Principal Component Analysis Applied to Digital Pulse Shape Analysis for Isotope Discrimination

Abstract

Digital pulse shape analysis (DPSA) techniques are becoming increasingly important for the study of nuclear reactions since the development of fast digitizers. These techniques allow us to obtain the (A, Z) values of the reaction products impinging on the new generation solid-state detectors. In this paper, we present a computationally efficient method to discriminate isotopes with similar energy levels, with the aim of enabling the edge-computing paradigm in future field-programmable gate-array-based acquisition systems. The discrimination of isotope pairs with analogous energy levels has been a topic of interest in the literature, leading to various solutions based on statistical features or convolutional neural networks. Leveraging a valuable dataset obtained from experiments conducted by researchers in the FAZIA Collaboration at the CIME cyclotron in GANIL laboratories, we aim to establish a comparative analysis regarding selectivity and computational efficiency, as this dataset has been employed in several prior publications. Specifically, this work presents an approach to discriminate between pairs of isotopes with similar energies, namely, 12,13C, 36,40Ar, and 80,84Kr, using principal component analysis (PCA) for data preprocessing. Consequently, a linear and cubic machine learning (ML) support vector machine (SVM) classification model was trained and tested, achieving a high identification capability, especially in the cubic one. These results offer improved computational efficiency compared to the previously reported methodologies.