Blind Source Separation Algorithms Of Cyclostationary Signal Based On Minimum Mutual Information

Blind Source Separation (BSS) is a method which aimed at recovering and extracting desired signal from observed signals without the priori information of channel and the source signals. As a new branch of digital signal processing(DSP), BSS which has important theoretical significance and practical value has been used in many fields such as wireless communications, biomedical signal processing, geological exploration, speech recognition, image signal processing and etc. This technology has become an important hot research topic in the fields of signal processing and neural network.The traditional blind source separation algorithm was proposed on the basis of the transmitted signal stationary. But most of the natrual signals have the characteristics of cyclostationary (CS). This paper is mainly to process the mixed signal based on cyclostationary theory, and extract the useful information from the received mixed signals, which greatly reduced the computation complexity.The major contribution of this paper is summarized as follow:Introduced the Concepts, principles and techniques of the classical blind source separation algorithms, and analyzed the advantages and disadvantages of the algorithm.Comprehensive analyzed the cyclostationary theory and its application in the field of signal separation. Summarized several blind source separation algorithm based on cyclostationarity theory and analyzed several major cost function and learning algorithm.Through analyzing the characteristic of cyclostationarity signal and traditional blind source separation algorithm, proposed a new blind source separation algorithm. The new algorithm takes the mutual information as the measurement to illustrate the similarity between cyclic correlation matrix and the matrix units. According to the natural gradient optimization algorithm to achieve a minimum of mutual information to get the ideal separation matrix.The blind source separation algorithm based on minimum mutual Information used fixed step, which caused it can not satisfy the requirement of both convergence rate and steady-state error. In order to overcome the inner contradictions of above algorithms, this paper presents a modified variable step size algorithm. The simulation experiments show its advanteges of fast convergence, high approximation accuracy and better separation results.