The Effects of Instrumental Deadtime on NICER Timing Products

Technical Summary: The Effects of Instrumental Deadtime on NICER Timing Products

Problem Statement
The Neutron Star Interior Composition Explorer (NICER) X-ray Timing Instrument (XTI) offers unprecedented timing resolution (~300 ns) and effective area, surpassing previous missions like RXTE and XMM-Newton. However, high count rates in X-ray binaries introduce instrumental deadtime, a period during which the detector cannot process new events. While previous missions (e.g., RXTE, NuSTAR) required specific frequency-dependent corrections to power spectra to account for deadtime-induced Poisson noise distortions, it remains unclear whether these established methodologies are necessary or appropriate for NICER's multi-detector architecture. Specifically, the authors investigate whether the high count rates observed in NICER data necessitate correcting lightcurves for deadtime prior to Fourier analysis, or if the standard approach of subtracting frequency-dependent noise from uncorrected lightcurves remains sufficient.

Methodology
The authors analyzed nine observations of the X-ray binary GX 339-4, selected to represent low, medium, and high count rate regimes (ranging from ~31 to ~7000 counts/s). The study utilized standard NICER processing pipelines to generate Good Time Intervals (GTIs), filtering out telemetry saturation, South Atlantic Anomaly passages, and background events (undershoot, overshoot, forced triggers).

The analysis proceeded in three stages:

1. Detector-Level Deadtime Characterization: The authors examined deadtime on a per-Focal Plane Module (FPM) basis. They calculated the average deadtime per recorded event and per X-ray event (post-screening) across the 52 active detectors. They tested the hypothesis that deadtime is consistent across detectors by fitting a constant value and evaluating the reduced chi-square statistic.
2. Lightcurve Correction: A custom correction was applied to the lightcurves. For each time bin, the total deadtime was summed based on event-by-event records, accounting for events where deadtime overlapped bin boundaries. The observed count rate ($r_{Obs}$) was corrected to a "true" rate ($r_{True}$) using the formula $r_{True} = r_{Obs} / (1 - \tau_{Dead,prop})$, where $\tau_{Dead,prop}$ is the proportion of the time bin lost to deadtime.
3. Fourier Domain Comparison: Power spectra were generated for both the raw (uncorrected) lightcurves and the deadtime-corrected lightcurves.
   • For the uncorrected lightcurves, the authors subtracted a frequency-dependent Poisson noise level defined by the Zhang et al. (1995) amendment to the van der Klis (1989) formalism (Equation 5 in the paper).
   • For the corrected lightcurves, a flat Poisson noise level ($2/r_e$) was subtracted.
   • The resulting power spectra and noise levels were compared across the nine observations to identify systematic deviations.

Key Results

• Detector Consistency: Across all nine observations, the average deadtime per recorded event was found to be roughly constant at ~22 $\mu$s across all functional FPMs. Statistical fitting confirmed that the deadtime distribution across detectors is consistent with a constant value (reduced chi-square $< 0.1$ for most observations), indicating no single detector suffers from a significantly higher proportion of large deadtime events.
• Proportional Deadtime: The fractional deadtime in lightcurve bins increases linearly with the observed count rate. Even at the highest count rates analyzed (~7000 counts/s), the proportional deadtime remained below 0.6% (maximum observed ~0.006). The authors note that detector deadtime would likely only reach significant levels (e.g., 1%) at count rates approaching ~23,000 counts/s, a regime where telemetric deadtime (buffer saturation) would already fragment the observation.
• Power Spectrum Comparison: When comparing the power spectra derived from deadtime-corrected lightcurves against those derived from uncorrected lightcurves with frequency-dependent noise subtraction, the differences were found to be minor.
  • At frequencies $f \lesssim 1$ Hz, the spectra were nearly identical.
  • At higher frequencies, deviations emerged but generally remained within the error bars of the rebinned spectra.
  • The deviations were negative, meaning the power from the deadtime-corrected lightcurve was slightly higher than the frequency-dependent corrected spectrum, but the magnitude of this difference was not statistically significant for the count rates considered.
• Noise Modeling: The study found that for NICER, the frequency-dependent noise terms (cosine and sine dependent terms in Eq. 5) are significantly smaller than for RXTE due to the larger number of detectors (50 vs. 5) and lower deadtime for large events. Consequently, the "ideal" flat Poisson noise ($2/r_e$) is a closer approximation to the actual noise floor for NICER than for previous instruments.

Significance and Claims
The paper concludes that for the count rates currently accessible to NICER (up to ~7000 counts/s), the effects of instrumental detector deadtime on timing and Fourier-domain products are not significant enough to warrant complex corrections to the lightcurves prior to analysis.

The authors assert that:

1. Standard Practices Remain Valid: The established approaches to handling deadtime, as applied to RXTE and other detectors (specifically subtracting frequency-dependent noise from uncorrected lightcurves), are still appropriate for NICER data.
2. No Necessity for Pre-correction: Performing deadtime corrections on lightcurves before creating Fourier products is not necessary at the count rates considered in this analysis. The computational cost and data storage requirements of processing uncalibrated event files to create deadtime-corrected lightcurves do not yield a material benefit in the resulting timing products.
3. Detector Uniformity: There are no significant inconsistencies across the functional detectors in the instrument, validating the use of ensemble data without individual detector deadtime weighting.

The study provides confidence that NICER data can be processed using standard pipelines to produce Fourier-domain products comparable to those obtained from historical instruments, facilitating the long-term monitoring of X-ray binary evolution. The authors also note that while machine learning approaches to infer underlying spectral shapes are emerging, the current analysis supports the continued use of traditional methods for NICER timing analysis.