Thesis

Towards highly efficient algorithms and hardware architecture design for single-photon signal processing

Downloadable Content

Download PDF
Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2025
Thesis identifier
  • T17181
Person Identifier (Local)
  • 201982935
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • In recent decades, single-photon detectors have emerged as crucial technologies for 3D remote sensing and biomedical applications. However, extracting essential parameters from encoded single-photon data presents significant challenges due to the ill-posed nature of parameter reconstruction, leading to high computational complexity. Robust yet compact algorithms are essential to ensure accuracy and computational efficiency. Furthermore, implementing these efficient algorithms on reconfigurable hardware processors promotes portability and real-world applicability. This thesis addresses these challenges through three interrelated topics that integrate signal processing on single-photon data, deep learning, and hardware implementation. For each topic, quantitative comparisons are conducted between our algorithms and state-of-the-art methods, demonstrating the superiority of our compact algorithms and hardware architectures. In the first topic, depth images are reconstructed from 3D point cloud data captured by a single-photon avalanche diode (SPAD) array, even under extreme low signal-to-background ratios (0.2, 0.04, and 0.02) per pixel. A low-bit quantization strategy is applied to the 3D DL model to achieve a small model size while maintaining accuracy. The second topic focuses on fluorescence lifetime reconstruction of both synthetic data and real data from experiments, utilizing 1D temporal point spread functions (TPSF) acquired by a photomultiplier tube (PMT) coupled with a timecorrelated single-photon counting (TCSPC) system. A lightweight 1D DL model, in conjunction with TPSF compression and a customized hardware processor, facilitates rapid and accurate lifetime image reconstruction. This approach surpasses conventional DL models and non-linear fitting methods in performance. In the third topic, we accurately reconstruct the blood flow index and coherence factor from autocorrelation functions for a diffuse correlation spectroscopy (DCS) system using customized 1D DL. A processor implemented on a reconfigurable device paves the way for integrating portable DCS systems in the future. Through these contributions, this thesis advances the field of single-photon signal processing by providing compact algorithms and hardware implementations that improve accuracy, computational efficiency, and portability in single-photon applications for 3D sensing and biomedical imaging.
Advisor / supervisor
  • Li, David Day Uei
Resource Type
DOI
Funder

Relations

Items

Thumbnail Title Date Uploaded Visibility Actions
file details PDF of thesis T17181 2025-01-21 Public Download