Speckle-learned Orbital Angular Momentum De-multiplexing

Abstract

Orbital Angular Momentum (OAM) multiplexing is a promising technique for enhancing optical communication capacity. Here we present a deep learning model to demultiplex the encoded OAM superposition modes using their corresponding speckle patterns. We employed a speckle-learned demultiplexing technique to accurately recognize the encoded OAM modes. A Convolutional Neural Network (CNN) is trained to recognize superimposed Laguree-Gaussian modes through their far-field intensity speckle patterns. Our approach allows for accurate recognition of encoded OAM modes through speckle-learned classification. The trained CNN achieved a classification accuracy of > 99 % in reconstructing a 4-bit grey image of 100×100 pixels.