Data transmission in a multimode optical fiber using a neural network

Abstract

In digital data transmission, single mode optical fibers are commonly used since they can carry very short optical pulses without any significant distortions. In contrast, multimode fibers support many propagation modes that travel with different speeds; thus, they cannot maintain the shape of a light pulse. This feature of multiple propagation modes can be a benefit since it makes possible the transmission of data through several channels simultaneously. We demonstrate how multimode fibers can be used to transmit images. Because of the different propagation constants of the modes, the transmitted image is scrambled to apparently random speckle patterns. A simple neural network can be used to model the transmission through the multimode fiber. We show how the neural network can be trained to recognize a set of patterns with high accuracy.