Shear-interference assisted deep-learning for enhancing spatially multiplexing capacity of free-space communication

Abstract

The information capacity is a critical measure in free-space optical (FSO) communication. Traditional mode-multiplexing techniques employed in optical communication utilizing Laguerre-Gaussian beams (LGBs) necessitate proper topological charge interval to prevent mode-crosstalk. In this paper, we propose a novel approach that integrates deep learning (DL) into the LGB-based optical communication model, incorporating shearing interference techniques. This combination enables the FSO system to achieve the theoretical limit of capacity, resulting in a remarkable 300 % increase in information capacity compared to traditional techniques. Experimental results demonstrate that the constructed neural network attains a flawless pattern recognition accuracy of 100 %. Additionally, we evaluate various partially coherent light scenarios under the influence of beam decoherence. The results reveal that accurate discrimination of multiplexing patterns is maintained when the fringe contrast exceeds 0.43. Even at a degraded contrast level of 0.18, the recognition accuracy remains impressively high at 91.16 %. To demonstrate practicality, we successfully encode and transmit a cyan-magenta-yellow-black (CMYK) color image using the proposed system, achieving a low bit error rate of 7.5 × 10-5. These findings highlight the potential of DL-assisted mode-division multiplexing for future optical communication applications.