Learning Enabled Dense Space-division Multiplexing through a Single Multimode Fibre
Space-division multiplexing is a promising technology in optical fibre communication to improve the transmission capacity of a single optical fibre. However, the number of channels that can be multiplexed is limited by the crosstalks between channels, and the multiplexing is only applied to few-mode or multi-core fibres. Here, we propose a high-spatial-density channel multiplexing framework employing deep learning for standard multimode fibres (MMF). We present a proof-of-concept experimental system, consisting of a single light source, a single digital-micromirror-device modulator, a single detection camera, and a deep convolutional neural network (CNN) to demonstrate up to 400-channel simultaneous data transmission with accuracy close to 100% over MMFs of different types, diameters and lengths. A novel scalable semi-supervised learning model is proposed to adapt the CNN to the time-varying MMF information channels in real-time, to overcome the environmental changes such as temperature variations and vibrations, and to reconstruct the input data from complex crosstalks among hundreds of channels. This deep-learning based approach is promising to maximize the use of the spatial dimension of MMFs, and to break the present number-of-channel limit in space-division multiplexing for future high-capacity MMF transmission data links.