Synthesis of Aperture Fields for Transmissive Metasurface Design from 3D Masks Using Deep Learning

For transmissive metasurfaces, we develop a deep learning (DL) approach that takes user-defined three-dimensional (3D) far-field (FF) masks and the incident field impinging on the metasurface as input. This DL approach then aims to infer the required aperture fields of the metasurface such that its resulting power pattern (PP) resides within the 3D lower and upper masks and the metasurface remains lossless and passive.