Deep Learning Based Pixelized Forward Simulator and Inverse Designer of the Frequency Selective Surface

In this paper, two deep learning frameworks based on Wideresnet40×2 and Reverse-CNN are proposed to serve as the forward simulator and inverse designer of the pixelated frequency selective surface (FSS). Compared to the classical parametric modeling, the proposed framework can realize the pixelized design with high degrees of freedom. Besides, a fully trained network can achieve considerable precision with tiny time cost, which has potential to be used in various real-time FSS designs.