Turbulence-Resilient Multicasting of OAM Beams Using a Hybrid Optical-Electronic Neural Network

While orbital angular momentum (OAM) enhances free-space optical capacity, ensuring high-security transmission under turbulence remains challenging. We propose a hybrid optical-electronic neural network for turbulence-resilient OAM multicasting and encryption. The architecture integrates a physical diffractive layer, utilizing a spatial light modulator (SLM) to load a pre-trained compensation phase, with a digital U-Net backend. We introduce a computational multicasting protocol based on asymmetric neural decoding. Authorized users reconstruct high-fidelity complex fields from intensity-only captures, decoding a single physical mode into distinct logical states, thereby mitigating eavesdropping risks. Joint training exhibits rapid convergence, dropping the training loss from an initial 0.355 to a stable plateau between 0.008 and 0.013, with the peak signal-to-noise ratio (PSNR) reaching a peak of 33.1 dB under favorable conditions, and experiments confirm highly accurate sim-to-real phase retrieval. This provides a scalable solution for next-generation optical wireless networks.