Multichannel blind deconvolution and equalization using the natural gradient

Abstract

Multichannel deconvolution and equalization is an important task for numerous applications in communications, signal processing, and control. In this paper, w e extend the efficient natural gradient search method in [I] to d e r i v e a set of on-line algorithms for combined multichannel blind source separation and time-domain deconvolution/equalization of additive, convolved signal mixtures. We prove that the doubly-infinite multichannel equalizer based on the maximum entropy cost function w i t h natural gradient possesses the so-called "equivariance p r o p e r t y " such that its asymptotic performance depends on the normalized stochastic distribution of the soursce signals and not on the characteristics of the u n k n o w n channel. Simulations indicate the ability of the algorithm to perform efficient simultaneous multichannel signal deconvolution and source separatiam.

Speech intelligibility improvement using convolutive blind source separation assisted by denoising algorithms

“…As there is a convolution in equation (2), the performance of the algorithms in the time domain is computationally expensive and time consuming (Amari et al, 1997;Kawamoto, 1998;Douglas and Sun, 2003;Buchner et al, 2004). However, they present good results for instantaneous mixtures.…”

Gradient Adaptive Paraunitary Filter Banks for Spatio-Temporal Subspace Analysis and Multichannel Blind Deconvolution

“…The system's overall computational complexity usually can then be reduced by assuming that the system adapts slowly, so that delayed versions of previously-computed signals are used in place of similar signals appearing within the parameter updates. Similar approximations have been used in other natural gradient methods for singleand multichannel deconvolution tasks [43,45,50]. Details of these approximations are given in the specific applications that follow.…”

“…In [43][44][45]50], a procedure is described and used to simplify related algorithms for multichannel blind deconvolution. This procedure assumes that the adaptive system is slowly varying, such that (40) such that delayed coefficient values can be used in place of more-recent coefficient values within the filtered gradient calculations.…”

“…Solutions to the above task are useful for separating digital communications signals in an unknown multiuser environment [49,50] as well as separating acoustic signals from multisensor recordings in reverberant room environments [51][52][53]. We now develop algorithms for multichannel blind deconvolution for prewhitened signal mixtures.…”

Speech dereverberation algorithm using transfer function estimates with overestimated order

“…A class of techniques has been proposed based on the assumption that the source signal has independent and identically distributed characteristics [1,2]. In [2], for example, an algorithm is proposed for multichannel deconvolution and the equalization of additive and convolutive signal mixtures.…”