Two-Level Antenna Modeling with Domain Confinement and Explicit Dimensionality Reduction

Abstract

Full-wave electromagnetic (EM) simulations are essential to reliable evaluation of modern antenna structures. Yet, high computational cost of EM analysis poses a serious challenge for EM-driven antenna design. A possible workaround is to utilize fast surrogates. However, the curse of dimensionality hinders the applicability of the conventional surrogate-assisted frameworks beyond simple structures described by a few parameters. A workaround is offered by the recently proposed nested kriging technique, where the surrogate domain is confined to the parameter space region containing high-quality designs, which allows for setting up reliable models at a low cost. Here, this expenses are further reduced through an explicit domain dimensionality reduction by means of principal component analysis.