A Blind Equalization by Cascaded Hybrid Wavelet Neural Network

A cascaded hybrid wavelet neural network(WNN) adaptive blind equalizer is proposed for severe nonlinear distortion channel.This equalizer cascades a transversal filter before the input layer of WNN,and the input signal of WNN is inferred from the output of the nodes of the transversal filter.The gradient information of the transversal filter and WNN can be obtained by constant modulus cost function respectively,and then the gradient information for updating the parameters of the hybrid WNN can be obtained by weighted fusion of the two gradient information.The cascaded hybrid WNN blind equalizer realizes the combination of linear and nonlinear optimization on non-convexity error surface.Simulation results with ordinary telephone channel and nonlinear channel show that the cascaded hybrid WNN blind equalizer has a better equalization performance comparing with FNN(feedforward neural network) and traditional WNN blind equalizers.