Substrate Integrated Waveguide Microwave Filter Tuning Through Variable-Fidelity Feature Space Optimization

摘要

The paper presents a fast optimization procedure for substrate integrated waveguide microwave filters. A set of characteristic points (or features) is identified in a filter response, and used to determine violation/satisfaction of given performance requirements. A linear response surface model of these features is constructed using the data extracted from the filter response at the current design, as well as the feature gradients estimated based on coarse-discretization EM model (the latter utilized to reduce the computational cost of the optimization process). The model is subsequently used as a prediction tool, guiding towards an optimum design. The optimization process is embedded in the trust-region framework. By exploiting close-to-linear dependence of the filter features on designable parameters, the optimum filter design can be obtained - as demonstrated using an example of an SIW filter - at the cost of a few evaluations of the high-fidelity EM-model.