Study On Methods And Applications Of The Blind Source Separation Based On Stable Distribution

Blind source separation(BSS) has become a widely concerned subject in recent years and has been widely used in the fields of speech source separation, geological exploration, underwater acoustic source separation, spatial array source separation, biomedical source separation, and so on. In practical applications, we offen meet some signals whose probability density function have significant peak pulses and thick tailings. The characteristics of these signals result in their deviation from Gaussian distribution. Alpha-stable distribution was introduced to model the signals and noise for its good performance. The research scope of the traditional blind separation is extended and the theoretical framework of the blind separation is gradually improved by it.This dissertation concentrates on the study of blind source separtion with alpha stable distribution. The main work and conclusion of this dissertation are listed as follows:(1) The purpose and significance of research on nonlinear blind source separation with alpha stable distribution is introduced in this dissertation. The prospect and derivation is described here. On the foundation of analyzing the basic theory, the basic algorithm especially the algorithm based on post nonlinear model and the nonlinear blind source separation model is summarized in this paper.(2) A nonliner blind source separation algorithm based on alpha stable model is presented. This algorithm includes two parts:linear processing and linear separation. The geometrical linearized processing is used in the first part. The blind source separation based on mininum dispersion coefficient and revolving transform algorithm is used in the second part. The new observed signals produced after the first part are input into the linear separation part in order to realize the source separation. The good performance and the high practical significance is proved by computer simulation from two points.(3) The distribution characteristic of the biomedical signals is analized and its parameter is estimated in this dissertation. The stable characteristic of this kind of signals is proved here. What's more, a kind of adaptive filter is designed based on the minimum deviation criterion in the condition of the stable distribution. This filter has a good performance compared to the LMS filter through the proof of the simulation. The nonlinear blind source separation algorithm proposed here is used into the biomedical signal processing field. The good performance and separation character in this special environment is proved by computer simulation.(4) The further development and the probable research direction of the blind source separation is induced after summarizing the advantages and disadvantages of this algorithm.