M.-B. Li, P. Saratchandran, G.-B. Huang, and N. Sundararajan (Singapore)
Complex Growing and Pruning, channel equalization, neural networks
In this paper, a complex-valued growing and prunging radial basis fuction (CGAP-RBF) network is proposed for communication channel equalization problems. The algorithm makes use of the concept of "significance" of a hidden neuron which is defined as the average information content of that neuron. By linking the significance of the "nearest" neuron to the learning accuracy, a growing and pruning compact RBF netwwork is developed. When there is no growing or pruning the network uses a comlex Extended Kalman Filter to adjust the network parameters. Simulation results show the better performance of CGAP-RBF equalizer compared to that of CRBF equalizer of Cha and Kassam and CMRAN equalizer of Deng et.al. for a QAM equalization problem in terms of symbol error rate and complexity.