Unknown stochastic signal detection via non-Gaussian noise modeling

Detection of weak stochastic signal under non-Gaussian background is a difficult problem, especially when the prior knowledge of the background as well as the signal is lacking. Traditional detection methods hardly consider both non-Gaussian background and lack of prior knowledge condition simultaneously. This paper proposes an unknown stochastic signal detection algorithm using information geometry tools. Firstly, we use Gaussian Mixture Model (GMM) to model the signals under detected. Secondly, the Kullback-Leibler divergence (KLD) between the GMMs of signal and noise is calculated to measure the difference between the signal and noise. Thirdly, the signal is detected by comparing the KLD with the threshold. Compared to the previous detection approaches, the proposed algorithm is independent of the prior hypothesis, so that it is adaptive for non-Gaussian detection background with deficiency of prior knowledge condition. Simulation results are presented to show the effectiveness and performance advantage of the proposed algorithm.