BR-GAN: structural-enhanced GAN with hybrid loss optimization for orbital angular momentum beam recovery

We propose a novel orbital angular momentum (OAM) beam recovery method based on a structural-enhanced generative adversarial network (GAN) with hybrid loss optimization (BR-GAN). Through updating the network structure, the training design, and the loss function from original GANs, the proposed BR-GAN can perform the recovery functions of distorted OAM beams in both amplitude and phase domains originated from various distortion-strengths, inhomogeneous atmospheric turbulence (AT). Comprehensive numerical simulations have been carried out to evaluate the recovery performance, including the peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), and mode purity of the OAM mode, and the recovered results of 41.76 dB, 0.9989, and 0.9815 obtained for the OAM + 1 as an example. A similar recovery performance has also been observed in weak, moderate, and strong turbulence scenarios. Compared with other conventional methods such as VAE, original GAN, VGG-11, or our old network, the proposed BR-GAN outperforms them in PSNR, SSIM, and mode purity, with a lead of at least 0.6045 in mode purity. Moreover, to evaluate the mode division multiplexing (MDM) transmission of OAM beams, the propagating and recovery of up to five modes have been tested by using the proposed BR-GAN, confirming its validation of the potential high-capacity free-space laser systems.