Decomposition and mode-weight estimation of mixed-mode OAM beams via diffractive neural networks

Beams carrying orbital angular momentum (OAM) have attracted considerable interest in high-capacity optical communication owing to their infinite-dimensional state space. Conventional methods for detecting OAM modes face significant limitations, including bulky systems, slow response times, and restricted detection ranges. Although deep learning algorithm have shown promise in mitigating some of these challenges, the characterization of mode distributions within mixed-mode OAM beams has received limited attention. We propose an all-optical, end-to-end approach for decomposing mixed-mode OAM beams and estimating their mode weights using diffractive deep neural network (D²NN). The network directly maps the incident optical field to outputs that both identify the constituent OAM modes and estimate their relative contributions. Numerical simulations demonstrate that the method can accurately recover the weights of up to 21 hybrid modes. Moreover, it maintains strong robustness under atmospheric perturbations, with the relative error remaining below 7%. This approach enables precise and efficient reconstruction of the OAM beams across varying numbers of modes, offering broad potential in multidimensional information encoding, quantum information processing, and optical computing.