An algorithm combining convolutional neural networks with SPGD for SLAO in FSOC

The wavefront distortion in the free space optical communication (FSOC) can be mitigated with sensor-less adaptive optics (SLAO). The SLAO controlling algorithm plays an important role for system performance, but the conventional methods still cannot reach satisfaction. This paper presents a hybrid method to improve the calibration efficiency and accuracy, in which convolutional neural networks (CNN) model coarsely classifies and corrects the aberrations, and then stochastic parallel gradient descent (SPGD) algorithm finely corrects aberrations. Simulations show that it can improve the speed for the distortion compensation and avoid the problem falling into local optimum in traditional SPGD. This hybrid method can improve coupling efficiency of the optical carrier in the SLAO and enable FSOC to achieve a higher Strehl ratio (SR) under atmospheric turbulence.