Hermite-Gaussian Superposition Modes for Speckle-Guided Demultiplexing

Abstract

Free-space optical communication is a cutting-edge technology for high-speed data transfer over long distances. Structured light modes like Hermite-Gaussian (HG) modes improve information transfer [1,2]. To enhance channel capacity and reduce cross-talk among higher-order modes, we use lower-order HG superposition (HG-SP) modes, which are more resilient to perturbations [3]. The light field of HG-SP is described by 𝐸(𝑥,𝑦,𝑧)=Σα𝑖𝑖 𝐻𝐺𝑚𝑖,𝑛𝑖(𝑥,𝑦,𝑧)exp (𝑖Δφ𝑖) where the three independent parameters, (𝑚,𝑛) modal indexes of HG modes, exp (𝑖Δφ𝑖) relative initial phases between the ith and 1st HG mode, and α𝑖 scale coefficients between modes, can obtain a large number of effective coding modes at a low mode order. From the large set of possible HG-SP modes, we have generated distinguishable HG-SP modes for better classification accuracy. Traditional machine learning methods rely on direct mode intensity images, which are sensitive to alignment and require capturing the entire mode for classification. This poses challenges in accurately identifying original modes and decoding encoded information. To overcome this, we utilize the more stable and noise-robust far-field speckle patterns of HG-SP modes We used a deep learning approach with a Convolutional Neural Network (CNN) to decode encoded information from far-field speckle patterns of HG-SP modes[3-5].. The CNN achieved >99% accuracy in distinguishing between modes. We selected 37 HG-SP modes to encode alphabets and digits. In simulations of an optical communication link, our method successfully reconstructed encoded phrases with >98% accuracy. This demonstrates the potential for increasing channel capacity and improving reliability in free-space optical communication