Improving Performance of an Adaptive Equalizer using EKF Trained Multilayered Neural Networks

摘要

In this paper, the Extended Kalman Filtering (EKF) based learning algorithm which has much faster convergence speed than the backpropagation algorithm has been extended for training the neural network used for Adaptive Equalization of communication channel. Further, it has been shown that the bit error rate obtained using the EKF algorithm is much superior to that was obtained with the backpropagation algorithm. But due to its computational intensity, parallel processing of the EKF based learning algorithm is indispensable for real time implementation. Hence, parallel implementation of the EKF based learning algorithm on a network of three transputers has been developed.