Dynamic sampling for SAXSTT: towards real-time measurement adaptation

Small-angle X-ray scattering tensor tomography (SAXSTT) is a powerful technique for non-destructive 3D characterization of nanoscale structures by reconstructing q -resolved reciprocal space maps. Yet this approach requires extensive raster scanning from multiple directions, leading to long acquisition times that challenge synchrotron-based experiments. To overcome this, we propose a dynamic sampling strategy combined with an online error feedback mechanism. This strategy ensures efficient real-time acquisition by adaptively terminating measurements based on evolving reconstruction quality. By applying this strategy to experimental datasets, we demonstrate up to a sevenfold reduction in the required number of projections while maintaining reconstruction quality, with the achievable reduction remaining strongly sample-dependent. Furthermore, we conduct quantitative evaluations using orientation alignment and normalized cross-correlation, confirming the robustness of the presented strategy. Compatible with existing beamline systems, this strategy offers a practical solution for time-efficient and adaptive SAXSTT acquisition.