Stability and performance of adaptive algorithms for multichannel blind separation and deconvolution

The problem of blind source separation is reviewed and the properties of stability and performance of the classic adaptive algorithms are derived. In case of an unstable separating solution, a stabilization procedure is proposed. These findings are then extended to the problems of single-channel and multichannel blind deconvolution. It is shown that the algorithms for all these problems can be stabilized for any situation of nonlinearities and source distributions.