EM-simulation-driven design optimization of compact microwave structures using multi-fidelity simulation models and adjoint sensitivities

Reliable design of miniaturized microwave structures requires utilization of full-wave electromagnetic (EM) simulation models because other types of representations such as analytical or equivalent circuit models are of insufficient accuracy. This is primarily due to considerable cross-coupling effects in tightly arranged layouts of compact circuits. Unfortunately, high computational cost of accurate EM analysis makes the dimension adjustment process challenging, particularly for traditional methods based on parameter sweeps, but also for conventional numerical optimization techniques. In this article, low-cost simulation-driven designs of compact structures were demonstrated using gradient search with adjoint sensitivities as well as multi-fidelity EM simulation models. The optimization process was arranged sequentially, with the largest steps taken at the level of coarse-discretization models. Subsequent fine tuning was realized with the models of higher fidelity. Switching between the models was realized by means of adaptively controlled termination conditions. This allowed for considerable reduction of the design cost compared with single-level optimization. The approach was illustrated using a compact microstrip rat-race coupler with two cases considered, that is, (i) bandwidth enhancement, and (ii) minimization of the structure size. In both cases, the optimization cost corresponded to a few high-fidelity EM simulations of the coupler structure.