Frequency domain realization of a multichannel blind deconvolution algorithm based on the n.pdfVIP

FREQUENCY DOMAIN REALIZATION OF A MULTICHANNEL BLIND DECONVOLUTION ALGORITHM BASED ON THE NATURAL GRADIENT Marcel Joho Philip Schniter
Phonak Inc., Champaign, IL, USA, (joho@ieee.org)
Dept. of EE, The Ohio State University, Columbus, OH, USA (schniter.1@osu.edu) Accepted for Independent Component Analysis and Blind Signal Separation ICA 2003, April 4-6, 2003, Nara, Japan ABSTRACT This paper describes two efficient realizations of an adaptive mul- tichannel blind deconvolution algorithm based on the natural gra- dient algorithm originally proposed by Amari, Douglas, Cichocki, and Yang. The proposed algorithms use fast convolution and cor- relation techniques and operate primarily in the frequency do- main. Since the cost function minimized by the algorithms is well- defined in the time domain, the algorithms do not suffer from the so-called frequency-domain permutation problem. The proposed algorithm can be viewed as an multi-channel extension of a single- channel blind deconvolution algorithm recently proposed by the authors. 1. INTRODUCTION 1.1. Problem description and preliminary work A well-known problem in acoustics is the so-called cocktail party problem, where the primary task is separation of a convolutive mixture of independently generated speech signals. A similar sce- nario occurs in data communications when an antenna array re- ceives a multi-path mixture of signals transmitted from several sources. Independent of any application, this task is generally known as the multichannel blind deconvolution (MCBD) problem. The general setup is shown in Fig. 1. The sequence of the source signals     is mixed and filtered by a causal convolutive mix- ing system    . This process can be described by a convolutional sum    !#%$(*) +-,  + /. + 0$ (1) where  1    is the sequence of the sensor signals and $   $   is an additive sensor noise sequence. In a blind setup, only  is accessible to the algorith