Adaptively constrained parameter extraction for robust space mapping optimization of microwave circuits

The performance of the space mapping (SM) optimization algorithm depends both on approximation and generalization capabilities of the underlying surrogate model. Often, the surrogate is selected by trial and error which may lead to excessive computational overhead and poor quality of the optimization outcome. Here, we introduce an adaptively constrained parameter extraction process to automatically find an approximation-generalization trade-off through the adjustment of the surrogate model parameter space. As a result, we obtain improved performance of the SM algorithm both in terms of its convergence properties and the quality of the optimized design. Verification using several microwave design problems is provided.