[Paper Review] Deep learning the high variability and randomness inside multimode fibres

Pengfei Fan, Tianrui Zhao+1 more|arXiv (Cornell University)|Jul 18, 2018

Optical Network Technologies3 references3 citations

TL;DR

This paper demonstrates that a single convolutional neural network can accurately reconstruct binary images transmitted through multimode fibers (MMFs) despite their highly variable and random transmission characteristics. By training on multiple MMF states, the deep learning model generalizes across unknown channel conditions without prior knowledge of the fiber's current state, enabling robust image recovery in dynamic environments.

ABSTRACT

Multimode fibres (MMF) are remarkable high-capacity information channels owing to the large number of transmitting fibre modes, and have recently attracted significant renewed interest in applications such as optical communication, imaging, and optical trapping. At the same time, the optical transmitting modes inside MMFs are highly sensitive to external perturbations and environmental changes, resulting in MMF transmission channels being highly variable and random. This largely limits the practical application of MMFs and hinders the full exploitation of their information capacity. Despite great research efforts made to overcome the high variability and randomness inside MMFs, any geometric change to the MMF leads to completely different transmission matrices, which unavoidably fails at the information recovery. Here, we show the successful binary image transmission using deep learning through a single MMF, which is stationary or subject to dynamic shape variations. We found that a single convolutional neural network has excellent generalisation capability with various MMF transmission states. This deep neural network can be trained by multiple MMF transmission states to accurately predict unknown information at the other end of the MMF at any of these states, without knowing which state is present. Our results demonstrate that deep learning is a promising solution to address the variability and randomness challenge of MMF based information channels. This deep-learning approach is the starting point of developing future high-capacity MMF optical systems and devices, and is applicable to optical systems concerning other diffusing media.

Motivation & Objective

• To address the challenge of high variability and randomness in multimode fiber (MMF) transmission channels caused by environmental perturbations and geometric changes.
• To enable reliable image transmission through MMFs without requiring knowledge of the current transmission matrix or channel state.
• To develop a deep learning framework that generalizes across diverse MMF configurations and dynamic shape variations.
• To demonstrate that a single trained model can recover information accurately across unknown, varying MMF states.
• To establish a foundation for future high-capacity optical systems using MMFs in communication, imaging, and sensing.

Proposed method

• A single convolutional neural network (CNN) is trained on multiple transmission matrices corresponding to different MMF states, including both static and dynamically varying configurations.
• The network learns the mapping from input binary images to output intensity patterns at the fiber's output, generalizing across unknown channel states.
• Training data includes diverse MMF configurations, ensuring the model learns invariant features despite changes in the fiber’s transmission behavior.
• The model is tested on unseen MMF states, including those not present during training, to evaluate generalization performance.
• The architecture is designed to be robust to channel variations, avoiding the need for retraining or channel state estimation.
• The approach leverages the representational power of deep learning to implicitly model complex, nonlinear mode coupling in MMFs.

Experimental results

Research questions

• RQ1 Can a single deep neural network generalize across multiple, highly variable and random transmission states in multimode fibers?
• RQ2 To what extent can a deep learning model recover binary images through MMFs without prior knowledge of the fiber’s current state?
• RQ3 How does the model perform when the MMF undergoes dynamic shape variations during transmission?
• RQ4 Can the trained model accurately predict outputs for MMF states not included in the training data?
• RQ5 Is deep learning a viable solution for overcoming the inherent randomness and variability in MMF-based optical information channels?

Key findings

• A single convolutional neural network successfully reconstructs binary images transmitted through a multimode fiber across multiple distinct transmission states.
• The model generalizes effectively to unknown MMF configurations, including dynamically changing states, without requiring retraining or channel state feedback.
• The network achieves accurate image recovery even when the transmission matrix changes completely due to fiber perturbations or geometric variations.
• The approach eliminates the need for measuring or estimating the transmission matrix in real time, overcoming a major bottleneck in MMF applications.
• The results demonstrate that deep learning can effectively model the complex, nonlinear, and random behavior of multimode fiber channels.
• This method enables reliable information transmission through MMFs in practical scenarios where channel conditions are unpredictable and unstable.