Direct construction of an optical linear transform and its application on optical complex data generation

Abstract

Optical computing technique has emerged as a promising platform for highly parallel data processing. In most optical computing architectures, optical linear transform is the basic composition, which is commonly designed by the established deep learning-based methods or general-purpose optimizers. There still lacks in-depth study to construct a solver targeted at optical linear transform applications. In this work, we propose a new algorithm that solves the transformation matrix of a linear optical system consisting of cascaded phase masks directly and show that its efficiency is significantly higher than those common solvers. As a direct application of this method, we can create target array of optical beams even with a single layer of phase mask in the experiment. The amplitude and phase of each beam in the array can be controlled independently without affecting each other. The optical system requires only one light source and one programmable phase mask. This setup can be readily incorporated into most current optical computing configurations. Our method may find broad applications in classic and quantum optical information processing.