Neural Network Method for Dual-Mode Dual-Band DSPSL Bandpass Filter Design

Abstract

A dual-mode dual-band bandpass filter using double-sided parallel-strip lines (DSPSL) is proposed in this article. This design is based on vertical coupling, which doesn’t require any via holes. A method based on neural-network is developed and applied to analyze the DSPSL filter design. Through the neural network analysis of nonlinear circuits, it can enhance the efficiency of electro-magnetic (EM) analysis techniques for DSPSL filter design. A multilayer perception neural network (MLPNN) is implemented to model the microwave DSPSL filter circuit. Quasi-Newton method is adopted, which has shorter training period and the faster convergence. The trained artificial neural network (ANN) model maps a set of scatting parameters for the DSPSL filter in terms of its geometric parameters. A good agreement between ANN results and EM simulations verifies the validity of this proposed MLPNN model.