Inverse Design of High-Order Bessel Vortex Wave Generator Based on Deep Learning

This paper proposes an inverse design method for amplitude-phase control metasurface units based on deep learning to achieve independent control of the amplitude and phase of electromagnetic waves. To solve the problem that the traditional parameter scanning design method requires a large amount of electromagnetic simulation and consumes a lot of computing resources and time, this paper uses the simulation results of a small number of amplitude-phase controlled metasurface units as the training set of the deep neural network and inputs high-order Bessel vortices. The target amplitude and phase of each unit of the spin wave generator, and the network output predicted size parameters. Through verification of 900 units of the inverse design, the average errors of the amplitude and phase responses were 0.81% and 0.42% respectively, which greatly improved the design accuracy and efficiency of the metasurface.