Phase diversity wavefront sensing in deep turbulence via metasurface optics and noise-tolerant CNN

Free-space optical communication (FSOC) systems offer high-bandwidth and secure communication with minimal capital costs. Adaptive optics (AO) are typically added to these systems to decrease atmospheric channel losses; however, the performance of traditional AO wavefront sensors degrades in long-range, deep turbulence conditions. Alternative wavefront sensors using phase diversity can successfully reconstruct wavefronts in deep turbulence, but current implementations require bulky setups with high latency. In this work, we present a compact implementation of phase diversity wavefront sensing using a birefringent metasurface optic in conjunction with a convolutional neural network (CNN) for accurate wavefront reconstruction in deep turbulence. Our metasurface leverages both spatial and polarization multiplexing to simultaneously generate eight PSFs at the detector with various defocus levels. These PSFs are used as inputs to train the CNN in an end-to-end approach where the figure of merit is taken to be the power in bucket obtained at the receiver after propagating the predicted wavefront through the turbulent atmosphere. Our results show high accuracy reconstructions of scintillated wavefronts with Rytov numbers ranging from 0.2 to 0.6, surpassing the performance of traditional wavefront sensors like the Shack-Hartmann in these turbulence regimes. This approach provides new opportunities for compact, reliable wavefront sensing in deep turbulence, enabling extended range and improved accuracy for FSOC systems.