Timing and characterization of shaped pulses with MHz ADCs in a detector system: a comparative study and deep learning approach

Abstract

A: Timing systems based on Analog-to-Digital Converters are widely used in the design of previous high energy physics detectors. In this paper, we propose a new method based on deep learning to extract the time information from a finite set of ADC samples. Firstly, a quantitative analysis of the traditional curve fitting method regarding three kinds of variations (long-term drift, short-term change and random noise) is presented with simulation illustrations. Next, a comparative study between curve fitting and the neural networks is made to demonstrate the potential of deep learning in this problem. Simulations show that the dedicated network architecture can greatly suppress the noise RMS and improve timing resolution in non-ideal conditions. Finally, experiments are performed with the ALICE PHOS FEE card. The performance of our method is more than 20% better than curve fitting in the experimental condition. K: Analysis and statistical methods; Pattern recognition, cluster finding, calibration and fitting methods; Front-end electronics for detector readout; Timing detectors 1Corresponding author.

Data-driven detector signal characterization with constrained bottleneck autoencoders

“…On one hand, AEs can be used to compress data by storing the much lighter latent space representation [4][5][6]. On the other hand, AEs average out the noise from different data examples during training and, therefore, they have great potential to denoise signals [7][8][9]. AE transformations are greatly interesting by themselves.…”

Universal uncertainty estimation for nuclear detector signals with neural networks and ensemble learning

“…In this setting, a random Gaussian white noise with 0.05 standard deviation is added to the sampling points. Since noises at different times are completely uncorrelated, nonlinear least squares curve fitting gives the maximum likelihood estimation [11] with reasonable standard errors of fitting parameters. It should be noted that this condition is a strong assumption and very idealized since the underlying mathematical function is unknown and variable in reality.…”

“…Because of correlation, curve fitting is no longer the maximum likelihood estimation so that it may give sub-optimal fitting results and unreasonable standard errors. However, because of the highly nonlinear mapping, neural networks are competent to work well even though the input data is correlated [11].…”

“…However, one should bear in mind that dedicated structure and enough model capacity are essential to work in more complicated situations. In [11], it was demonstrated that a fully-connected feedforward neural network is unable to predict time in intense random Gaussian noise. Usually timing prediction is more challenging than energy prediction, and a well-structured network architecture is vital for network convergence in harsh situations.…”

“…It also provides a great opportunity to apply novel machine learning algorithms to traditional problems in nuclear instrumentation. In [11], a neural network based on autoencoders was used for timing and characterization of pulse signals from a photon spectrometer. It achieved significantly better results than curve fitting.…”