Study On Fast-ICA Algorithm For Blind Source Separation

Blind source separation (BSS) aims at recovering source signals without any prior knowledge of the source signals and that of the channel. Especially when the source signals are statistically independent, BSS is termed as independent component analysis (ICA). Over the past years, BSS has attracted growing attention for its broad application in various fields, such as wireless telecommunication, audio and image signals processing, financial application, and so on. In this paper, we focus on the linear mixture model, a new fast-ICA algorithm without prewhitening is proposed.We first introduce the development and application of blind source separation problem. Based on the mathematical model, we analyze two indeterminacies inherent in BSS and several assumptions of this problem. Then we discuss the optimization theory and summarize typical contract functions and existing algorithms. Computer simulation show how these algorithms work in BSS problem. Existing fast-ICA algorithm has been reviewed by discussing the concept of kurtosis, nongaussian property and gradient method. After that, by combining the separating vector in the existing fast-ICA algorithm with the prewhitening matrix as well as considering the autocorrelation matrix, we proposed a new separating vector, which is used to recover statistically independent component from the observed data without prewhitening. The fixed point algorithm of the proposed separating vector is developed. Finally, the effectiveness of this new method is verified by computer simulations.Since fast-ICA algorithm is kind of fixed point approaches, it converges much faster than many adaptive algorithms. Substituting the prewhitening process with orthogonal constraint, the convergence speed of the algorithm can be increased and the post forward propagation of the estimation error in prewhitening process is evaded, the proposed algorithm performs well than the existing fast-ICA algorithm and many adaptive algorithms.