Learning-enabled data transmission with up to 32 multiplexed orbital angular momentum channels through a commercial multi-mode fiber

Abstract

Multiplexing orbital angular momentum (OAM) modes enable high-capacity optical communication. However, the highly similar speckle patterns of adjacent OAM modes produced by strong mode coupling in common fibers prevent the utility of OAM channel demultiplexing. In this paper, we propose a machine learning-supported fractional OAM-multiplexed data transmission system to sort highly scattered data from up to 32 multiplexed OAM channels propagating through a commercial multi-mode fiber parallelly with an accuracy of >99.92%, which is the largest bit number of OAM superstates reported to date (to the best of our knowledge). Here, by learning limited samples, unseen OAM superstates during the training process can be predicted precisely, which reduces the explosive quantity of the dataset. To verify its application, both gray and colored images, encoded by the given system, have been successfully transmitted with error rates of <0.26%. Our work might provide a promising avenue for high-capacity OAM optical communication in scattering environments.