Self-supervised dynamic learning for long-term high-fidelity image transmission through unstabilized diffusive media.

Area of Science:

• Optical physics
• Machine learning
• Information theory

Background:

• Multimode fibers (MMF) enable parallel data transmission for optical communication and remote imaging.
• Channel variability in MMFs degrades static transmission models, limiting long-term accuracy.
• Existing methods struggle with unstabilized MMFs over extended periods.

Purpose of the Study:

• To develop a robust method for long-term, high-fidelity optical field transmission through unstabilized MMFs.
• To address the limitations of static modeling and data-driven learning in dynamic MMF channels.
• To enable reliable remote imaging and enhance capacity in optical communication systems.

Main Methods:

• A self-supervised dynamic learning approach utilizing multiple adaptive networks with long- and short-term memory.
• Ensemble learning to combine network outputs for robust image recovery.
• Demonstration of transmission over 1 km MMF for over 1000 seconds.

Main Results:

• >99.9% accuracy in transmitting 1024 spatial degrees-of-freedom over 1 km MMF.
• Sustained high-fidelity transmission over extended durations (>1000 seconds).
• Enabled compressive encoded transfer of high-resolution video with significant throughput enhancement.

Conclusions:

• The proposed self-supervised dynamic learning approach ensures long-term, high-fidelity transmission in unstabilized MMFs.
• This method overcomes MMF channel variability, paving the way for practical AI-driven spatial transmission.
• Significant advancements in remote imaging and optical communication capacity are achievable.