A neutron-gamma pulse shape discrimination method based on pure and mixed sources

We present a novel trainable approach to distinguish neutrons from gammas using a particle detector. Traditionally, Pulse Shape Discrimination (PSD) methods for this problem utilize an ad-hoc computation of tail signal energy to perform the detection. Our first contribution is a rigorous analysis of the performance of this existing approach on gold standard Time of Flight (TOF) data. While this approach performs well for high energy pulses, its accuracy drops dramatically as the pulse energy decreases. Our second contribution is a novel data driven classifier that is trained from two readily available sources: one that emits gamma particles (Cs-137), and another that emits a mixture of gamma and neutron particles (Cf-252). We test our approach using TOF experiments and show a marked improvement in accuracy over the traditional method for low false positive rates and low energies.