Low-cost surrogate modeling for rapid design optimization of antenna structures

Availability of fast yet reliable replacement models (also referred to as surrogates) is essential to reduce the computational cost of antenna design process. Unfortunately, conventional approximation (or data-driven) modeling is not well suited for modeling of highly nonlinear responses of antenna structures, especially for larger number of geometry parameters. The latter is due to the so-called curse of dimensionality, i.e., a rapid growth of the number of necessary training samples with the problem dimensionality. In this work, we propose a novel approach where the region of surrogate model validity is restricted to a manifold spanned by several reference designs corresponding to antenna optimized for various operating frequencies and dielectric permittivity of the substrate material. This allows us to focus the modeling process in the region that only contains designs that are close-to-optimum from the point of view of the aforementioned operating/material criteria. Rigorous analytical formulation of the technique is supported by a case study of a ring slot antenna. Considerable reduction of the number of training points compared to conventional modeling methods is also demonstrated.