Support Vector Machines to Accelerate Reflectarray Analysis and Optimization

Abstract

The analysis of reflectarray antennas is substantially accelerated by using Support Vector Machines (SVMs) to model the matrix of reflection coefficients in substitution of a full-wave analysis based on local periodicity, the Method of Moments (MoM-LP) in this work. The SVM model takes as input variables two geometrical variables which control the phase-shift for two different polarizations. Thus, the employed model is able to analyze shaped-beam dual-linear polarized reflectarrays. As test case, a shaped-beam reflectarray radiating a pattern for Local Multipoint Distribution Service applications, presenting a squared-cosecant cut in elevation and sectored-beam in azimuth, is presented, showing a high degree of agreement in both the copolar and crosspolar patterns between the simulations of SVM and MoM-LP. Furthermore, the acceleration with regard to the MoM-LP is between three and four orders of magnitude, which demonstrates the suitability of the SVM model approach for reflectarray direct optimization.