Advanced Surrogate-Based EM Optimization Using Complex Frequency Domain EM Simulation-Based Neuro-TF Model for Microwave Components

Abstract

Surrogate-based electromagnetic (EM) optimization techniques have become popular for microwave design. Neuro-transfer function (neuro-TF) is one of the effective surrogates to represent the EM behaviors for design optimization. In this article, we propose an advanced surrogate-based EM optimization using neuro-TF developed by complex frequency domain (CFD)-based EM simulations. For the first time, we develop and introduce the CFD-based EM simulation using fast frequency sweep. We propose a novel transfer function zero/pole extraction technique based on the magnitude of the S-parameter of CFD-based EM simulations. Two-step training process is used for developing the neuro-TF surrogate model with the extracted zero/poles. Using the proposed zero/pole extraction technique, the developed neuro-TF surrogate model can have larger geometrical range than that using vector fitting. Consequently, the surrogate-based EM optimization using neuro-TF developed by CFD-based EM simulation can achieve a speedup over the standard neuro-TF optimization. Two examples of EM optimizations of microwave components are used to demonstrate the proposed technique.