High-capacity and robust information transmission using generalized random structured beams and deep learning-based decoding

With the rapid growth of data traffic, achieving high-capacity, stable, and secure information transmission has become a critical challenge for free-space optical communication systems. This paper proposes an information transmission scheme based on generalized random structured beams and deep learning-based decoding to address these challenges. By exploiting extended optical coherence engineering, image information is encoded into random modes, enabling a substantial enhancement in channel capacity owing to the theoretically unbounded number of available random modes. At the receiver, a convolutional neural network is employed to decode the transmitted information directly from the intensity distributions of the random modes. Our results demonstrate decoding accuracies exceeding 99% for 256-grayscale image transmission. By incorporating a random pixel-indexing encryption mechanism, the proposed scheme further enhances transmission security. Moreover, reliable information recovery is maintained under strong noise interference, highlighting the robustness of the proposed approach in complex channel environments. We anticipate that integrating optical coherence engineering with deep learning will provide a promising pathway for advancing free-space optical communication systems.