Scintillation pulse characterization with spectrum-inspired temporal neural networks: case studies on particle detector signals

TL;DR

This paper tackles the challenge of extracting precise scintillation-pulse characteristics from detector signals by exploiting spectral-temporal structure with a spectrum-informed neural network backbone, TimesNet-LE. The authors introduce spectrum-inspired multi-period convolution, modify TimesNet to emphasize low-frequency content and DC information, and evaluate representation learning strategies (SimSiam and encoder-predictor) to boost performance. Across two case studies—the LUX-like simulated scintillation data and NICA/MPD calorimeter experiments—the TimesNet-LE backbone with a nonlinear head achieves superior accuracy with markedly fewer parameters than baseline models and conventional methods. The findings suggest practical benefits for online data processing and scalable pulse characterization in diverse detector systems, with potential hardware deployment for low-latency inference.

Abstract

Particle detectors based on scintillators are widely used in high-energy physics and astroparticle physics experiments, nuclear medicine imaging, industrial and environmental detection, etc. Precisely extracting scintillation signal characteristics at the event level is important for these applications, not only in respect of understanding the scintillator itself, but also kinds and physical property of incident particles. Recent researches demonstrate data-driven neural networks surpass traditional statistical methods, especially when the analytical form of signals is hard to obtain, or noise is significant. However, most densely connected or convolution-based networks fail to fully exploit the spectral and temporal structure of scintillation signals, leaving large space for performance improvement. In this paper, we propose a network architecture specially tailored for scintillation pulse characterization based on previous works on time series analysis. The core insight is that, by directly applying Fast Fourier Transform on original signals and utilizing different frequency components, the proposed network architecture can serve as a lightweight and enhanced representation learning backbone. We prove our idea in two case studies: (a) simulation data generated with the setting of the LUX dark matter detector, and (b) experimental electrical signals with fast electronics to emulate scintillation variations for the NICA/MPD calorimeter. The proposed model achieves significantly better results than the reference model in literature and densely connected models and demonstrates higher cost-efficiency than conventional machine learning methods.

Figures (14)

• Figure 1: The functional diagram of multi-period convolution blocks. The symbol in the parentheses at the bottom of each square denotes the tensor shape in each step. The red paths and squares are components not in original TimesNet.
• Figure 2: The functional diagram of TimesNet-LE. The path and texts in red color are main differences between TimesNet and TimesNet-LE.
• Figure 3: The functional diagram of representation learning schemes.
• Figure 4: Simulation data of scintillation signals in the LUX experiment. Data with maximum correlation between parameters are visualized. We select the first 20 ns and plot 30 curves in each sub-figure.
• Figure 5: Performance measures (standard deviation of residuals between the ground-truth and the predicted) of parameter estimation on simulation data with correlated or uncorrelated noise at different signal-to-noise ratios. Network architectures with TimesNet-LE are marked with stars. Sub-figures in the bottom row share the same legend with those in the top row.