A new eigenvector-based algorithm for multichannel blind deconvolution of input colored signals

Abstract

The development of a new algorithm is presented for the deconvolution of an unknown, possibly colored signal which is observed through two or more unknown channels described by rational system transfer functions. It is assumed that the channel orders as well as the root locations are unknown. The proposed algorithm starts with over-estimated channel orders and iteratively reduces them. The root locations are simultaneously estimated. This algorithm may also be used in multichannel system identification problems. >