Based On Information Maximization Method Of Multi-channel Blind Speech Signal Separation

Blind signal separation is an emerging technique of array processing and data analysis that aims to recover unobserved signals or ources?from observed mixtures. There are many potential exciting applications of blind signal separation in science and technology, especially in wireless communication, medical diagnosis, image enhancement and radar signal processing.In this thesis the emphasis is given to information-maximization approach. We can derive a self-organizing learning algorithm that maximizes the information transferred in a network of nonlinear units. This approach makes use of the information theory and the charateristic of speech signal , and then transfer estimating the mixing coefficient to minimizing the mutual-information or maximizing the entropy. So we can solve this problem by using the standard stochastic gradient descent method.Firstly, the concept of blind signal separation and some approaches are introduced . And the principle, derivation and computation steps of information-maximization approach are presented . Then we bring a improved algorithm and apply it to the speech signal separation. The simulation results show that the improved algorithm is effective.