Analysis of Signal Separation and Signal Distortion in Feedforward and Feedback Blind Source Separation Based on Source Spectra

摘要

Source separation and signal distortion in three kinds of blind soure separation (BSS) systems with convolutive mixture are analyzed. They consist of two feedforward BSS system, one trained in the time domain and the other trained in the frequency domain, and a feedback BSS system, trained in the time domain. An evaluation measure of the signal distortion has been investigated and conditions for source separation and distortion free have been derived. Based on these conditions, source separation and signal distortion have been analyzed. The feedforward BSS systems have some degree of freedom and the output spectrum can be changed. The feedforward BSS system, trained in the frequency domain, has a weighting effect, which can suppress signal distortion. However, this weighting effect is only effective only when the source spectra are similar to each other. Since, the feedforward BSS system, trained in the time domain, does not have any constraints on signal distortion free, its output signals can be easily distorted, A new learning algorithm with a distortion free constraint has been proposed. On the other hand, the feedback BSS system can satisfy both source separation and distortion free conditions simultaneously. Performed simulation results support our theoretical analysis.