Neural network--featured timing systems for radiation detectors: performance evaluation based on bound analysis

TL;DR

This paper evaluates neural network-based timing algorithms for radiation detectors, analyzing their performance limits through bound analysis and simulations, to guide future resolution optimization.

Contribution

It introduces a systematic bound analysis framework for neural network feature extraction in radiation detection timing, validated by case studies.

Findings

Neural networks can approach the Cramér-Rao lower bound in timing accuracy.

The proposed bound analysis effectively guides the assessment of feature extraction algorithms.

Simulation results demonstrate the potential of neural networks to improve timing resolution.

Abstract

Waveform sampling systems are used pervasively in the design of front end electronics for radiation detection. The introduction of new feature extraction algorithms (eg. neural networks) to waveform sampling has the great potential to substantially improve the performance and enrich the capability. To analyze the limits of such algorithms and thus illuminate the direction of resolution optimization, in this paper we systematically simulate the detection procedure of contemporary radiation detectors with an emphasis on pulse timing. Neural networks and variants of constant fraction discrimination are studied in a wide range of analog channel frequency and noise level. Furthermore, we propose an estimation of multivariate Cram\'er Rao lower bound within the model using intrinsic-extrinsic parametrization and prior information. Two case studies (single photon detection and shashlik-type calorimeter) verify the reliability of the proposed method and show it works as a useful guideline when assessing the abilities of various feature extraction algorithms.