The Research And Implementation Of Blind Signal Separation System

In recent years, the research on the blind signal separation becomes more and more important in signal processing areas, and has made rapid development. Requiring very little prior knowledge of the original signal and hybrid systems, the blind signal separation can recover the original signal from the observed mixed-signal, used in a wide range of applications such as in radar, sonar, radio communications, biomedical and image processing.This thesis focuses on the blind signal separation as our research object. Studying the blind source separation algorithm instantaneous linear aliasing situations, main tasks are as follows:In the first place, some of the basic theory of blind signal separation is studied. The basic model of blind signal separation is described, and the constraints of blind signal separation and pretreatment are illustrated. The criteria to establish cost function and optimization methods, and the method to evaluate the performance of separation are proposed.Secondly, it studied the basic algorithm of blind signal separation and has made some improvements, and conducted a theoretical analysis of the ICA algorithm, PCA algorithm, Fast ICA algorithm, at the same time using the MATLAB for simulation of the algorithms. For the problem of bad convergence when the Fast ICA algorithm in the time of the not strong original data independence, we proposed an improved method, and conducted a simulation study.Voice separation problem is an important component of blind signal separation technical studies. It not only inextricably linked to "cocktail party problem", which is the origin of the study of blind signal separation, but also the solution to the problem has a very wide range of practical significance. For this reason, this thesis also studied voice separation technique and had developed a blind signal separation system for the voice signal. Blind signal separation system can capture the true voice signals and use a different algorithm to separate. Analyzing the strengths and weaknesses as well as application areas of various algorithms, some valuable conclusions are presented as a reference for further study. This thesis provides certain reference value to the actual works in the blind source separation.