A 1000x acceleration for asynchronous single-photon LiDAR timestamp modeling, with convergence guarantees.
Asynchronous single-photon LiDAR streams timestamps continuously rather than accumulating a histogram over a fixed gate. It is the right architecture for navigation and high-frame-rate 3D sensing — but modeling the steady-state timestamp distribution in the presence of dead time has been the bottleneck.
(Zhang et al., 2026) reformulates the problem as a Markov chain and computes its stationary distribution roughly 1000× faster than the previous approach. Just as importantly, the paper provides a convergence analysis: a characterization of how quickly the chain reaches steady state, which tells you how much data a real system actually needs before the model applies.
Speed alone would be a useful engineering result. Speed with a convergence bound is what makes the model usable inside a real-time estimation loop.
Asynchronous single-photon LiDAR (SP-LiDAR) is an important imaging modality for high-quality 3D applications and navigation, but the modeling of the timestamp distributions in the presence of dead time remains a challenge. We present a Markov-chain formulation that accelerates the computation of the steady-state timestamp distribution by three orders of magnitude, together with a convergence analysis that characterizes how quickly the chain reaches its stationary regime.
@inproceedings{zhang2026realtime,title={Real-Time Markov Modeling for Single-Photon LiDAR: 1000x Acceleration and Convergence Analysis},author={Zhang, Weijian and Weerasooriya, Hashan K. and Chennuri, Prateek and Chan, Stanley H.},booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},year={2026},}