Blind signal separation using overcomplete subband representation

This paper discusses a multirate filterbank-based extended infomax algorith m for real-world signal separation, i,e,, convolved mixtures separation. Si nce convolution in the time domain corresponds to instantaneous mixing in t he frequency domain, polyphase subband projection naturally becomes an effi cient alternative to the Fourier transform based frequency domain approach. The online implementation proposed is featured by a simultaneous inverse c hannel identification in the frequency domain and signal filtering in the t ime domain. It is shown that an over-representation structure reduces alias ing between different bands and results in more accurate inverse channel es timates. Therefore, it provides better performance than the Fourier transfo rm based structure in the measures of both separation and distortion. The p erformance limitation of the method is also evaluated in terms of the Wiene r solution.