRT-baseddetector non-applicable in the GMD noise. To solve this problem, a Rao-based detector is proposedfor the detection of a linearly modeled PU signal in the GMD noise. In order to simplify the computation, the moment-based estimation method instead of MLE is used for the unknown noisevariance estimation, and the time mean is used to calculate the Fisher information array elementinstead of the complex integral operation. Then the detection performances in terms of theprobabilities of detection and false alarm are derived. In addition, multi-user cooperative detectionschemes based on the OR rule and Chair-Varshney rule are proposed to improve the performance ofthe single user detection. Through theoretical analysis and Monte Carlo simulations, it is shown thatthe performance of the energy detector is drastically decreased in the GMD noise, while the Raodetector has a better performance in the GMD noise even in low SNR. The probability of detectionis a monotonically increasing function of the mixture parameter between0.01and0.03. Besides, thecooperative detection scheme based on the improved OR rule and Chair-Varshney rule can improveobviously the global probability of detection, which can greatly decreases the interference to theprimary user.(2) Cooperative spectrum sensing based on Rao detector in the generalized Gaussiandistribution (GGD) noise is researched. Several spectrum sensing detectors have been proposed forthe GGD noise, depending upon a priori knowledge of various information, which may not bereadily available in practice. A novel Rao-based detector is proposed for the detection of a primaryuser in GGD noise. The statistic of the proposed Rao detector does not require any a prioriknowledge about the primary user signal and channels, and is only the function of the shapeparameter of GGD noise. At the same time, the Rao detector can improve the utilization rate of theblank frequency spectrum through nonlinear power operation of the observed signal and effectivelyreducing the noise non-Gaussianity. The detection performances of the Rao detector and thetraditional energy detector are derived for the low signal-to-noise ratio regime. Furthermore, wealso analyze and compare the performances of the two detectors in terms of the asymptotic relativeefficiency. In order to overcome the harmful fading and shadowing effects, the Rao-based detectionis extended to a multi-user cooperative framework by using the "K-out-of-M" decision fusion ruleand considering erroneous reporting channels due to Rayleigh fading. Analytical and computersimulation results show that the Rao detector can significantly enhance the spectrum sensingperformance over the conventional energy detection and the polarity-coincidence-array method inGGD noises. As the degree of non-Gaussianity increases, the performance of the energy detectordecreases while the performance of the Rao detector is greatly improved. Furthermore, the Majorityrule should be used to obtain a high performance over the Rayleigh fading reporting channels, which can improves the spectrum efficiency and decreases interference to primary users.(3) A FLOM-based spectrum sensing in the symmetric α-stable distribution (S α S) noise isresearched. Since the S α Sdistribution has no finite second and high order statistic and does notpossess a compact analytical form of PDF, the traditional spectrum sensing methods in general notapplicable. A FLOM (fractional lower order moment)-based scheme is proposed, which does notrequire any a priori knowledge about the primary user signal, channel gain and the noisedistribution and is a non-parametric method. The FLOM detector is a simple and easy-to-implementdetection device. The statistic is defined as the fractional lower order moment of the observed signalinstead of the two-order or high-order moment and the fractional power operation can reduce noisespikes, so the FLOM detector can effectively solve the spectrum sensing problem in stabledistribution noise. Its detection performance for non-fading sensing channels and Rayleigh fadingchannels is derived by using the central limit theorem. Through theoretical analysis and numericalsimulations, it is observed that the FLOM detector can significantly enhance the detectionperformance over the energy detector and the Cauchy detection in the S α Snoise and thecooperative spectrum sensing scheme has a significantly higher global probability of detection thanthe non-cooperative one.