Physics-Constrained GAN Boosts OAM Correction in Ocean Turbulence

Xiaoji LiZhiyuan Wang^*

• School of Information and Communication, Guilin University of Electronic Technology, Guilin, China

This study addresses the challenge of improving wavefront correction for Orbital Angular Momentum (OAM) in oceanic turbulence using a physics-constrained Generative Adversarial Network (GAN). Degraded input images, with a Structural Similarity Index (SSIM) of 0.62, were reconstructed using different loss functions. The baseline model improved the SSIM to 0.84, while adding spectral constraints (+Spec) increased the SSIM to 0.86, and spatial constraints (+Ortho) raised it further to 0.95. The combination of both spatial and spectral constraints (+Ortho+Spec) resulted in an almost optimal SSIM of 0.98. Ablation studies showed that the baseline model had a modal purity of 85.0%. Adding spatial constraints (+Ortho) improved purity to 95.7%, while spectral constraints (+Spec) raised it to 86.5%. The dual-constraint approach achieved the highest purity of 98.4%, a 13.4% improvement. Power spectral density analysis using Kullback-Leibler (KL) divergence confirmed the superiority of the dual-constraint model (KL=0.56), compared to the baseline (KL=2.47), spatial-only (KL=2.08), and spectral-only (KL=0.72) models. Additionally, training loss analysis revealed that the dual-constraint model consistently reduced loss and enhanced convergence, generalization, and performance. These results demonstrate that integrating both spatial and spectral constraints effectively optimizes reconstruction, purity, and spectral fidelity, offering a robust solution for OAM correction in underwater optical communication systems.