Robust space mapping optimization exploiting EM-based models with adjoint sensitivities

A robust space mapping algorithm for microwave design optimization is presented. Our approach utilizes low-cost derivative information obtained through adjoint sensitivities to improve the matching between a high-fidelity model evaluated using expensive EM simulation and its surrogate. Here, the surrogate is obtained through coarse-discretization EM simulation (low-fidelity model). Adjoint sensitivities are also exploited to speed up the parameter extraction and surrogate optimization processes. As a result, design can be performed at low cost in terms of the number of both high- and low-fidelity EM simulations. The efficiency is demonstrated using two examples: a monopole UWB antenna and a microstrip filter.