| dc.description.abstract | In this work, an open source code RadSigPro 1.0 is developed and used for fast processing
of nanosecond long pulses from scintillation detectors. The pulse processing and identification
involves pulse height distribution (PHD), pulse shape discrimination (PSD), and time-of-flight
(TOF). For the goal of better particle segregation, processed particle waveforms are supplied to
test machine learning techniques along with TOF labeled neutrons and gamma-rays to train the
data. The code is used to model the programmable logic design of an field programmable gate array (FPGA) design for on-the-fly processing of neutron and gamma-ray pulses, along with testing the results. Finally, a comparison between CAEN’s CoMPASS Data Acquisition (DAQ) Software and RadSigPro’s resulting tallies is attempted.
When trained on pulse waveform data, classification accuracy of 96% could be achieved with
less than 100 ns of data, but 400 ns were required to get the accuracy to 97%. This indicates the
information relevant to labeling a pulse as a neutron or gamma-ray is mostly found at the pulse’s start. Principle component analysis (PCA) extracts information from the entire pulse, so relevant information is not lost when the number of components is trimmed. As a result, support vector machine (SVM) models trained on two principal components could accurately classify pulses over 94% of the time. To achieve 97% accuracy, models with nonlinear kernels required fewer than 50 principal components for training. Misclassification results displayed a 1.97% false gamma-ray rate and a 2.27% false neutron rate.
A weighted average of the percent difference of the results for RadSigPro 1.0 implemented on a
central processing unit (CPU) and an FPGA logic design is calculated. This shows a 0% difference
for the PHD data sets, a 0.458% and 0.344% difference for the designated gamma-detector and
neutron-detector’s PSD data sets respectively, and a 0% difference for the TOF data set. When the FPGA logic design is applied and simulated, it computes the total and tail pulse areas within 5 nanoseconds of the arrival of the final data point used for accumulation and also captures the pulse height value within 2 nanoseconds of the arrival of the pulse maximum’s data point. | |
