Embedded instrumentation platform on SoC/FPGA for mixed radiation fields

This thesis documents a series of developments that resulted on an instrumentation platform for mixed radiation field measurements, enabled by field-programmable gate arrays (FPGA) and System-on-chip (SoC) technologies. This platform is capable of reliably discriminate online individual events from diverse radiation sources using a single detector, even under challenging distortion conditions caused by high radiation intensity or low signal-to-noise ratio. Many of the developments in the series required experimental data from mixed radiation environments in order to optimize the implementations. In this regard, an open-source remote diagnostics platform was created to control and validate instrument prototypes in nuclear and HEP experiments. This tool permits developers to test in real-time data acquisition systems (DAQs) and detectors located in radiation-controlled areas, while safely operating the DAQ remotely from a control room. This diagnostics platform enabled agile experiment deployments in constrained time slots, saving valuable time in the data recording sessions during the research. Once tested in the laboratory, a sequence of novel methods has been devised to implement the flexible embedded instrument for mixed radiation fields. First, a method for fast pulse shape recognition targeting SoC/FPGA was formulated. This technique is based on the Pearson’s correlation to create a matching filter, optimized to avoid expensive computational operations. Equivalent detection performance under low signal-to-noise ratio was demonstrated in a SoC/FPGA deployment compared to the Pearson’s algorithm, while substantially reducing latency and hardware utilization. Then, a method based on frequency-domain analysis for gamma/neutron discrimination (FCI) in mixed radiation environments was developed, envisioning embedded real-time applications with single crystal detectors. Superior performance was experimentally validated when compared to pulse-shape discrimination (PSD), especially in the lower energy ranges. The method was optimized to be computationally efficient and was deployed in a SoC/FPGA device, showing excellent resource utilization at sustained operation frequencies compatible with online processing. Besides, a public dataset was released, featuring the first data collection with gamma/neutron labels supported by the neutron reactions of the scintillator. A simulation also proved that FCI is very likely to outperform PSD with organic scintillators, motivating further experiments for diverse detector types. Finally, by gathering several contributions from the aforementioned developments, an embedded system was designed to overcome the limitations of FCI under continuous pile-up distortion. A low-SWaP (size, weight, and power) and high performance event discrimination instrument was created for mixed radiation fields, supported by a real-time machine learning model running on a low-end FPGA. The final product achieved the best performance/SWaP ratio among the latest developments. This optimization milestone enabled top-tier capabilities comparable with benchtop instruments, such as high event-rate continuous pile-up recovery and discrimination, in a flexible embedded platform with outstanding portability features. Experimental results demonstrated such metrics with a commercial integrated monocrystal detector designed for gamma/neutron discrimination.