A new variable step-size equivariant adaptive source separation algorithm

Abstract

In this paper, variable step-size blind source separation (BSS) algorithms are investigated. Since it is hard to achieve both fast convergence and stable tracking performance for a given step-size, step-size is crucial for the equivariant adaptive source separation (EASI) algorithms. Firstly, the measurement of the independence for the output signals is analyzed, then, a new EASI algorithm whose step-size is changed adaptively by mutual information is proposed. Computer simulation results show that the new algorithm has satisfactory convergence and stable tracking performance.