Extend neural-network timewalk correction to additional detector inputs

Investigate whether information from neighbouring pixels and different track incidence angles can improve the Timepix4 neural-network timewalk correction, and determine how incorporating these variables affects the required calibration sample size and the robustness of the correction.

Background

The study trains the Timepix4 timewalk-correction neural network using only the measured charge and reconstructed intrapixel impact position, matching the inputs used by the fit-based baseline. The authors identify neighbouring-pixel information and varying track incidence angles as potentially relevant variables that could be supplied as additional neural-network inputs without introducing further explicit binning dimensions.

The paper does not resolve whether these additional inputs improve the correction, nor how their inclusion changes the amount of calibration data required or the robustness of the resulting correction. These questions are therefore stated as directions for future investigation.

References

Future studies are planned to investigate whether information from neighbouring pixels provides further improvements, and if the approach can be extended to take into account different track incidence angles. These variables can be included as additional NN inputs without introducing further explicit binning dimensions. However, their effect on the required calibration sample size and the robustness of the correction remains to be evaluated.

— Neural network-based timewalk correction for the Timepix4 ASIC  (2610.10328 - Bacher et al., 7 Oct 2026) in Conclusion, final discussion paragraph