Single‐Shot Optical Wavefront Distortion Recovery and Prediction via Digital Twin Neural Network

ABSTRACT Rapidly varying wavefront distortions impose a critical challenge to free‐space optical communication and in vivo imaging, which highlights the urgency of real‐time and high‐fidelity wavefront compensation. Deep‐learning‐based wavefront shaping offers a powerful tool to alleviate wavefront distortion. In this work, single‐shot‐based wavefront distortion recognition and compensation are realized by developing a retrieval neural network with digital twin generated data. The network can simultaneously predict the incoming distortions with speckles of the current frame. In a dynamic distortion channel, the wavefront transmission fidelity is significantly improved from 0.14 to 0.76 and the single‐mode‐fiber coupling efficiency of the transmitted field approaches (53 ± 3)% with single‐shot compensation. This work alleviates the burden of experimental data acquisition, offers a potential real‐time distortion correction technique and paves the way for quantum communication and biological imaging in dynamic complex environments.