NEURAL NETWORKS PROCESSING OF DIGITAL COMMUNICATION SIGNALS

Abstract

In digital communication systems, practical problems arise, some of them are related to the channel and others are related to the receiver. These problems degrade the performance of the receiver. In this thesis, a new technique for solving the conventional problems of estimation and detection is proposed using neural networks (NN) approach. Recently NN has attracted the attention of scientists and engineers in many fields based on their many computational features such as massively parallel and distributed structures for signal processing, and give new way to solve many problems. So in this thesis, we make use of multilayer perception (MLP) as well as recurrent NN models trained according to the Back-propagation and Hopfield learning algorithms respectively. The contributions in this thesis are summarized in the following paragraphs. The detection of the data with channel encoding using MLP NN decoder and showing that the performance is much better than hard desion and soft decision decoders conventionally used. {;f The design of synchronization system using MLP NN for spread spectrum communications that minimiz es the effect of ranging problem between two mobile users and search the received code phase faster than· the traditional methods. The design of matched filter using NN techniques that is matched to one code and minimiz es the cross correlation between that code and any set of different given codes. This may be the key of implementing what we can call neural matched filters.