Turbulence-tolerant 12-bit/symbol OAM shift keying free space optical communication using a two-stage neural network

Abstract

Orbital angular momentum-shift keying free space optical (OAM-SK FSO) communication inherently suffers from disturbances in atmospheric turbulence (AT), which significantly degrades the communication performance. In this manuscript, we demonstrate a turbulence-tolerant 12 bit/symbol OAM-SK FSO communication system using a two-stage neural network, which features a modified U-Net for compensating OAM modes, and a two-step convolutional neural network for decoding. In simulation, the full capability (accuracy $>98.87\text{\%}$) of the proposed model in recognizing all types of OAM modes disrupted by different levels of AT is demonstrated. After transfer learning on experiment data, the model still exhibits good generalization in the practical system. The transmission of a 12,000-depth pseudo-random binary sequence encoded by 1000 12-bit symbols is displayed in a mock-up experiment, and bit error rates of $2.50\times 10^{-4}$, $4.17\times 10^{-4}$, $4.75\times 10^{-3}$ under weak, moderate and strong AT are observed respectively. This study provides a new, to the best of our knowledge, platform for high-capacity, robust FSO communication.