Mitigating photoionization effects on time series data for noble element time projection chambers

Noble liquid-gas time projection chambers (TPCs) have set the standard for direct detection dark matter experiments. These experiments are extremely sensitive, so much so that they can detect the light produced by single electrons within their volume. However, this sensitivity comes at a price; these experiments are strongly impacted by any impurities present in them. In particular, when photons produced by an initial interaction interact with an impurity, photoionization occurs in the purified noble liquid and gas volumes or from detector material surfaces, leading to released electrons that generate further light signals long after the initial interaction. This is a problem for high energy (MeV scale) interactions, in which these photoionization backgrounds overlap with the signals originating from the main interaction. To remedy this, we propose developing sequential machine learning methods trained on simulated photosensor waveform data, through which the effect of photoionization can be mitigated and the corrected signal waveform can be recovered. These methods can be applied to improve the energy resolution for measuring interactions at the MeV scale and are broadly applicable for time series data from TPC type detectors.