Adaptive non linear system identification and channel equalization usinf functional link artificial neural network

Abstract

In system theory, characterization and identification are fundamental problems. When the plant behavior is completely unknown, it may be characterized using certain model and then, its identification may be carried out with some artificial neural networks(ANN) like multilayer perceptron(MLP) or functional link artificial neural network(FLANN) using some learning rules such as back propagation (BP) algorithm. They offer flexibility, adaptability and versatility, so that a variety of approaches may be used to meet a specific goal, depending upon the circumstances and the requirements of the design specifications. The primary aim of the present thesis is to provide a framework for the systematic design of adaptation laws for nonlinear system identification and channel equalization. While constructing an artificial neural network the designer is often faced with the problem of choosing a network of the right size for the task. The advantages of using a smaller neural network are cheaper cost of computation and better generalization ability. However, a network which is too small may never solve the problem, while a larger network may even have the advantage of a faster learning rate. Thus it makes sense to start with a large network and then reduce its size. For this reason a Genetic Algorithm (GA) based pruning strategy is reported. GA is based upon the process of natural selection and does not require error gradient statistics. As a consequence, a GA is able to find a global error minimum. Transmission bandwidth is one of the most precious resources in digital communication systems. Communication channels are usually modeled as band-limited linear finite impulse response (FIR) filters with low pass frequency response. When the amplitude and the envelope delay response are not constant within the bandwidth of the filter, the channel distorts the transmitted signal causing intersymbol interference (ISI). The addition of noise during propagation also degrades the quality of the received signal. All the signal processing methods used at the receiver's end to compensate the introduced channel distortion and recover the transmitted symbols are referred as channel equalization techniques.When the nonlinearity associated with the system or the channel is more the number of branches in FLANN increases even some cases give poor performance. To decrease the number of branches and increase the performance a two stage FLANN called cascaded FLANN (CFLANN) is proposed.This thesis presents a comprehensive study covering artificial neural network (ANN) implementation for nonlinear system identification and channel equalization. Three ANN structures, MLP, FLANN, CFLANN and their conventional gradient-descent training methods are extensively studied. Simulation results demonstrate that FLANN and CFLANN methods are directly applicable for a large class of nonlinear control systems and communication problems.