Deep learning-based general beam synthesis for atmospheric propagation.

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Optical Communications

Background:

• Free-space optical systems require optimized light beams for effective transmission.
• Designing beams for propagation through random media like atmospheric turbulence is challenging.
• Current methods struggle to generate application-specific beams for turbulent environments.

Purpose of the Study:

• To propose a deep learning-based method for generating optimal light beams for free-space optical systems.
• To address the challenge of designing beams that can effectively propagate through atmospheric turbulence.
• To optimize beams based on statistical performance metrics at the receiver.

Main Methods:

• A convolutional neural network (CNN) was trained using batch-wise computation to approximate receiver statistics.
• Pseudo-modes synthesized via weighted superposition of Hermite-Gaussian eigenmodes created general beams.
• An end-to-end framework enabled self-supervised learning without pre-calculated datasets.
• Optimization considered average received power, scintillation index, and mean signal-to-noise ratio (SNR).

Main Results:

• The synthesized beams demonstrated lower scintillation and higher intensity compared to Gaussian Schell-model beams.
• Wave optics simulations and experiments validated the effectiveness of the generated beams.
• A markedly enhanced receiver SNR was achieved with the novel beam design.
• The advantages persisted across various link configurations, broadening the applicability of stochastic beams.

Conclusions:

• Deep learning offers a powerful approach to designing optimal beams for free-space optical communications in turbulent conditions.
• The proposed method successfully generates beams with improved performance metrics, including reduced scintillation and enhanced SNR.
• This work extends the application range of stochastic beams and advances the capabilities of free-space optical systems.