Optimal NaI(Tl) energy resolution using a neural network and tagged gamma photons

Abstract This paper presents a novel method for recovering optimal energy resolution in a NaI(Tl) detector directly from raw photomultiplier tube (PMT) signals, bypassing the need for traditional hardware signal processing modules such as preamplifiers, amplifiers and analog-to-digital converter. The approach utilizes a digital waveform recorder and an Artificial Neural Network (ANN) trained with experimental data derived from Compton scattered photons occuring in a High-Purity Germanium (HPGe) detector. The ANN method, when applied to the raw PMT pulses, yields similar energy resolutions to the conventional method where the PMT signal is processed by a preamplifier and a spectroscopy amplifier. However, it consistently outperforms the raw PMT pulse amplitude and integral methods, demonstrating superior energy resolution, especially given the low amplitude and high background fluctuations inherent in the raw PMT signals. Notably, the ANN achieved a photoelectric peak resolution of 7.4% at 661.7 keV for 137 Cs, which is comparable to the optimal resolution obtained with full analog electronics. This technique, demonstrated with a 137 Cs radioactive source for primary photons, can allow for the precise determination of energy deposition in NaI(Tl) or similar detectors from photons having energies equal or smaller than the energy of the primary photons which undergoes the Compton scattering. This work highlights the potential of ANNs, when coupled with waveform digitizers, for streamlining detector setups and optimizing performance.