Xenon Signal Denoising via Supervised, Semi-Supervised, and Unsupervised Models

Abstract

This study presents a denoising algorithm trained using machine learning to improve the energy resolution of a single-phase liquid xenon time projection chamber for neutrinoless double beta decay detection. Supervised, unsupervised, and semi-supervised models are demonstrated to significantly remove noise from simulated measurements while preserving signal information. The supervised model achieves an energy resolution of $<1\%$, while the semi-supervised models achieve energy resolutions of $\sim 1\%$, and the unsupervised model performance is $\sim 1.5\%$. This work is evidence that machine learning denoising can improve energy resolution compared to traditional algorithms, even when experimentalists lack perfect a priori knowledge of the signals. Such models provide a realistic path toward next-generation sensitivity in $0νββ$ searches.

Figures (5)

• Figure 1: Example readout of a single 32-channel tile for a simulated event with added instrumentation noise.
• Figure 2: Simulated charge readout for a single tile channel. (Upper) Model-predicted (blue) charge readout for noisy signal (red) and true signal (black) simulation data for one tile channel. (Lower) Residuals for noisy (red) and model prediction (blue).
• Figure 3: Example of the Gaussian filter applied to generate the 55%-distortion training set. The recovered signal is an attempt to invert the distortion, demonstrating that the distortion represents significant loss of the original information.
• Figure 4: Charge resolutions across models of various data degradations.
• Figure 5: Projected 90% median sensitivity as a function of energy resolution, from Ref.nexo_main.