Date of Award

Spring 1994

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Electrical Engineering - (Ph.D.)

Department

Electrical and Computer Engineering

First Advisor

Yeheskel Bar-Ness

Second Advisor

Denis L. Blackmore

Third Advisor

Alexander Haimovich

Fourth Advisor

J. Mazo

Fifth Advisor

Zoran Siveski

Abstract

In high speed digital transmission over bandlimited channels, one of the principal impairments, besides additive white Gaussian noise, is intersymbol interference. For unknown channels, adaptive equalization is used to mitigate the interference. Different types of equalizers were proposed in the literature such as linear, decision feedback equalizers and maximum likelihood sequence estimation. The transmitter embeds sequences with the data regularly to help the equalizer adapt to the unknown channel parameters.

It is not always appropriate or feasible to send training sequences; in such cases, self adaptive or blind equalizers are used. The past ten years have witnessed an interest in the topic. Most of this interest, however, was devoted to linear equalization

In this dissertation we concentrate on blind decision feedback equalization and blind maximum likelihood sequence estimation. We propose a new algorithm: the "decorrelation algorithm," for controlling the blind decision feedback equalizer. We investigate properties such as convergence and probability of error.

A new algorithm is also proposed for blind maximum likelihood sequence estimation. We use two trellises: one for the data and the other for the channel parameters. The Viterbi algorithm is used to search the two trellises for the best channel and data sequence estimates. We derive an upper bound for this scheme.

We also address the problem of ill convergence of the constant modulus algorithm and propose a technique to improve its convergence. Using this technique, global convergence is guaranteed as long as the channel gain exceeds a certain critical value.

The question of the Viterbi algorithm's complexity is important for both conventional and blind maximum likelihood sequence estimation. Therefore, in this dissertation, the problem of reducing the complexity of the Viterbi algorithm is also addressed. We introduce the concept of state partitioning and use it to reduce the number of states of the Viterbi algorithm. This technique offers a better complexity/performance tradeoff than previously proposed techniques.

Share

COinS