Design of Sparse Cosine-Modulated Filter Banks Using BP Neural Network

Abstract

This paper presents a design paradigm for sparse nearly perfect reconstruction cosine-modulated filter banks using BP neural network. Sparse FIR filter banks have lower implementation complexity than full filter banks with keeping a good performance level. First, a series of frequency response data satisfying perfect reconstruction condition are being selected. Second, the desired sparse linear phase FIR prototype filter is derived through the orthogonal matching pursuit performed under the weighted l2 norm, and the training function and hidden layer nodes in BP neural network. The simulation results fully testified the proposed scheme for the design sparse NPR cosine-modulated filter banks is reviewed.