Transmissive Metasurface Synthesis From Far-Field Masks Using Unsupervised Learning
Chen Niu, Mario Phaneuf, Puyan Mojabi
Abstract
Designing lossless and passive transmissive metasurfaces requires the knowledge of tangential electromagnetic fields on the metasurface aperture and the enforcement of local power conservation (LPC). In design scenarios where the desired radiation patterns are specified by lower and upper masks, we have developed a deep learning (DL) approach to infer the required metasurface aperture fields, while favouring the LPC constraint. In the cases examined here, the aperture fields obtained through the DL approach lead to power patterns that exhibit good alignment with the specified far-field masks, while adhering to the LPC constraint. When simulating the resulting metasurfaces using impedance sheets, this alignment slightly degrades, partly due to the non-zero thickness of the metasurface and the local periodicity assumption used in the unit cell design.