Improved Modeling of Microwave Structures Using Performance-Driven Fully-Connected Regression Surrogate

Fast replacement models (or surrogates) have been widely applied in the recent years to accelerate simulation-driven design procedures microwave engineering. The fundamental reason is a considerable-and often prohibitive-CPU cost of massive full-wave electromagnetic (EM) analyses related solving common tasks such as parametric optimization or uncertainty quantification. most popular class surrogates are data-driven models, which fast evaluate, versatile, and easy handle. Notwithstanding, curse dimensionality well utility demands (e.g., so that model covers sufficiently broad ranges system operating conditions), limit applicability conventional methods. A performance-driven modeling paradigm allows for mitigating these issue by focusing surrogate setup process constrained domain encapsulating designs being high quality w.r.t. assumed figures interest. nested kriging framework capitalizing on this idea, renders using interpolation, has shown surpass traditional approaches. In pursuit further accuracy improvements, work incorporates concept into fully-connected regression (FRCM). latter recently introduced context frequency selective surfaces, combined deep neural networks with Bayesian optimization, employed determine network architecture hyper-parameters. Using two examples miniaturized microstrip couplers, our methodology demonstrated outperform both techniques kriging, reliable constructed over multi-dimensional parameters spaces just few hundreds samples.