A new method for source separation
Russell H. Lambert · 2002
A new adaptive filtering approach to multichannel source separation is presented. The method is based on an extension of traditional single channel blind equalization. This multichannel extension is presented theoretically and the results are demonstrated by simulation using both communications data and speech. The new source separation algorithm is compared to existing adaptive separation algorithms which handle multipath: the Herault-Jutten (see Signal Processing, vol.24, p.1-10, 1991) source separation algorithm, and the time-recursive Weinstein et. al. (see IEEE Trans. Speech and Audio Processing, vol.1, no.4, p.405-413, 1993) algorithms. The new method offers dramatic performance improvement over existing methods and the ability to handle different source data types in an optimal fashion.