Comparison of Deep Learning Methods for Solving the Phase Problem in X-ray Diffraction Analysis

An approach is proposed that can allow ab initio solution of the phase problem in X-ray diffraction studies. A synthetic diffraction data generator has been created that can be used to solve applied problems using machine learning. An automated container is presented that allows reproducible experiments aimed at solving the problem within the framework of the proposed approach. The FFT_UNet and XRD_Transformer models were developed, taking into account the physical specifics of the problem, and a comparative analysis and interpretation of their work on real data were carried out.