Advanced learning-based approaches for reflectarrays design

The problem of efficiently and effectively compute the response (i.e., reflection coefficients) of next-generation reflectarray elements with wide number of degrees-of-freedom is addressed in this work. Towards this end, a machine learning-based approach based on advanced Kriging strategies is exploited (instead of classical full-wave solvers) in order to predict the response of complex unit cells of interest for the design of high-performance reflectarrays. Preliminary numerical results aimed at comparing the accuracy and efficiency of the proposed methodology with respect to standard full-wave approaches are illustrated.