Title: Equalisation of a wireless ATM channel using a pruned recurrent neural network

Authors: Dong-Chul Park

Addresses: Intelligent Computing Research Lab., Department of Electronics Engineering, Myong Ji University, 449-729, Republic of Korea

Abstract: A equalisation method of a wireless Asynchronous Transfer Mode (ATM) communication channel using a Complex BiLinear Recurrent Neural Network (CBLRNN) is proposed in this paper. A Genetic Algorithm (GA) is used for the pruning process of the trained CBLRNN. As a result, a pruned Bilinear Recurrent Neural Network (BLRNN) is obtained and the pruned BLRNN can reduce the computational cost by 29.9% in terms of the number of weights. The equaliser based on CBLRNN pruned by the GA is compared with Decision Feedback Equaliser (DFE), Volterra filter based equaliser, and Multilayer Perceptron Neural Network Equaliser. Experiments show that the pruned CBLRNN equaliser for 8PSK signals gives favourable results in the Symbol Error Rate (SER) criterion over conventional equalisers.

Keywords: equalisation; pruning; asynchronous transfer mode; recurrent neural networks; RNNs; wireless ATM channels; genetic algorithms; symbol error rate.

DOI: 10.1504/IJSCC.2010.035416

International Journal of Systems, Control and Communications, 2010 Vol.2 No.4, pp.337 - 348

Published online: 30 Sep 2010 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article