Research Of Blind Equalization Algorithm Based On Neural Networks

Artificial neural network is a non-linear signal processor. It is made up of neurons which are interconnected, and it can do complicated operations. In digital communications, the severe inter-symbol interference is caused by time-delay, multipath transmission and coupled effects when signals transfer in channels. The severe inter-symbol interference degrades the reliability and decreases the rate of propagation. Equalization technologies can remove the effects of them. The conventional equalizer resorts to a training sequence in order to overcome inter-symbol interference. The blind equalization technology without the training sequence is possible to track the channel change, and gains signals by characteristics. So the reliability of communication is enhanced. Via neural network designing equalizer has many merits such as fast convergence rate, low bit error rate and so on. They are worth for us to studying.This paper mainly studies the blind equalization algorithm based on BP neural networks. The primary works of this paper can be summarized as follows:(1) This paper summarizes the basic theory of neural network and blind equalization. Do a research of blind equalization algorithm based on BP neural networks and it is deduced in details.(2) This paper analyzed the contradiction between convergence and steady error which is caused by fixed step size in blind equalization algorithm based on BP neural networks, and gives two kinds of adaptive variable step size algorithms based on BP neural networks. Then have comparisons with the fixed step size algorithm by computer simulations in different two channels. These simulations approve that the two adaptive variable step size algorithms have the better convergence performance, steady error and bit error rate. And do a further study on the effects of different error functions on the performance of the two adaptive variable step size algorithms.(3) This paper has an introduction of bilinear recursive neural networks, combine bilinear recursive neural networks with BP neural networks. Then propose an improvement algorithm of blind equalization algorithm which named BP-BRNN algorithm. The weights in every layer of this algorithm are deduced in details and this new improvement algorithm also has increasing in performances than the two adaptive variable step size algorithms by simulations in different channels.