Atmospheric Turbulence Phase Reconstruction via Deep Learning Wavefront Sensing.

Area of Science:

• Optical Communications
• Atmospheric Optics
• Deep Learning

Background:

• Atmospheric turbulence severely degrades free-space optical communication performance.
• Accurate phase reconstruction is vital for mitigating these disturbances.
• Existing methods face challenges with slow convergence and limited accuracy.

Purpose of the Study:

• To develop a fast and accurate method for atmospheric turbulence phase reconstruction.
• To enable effective phase compensation in free-space coherent optical communication.
• To overcome limitations of traditional phase reconstruction techniques.

Main Methods:

• A deep learning approach using a U-Net model was employed.
• The model was trained on extensive light intensity-phase samples across various turbulence intensities.
• Phase reconstruction was performed directly from turbulence-affected light intensity images.

Main Results:

• The U-Net model achieved rapid phase reconstruction with an average processing time of 0.14 seconds.
• Simulations showed a low average loss function value (0.00027) and mean squared error (0.0003).
• Experimental validation confirmed a mean squared error of 0.0007 for single turbulence reconstruction.

Conclusions:

• The proposed deep learning method offers fast convergence, robust performance, and strong generalization capabilities.
• This approach provides a novel and effective solution for atmospheric disturbance correction in optical communication.
• The technique successfully reconstructs turbulence phases for varying turbulence strengths.