A novel eigenmode-based neural network for fully automated microstrip bandpass filter design

Abstract

A novel eigenmode-based neural network (NN) for a fully automated design of microstrip bandpass filter (BPF) is proposed in this paper. The NN is now useful for BPF designs because a part of design procedure can be automated. Although the design time is reduced by the NN, an extra structural optimization is still needed as post processing. This is because a passband response is degraded by undesired but intrinsic cross couplings that are not considered in filter circuit synthesis. No fully automated BPF design techniques have been developed yet. In the proposed method, the NN is constructed based on the coupling matrix of transversal array filter, which can evaluate all the couplings between resonators as eigenmodes appearing in BPF. As examples, two third-order parallel-coupled microstrip BPFs are automatically designed with the proposed NN. The effectiveness of the proposed NN is verified numerically and experimentally.