Research On Blind Source Separation In Reverberation Background

Target detection and recognition under reverberation background has been one of the important and difficult problem in underwater acoustic signal processing. The main reason is the uncertainty of the target and the complexity and the variability of the marine environment. Blind separation algorithm basically does not require prior knowledge of source signals and ocean channel. Based on this advantage, the blind separation algorithm can be applied to target detection in reverberation background.Combined the algorithm of blind source separation with the model of active sonar receiving data, some basic assumption of target echo, reverberation and ambient noise are made. A separability method based on the spatial correlation characteristics is improved to enhance its universality. In this way the number of source signals of blind separation can be determined.Considering the properties and mixture model of the target echo and reverberation, in the paper four blind separation methods are introduced:second order nonstarionary source separation(SONS), Time-Frequency Separation(TFS), the max-SNR blind separation and blind source switch algorithm based on kurtosis, and compared their advantages and disadvantages. In order to better evaluate the separation effect of the signal containing linear frequency modulation characteristics, the evaluation criterion based on WVD figures is proposed. The result of simulation proves the correctness of the algorithms. By the processing and analysis of sea trial data, the feasibility and effectiveness of the first three methods in the practical application are proved, which enable the target echo having clearer chirp characteristics and improving correlation coefficient, but the fourth method in practical applications fails. And combined with the engineering practice a multi-stage blind separation algorithm is put forward in the paper. In smaller angle separation effect of it is still very evident.