Optimal design of digital FIR filters based on back propagation neural network

Abstract

In the traditional digital finite impulse response (FIR) filter design, there exist some limitations. For instance, unwanted signals cannot be filtered out by using these FIR filters. Therefore, to break these limitations, this paper proposes an optimization method for designing digital FIR filters based on the back propagation neural network (BPNN) algorithm. Firstly, an amplitude response model has been established based on the linear properties of the digital FIR filters. Then, the BPNN algorithm has been used to minimize the estimation error between the ideal and the actual amplitude response such that the optimal coefficients of the digital FIR filter can be obtained. Finally, several design examples are used to verify the performance of our proposed optimal design based on BPNN. The simulation results show that, compared with the optimal designs based on the sequential and rectangular window, our proposed optimal design based on BPNN can achieve better filtering effectiveness but at the cost of larger computational complexity.