Abstract
A typical problem in low-level liquid scintillation (LS) counting is the identification of α particles in the presence of
a high background of β and γ particles. Especially the occurrence of β-β and β-γ pile-ups may
prevent the unambiguous identification of an α signal by commonly used analog electronics. In this case, pulse-shape
discrimination (PSD) and pile-up rejection (PUR) units show an insufficient performance. This problem was also observed in own earlier
experiments on the chemical behaviour of transactinide elements using the liquid-liquid extraction system SISAK in combination with
LS counting. α-particle signals from the decay of the transactinides could not be unambiguously assigned. However, the
availability of instruments for the digital recording of LS pulses changes the situation and provides possibilities for new
approaches in the treatment of LS pulse shapes. In a SISAK experiment performed at PSI, Villigen, a fast transient recorder, a PC card
with oscilloscope characteristics and a sampling rate of 1 giga samples s−1 (1 ns per point), was used for the first
time to record LS signals. It turned out, that the recorded signals were predominantly α, β-β and β-γ pile up, and fission events. This paper describes the subsequent development and use of artificial neural networks (ANN) based on the
method of “back-propagation of errors” to automatically distinguish between different pulse shapes. Such networks can “learn”
pulse shapes and classify hitherto unknown pulses correctly after a learning period. The results show that ANN in combination with fast
digital recording of pulse shapes can be a powerful tool in LS spectrometry even at high background count rates.
High‐Performance Single‐Crystal Diamond Detector for Accurate Pulse Shape Discrimination Based on Self‐Organizing Map Neural Networks
