Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Synthesis of pulses from particle detectors with a Generative Adversarial Network (GAN)

Alberto Regadío, Luis Esteban, Sebastián Sánchez-Prieto

TL;DR

The paper addresses the challenge of developing readout electronics when detector pulses are scarce by proposing a 1D DCGAN that generates realistic detector pulses. The Generator, trained with cross-entropy loss to fool the Discriminator, produces pulses of fixed length that mimic both waveform shapes and pulse-height distributions, as validated by per-pulse distances and histogram-based metrics including FWHM. The model is trained on real pulses from a NaI scintillator using $N=128$ and $L=512$, and the trained Generator is deployed on a PYNQ-Z2 FPGA-based SoC, demonstrating detector-free testing capability. Results show convergence of losses, high similarity between real and synthetic pulses, and matching pulse-height histograms for sources such as $^{137}$Cs and $^{22}$Na, with significant potential to reduce testing costs and accelerate electronics development across detector platforms.

Abstract

To address the possible lack or total absence of pulses from particle detectors during the development of its associate electronics, we propose a model that can generate them without losing the features of the real ones. This model is based on artificial neural networks, namely Generative Adversarial Networks (GAN). We describe the proposed network architecture, its training methodology and the approach to train the GAN with real pulses from a scintillator receiving radiation from sources of ${}^{137}$Cs and ${}^{22}$Na. The Generator was installed in a Xilinx's System-On-Chip (SoC). We show how the network is capable of generating pulses with the same shape as the real ones that even match the data distributions in the original pulse-height histogram data.

Synthesis of pulses from particle detectors with a Generative Adversarial Network (GAN)

TL;DR

The paper addresses the challenge of developing readout electronics when detector pulses are scarce by proposing a 1D DCGAN that generates realistic detector pulses. The Generator, trained with cross-entropy loss to fool the Discriminator, produces pulses of fixed length that mimic both waveform shapes and pulse-height distributions, as validated by per-pulse distances and histogram-based metrics including FWHM. The model is trained on real pulses from a NaI scintillator using and , and the trained Generator is deployed on a PYNQ-Z2 FPGA-based SoC, demonstrating detector-free testing capability. Results show convergence of losses, high similarity between real and synthetic pulses, and matching pulse-height histograms for sources such as Cs and Na, with significant potential to reduce testing costs and accelerate electronics development across detector platforms.

Abstract

To address the possible lack or total absence of pulses from particle detectors during the development of its associate electronics, we propose a model that can generate them without losing the features of the real ones. This model is based on artificial neural networks, namely Generative Adversarial Networks (GAN). We describe the proposed network architecture, its training methodology and the approach to train the GAN with real pulses from a scintillator receiving radiation from sources of Cs and Na. The Generator was installed in a Xilinx's System-On-Chip (SoC). We show how the network is capable of generating pulses with the same shape as the real ones that even match the data distributions in the original pulse-height histogram data.
Paper Structure (7 sections, 2 equations, 13 figures)

This paper contains 7 sections, 2 equations, 13 figures.

Table of Contents

  1. Introduction
  2. Model design
  3. Generator architecture
  4. Discriminator architecture
  5. Evaluation metrics
  6. Results
  7. Conclusions

Figures (13)

  • Figure 1: Architecture of the proposed GAN. $N$ is the dimension of the random vectors and $L$ is the pulse length.
  • Figure 2: Generator architecture. The arrows stand for layers; on the left of the bar the type of layer is shown and on the right the activation function; 'conv' stands for convolutional layer of kernel size equal the number that accompanies it. The rectangles represent the layer input/output matrices together with their dimensions.
  • Figure 3: Discriminator architecture. The arrows stand for layers; on the left of the bar the type of layer is shown and on the right the activation function; 'conv' stands for convolutional layer of kernel size equal the number that accompanies it. The rectangles represent the layer input/output matrices together with their dimensions.
  • Figure 4: Diagram of the detection chain used for the experimental test. PM stands for photomultiplier and PA-12 is the preamplifier.
  • Figure 5: Learning curve for ${}^{137}$Cs. The y-axis stands for loss functions according to Eq. (\ref{['EqJg']}) and (\ref{['EqJd']}) for the Generator and the Discriminator, respectively. Note that both functions of the discriminator are overlapping.
  • ...and 8 more figures