Machine Learning Framework Combining Radial Phase Grating and Channel Information-Assisted Underwater Wireless Optical OAM Communications

Abstract

In this letter, an enhanced machine learning (ML) framework combining radial phase grating (RPG) and channel information has been proposed for underwater wireless optical orbital angular momentum (OAM) communications. The optical irradiance profile and diffraction pattern are simulated by the combination of split-step angular-spectrum propagation and Monte Carlo methods, then they were embedded by the convolution and fully connected layers. The transfer learning-based ML architectures are trained by the image features and the embedding channel information with fine-tuning. The impacts of oceanic turbulence, seawater types, and transmission distance are considered. Results show that the transfer learning-based OAM detectors could achieve the order-of-magnitude gain in the bit error rate. Besides, the proposed ML framework could combine with a variety of state-of-the-art methods and thus improve the system performance. This work indicates transfer learning-based ML frameworks have great potential in ML-based OAM communication systems.