Turbulence compensation with pix-to-pix generative adversarial networks in vector vortex beams

Abstract

Orbital angular momentum (OAM) has significantly propelled free space optical communication (FSOC) towards achieving ultra-large transmission capacities, but mode-crosstalk in atmospheric turbulence limits its application. Here, we propose a proof-of-concept turbulence compensation approach utilizing pix-to-pix generative adversarial networks (pix2pixGAN) that does not rely on the wavefront sensor. The model captures the complex relationships between distorted optical fields and phase screens through extensive training, after which the phase screen is directly recovered from the well-trained model by identifying the corresponding distorted image to compensate for distortions. Using this model, the Strehl ratio improvement is measured at 35.7%, 8.9%, and 1.7% under three distinct turbulence conditions, respectively. Furthermore, the recognition of vector vortex beams (VVBs) integrating with the pix2pixGAN significantly improves average mode accuracy from 2% to over 99%. Additionally, the exploration of VVB-based communication further elucidates pix2pixGAN's role in enhancing communication quality. These findings suggest a potential advancement in developing a novel neural network-based strategy to compensate for transmission distortions under intense turbulence.