Low-cost parameter extraction and surrogate optimization for space mapping design using em-based coarse models

Space mapping (SM) is one of the most popular surrogate-based optimization techniques in microwave engineering. The most critical component in SM is the low-fidelity (or coarse) model -a physically-based representation of the structure being optimized (high-fidelity or fine model), typically evaluated using CPU-intensive electromagnetic (EM) simulation. The coarse model should be fast and reasonably accurate. A popular choice for the coarse models are equivalent circuits, which are computationally cheap, but not alw ate, and in many cases even not available, limiting the practical range of applications of SM. Relatively accurate coarse models that are available for all structures can be obtained through coarsely-discretized EM simulations. Unfortunately, such models are typically computationally too expensive to be efficiently used in SM algorithms. Here, a study of SM algorithms with coarsely-discretized EM coarse models is presented. More specifically, novel and efficient parameter extraction and surrogate optimization schemes are proposed that make the use of coarsely-discretized EM models feasible for SM algorithms. Robustness of our approach is demonstrated through the design of three microstrip filters and one double annular ring antenna.