Photoelectric hybrid deep neural network based adaptive optics scheme for vortex beam in oceanic turbulence

Abstract

Underwater wireless optical communication (UWOC) systems are essential for underwater applications due to their superior performance, with orbital angular momentum (OAM) boosting channel capacity via mode orthogonality. However, oceanic turbulence (OT) distorts OAM-carrying vortex beams, and existing adaptive optics (AO) methods and single-type neural network approaches are limited by cost, optimization issues, or inadequate precision. In this paper, we propose a photoelectric hybrid deep neural network (PHDNN) based AO scheme to correct the distortion caused by OT to vortex beam carrying OAM. The PHDNN model is composed of diffractive deep neural network and convolutional neural network, which are trained jointly to predict the coefficients of Zernike polynomial representing OT. The phase screen reconstructed based on the Zernike coefficients is conjugated and used to eliminate OT. The proposed scheme achieves fast convergence at 25 epochs with a final loss of 0.096, low latency of 4.113s under strong OT, and robust adaptability to varying turbulence intensities. Experimental results show that it reconstructs high-quality phase screens with a peak signal-to-noise ratio up to 35.548, and significantly enhances OAM mode purity to 0.936 under strong OT, in sharp contrast to the uncorrected value of 0.024. The proposed scheme provides the benefits for the realization of the fusion of all-optical neural network and electronic neural network, and contributes to the development of UWOC system.