Physics-constrained GAN boosts OAM correction in ocean turbulence.

Area of Science:

• Optical Communications
• Signal Processing
• Machine Learning

Background:

• Oceanic turbulence degrades wavefronts, challenging underwater optical communication systems.
• Orbital Angular Momentum (OAM) offers potential for increased data capacity but requires robust wavefront correction.

Purpose of the Study:

• To develop and evaluate a physics-constrained Generative Adversarial Network (GAN) for improved OAM wavefront correction in oceanic turbulence.
• To investigate the impact of spatial and spectral constraints on the performance of the deep learning model.

Main Methods:

• A deep learning framework incorporating physical constraints was developed.
• The model was trained with varied loss settings: baseline, spectral constraints (+Spec), and spatial constraints (+Ortho).
• Performance was evaluated using metrics like Structural Similarity Index Measure (SSIM), modal purity, and Kullback-Leibler (KL) divergence for power spectral density analysis.

Main Results:

• The dual-constraint approach (+Ortho+Spec) achieved a near-optimal SSIM of 0.98, significantly outperforming the baseline (SSIM = 0.62).
• Modal purity reached 98.4% with dual constraints, compared to 95.7% with spatial constraints alone.
• Power spectral density analysis confirmed the superiority of dual constraints (KL divergence = 0.56) over the baseline (KL divergence = 2.47).

Conclusions:

• Integrating both spatial and spectral physical constraints into a GAN framework effectively optimizes wavefront reconstruction, modal purity, and spectral fidelity.
• This physics-constrained GAN provides a robust solution for OAM correction in challenging underwater optical communication environments.
• The findings highlight the potential of physics-informed deep learning for advancing underwater optical communication technologies.