Blind Source Separation Algorithm Based On Independent Component Analysis

Blind signal processing has been in development for decades, it can be divided into three categories:blind identification, blind deconvolution, blind source separation. At present, the main method for blind source separation is a method of independent component analysis (ICA), it is an efficient method for blind signal separation which has developed last two years. It means that it can estimate the source signal components only by source signals’prior conditions, even if the source signals and mixing matrix are unknown. Independent component analysis method has been widely used in speech signal processing, medical signal processing, image processing, array signal processing and so on. This paper introduces the blind source separation research present situation and the study of independent component analysis current situation, elaborated the concept of independent component analysis, basic model, the restrict conditions of solving and the solution uncertainty. Then, according to the general steps of independent component analysis, describing preprocessing method, objective function, optimization algorithm, separation performance evaluation function. Then, introducing the application of the wavelet ICA method used in blind image separation, and using wavelet ICA to one-dimensional voice or manual signal blind source separation problem, through computer simulation we know that wavelet ICA also improved voice or manual signal blind source separation’s performance. In view of the natural gradient algorithm for step size dependence, this paper proposes an improved adaptive step-size method based on hyperbolic function. With the idea of Performance\nde\(PI), we set an improved criteria to judge whether the objective function L(w) is optimal.And through the wavelet ICA blind image separation computer simulation experiments, we confirmed the superiority of wavelet ICA. The natural gradient algorithm has a poor climbing ability, and easy to fall into local extreme values.Taking into account the chaos has the characteristic of randomness, regularity, ergodicity, etc. So we obtain chaos ICA method by combining chaos theory and natural gradient algorithm.We verified the separation performance of chaos ICA through image blind separation and blind speech separation experiments. And comparing the separation performance with the natural gradient algorithm’s, the results show that chaos ICA method for separating effect is more superior.