NONLINEAR CHANNEL EQUALIZATION WITH MAXIMUM COVARIANCE INITIALIZED CASCADE-CORRELATION LEARNING (TuePmPO3)

Author(s) : Arto Kantsila (Tampere University of Technology, Institute of Digital and Computer Systems, Finland)

Abstract : In this paper we have studied maximum covariance initiali-zation scheme and cascade-correlation learning to improve the performance of a multilayer perceptron network equal-izer in nonlinear channel environment. The initialization scheme enables faster convergence and the cascade-correlation learning provides adaptive network size. These methods are compared to a traditional MLP network equal-izer and to a simple linear equalizer.

Menu