A parallel implementation of a smoothed adaptive deconvolution filter

Abstract

This paper considers a multiple processor adaptive approach to deconvolution estimation obtained by applying the extended least squares (ELS) technique. The approach adopted within the system only requires one input (the corrupted signal). This contrasts with the more common use of a FIR filter coupled with the least mean square estimation algorithm. The adaptive deconvolution smoothing filter has been implemented on a three-transputer system. The network uses a transputer to act as a digital deconvolution smoothing filter while a second transputer handles the extended least squares (ELS) parameter estimation algorithm. A third transputer calculates the gain factor for the deconvolution smoother. The three transputers operate in parallel. The example considered in this paper is of a speech signal which has passed through a distorting communication channel resulting in a narrow band signal. The distorting channel is assumed to be represented as an all-pole model. This signal is then corrupted by additive wide band noise. The transputer network is used to estimate the original speech signal.