Efficient and Accurate Modeling of Reflectarray Unit Cells Using SVMs

Abstract

This paper presents a technique for an efficient and accurate characterization of dual-polarized reflectarray unit cells by means of Support Vector Machines (SVMs) to accelerate the analysis and optimization of this type of antenna. The goal is to accurately obtain the matrix of complex reflection coefficients using geometrical data of the unit cell. To this end, SVMs are employed to obtain surrogate models of those coefficients in an efficient fashion by employing a novel grid search which accelerates by three orders of magnitude the training process with regard to an exhaustive grid search. In addition, the precision of the surrogate model is tested against full-wave simulations assuming local periodicity obtaining a high degree of agreement.