Research On Signal Detection Algorithms In Non-Gaussian Alpha-Stable Distributed Noise

In traditional signal detection,it is usually assumed that the background noise obeys Gaussian distributions.However,in many applications,background noise is non-Gaussian and exhibits impulsive behavior,e.g.,man-made noise,atmospheric noise,underwater acoustic noise,interference from other users and so on.In impulsive noise,traditional detectors under the Gaussian noise assumption suffer from a drastic performance degradation due to the mismatch between the Gaussian noise model and actual impulsive noise.To obtain better detection performance,the detection in impulsive noise comes into consideration.Practice shows that alpha-stable distributions are successful models for the impulsive noise,and it catches researches' increasing attention in the literature.Aiming at solving the communication reconnaissance problems,this paper studies non-cooperative signal detection in alpha-stable distributed noise.First,the measured data collected in the seawater is fitted using the alpha-stable distribution,which shows a good fitness.Then,this paper investigates the detection methods in alpha-stable distributed noise from three aspects.First,the Kolmogorov-Smirnov(K-S)test is applied for signal detection in alpha-stable distributed noise.The algorithm based on the two-sample K-S test is proposed,its performance under different signals and characteristic exponents is evaluated through numerical simulations,and the effect of noise uncertainty on the detection performance is discussed.A huge gap between the detection performance for the DC signal and other signals does exist,and the reason is explored.To improve the performance for the Gaussian and BPSK signals,an improved algorithm based on the data preprocessing is proposed,and the performance relationship between different data preprocessing algorithms is proved.The square preprocessing algorithm is taken as an example,and its performance is analyzed.Simulation results prove the efffectiveness of the data preprocessing.Second,this paper studies the detection methods based on the statistic moments.Firstly,the performance limitation for the energy detector and fractional lower order moments-based detector is investigated.To obtain the optimal exponent of the fractional lower order moments-based detector,the asymptotical expression for detection probability is derived.Based on the result,the optimal exponent can be calculated by solving an optimization problem in an iterative optimization algorithm.The numerical simulation proves the effectiveness of the obtained optimal exponent.Then,a detector based on the logarithmic moment is proposed.The theoretic expressions of the detection and false alarm probabilies are derived,and the performance under different channels and noise uncertianty is discussed.Simulation results show that the logarithmic detector can perform better than the fractional lower order moments-based detector in impulsive nosie.Finally,this paper proposes a detector framework based on the nonlinear transformations,and gives the basic principle for choosing a suitable transformation function.Then,a soft limiter-based detector and a Gaussian function-based detector are constructed,which represent the detectors based on the increasing and decreasing bounded functions,respectively.The detection and false alarm probabilities are derived,the parameter optimization for the detectors is discussed,and the computational complexity is compared with other detectors.Theoretic analysis and simulation results show that detectors based on the bounded nonlinear transformation can achieve good performance with low computational complexity.Especially,the Gaussian function-based detector performs much better than traditional detectors for both the computaional complexity and detection probability,.