Learning rotated speckle patterns projected from step-index multimode fiber under random external disturbances

During multimode fiber (MMF)-based information transmission, strong and persistent external disturbances can readily induce mode coupling, significantly degrading image reconstruction and posing major challenges for practical applications. To ensure stable transmission, this paper analyzes mode cross-coupling and inter-mode interactions and then proposes a conditional generative adversarial network to learn rotationally speckle patterns projected through a step-index MMF under random bending or swaying perturbations. The method demonstrates exceptional robustness and stability in reconstructing various types of test images experimentally. Those findings contribute to critical theoretical and technological foundations for MMF information transmission, paving the way for further research and practical application deployment.