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

Abstract Free space optical communication is increasingly important for meeting rising data demands, yet achieving high capacity, stability, and security simultaneously remains difficult. Current approaches lack a unified strategy to address these competing requirements. Here we show an information transmission scheme based on generalized random structured beams and deep learning decoding. Information is encoded into random modes through optical coherence engineering, increasing channel capacity, and a convolutional neural network recovers data directly from intensity patterns. The method achieves over 99% accuracy for 256 grayscale image transmission, maintains reliable performance under strong noise, and enhances security through random pixel indexing. This approach provides a practical route toward high capacity, robust, and secure optical communication systems.