Automated Microstrip Bandpass Filter Design Using Feedforward and Inverse Models of Neural Network

Abstract

In this paper, a new bandpass filter (BPF) design method fully automated by both feedforward and inverse models of neural network (NN) is proposed for an efficient design. A conventional inverse model can be applied only for the design of a substructure in BPF to evaluate its coupling coefficient, thus failing to design a multicoupled higher-order BPF like a microstrip filter. In the proposed automated design, a transversal coupling matrix is introduced to NN to evaluate all the couplings of BPF. As a result, the inverse model can instantaneously guess initial structural parameters with high accuracy by inputting an ideal transversal coupling matrix synthesized from design specifications. Then, using the feedforward model in conjunction with optimization algorithm enables to rapidly find optimal structural parameters from initial guess. The effectiveness of the proposed automated design scheme is verified through a structural design of a typical fifth-order microstrip BPF.