Deep learning-based multi-ring optical perfect vortex beam for audio communication

Abstract An audio coding method based on multi-ring optical perfect vortex beam (MR-OPVB) is proposed, which uses its orbital angular momentum (OAM) as the information carrier, and realizes efficient signal coding through MR-OPVB. The coaxial interference method of fundamental-mode Gaussian beams was used to extract OAM information, focusing on the interference effects between perfect vortex waves of different radii and their impact on OAM recognition, as well as strategies to mitigate such interference disturbances. Convolutional neural network was used to recognize the mode of MR-OPVB affected by atmospheric turbulence, which can achieve 99.8% recognition accuracy in moderate turbulence conditions. The trained resnet-18 model was utilized to identify and decode the MR-OPVB affected by atmospheric turbulence, and the distortion rate of the recovered audio signal was 0.954%. This study offers a coding concept for optical perfect vortex beam communication and validates that machine learning can effectively enhance the quality of free space optical communication.