Blind Signal Separation Based On Independent Component Analysis Algorithm

The independent component analysis (ICA) addresses non-gaussian source signals under amusing independent each other, it performs blind separation for multichannel observed signals. It has widely applied in wireless communication, biological signals extraction, audio signals processing, image processing, noise reducing and so on.In this dissertation, the basic principle of the ICA has firstly been introduced. Based on it, the ICA is studied under instantaneous mixtures and convolutive mixtures blind separation models.The main works are as follows:1. Based on Herault-Jutten algorithm separating signals blindly, we obtain an algorithm using natural gradient under instantaneous mixtures model.The equivariance property has been discussed too.2. The fast fixed-point algorithm has been studied.We firstly consider one unit algorithm, extracting one original signal every time under instantaneous mixtures model. Several unit algorithms have been further obtained.3.The ICA algorithms have been studied under convolutive mixtures model in time and frequency domain. A contrast function has been built using orthogonalization time-delayed correlation matrices in time domain.The separation algorithm has been obtained by minimizing it with the Frobenius norm. Then the blind signal separation algorithm has been also discussed in frequency domain.We focus on discussing scaling and permutation problems.4. Finally we study a modified algorithm, which applies the fast fixed-point algorithm for complex signals to blind signals eparation in frequency domain. It's a one-unit algorithm. To prevent converging to the same signal, the deflation algorithm is used to separate signals one by one.Computer simulations shows the validity on the mentioned algorithms.