Image reconstruction through unknown random configurations of multimode fibers using deep learning

[S. Resisi et al, arxiv, 2011.05144 (2020)]

Multimode fibers tend to scramble input image due to intermodal dispersion and random mode coupling. While this scrambling effect can be learned, for instance by measuring the transmission matrix of the fiber, slight changes of the geometrical conformation of the fiber modify its response, making the calibration obsolete. In the present paper, the authors use deep-learning using data sets acquired over a wide range of deformations to reconstruct images sent through unknown configurations of multimode fibers. This is possible thanks to the presence of invariant properties that the numerical model learns.

Fast numerical estimations of axisymmetric multimode fibers modes

Estimating the propagation constants and the transverse mode profiles of multimode fibers is not as easy as it sounds. In our recent work we highlighted here, we needed to estimate the mode profiles for a standard graded-index fiber. It turned out that many standard approximations done in the literature to estimate the propagation constants do not give results accurate enough for the mode profile. The general approach we introduced in a previous tutorial to numerically find the fiber modes for any index profile using a 2D scalar finite differences approach is still valid. However, to provide accurate results, it needs a fine discretization of the space that leads to important memory and computational time requirements when the fiber core increases. If we consider an axisymmetric fiber, we can obtain a 1D formulation of the problem, that is unfortunately unstable under naive finite differences approaches. We detail here a stable formulation that leads to accurate and fast estimations of the mode profiles.

Learning and avoiding disorder in multimode fibers