A neural network based algorithm for MRPC time reconstruction

Abstract

Multi-gap Resistive Plate Chamber(MRPC) is a widely used timing detector with a typical time resolution of about 60 ps. This makes MRPC an optimal choice for the time of flight(ToF) system in many large physics experiments. The prior work on improving the time resolution is mainly focused on altering the detector geometry, and therefore the improvement of the data analysis algorithm has not been fully explored. This paper proposes a new time reconstruction algorithm based on the deep neural networks(NN) and improves the MRPC time resolution by about 10 ps. Since the development of the high energy physics experiments has pushed the timing requirements for the MRPC to a higher level, this algorithm could become a potential substitution of the time over threshold(ToT) method to achieve a time resolution below 30 ps.

The study of a new time reconstruction method for MRPC read out by waveform digitizer

“…This method is an end-toend solution of reconstructing the first interaction time between the incident particles and the working gas. A preliminary study of the neural network has proved that it is more robust with the noise and can achieve a better performance [3]. Therefore, two sets of the neural networks are studied and compared in this paper.…”

Improving the time resolution of the MRPC detector using deep-learning algorithms

“…These kinds of algorithms have undergone tremendous innovations in the past 10 years and have already received wide attentions from the field of particle physics [11,12]. Prior work that utilized the simple fully-connected (FC) or long short term memory (LSTM) network to reconstruct MRPC detection time has shown promising results, and therefore they are extended and improved in this work [9,10].…”

“…Based on the new system, we proposed a time reconstruction algorithm using deep learning and neural networks. A combined long short term memory (ComLSTM) network which is an extension of our previous work [9,10] is designed and implemented to improve the MRPC time resolution from the perspective of the software. The simulation data used to train the network are optimized so that the most useful information is passed into the network and extracted by it.…”

Next frontiers in particle physics detectors: INSTR2020 summary and a look into the future

“…New studies on MPRCs with 20 gas gaps using a low-resistivity 400 μm-thick glass aims to achieve ~ 20 ps resolution per MIP for an upgraded detector [120]. An interesting development for the future TOF applications is represented by the implementation of neural networks for MRPC time reconstruction, instead of the standard tile-over-threshold (TOT) technique [121]. In the context of the PICOSEC-MM project, a two-stage amplification Micromegas coupled to a Cherenkov radiator coated with a semitransparent CsI photocathode achieved a time resolution of 24 ps per MIP for a 1 cm prototype [122].…”