Machine learning applied to fiber-fed focal plane wavefront sensing: a study of aberrated wave transmission through multimode optical fibers

This research explores the potential of machine learning and neural networks in recognizing the input features of aberrated wavefronts transmitted through multimode optical fibers, in view of applications for wavefront sensing in ground-based telescopes. Recent studies highlight the efficacy of multimode fibers for imaging and sensing, suggesting neural networks effectiveness in mapping relationships between output distortions and input wavefront aberrations. The initial step of our study concerned multimode fiber propagation simulations where an input Gaussian beam was distorted with known aberrations and then sent through the fiber, to analyze the effects on the output. This groundwork is used to train and validate a convolutional neural network architecture, providing a preliminary reconstruction accuracy from the simulated output images. The subsequent phase is experimental, and it aims to validate and potentially optimize neural network training based on the new outcomes.