Support this project and keep always updated by following on github.
Copyright (c) 2021 Danny Petschke ([email protected]). All rights reserved.
deepIRF - A deep learning approach for reverse-broadening the instrumental response (IRF) in lifetime spectra.
Is it generally possible to correct for the timing uncertainties inherently produced by the photomultiplier tubes?
- 1 Mio. pulse-pairs recorded from an isotope 60-Co have been stored along with the obtained timing differences (target/label value), i.e. the PMT uncertainty between them
- 70 % of the pulse-pairs have been used for training the model
- the other 30 % of the pulse-pairs have been used to test the trained model
- the results as shown in the figure below indicate that in principle deep learning can significantly reduce the uncertainty originating from the PMTs
- blue curve = initial "unseen" data (30 %), red curve = deepIRF-corrected initial "unseen" data (30 %)
https://drive.google.com/file/d/1S5gFA1I52M94mFA4AACai2Gp8C3BHxtR/view?usp=sharing