Signal Processing Based Pile-up Compensation for Gated Single-Photon Avalanche Diodes

Adithya Pediredla, Alberto Tosi, Mauro Buttafava, Aswin C. Sankaranarayanan, Ashok VeeraraghavanarXiv 2018 · DOI · arXiv

Abstract

Single-photon avalanche diode (SPAD) based transient imaging suffers from an aberration called pile-up. When multiple photons arrive within a single repetition period of the illuminating laser, the SPAD records only the arrival of the first photon; this leads to a bias in the recorded light transient wherein the transient response at later time-instants are under-estimated. An unfortunate consequence of this is the need to operate the illumination at low-power levels to reduce the probability of multiple photons returning in a single period. Operating the laser at low power results in either low signal-to-noise ratio (SNR) in the measured transients or reduced frame rate due to longer exposure durations to achieve a high SNR. In this paper, we propose a signal processing-based approach to compensate pile-up in post-processing, thereby enabling high power operation of the illuminating laser. While increasing illumination does cause a fundamental information loss in the data captured by SPAD, we quantify this information loss using Cramer-Rao bound and show that the errors in our framework are only limited to this information loss. We experimentally validate our hypotheses using real data from a lab prototype.

Introduction: The Pile-up Problem

Transient imaging captures "light-in-flight," visualizing light propagation at picosecond scales. This is typically achieved using Single-Photon Avalanche Diodes (SPADs), which are sensitive enough to detect individual photons. However, SPADs operate on a "first-come, first-served" basis. In a single laser cycle, a SPAD records only the arrival time of the first photon it detects. If multiple photons return from a scene, subsequent ones are ignored.

This phenomenon, known as pile-up, creates a significant bias: the resulting histogram of photon arrivals is skewed toward earlier times, causing distant objects or secondary reflections to appear dimmer or disappear entirely. Traditionally, researchers mitigate this by operating lasers at extremely low power to ensure that the probability of multiple photons arriving per cycle is negligible. However, this leads to low signal-to-noise ratios (SNR) and necessitates long exposure times.

Figure 1: Comparison of ideal, piled-up, and de-biased transients across different imaging scenarios.

Methodology: Shifting the Burden to Software

The proposed framework moves the solution from hardware constraints to signal processing. By modeling the physics of the sensor, the "piled-up" data can be mathematically inverted to recover the true transient.

Probabilistic Modeling

The SPAD histogram is modeled as a multinomial distribution. The key insight is that the probability of detecting a photon in a specific time bin is conditional on no photons having been detected in any preceding bins. This dependency allows for the derivation of statistical estimators.

Statistical Estimation

Two primary estimators were developed:

  1. Maximum Likelihood (ML) Estimator: A formula that calculates the most probable "true" light transient from the distorted histogram.
  2. Maximum A Posteriori (MAP) Estimator: This incorporates "priors" (educated guesses about the scene, like the fact that transients are usually sparse) to improve reconstruction when data is sparse or noisy.

To quantify the performance, the researchers utilized the Cramer-Rao Bound (CRB), which defines the theoretical limit of information recovery. This allowed them to measure the information loss incurred by increasing laser power and show that their algorithm performs at the mathematical limit of what is possible.

Experimental Setup and Validation

To validate the framework, a custom laboratory prototype was constructed. The setup includes a supercontinuum laser with a pulse-picker for repetition rate control and a fast-gated CMOS SPAD.

Figure 2: Schematic of the experimental laboratory prototype used for data acquisition and validation.

The experiments focused on a "Two-Path" scene consisting of two cardboard patches at different distances. This setup generates two distinct peaks in the light transient. Under high-power illumination, the second (later) peak is suppressed by pile-up from the first. The algorithm's success was measured by its ability to accurately recover the relative amplitudes of both peaks.

Key Results and Insights

The research yielded several critical findings:

  • Deterministic Bias Recovery: Pile-up is a predictable statistical pattern. By inverting the sensor's distortion model, the "lost" photons can be effectively recovered.
  • High Power Advantage: While high-power illumination causes some information loss due to sensor saturation, the resulting boost in total photon counts often yields a higher overall SNR than low-power operation, provided the pile-up is compensated.
  • Robustness: The framework remains effective even when up to 90% of incoming photons are "dropped" due to pile-up.
  • Background Rejection: The method effectively cleans transients corrupted by high ambient light, which typically masks the signal in standard SPAD imaging.

Conclusion and Future Work

This signal-processing approach enables SPADs to operate at much higher power levels, significantly reducing acquisition times for transient imaging. Future work will address hardware non-idealities such as "after-pulsing" and "gate-ringing," and explore the application of these algorithms to real-time video of dynamic scenes using advanced spatial priors or deep learning techniques.

Summary updated

Abstract

Single-photon avalanche diode (SPAD) based transient imaging suffers from an aberration called pile-up. When multiple photons arrive within a single repetition period of the illuminating laser, the SPAD records only the arrival of the first photon; this leads to a bias in the recorded light transient wherein the transient response at later time-instants are under-estimated. An unfortunate consequence of this is the need to operate the illumination at low-power levels to reduce the probability of multiple photons returning in a single period. Operating the laser at low power results in either low signal-to-noise ratio (SNR) in the measured transients or reduced frame rate due to longer exposure durations to achieve a high SNR. In this paper, we propose a signal processing-based approach to compensate pile-up in post-processing, thereby enabling high power operation of the illuminating laser. While increasing illumination does cause a fundamental information loss in the data captured by SPAD, we quantify this information loss using Cramer-Rao bound and show that the errors in our framework are only limited to this information loss. We experimentally validate our hypotheses using real data from a lab prototype.
BibTeX
@misc{pediredla2018signal,
  author = {Adithya Pediredla and Alberto Tosi and Mauro Buttafava and Aswin C. Sankaranarayanan and Ashok Veeraraghavan},
  title = {{Signal Processing Based Pile-up Compensation for Gated Single-Photon Avalanche Diodes}},
  year = {2018},
  eprint = {1806.07437},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1806.07437},
}
Figure