Recent Advances in Multi-Feature Assisted Neuro-Transfer Function Surrogate Based EM Optimization for Microwave Filter Design

摘要

This paper provides an overview of the multi-feature assisted neuro-transfer function surrogate based electromagnetic (EM) optimization method for microwave filter design. This optimization method addresses a situation that the starting point for the design optimization is far away from the design specifications as well as the filter response does not have clear feature frequencies. Multiple feature parameters are used to help move the pass-band of the filter response into the range of design specifications. The feature parameters used in this paper consist of feature frequencies and feature heights. Feature frequencies in this paper are calculated from the zeros of the transfer functions extracted from the EM responses. Feature heights are the magnitude of the responses at the mid-frequencies between two successive feature frequencies. The EM data samples for establishing the feature model are generated using parallel EM simulations. This optimization method has a better chance of avoiding local minima and reaches optimal EM solution faster.