Origami Metasurface Hologram Based on Deep Learning

Abstract

An origami metasurface structure is designed for hologram imaging based on deep learning. The metasurface unit cell is folded to form an origami structure, which is a straightforward way to change metasurface shape. The transmission electromagnetic performance of the unit cell is steady while the angle of incident waves changes from -60 to 60 degrees. In this paper, the folded angle of 45 degree is chosen. Every four cells fold in different ways to form a supercell. The total amounts of supercells are $30\times 30=900$. Next, deep learning is utilized to compute the compensation phases of 3600 elements for generating hologram imaging. These phases determine the rotation angles of the unit cells. A left-handed circular polarized (LHCP) wave transmits the hologram structure and subsequently generates an image with the shape of the letter J on the imaging plane. This is validated by the simulations in the commercial electromagnetic software CST.