Research On Optimization Of Blind Source Separation Algorithms Based On Independent Component Analysis

Blind source separation technology is one of the hot research direction in the field of information and signal processing in the 1990s, this technology can separate individual source signals by the mixed signals according to the statistical characteristics of source signals.when the prior knowledge about the mixed source signals and the mixed system completely is unknown or only knew a few case. Nowadays, independent component analysis is one of the most effective methods to solve the problem, it is as a method of signal processing developed by the blind source separation technology and its main task is to extract the source signals from the mixed signals. Blind source separation technology can be widely used in speech signal processing, medical signal processing, image processing, wireless communication, seismic signal survey and so on. It has the very important practical value in the society.This paper is main to optimize the convergence speed and signal separation performance of the algorithm based on the independent component analysis and the effectiveness of the algorithm is verified by the speech signal. The research contents of this paper can be divided into the following five aspects:(1).This paper elaborate the basic theory of blind source separation problem, that includes different mixed mathematical model in the linear instantaneous mixture and convolutived mixture conditions, some constraint conditions according with the realistic situations in independent component analysis, the pretreatment processing of the signals, cost function of different separation criterion and evaluation standard of the separation performance in the algorithm.(2).In view of the contradiction between convergence speed and separation performance in blind source separation algorithm, a new adaptive algorithm based on improved separation performance index is proposed. The proposed algorithm restructures the traditional separation system and define a new separation performance index parameter, the algorithm takes this new parameter as the independent variable and attains the adaptive updating rule of the step-size using the Rayleigh distribution function. Then the simulation experiment will be carried on combined with the natural gradient algorithm and EASI algorithm in the new improved structure. The simulation results show that the convergence speed and separation performance of the proposed method is better than the traditional separation algorithm.(3).In view of the big calculation problem in the natural gradient algorithm, a symbolic function is imported, then the natural gradient blind separation algorithm can be changed as the sign gradient separation algorithm. At the same time, the momentum items are added into the sign gradient separation algorithm through studying the momentum items in neural network, this paper can reconstruct updating formula and form a new sign gradient blind source separation algorithm added the momentum items. Then the simulation experiment can be carried on through the comparison of different momentum items step factor and analyse the experimental results.(4).In view of the contradiction between convergence speed and performance index in blind source separation algorithm,a new separation algorithm based on an improved artificial bee colony algorithm is proposed. This proposed algorithm uses the absolute value of mixing signal kurtosis as the optimization objective function,the artificial bee colony algorithm has improved the search process at the onlooker stage. Then the improved artificial bee colony algorithm can converge quickly at the initial stage and improve the global convergence performance at the last stage, the mixed signals can be separated after using the improved artificial bee colony algorithm to optimize the initial separation matrix. Simulation results show that the improved algorithm can significantly speed up the convergence rate and maintain a good convergence performance, and that it can solve the contradiction between convergence speed and performance index in blind source separation algorithm.(5).This paper studys the convolutive mixed model that is more in line with the actual situation in blind source separation. Firstly, this paper improves the fast independent component analysis method based on negative entropy maximization, then it uses the three order convergence speed of Newton iteration form instead of two order convergence rate of modified Newton iteration in FastICA and a new blind source separation updating formula can be get. Besides, the paper should process the signal of convolution mixture and the convolution mixed signals can be separated in time domain by the improved algorithm. The simulation experiment can be carried on by the improved algorithm in the instantaneous mixture and convolution mixture case and analyse the experimental results.