The Blind Source Separation Algorithms Based On Cyclostationary

Blind Source Separation (BSS) was proposed at the end of 1990s, which is a completely new way of signal processing, and it can recover the desired signal without the priori channel information and the source signals. This technology is mainly used to recover and extract the Potential components in the multi-channel, then analyze the desried signal from the mixed signals. BSS can separate the desired signal from the mixed signals without the Training sequence, it has been widely used in such fields as: communication, image processing, Biomedical, geological exploration, and etc. BSS has become an important hot research topic in the fields of signal processing.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 simulation experiments show that the new algorithms have better separation results.The major contribution of this paper is summarized as follow:1.Comprehensive analyzed the cyclostationary theory and its application in the field of signal separation. Introduced the Concepts, principles and techniques, and the classification of blind source separation algorithm, mainly analyzed the BSS algorithms based on the second order and the high order cyclostatistics.2. Described the principle of controlling the BSS matrix based on degree of cyclostationary (DCS), derived the second-order DCS criterion, and a new BSS algorithm based on the criterion was proposed. Then, extend the concept of DCS to high-order statistics, derived the 3-order cyclic cumulant DCS criterion, proved the effection of the new criterion for the signal separation, and proposed a new BSS algorithm based on the 3-order cyclic cumulant.3. Described the optimization rules to establish objective function extremum and analyzed the joint approximate diagonalization matrix (JADE), and proposed a new BSS algorithm of CS JADE. This algorithm whitening processed the received mixed signals firstly and orthogonal transformated the components according to decorrelation principle to transformated the components according to decorrelation principle to realize the separation effect. Simultaneously, introduced the cyclic cumulant matrix into the JADE principle, proposed the blind source separation algorithm of cyclic cumulant JADE principle.4. According to the theory of information theory, analyzed the nature of KL divergence and the minimum mutual Information (MMI), proposed a new blind source separation of minimum mutual Information. The proposed 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.