SOFNN-based equalization using rival penalized controlled competitive learning for time-varying environments

Abstract

A self-organizing fuzzy neural network (SOFNN)-based equalization is presented for time-variant environments. A rival penalized controlled competitive learning (RPCCL) is adopted to locate global minimum for mean vectors of fuzzy rules and organize the ideal structure of the fuzzy neural network (FNN) simultaneously. Then a supervised learning by means of the backpropagation (BP) algorithm is used for adjusting all parameters of the FNN. Results show that the performance of the newly designed strategy is much improved for adaptive filters with conventional FNN or least mean square (LMS) scheme.