Event based signal processing for isotope identification at reduced power
Abstract
Conventional radiation detection systems comprise analog sensors, data digitization and digital data processing systems driven by synchronous clocks. Such systems continuously consume power even in environments where radiation interactions may be rare and are always Poisson distributed. Conventional computing, using iterative minimization and optimization approaches to deconvolve actual radiation sources from detector responses, does not exploit the opportunities for parallel event processing which may offer significant power savings. Kromek proposes to reconfigure the “sensor to result” system chain to provide an entirely event driven chain. Event driven chains only consume power where events arrive rather than in response to a continuous clock. Kromek are already working on event driven digitization with work on digital silicon photomultipliers with event as opposed to frame readout. In this fundamental research program, we aim to develop understanding and demonstrate early prototypes of two further parts of the event processing chain. Firstly, the conversion of analogue sensor data into asynchronous event data. Such techniques are now being demonstrated in video processing to avoid the need to store unchanging background data. Secondly the use of analogue continuous time, discrete processors to derive true radiation flux from detector response. Solutions may include both conventional signal processing and neuromorphic processors in the event domain. Our approach will involve abstraction of the isotope identification process to a general base case. We will benchmark current processing with this base case. Event processing approaches, selected against this benchmark, will then be tested in both simulation and conventional neuromorphic hardware before embodiment in dedicated test very-large scale integration (VLSI).The project contains partners from the University of Liverpool, the University of Edinburgh and the University of Manchester, and provides for 4 PhD students.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 16, 2019
- Source ID
- HDTRA11810034
Entities
People
- Edward Marsden
Organizations
- Defense Threat Reduction Agency