Nonlinear Feedback Equalization of Digital Signals Transmitted Over Dispersive Channels

Abstract

<p> This thesis deals with the problem of digital communication over noisy dispersive channels. The dispersion causes the overlapping of successive received pulses thus creating intersymbol interference which severely limits the performance of conventional receivers designed to combat only additive interference or noise.</p> <p> In this thesis Bayes estimation theory has been applied to obtain a new, optimum, unrealizable receiver structure for the improved reception of noisy, dispersed, pulse amplitude-modulated (PAM) signals. By making certain approximations, a realization of this structure, known as the estimate feedback receiver or equalizer, is obtained. It consists of the combination of a matched filter and a nonlinear, recursive equalizer having, in the case of binary signals, a hyperbolic tangent nonlinearity in the feedback path. The well known decision feedback equalizer is shown to be a small noise limiting case of the estimate feedback equalizer. A saturating limiter is also considered as an approximation to the hyperbolic tangent nonlinearity.</p> <p> A new adaptive algorithm for the iterative adjustment of the estimate feedback equalizer is derived. It incorporates an extrapolation process which has the purposes of accelerating convergence of the equalizer's parameters to their optimum values and of maintaining the equalizer's frame of reference. It is constrained so that the equalizers parameters always move toward their optimum values.</p> <p> Computer simulations are used to demonstrate the properties of the adaptive estimate feedback equalizer and to compare them to those of presently known equalizers. When the estimate feedback equalizer is used, without a matched filter preceding it, to equalize phase distorted channels, its performance is seen to be superior to that of existing equalizers. The performance of an equalizer using a saturating limiter in place of the optimum hyperbolic tangent nonlinearity is seen to be almost as good as that of the estimate feedback equalizer.</p>