Pty-co-SAXSNN: CNN-based deconvolution for simultaneous X-ray ptychography and SAXS nanostructure characterization

We present pty–co–SAXSNN, a convolutional neural network approach enabling simultaneous ptychography and small-angle X-ray scattering (SAXS). Beam divergence from focusing optics often blurs SAXS features, limiting the analysis of reciprocal-space structures within ptychographic diffraction patterns. To overcome accuracy and speed limitations of traditional deconvolution, this framework uses ptychography-reconstructed probe to deconvolve and correct these effects. Trained on realistic simulations from experimentally reconstructed probes, the network is applied to gold nanoparticle clathrate assemblies, rapidly recovering sharp Bragg peaks and spatially resolved orientations. This framework enables fast, robust, and dose-efficient structural characterization, linking nanoscale crystallinity with mesoscale organization in complex materials. Graphical abstract