Learning and avoiding disorder in multimode fibers

TL;DR

This paper presents a framework to identify optical channels in multimode fibers that are resilient to disorder, enabling improved data transmission and endoscopic applications through deep learning-based modal transmission matrix estimation.

Contribution

It introduces a novel method to find disorder-resistant modes in multimode fibers using deep learning, enhancing robustness and potential practical use.

Findings

Discovered nearly complete sets of disorder-resilient optical channels.

Demonstrated that a few key properties can characterize light propagation despite high disorder.

Developed a fast, model-based deep learning approach for modal transmission matrix estimation.

Abstract

Multimode optical fibers (MMFs) have gained renewed interest in the past decade, emerging as a way to boost optical communication data-rates in the context of an expected saturation of current single-mode fiber-based networks. They are also attractive for endoscopic applications, offering the possibility to achieve a similar information content as multicore fibers, but with a much smaller footprint, thus reducing the invasiveness of endoscopic procedures. However, these advances are hindered by the unavoidable presence of disorder that affects the propagation of light in MMFs and limits their practical applications. We introduce here a general framework to study and avoid the effect of disorder. We experimentally find an almost complete set of optical channels that are resilient to disorder induced by strong deformations. These deformation principle modes are obtained by only exploiting measurements for weak perturbations. We explain this effect by demonstrating that, even for a high level of disorder, the propagation of light in MMFs can be characterized by just a few key properties. These results are made possible thanks to a precise and fast estimation of the modal transmission matrix of the fiber which relies on a model-based optimization using deep learning frameworks.