An improved cross-correlation based method of blind source separation in frequency domain

Abstract

A blind source separation algorithm based on cross-correlation in frequency domain is considered in this paper. Firstly we describe that the variation of the length of signal frame will influence the result of separation. Then the multi-frame size method through multi-stage is introduced to meet the requirements of the length of signal frame in all aspects. However we find that another problem called pseudo-solution also could lead separation to fail, so possible treatments are proposed. Experiments are provided to confirm the validity of this improved method.