Deep learning the high variability and randomness inside multimode fibers.

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Information Transmission

Background:

• Multimode fibers (MMF) offer high-capacity information transmission.
• MMF performance is significantly degraded by external perturbations and environmental changes, leading to signal instability.
• Existing methods struggle to maintain reliable data transmission in dynamic MMF environments.

Purpose of the Study:

• To demonstrate the feasibility of transmitting binary images through a single multimode fiber (MMF) under dynamic shape variations using deep learning.
• To evaluate the generalization capability of a convolutional neural network (CNN) for MMF information recovery.
• To establish deep learning as a viable solution for overcoming MMF transmission challenges.

Main Methods:

• Implementation of a deep learning framework, specifically a convolutional neural network (CNN), for image transmission.
• Experimental setup involving a single multimode fiber (MMF) subjected to controlled dynamic shape variations.
• Training and testing the CNN on binary image data transmitted through the perturbed MMF.

Main Results:

• Successful binary image transmission was achieved through a single MMF experiencing dynamic shape variations.
• The trained convolutional neural network (CNN) demonstrated excellent generalization capabilities across various MMF states.
• Accurate prediction of unknown transmitted information was achieved, even in previously unencountered MMF configurations.

Conclusions:

• Deep learning, particularly CNNs, provides a robust solution to the inherent variability and randomness in multimode fiber (MMF) communication.
• This approach represents a significant advancement for developing future high-capacity MMF optical systems.
• The deep-learning methodology is potentially applicable to other optical systems involving diffusing media.