Utilising machine learning algorithms for high speed data processing of a single photon counting 256 channel PMT with a timing resolution of 60 ps

Development of a 256 Micro-Channel Plate Photo-Multiplier Tube (MCP-PMT) to further increase the spatial resolution while maintaining the temporal resolution of 60ps using charge sharing techniques would enable a novel commercial camera system. This would be designed so that each channel is an independent photon detector allowing for single photon counting possibilities which would advance fields such as LiDAR, particle physics, Time Correlated Single Photon Counting (TCSPC) in biology and quantum information systems. Electronic capabilities to measure increasing photon rates have introduced a bottleneck in the path of real time data processing using current algorithmic software. This research explores a potential solution of a machine learning (ML) algorithm for performing the data processing and imaging, with the objective of reconstructing the photon event in both spatial and temporal coordinates with a time constraint 10μs per single photon event. In this paper optimisation of the model is detailed with discussion on various hyper-parameters of the model architecture. The results demonstrate the ML model's success in reconstructing photon events from real detector data with a training set consisting of simulated data. The research details approaches to training with real detector data and introduces a generative model, which aims to generate realistic photon events. Further work aims to assess the variance in performance of the ML model when trained with a hybrid training set containing both generated photon events and photon events from the MCP-PMT.