Parallel implementation of spectral pileup correction and Gaussian noise suppression using CUDA heterogeneous architecture
-
作者
Chen, Zikang; Lin, Congyu; Huang, Yiwei; Zhang, Yue; Trigano, Tom; Zheng, Xiaoying; Zhu, Yongxin
-
刊物名称
NUCLEAR INSTRUMENTS & METHODS IN PHYSICS RESEARCH SECTION A-ACCELERATORS SPECTROMETERS DETECTORS AND ASSOCIATED EQUIPMENT
-
年、卷、文献号
2025, ,
-
关键词
Chen, Zikang; Lin, Congyu; Huang, Yiwei; Zhang, Yue; Trigano, Tom; Zheng, Xiaoying; Zhu, Yongxin
-
摘要
Spectroscopy devices are affected by the pulse pileup phenomenon, which arises from the overlap of detected signals. The energy-domain-based pileup correction algorithm by Trigano et al. (2015) estimates the pulse energy distribution by directly measuring the duration and energies of overlapping pulses, without the need for individual pulse identification. The correction algorithm can efficiently recover the energy spectrum even under a high photon arrival rate. However, the correction algorithm is sequential in nature and is slow when the energy resolution is high. In this paper, a fast parallel implementation of the original correction algorithm is presented. The parallel version exploits state-of-the-art many-core system technology and achieves near-linear acceleration as the problem size scales. The operating efficiency increases by over 200-fold when the energy spectrum is split into 2,048 bins. Additionally, a low-complexity Gaussian noise elimination method is introduced, which operates directly on the energy histogram and enhances energy resolution in the presence of Gaussian noise. Applied to real-world data from the Shanghai Synchrotron Radiation Facility, the algorithm demonstrates robust false peak removal. Finally, the performance limitations of the algorithm under high event rate conditions are analyzed asymptotically.