Deep learning based optical orbital angular momentum demultiplexing under high scattering

Optical communication systems based on Orbital Angular Momentum (OAM) theoretically have great potential to increase the channel capacity of the system. When light passes through a high scattering medium, its phase and intensity are affected by scattering, which makes it difficult to demultiplex the OAM modes. In order to alleviate the mode crosstalk caused by scattering, this paper proposes a deep learning-based scheme for OAM modes demultiplexing. A simulation model of the optical communication system is built based on the scattering medium transmission matrix theory. The multiplexed OAM beam is transmitted through the system to generate the speckle pattern, matching the incident phase distribution as the data set. Based on this dataset, a U-Net type Deep Neural Networks (DNN) are trained to reconstruct the phase of the light distorted by the scattering medium, thereby the multiplexed OAM modes are identified by a Visual Geometry Group (VGG) type DNN. The simulation results show that at a Signal-to-Noise Ratio (SNR) of (1, 20) dB, the recognition rate of the demultiplexed OAM modes can reach beyond 97%. For grayscale image transmitting via OAM multiplexing under the high scattering, the Pearson correlation between the demultiplexed image and the original image is more than 0.98.