Characterization of Deconvolution-Based PMT Waveform Reconstruction Under Large Charge Dynamic Range and Varying Scintillation Time Profiles

This study validates the reliability of a deconvolution-based algorithm for reconstructing photomultiplier tube waveforms across a large charge dynamic range (0–200 photoelectrons) and varying scintillation time profiles, demonstrating stable performance with approximately 1% charge non-linearity and the capability to handle muon-induced large signals.

The Gist

The Big Picture: Listening to the Ghosts of Particles

Imagine you are trying to listen to a whisper in a very noisy room. Now, imagine that "whisper" is actually a tiny flash of light created when a subatomic particle (like a neutrino) hits a giant tank of liquid.

Scientists use special "ears" called Photomultiplier Tubes (PMTs) to catch these flashes. But here's the problem: the signal the PMT hears isn't a clean, perfect note. It's a messy sound that echoes, rings, and sometimes even dips below silence (called an "undershoot") before settling down.

If you try to measure the volume of that sound just by listening to the messy echo, you'll get the wrong answer. You might think the particle was huge when it was tiny, or vice versa. This paper is about teaching the computer how to "clean up" that messy sound so scientists can know exactly how big the particle was and when it happened.

The Problem: The "Echo Chamber" Effect

Think of a PMT signal like shouting in a canyon.

1. The Shout: The particle hits, creating a flash of light.
2. The Echo: The PMT hears the shout, but then the sound bounces around the canyon walls (this is the "undershoot" or "ringing").
3. The Mess: If you shout again before the echo from the first shout dies down, the sounds mix together. This is called pile-up.

In big experiments (like the JUNO experiment mentioned in the paper), scientists need to measure everything from a tiny "whisper" (a few flashes of light) to a massive "roar" (thousands of flashes from a cosmic ray muon). The challenge is that the "echo" behaves differently depending on:

• How loud the shout was (the charge).
• What kind of liquid the sound traveled through (different scintillators have different "ringing" times).
• Whether the sound came from a single point or a long line (like a train passing by).

The Solution: The "Noise-Canceling" Algorithm

The authors propose a method called Deconvolution.

The Analogy: Imagine you are trying to hear a song, but someone is playing a distorted, echoey version of it through a bad speaker.

• Old Method: You try to guess the original song by just turning up the volume or squinting at the waveform. This often leads to mistakes.
• The New Method (Deconvolution): You have a recording of exactly how that specific bad speaker distorts sound. You use a mathematical "reverse filter" (like a super-powered noise-canceling headphone) to subtract the distortion. Suddenly, the original, clean song appears.

In this paper, the "bad speaker" is the PMT's natural tendency to ring and undershoot. The "reverse filter" is a mathematical algorithm using something called Fourier Transforms (which is just a fancy way of breaking a sound wave down into its individual musical notes to fix them one by one).

What Did They Test?

The researchers wanted to make sure their "noise-canceling" algorithm works in the real world, not just in a perfect lab. They tested it in three tough scenarios:

1. The Volume Test (Dynamic Range):

   • The Challenge: Can the algorithm handle a whisper (0 flashes) and a scream (200+ flashes) equally well?
   • The Result: Yes. Whether the signal was tiny or huge, the algorithm kept the measurement error below 1%. It's like a microphone that works perfectly whether you are whispering or screaming.

2. The Liquid Test (Scintillation Profiles):

   • The Challenge: Different liquids "ring" for different amounts of time. Some stop ringing quickly; others take a long time to settle.
   • The Result: The algorithm didn't care. It worked just as well on fast-ringing liquids as it did on slow-ringing ones. It's like a translator who speaks every dialect perfectly, regardless of how fast the speaker talks.

3. The "Train" Test (Muon Events):

   • The Challenge: Sometimes, a cosmic ray (a muon) flies right through the detector like a train. This creates a massive, long signal that might not even finish "ringing" before the recording stops.
   • The Result: This was the hardest test. When the signal was so big that the "echo" hadn't finished by the time the recording stopped, the measurements got a little fuzzy.
   • The Fix: They found that if you just record for a little longer (extending the time window), the echo settles down, and the algorithm works perfectly again.

The Takeaway

This paper is essentially a "stress test" for a new way of cleaning up particle signals.

• Before: Scientists had to worry that their measurements might be wrong if the signal was too big, too small, or if the liquid changed.
• Now: They have a robust "digital filter" that cleans up the signal reliably, no matter the size of the particle or the type of liquid.

In short: They built a mathematical "de-echo" machine that allows scientists to hear the true voice of the universe's smallest particles, even when they are screaming in a noisy, echoey room. This is crucial for experiments trying to solve mysteries like dark matter or how the sun works.

Technical Summary

1. Problem Statement

Photomultiplier tubes (PMTs) are critical sensors in neutrino and dark matter experiments (e.g., JUNO, Daya Bay). Accurate event reconstruction relies on extracting precise charge and timing information from PMT waveforms. However, current reconstruction methods face significant challenges in large-scale liquid scintillator detectors:

• Large Charge Dynamic Range: Detectors must handle signals ranging from single photoelectrons (PE) to thousands of PEs (e.g., from muons or atmospheric neutrinos).
• Complex Waveform Features: PMT waveforms often exhibit "undershoot" (a negative dip following the main peak) and reflections. In high-charge scenarios, these features can prevent the waveform from returning to the baseline within the sampling window, leading to spectral leakage and reconstruction bias.
• Varying Scintillation Time Profiles: Different liquid scintillator formulations and incident particle types (e.g., $\alpha$ vs. $\gamma$) produce different scintillation decay time profiles. Algorithms must remain stable regardless of these variations.
• Track-like Events: Unlike point-like events, through-going muons create extended energy deposition tracks, resulting in complex, overlapping photon arrival times and significantly larger charge accumulations that challenge baseline recovery.

Existing methods like simple charge integration or CR-(RC)4 filters often introduce non-linearity or biases when handling undershoots and large signals. While deconvolution techniques (introduced by Daya Bay and refined by JUNO) show promise, their stability across these specific extreme conditions requires rigorous validation.

2. Methodology

The authors employed a comprehensive simulation and analysis framework to test a deconvolution-based reconstruction algorithm.

A. Simulation Framework:

• Waveform Generation: Simulated PMT waveforms were generated by convolving a photoelectron hit sequence $u(t)$ with a Single Photoelectron (SPE) response template $r(t)$ and adding electronic noise.
• SPE Templates: Three SPE templates were created with varying undershoot amplitudes relative to the main peak: 1.3%, 6.5%, and 13%.
• Scintillation Profiles: Eight distinct scintillation time profiles were used as inputs, representing various liquid scintillator formulations (e.g., LAB/PPO, PXE/PPO) and excitation types ($\gamma$, $\alpha$, $e^-$). These profiles were modeled using multi-exponential decay functions.
• Event Types:
  • Point-like Events: Simulated with charges distributed from 0 to 200 PEs.
  • Track-like Events: Simulated using a Geant4 model of a 15m radius spherical liquid scintillator detector with 10,650 PMTs. A 200 GeV muon traversed the detector, generating waveforms with charges ranging from 200 to 10,000 PEs.

B. Reconstruction Algorithm:

• Core Technique: Fast Fourier Transform (FFT) based deconvolution.
  • The raw waveform is transformed to the frequency domain.
  • A low-pass filter suppresses high-frequency noise.
  • Deconvolution is performed using a calibrated SPE response filter to remove undershoot and shape features.
  • Inverse FFT (IFFT) returns the signal to the time domain, yielding the reconstructed photoelectron hit sequence.
• Enhancements: The Short-Time Fourier Transform (STFT) was optionally applied to improve pile-up identification for point-like events. For large muon signals, STFT was omitted to reduce computational cost as pile-up identification became less critical compared to charge integration.
• Baseline Handling: The study specifically analyzed the "baseline recovery status" at the end of the 1000 ns sampling window ($t=1000$ ns). A threshold of amplitude deviation $< -0.8$ a.u. was defined as the point where reconstruction quality degrades significantly.

3. Key Contributions

• Systematic Validation of Dynamic Range: The study provides the first comprehensive characterization of deconvolution algorithms across a massive dynamic range (0–10,000 PEs), specifically addressing the transition from low-energy neutrino signals to high-energy muon tracks.
• Scintillation Profile Independence: It rigorously demonstrates that the algorithm's performance is independent of the specific scintillation time profile or particle type, a crucial requirement for detectors using mixed scintillator formulations or performing particle identification.
• Baseline Recovery Analysis: The paper identifies the specific conditions under which large undershoots and high charges cause baseline recovery failure within standard sampling windows (1000 ns) and proposes a quantitative metric ($Amplitude_{t=1000ns}$) to predict reconstruction failure.
• Mitigation Strategy: It demonstrates that extending the sampling window (e.g., from 1000 ns to 2000 ns) effectively mitigates reconstruction errors caused by incomplete baseline recovery in large signals.

4. Key Results

A. Point-like Events (0–200 PEs):

• Linearity: The deconvolution algorithm achieved a residual non-linearity of < 1% across the entire 0–200 PE range.
• Robustness: This performance was consistent across all three undershoot configurations (1.3% to 13%) and all eight scintillation time profiles.
• STFT Impact: The inclusion of STFT did not significantly alter the charge reconstruction accuracy for point-like events, confirming the robustness of the core FFT deconvolution.

B. Track-like Events (Muon-induced, up to 10,000 PEs):

• Baseline Recovery Issues: For muon events with high charge (hundreds to thousands of PEs) and large undershoots (13%), a significant fraction of waveforms (up to 82.6% in the 13% case) failed to return to the baseline by 1000 ns.
• Performance with Selection: When selecting only waveforms where the baseline deviation was within acceptable limits ($Amplitude \ge -0.8$ a.u.), the residual non-linearity remained < 2%.
• Window Extension: For waveforms with poor baseline recovery, extending the sampling window to 2000 ns restored the reconstruction accuracy, bringing the residual non-linearity back down to < 1% for charges up to 1000 PEs.
• Undershoot Sensitivity: The algorithm maintained stability even with 13% undershoot, provided the sampling window was sufficient to capture the full decay.

5. Significance

This work validates the deconvolution-based reconstruction method as a robust solution for next-generation large-scale liquid scintillator detectors like JUNO.

• Reliability: It confirms that the method can handle the extreme dynamic ranges required for diverse physics goals (from reactor neutrinos to supernova bursts and atmospheric neutrinos) without introducing significant non-linearity.
• Operational Guidance: The study provides practical guidelines for detector operation, specifically regarding the trade-off between sampling window length and data volume. It suggests that for high-energy muon events, extending the sampling window or implementing dynamic baseline correction is necessary to maintain precision.
• Future Applications: The findings support the use of this algorithm for precise energy reconstruction and vertex determination in complex environments, ensuring that variations in scintillator properties or particle types do not compromise data quality.