When orbital angular momentum (OAM) beams pass through scattering media, mode coupling and interference occur, distorting the phase front and converting the beam into a random speckle pattern, so high capacity and fidelity communication becomes challenging. To address this, we propose a few-shot learning scheme to demonstrate an unprecedented highest capacity 16-bit and 24-bit fractional OAM multiplexing communication system under complex environments. By decomposing the full classification task into 16 or 24 binary subtasks, the system requires only 20 000 and 30 000 training classes, corresponding to 30.5% and 0.18% of the total class space, respectively. This approach significantly reduces the required data and computational resources. Experimentally, OAM multiplexed beams are scattered by a rotating ground-glass diffuser to generate speckle patterns. After training on a small subset of superimposed OAM states, the model learns to extract latent features from the speckles and accurately predicts each OAM channel across the entire set. Additionally, the use of a space-to-depth mechanism for low-resolution OAM recognition enhances the accuracy of the OAM channels impacted by the environment. The system achieves average channel identification accuracies of 99.82% (16-bit) and 94.45% (24-bit). To validate the scheme, color images are successfully transmitted in both static and dynamic scattering environments. In short, this work provides a clever approach for extremely high-capacity OAM communication in harsh environments.
Zhou et al. (Mon,) studied this question.