Artificial Intelligence Empowering High‐Efficiency Recognition of Orbital Angular Momentum Modes for Optical Communication: A Review

Abstract

ABSTRACT Orbital angular momentum (OAM) based Free‐space optical (FSO) communication has gained considerable research interest due to its potential for increasing spectral efficiency and channel capacity. Nevertheless, recognition of OAM modes in realistic FSO communication systems suffers greatly from atmospheric turbulence (AT) effects, beam spreading, and even mode crosstalk. Resulting difficulties for conventional approaches to perform well under these kinds of distortions, leading to a turn toward the most advanced techniques such as machine learning (ML) and deep learning (DL). In this article, we provide a systematic review of recent progress in the ML‐based recognition methods for OAM modes and examine widely used modern neural networks such as feedforward neural network (FNN), convolutional neural network (CNN), vision transformers (ViT), and diffractive deep neural network (DDNN). We consider the architectures, training procedures, and parameters used for those models to provide evidence for their improved accuracy and robustness in simulations and experiments. We also present the critical difficulties, such as the real‐world long‐distance test and recognition of fractional OAM (F‐OAM) based on computational efficiency. Based on the state of the art and future perspectives, this work is intended to provide a good reference for promoting some practical, intelligent, reliable, and high‐capacity OAM‐FSO communication systems.