Recent advances in deep learning for imaging and communication through multimode and multicore fibers

Abstract

Multicore and multimode optical fibers have emerged as promising technologies for high-capacity data transmission and minimally invasive imaging due to their ability to support multiple spatial channels. However, light propagation in such fibers is inherently complex, as modal dispersion and mode coupling lead to severe distortions, often manifested as random speckle patterns at the output. These effects significantly limit the direct retrieval of transmitted information and degrade system performance. In recent years, deep learning has gained considerable attention as a powerful tool to model the complex and nonlinear relationships between input and output optical fields in such systems. This review presents a comprehensive overview of deep learning techniques applied to multicore and multimode fiber systems, focusing on image reconstruction, signal recovery, and mode control. This paper also describes the conventional approaches such as transmission matrix methods and highlights their limitations in dynamic and noisy environments. Furthermore, recent advances in data-driven and hybrid physics-informed deep learning models are analyzed, demonstrating improved reconstruction accuracy, reduced data requirements, and enhanced robustness.