Fast reconstruction of laser beam near-field and focal spot profiles using deep neural network and numerical propagation

Abstract

Inertial confinement fusion (ICF) experiments demand precise knowledge of laser beam parameters on high-power laser facilities. Among these parameters, near-field and focal spot distributions are crucial for characterizing laser beam quality. While iterative phase retrieval shows promise for laser beam reconstruction, its utility is hindered by extensive iterative calculations. To address this limitation, we propose an online laser beam reconstruction method based on deep neural network. In this method, we utilize coherent modulation imaging (CMI) to obtain labels for training the neural network. The neural network reconstructs the complex near-field distribution, including amplitude and phase, directly from a defocused diffraction pattern without iteration. Subsequently, the focal spot distribution is obtained by propagating the established complex near-field distribution to the far-field. Proof-of-principle experiments validate the feasibility of our proposed method.