An extended ACDC algorithm for the blind estimation of convolutive mixing systems

This paper presents an extension of the ACDC algorithm introduced by Yeredor for the instantaneous mixing problem to the more general convolutive mixing problem with non-white sources. Further assumptions made on the source signals are their mutual statistical independence, nonstationarity and smoothness of their power spectra. The algorithm iterates the estimation of the mixing system (AC step) and the source statistics (DC step) until convergence is achieved. The proposed algorithm operates in the frequency domain, but unlike most frequency domain algorithms, it carries out some of the operations jointly for all frequencies. This allows us to overcome frequency dependent permutation and scaling problems. Disciplines Physical Sciences and Mathematics Publication Details This paper originaly appeared as: Mertins, A & Russel, I, An extended ACDC algorithm for the blind estimation of convolutive mixing systems, Proceedings. Seventh International Symposium on Signal Processing and Its Applications, 1-4 July 2003, vol 2, 527-530. Copyright IEEE 2003. This conference paper is available at Research Online: http://ro.uow.edu.au/infopapers/252 AN EXTENDED ACDC ALGORITHM FOR THE BLIND ESTIMATION OF CONVOLUTIVE MIXING SYSTEMS Alfred Mertins University of Oldenburg Institute of Physics 261 11 Oldenburg, Germany [email protected] ABSTRACT This paper presents an extension of the ACDC algorithm introduced by Yeredor for the instantaneous mixing problem to the more general convolutive mixing problem with nonwhite sources. Further assumptions made on the source signals are their mutual statistical independence, nonstationarity and smoothness of their power spectra. The algorithm iterates the estimation of the mixing system (AC step) and the source statistics (DC step) until convergence is achieved. The proposed algorithm operates in the frequency domain, but unlike most frequency domain algorithms, it carries out some of the operations jointly for all frequencies. This allows us to overcome frequency dependent permutation and scaling problems.This paper presents an extension of the ACDC algorithm introduced by Yeredor for the instantaneous mixing problem to the more general convolutive mixing problem with nonwhite sources. Further assumptions made on the source signals are their mutual statistical independence, nonstationarity and smoothness of their power spectra. The algorithm iterates the estimation of the mixing system (AC step) and the source statistics (DC step) until convergence is achieved. The proposed algorithm operates in the frequency domain, but unlike most frequency domain algorithms, it carries out some of the operations jointly for all frequencies. This allows us to overcome frequency dependent permutation and scaling problems.