SPAD Dead-Time Linearization Model

• SPAD dead-time linearization models are frameworks that correct statistical distortions from detector dead time by inverting the nonlinear photon counting response.
• They utilize mathematical formulations such as Mandel–Srinivas inversion and first-order kernel convolution approximations to restore true Poissonian photon statistics.
• These methods are critical for applications in quantum optics, LiDAR imaging, and optical communications, enhancing measurement fidelity under diverse operating conditions.

A single-photon avalanche diode (SPAD) exhibits a dead time following each detected avalanche event, during which it is insensitive to subsequent photon arrivals. At moderate to high count rates, this dead time introduces significant nonlinearity and statistical distortion into the observed photon counts, precluding direct inference of the underlying photon flux or true event statistics. SPAD dead-time linearization models are quantitative frameworks and algorithmic procedures that invert this nonlinear transformation—restoring the linear relationship between photon flux and observed output, and recovering the true distribution of input photonic events from SPAD measurements. These models underpin high-fidelity data extraction in quantum optics, LiDAR, quantum communications, and optical sensing, ensuring statistical accuracy across a broad range of illumination intensities and signal conditions.

1. Mathematical Formulation of SPAD Dead-Time Effects

In the standard non-paralyzable dead time model, let $T$ denote the counting window and $\tau_d$ the fixed dead time per detection. The probability to record $n$ events in $[0, T]$ for an ideal, zero-dead-time SPAD is $P_{\rm true}(n)$. For a real SPAD with dead time, the recorded probability is $P_{\rm meas}(n)\equiv p(n,T,\tau_d)$.

The relationship between these distributions, known as the Mandel–Srinivas inversion, is

$$\begin{aligned} P_{\rm true}(0) &= P_{\rm meas}(0,T,\tau_d) \ P_{\rm true}(n) &= P_{\rm meas}\bigl(n,\,T+n\,\tau_d,\,\tau_d\bigr) + \sum_{r=0}^{n-1} \left[P_{\rm meas}\bigl(r,\,T+n\,\tau_d,\,\tau_d\bigr) - P_{\rm meas}\bigl(r,\,T+(n-1)\,\tau_d,\,\tau_d\bigr)\right], \quad n\ge1 \end{aligned}$$

where $P_{\rm meas}(n, T+k\tau_d, \tau_d)$ is estimated by histogramming the experimental data in a window of length $T+k\tau_d$ (Menkart et al., 2022).

An iterative procedure updates estimates of $P_{\rm true}(n)$ using measured histograms at expanded window lengths, converging in one pass because the update is not history-dependent.

For small $\tau_d$0, the kernel $\tau_d$1 allows a linear convolution approximation: $\tau_d$2 with

$\tau_d$3

yielding a closed-form first-order correction for $\tau_d$4 (Menkart et al., 2022).

2. Physical Assumptions and Model Regimes

SPAD dead-time linearization models operate under the following principal assumptions:

• Non-paralyzable dead time: Photons impinging during $\tau_d$5 are ignored, and the dead time interval is not extended (active quenching architecture).
• Negligible afterpulsing: For afterpulsing to be negligible, $\tau_d$6 must exceed relevant afterpulsing timescales (e.g., $\tau_d$7 in InGaAs SPADs).
• Sufficiently large counting windows: $\tau_d$8, where $\tau_d$9 is the largest observed count.
• Poissonian input statistics: Incident photon arrivals are Poisson-distributed unless stated otherwise; deviations due to the SPAD are entirely due to dead time.

The model is robust for $n$0 to maintain low pile-up and reliable statistics. Parameter choices in practice span from $n$1s (high-speed measurement) up to $n$2s (afterpulsing suppression), and window $n$3 must be ample enough to encapsulate observed event numbers (Menkart et al., 2022).

3. Algorithmic Implementation and Practical Linearization

The operational workflow for SPAD dead-time linearization consists of:

1. Data Collection: Acquire a timestamped event list under specified $n$4 and window $n$5.
2. Probability Estimation: For all $n$6, compute $n$7 via sliding-window histograms.
3. Distribution Inversion: Apply the Mandel–Srinivas recursion to recover $n$8 for all $n$9.
4. Normalization and Results: Optionally renormalize so $[0, T]$0, and derive statistics such as the corrected mean $[0, T]$1.

This algorithm restores Poissonian photon count statistics in measured data, as demonstrated with InGaAs SPADs detecting both dark counts and laser-illuminated signals. The dead-time corrected distributions collapse onto the appropriate Poissonian with high fidelity; raw data exhibiting clear sub-Poissonian narrowing revert to Poisson after correction (Menkart et al., 2022).

4. Extensions to Advanced SPAD Architectures and Correction Models

For SPADs exhibiting non-instantaneous, exponential recovery of quantum efficiency, the linearization framework generalizes by parameterizing the time-dependent efficiency $[0, T]$2 with a single-pole exponential time constant $[0, T]$3: $[0, T]$4 The mean steady-state count rate is given in closed-form by

$[0, T]$5

where $[0, T]$6 is the exponential integral. The true photon flux $[0, T]$7 is then analytically inverted using the principal branch of the Lambert $[0, T]$8 function: $[0, T]$9 This model remains accurate through the saturation and paralyzable regimes, provided that higher-order effects and non-exponential recoveries are negligible. All detector parameters $P_{\rm true}(n)$0 can be extracted from a single inter-detection histogram (Krause et al., 14 Jul 2025).

In passive-quenching SPAD arrays, the mean detected count per symbol interval $P_{\rm true}(n)$1 from incident rate $P_{\rm true}(n)$2 is $P_{\rm true}(n)$3, highlighting the exponential roll-off. The inverse mapping for flux pre-distortion uses the principal branch of the Lambert $P_{\rm true}(n)$4 function, enabling linearization of receiver statistics in communication systems (Huang et al., 2021).

5. Application Domains and Empirical Performance

SPAD dead-time linearization is critical for:

• Photon-counting statistics extraction: Correction of dark count and signal statistics to characterize novel light sources, quantum measurements, and entropy rates (Menkart et al., 2022).
• High-dynamic-range and high-flux regimes: Extension of SPAD usable flux range well beyond classical $P_{\rm true}(n)$5 saturation via analytic or numerical inversion models (Krause et al., 14 Jul 2025, Krause et al., 14 Jul 2025).
• LiDAR and dToF imaging: Accurate inversion of pile-up distorted histograms in time-of-flight measurements; histogram-less 3D imaging via linearized timestamp averaging, capable of operating at detection rates up to 90%, far beyond the classical 5% pile-up limit (Tontini et al., 2023).
• Optical wireless communication: Transmitter-side pre-distortion and receiver-side noise normalization to mitigate nonlinear distortion and signal-dependent noise, restoring conventional AWGN communication theoretic assumptions (Huang et al., 2021).

Empirical results confirm that dead-time corrected probability distributions attain Poissonian forms and mean rates closely matching true photon flux, as in the Menkart et al. study at various detection efficiencies. Correction algorithms consistently overlap corrected data with the appropriate Poisson envelope, both in simulation and experiment (Menkart et al., 2022, Tontini et al., 2023).

6. Theoretical Limits, Parameter Optimization, and Practical Constraints

The precision limits of SPAD-based measurements, especially in direct ToF LiDAR, are governed by the pile-up distortion induced by dead time. The Cramér–Rao lower bound (CRLB) for the retrieval of parameters (e.g., time of flight $P_{\rm true}(n)$6, photon flux $P_{\rm true}(n)$7) explicitly incorporates the dead-time-induced coupling through the Fisher information matrix: $P_{\rm true}(n)$8 with $P_{\rm true}(n)$9 encoding the pile-up correction per observation bin. The CRLB tightens for optimized signal flux (roughly $P_{\rm meas}(n)\equiv p(n,T,\tau_d)$0), pulse width ($P_{\rm meas}(n)\equiv p(n,T,\tau_d)$1), and counting bin ($P_{\rm meas}(n)\equiv p(n,T,\tau_d)$2). Excess flux induces heavy pile-up and strong parameter coupling, deleteriously impacting variance (Wu et al., 15 Jul 2025).

Model validity is limited when afterpulsing dominates, fluxes drive multi-trap effects, or SPAD arrays exhibit strong crosstalk. Accurate correction is further contingent on precision in $P_{\rm meas}(n)\equiv p(n,T,\tau_d)$3, window sizing, and mean count extraction.

7. Impact, Related Models, and Future Directions

SPAD dead-time linearization enables high-fidelity quantum optics, ranging, and communication measurements by restoring the statistical linearity between input photon flux and observed statistics. Methodologies generalized to non-instantaneous exponential recovery, paralyzable regimes, and large-array operation further broaden practical relevance. Empirical and theoretical validation consistently demonstrate restored Poisson statistics and range precision that approach the CRLB for a broad range of instrument configurations and environmental conditions (Menkart et al., 2022, Krause et al., 14 Jul 2025, Wu et al., 15 Jul 2025, Tontini et al., 2023).

Ongoing challenges involve robust modeling of multi-time-constant recovery, afterpulsing, non-stationary backgrounds, and high-crosstalk arrays. Advances in analytic inversion for complex, non-Markovian dead time processes and integration into real-time firmware architectures remain key future research directions.