Study and application of wavelet network parameter initialization
Mingyan Jiang · Information technology newsletter · 2005
Based on the wavelet neural network, this paper proposes a new network named varying scales wavelet neural network to reduce wavelet-neuron numbers and simplify network structure. In order to prevent the network from getting into the local minimal point, entropy function is used as penalty function. The new network is applied to channel equalization. Simulation demonstrates that this network has less wavelet-neurons and recursive steps while converging to global minimal point than usual wavelet neural network equalizer.