Meta-learning-based optical vector beam high-fidelity communication under high scattering.

Area of Science:

• Optical communication
• Signal processing
• Machine learning

Background:

• Free-space optical communication offers high bandwidth but suffers signal distortion from atmospheric scattering.
• Existing deep learning demodulators require extensive data and training for diverse scattering conditions.

Purpose of the Study:

• To develop a more efficient and generalizable demodulator for distorted optical communication channels.
• To reduce the data preparation and training costs associated with adaptive demodulation techniques.

Main Methods:

• Implemented a meta-learning-based auto-encoder demodulator.
• Leveraged prior theoretical knowledge for rapid adaptation.
• Trained the model with minimal realistic samples per class for distortion rectification.

Main Results:

• Achieved high-fidelity communication using hybrid vector beams with the developed demodulator.
• Demonstrated successful transmission of a 256 gray value image with an error ratio below 0.05% under severe scattering.
• Significantly reduced data costs when adapting to different scattering channels.

Conclusions:

• The meta-learning auto-encoder provides an effective solution for high-fidelity optical communication in random media.
• This approach enhances the generalization ability of demodulators, making optical communication more robust and cost-effective in atmospheric environments.