Blind signal separation via simultaneous perturbation method

Abstract

When independent plural signals are mixed and the mixed plural signals are measured, the blind signal separation technique is a very interesting approach to separate these signals only based on the measured signals. This technique is applicable to many fields including communication engineering, signal processing, image processing, analysis of organs inside a body and so on. We propose a recursive method to obtain a separating matrix based on the mutual information, via the simultaneous perturbation optimization method. The simultaneous perturbation method estimates a gradient of the mutual information with respect to the separating matrix, based on a kind of finite difference. Therefore, the separating matrix is updated by only two values of the mutual information. Some examples for image signals and audio signals are shown to confirm viability of the proposed method In these examples, our method separated some signals from mixed ones properly. This method is applicable to on-line separation because of simplicity of the algorithm.