Enhancing the sensitivity of next-generation X-ray imaging detectors with artificial intelligence and advanced event reconstruction algorithms
Advanced algorithms incorporating artificial intelligence and machine learning (AI/ML) enhance the sensitivity of X-ray imaging detectors and the scientific capabilities of future X-ray missions. In orbit, current instruments are limited in their sensitivity by (1) the instrumental background, induced by cosmic rays which produce signals that can be confused for genuine, astrophysical X-rays, and (2) the ability to reconstruct the detected photon events, degrading the quantum efficiency and energy resolution at the lowest energies, where much discovery space resides. We report on the development of prototype algorithms designed to operate on the raw frame-level data to provide improved identification of particle-induced background events and enhanced energy reconstruction. These algorithms consider the contextual information from all signals in a frame, and are built upon physics-motivated models of charge diffusion and signal generation within the detector. Using high fidelity simulations, we show that following recent developments, prototype ML algorithms can reduce the unrejected particle background by up to 68 per cent compared with traditional filtering methods when operated in an aggressive mode suitable for source detection in imaging surveys, or up to 40 per cent in a conservative mode designed to prioritize accurate measurements of the spectrum. We find that next-generation event reconstruction algorithms improve the sensitivity and energy resolution of CCD-like detectors at event energies below 1keV, and can aid in background filtering, and reduce the impact of photon pile-up. We present new laboratory data that demonstrates the performance of the algorithm on the MIT-LL CCID-93 CCD detector. Together with the capabilities of next-generation high-speed, low-noise detectors, these algorithms can satisfy the requirements for future X-ray flagship missions.