REVIEW 3 major objections 4 minor 17 references
Predictive model of persistence in H2RG detectors
T0 review · 3 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash
Pith's one-line read The paper claims a five-pole digital filter can predict image persistence in infrared detectors for arbitrary exposure histories.
desk verdict A genuinely useful IIR-filter model for H2RG persistence, but the 'arbitrary exposure histories' claim is broader than the validation supports. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the five-pole infinite impulse response (IIR) filter: a recursive digital filter in which each output sample is a weighted combination of the current input and the previous output, equivalent to an $RC$ network with an exponential memory. One pole is assigned to each measured trap time constant, $\tau = 1, 10, 100, 1000, 10000$ s. The filter implements the symbolic trap model of a capacitor charged by a current source proportional to photo-charge and discharged through a resistor, so the state variable is the trapped charge in each bin. The other required components are a per-pixel persistence map giving the relative total trap number and a five-element trap-density vector $\rho(\tau_i)$ giving the maximum fraction of photo-charge trapped at equilibrium; the paper assumes the time-constant distribution is uniform across the array and only the total trap number varies.
What would settle it
One decisive test would be to compare two exposure histories that deliver the same total fluence in different temporal patterns (for example, one long soak versus many short flashes). A linear time-invariant filter predicts identical trapped-charge evolution whenever the convolved input is identical, so any dependence of the persistence on the temporal order beyond the filter's prediction would show that trapping current depends on the trapped charge. A second, simpler check is to drive a pixel above the nominal full well and measure whether the filter's predicted persistence diverges from the data as leakage current becomes significant.
Extended reading notes
Core claim
The central discovery is that persistence in H2RG HgCdTe detectors can be modelled as the linear response of a bank of five first-order traps, and that the same model quantitatively accounts for reciprocity failure. In the model each pixel's trap population evolves by balancing a trapping current $\rho(\tau_i) E(t)/\tau_i$ proportional to the current photo-charge $E(t)$ against a detrapping current $Q(t)/\tau_i$ proportional to the charge already trapped, integrated over time. The authors demonstrate that the charge-up and detrapping time constants are equal, so one set of exponentials describes both the appearance and decay of persistence, and that below full well the trapped fraction is linear in exposure and independent of signal level. Fitted to three detectors (one 2.5 $\mu$m and two 5.3 $\mu$m cutoff), the model reproduces measured persistence after single long soaks, after six and forty 600 s up-the-ramp exposures, and in pixel-by-pixel corrected image sequences without noticeably increasing noise.
Load-bearing premise
The model's linearity rests on the assumption that trapping current is proportional to the instantaneous photo-charge and independent of how much charge is already trapped; the authors verify only indirectly that charge-up and detrapping time constants match, and they observe leakage currents that grow as the pixel approaches full well, where the linear behaviour is expected to break down.
Editorial extensions
If this is right
- An instrument running the filter can attach a predicted persistence frame to every exposure automatically, with no human request, because the trapped-charge ledger is always up to date.
- The same filter output gives a reciprocity-failure estimate during the exposure, so one engine corrects both persistence and count-rate nonlinearity.
- Because the filter is recursive and needs only the five bin states per pixel, it runs in a fraction of the frame time on ordinary hardware.
- Calibration of a detector reduces to measuring one time-constant vector and one per-pixel trap map; the same vector can be reused at fixed temperature and bias.
Reading between the lines
- A consequence the paper leaves implicit: because the model is linear and time-invariant, the full persistence behaviour is characterised by the filter's impulse response; measuring detrapping after a single short flash at several soak times is sufficient to predict any exposure sequence, so a very compact calibration standard could be defined.
- The paper does not model trapped holes, saturated-pixel spillover, or gaps in the exposure record; extending the filter with a state-dependent trapping term would be a natural next step and would be testable near full well.
- If the engine is deployed at the data-acquisition computer, archived raw frames plus the persistent ledger would let the correction be recomputed after any future model improvement, making the persistence history a standard data product.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper presents a trap-based model for image persistence in Teledyne H2RG HgCdTe detectors. The authors characterize trapping and detrapping currents in three H2RG detectors, represent the trap ensemble as five exponential time-constant bins, and construct a 5-pole Infinite Impulse Response (IIR) digital filter that tracks trapped charge per pixel as a function of exposure history. They further claim that the same trapping/detrapping currents explain reciprocity failure. Validation is performed on the same three detectors using below-full-well LED exposure sequences, and the paper argues that the model can provide near real-time persistence correction for arbitrary exposure histories.
Significance. If the model's scope claims are supported, this would be a valuable practical contribution: a simple, computationally efficient recursive filter that provides per-pixel persistence maps in near real time, improving on existing phenomenological power-law corrections (e.g., WFC3). The physical grounding in depletion-region trap dynamics, the per-pixel trap-density maps, and the characterization of three different H2RG devices are genuine strengths. However, the validation is narrow—conducted on the same detectors used for fitting, with no held-out device, no quantitative residuals, and no saturated or gapped exposure histories—so the headline claim of 'arbitrary exposure histories' goes well beyond what the current data demonstrate.
major comments (3)
- [Abstract and Sec. 7] The claim that the model handles 'arbitrary exposure histories' is contradicted by the paper's own stated limitations. Sec. 7 explicitly concedes that 'exposures that reach full-well within a small fraction of the exposure time will cause errors,' and Sec. 3.9 reports a secondary leakage current near full-well, of opposite sign and many times dark current, that is not included in the model. Since saturation and high-fill exposures are exactly the regime where persistence is most consequential in astronomy, the abstract's wording materially overstates the model's valid domain. The claim should be restricted to below-full-well, gapless exposure histories, and the manuscript should state this limitation in the abstract and conclusions.
- [Sec. 5 and Table 1] The model is validated only on the same three detectors from which the trap density vector ρ_i (Table 1) and per-pixel persistence maps (Sec. 3.7) were fitted. While the validation sequences are different exposure histories and therefore not a simple re-fit of the same curve, all predictions share the fitted parameters from those detectors. No independent detector is held out, no cross-validation is performed, and the agreement in Figs. 21–25 is presented without residuals, uncertainties, or quantitative goodness-of-fit metrics. This limits confidence that the model generalizes to other H2RG devices or to operating conditions outside the characterized range. The authors should either provide a held-out detector test or clearly frame the results as a demonstration of the method's internal consistency rather than its predictive generality.
- [Sec. 3.9] The reciprocity-failure claim is based on a single up-step/down-step comparison on one ERIS 5.3 μm detector (Fig. 17), after subtracting an assumed constant leakage current (Fig. 18). The statement that trapping currents 'must be a major contributor to Reciprocity Failure' is an extrapolation from one corrected dataset and does not quantitatively connect the measured trap parameters to the magnitude of reciprocity failure across the three detectors or over a range of flux levels. Given that Ref. 5 (Biesiadzinski et al.) finds traps cannot fully explain reciprocity failure, the manuscript should present a more quantitative test, such as comparing model-predicted signal loss with photometric measurements at several flux levels, before claiming that the same trap model explains both persistence and reciprocity failure.
minor comments (4)
- [Eq. 5] Equation (5) uses 'Q(t − dt)' inside an integral, which is notationally ambiguous; it should be expressed as a discrete update or as a previous-state variable in a recursive filter, matching the IIR description that follows.
- [Figs. 10 and 17] Several figures lack clear axis labels or units. In particular, Fig. 10 would benefit from explicit labels of 'soak time' and 'trap charge fraction', and Fig. 17 should identify the color codes in the caption or a legend, since the text refers to colors that are not visible in a monochrome print.
- [Table 1] Table 1 reports trap-density fractions ρ_i to two significant figures without uncertainties or fit-quality information; adding error estimates from the exponential fits would help the reader judge the significance of the per-detector and per-time-constant differences.
- [Sec. 5] The stability constraint that the frame time must be less than the smallest trap time constant, leading to removal of the τ = 1 s filter, is a practical limitation worth stating explicitly in the model description (Sec. 4) rather than only in the validation test.
Circularity Check
No significant circularity; the IIR persistence model is calibrated from characterization data and validated on distinct exposure histories, with only scope limitations explicitly acknowledged.
full rationale
The derivation is self-contained in the sense required by the circularity check. The trap parameters rho_i are measured quantities (Sec. 3.2, Table 1) obtained by fitting detrapping profiles to a fixed set of time-constant bins; they are not defined in terms of the model output. The dynamical model, Eq. 5, is a numerical integral of the stated trapping/detrapping ODEs (Eqs. 2-4), so the IIR filter is an implementation of the assumed linear dynamics rather than a restatement of the fit. The validation in Sec. 5 uses distinct exposure histories (6 and 40 up-the-ramp sequences, plus pixel-by-pixel movies) that were not the data used to set rho_i, so the agreement is not forced by construction. The first 8192 s soak test is closer to the characterization data, but it is not the basis of the central claim. The paper does contain explicit scope limitations: Sec. 4 requires complete exposure records, and Sec. 7 concedes that full-well/partial-frame exposures 'will cause errors' and that gaps reduce accuracy. Sec. 4 also notes that long-term stability of the trap-density vector has not yet been tested. These undercut the abstract's 'arbitrary exposure histories' but are external validity limitations, not circular derivation. The only self-citation (Ref. 16) is used for an incidental observation about cooling and persistence and is not load-bearing. No step reduces by construction to its inputs.
Assumptions & free parameters
free parameters (3)
- Trap density vector rho_i (i=1..5) =
Table 1: e.g., CRIRES+ 5.3um: 1.4e-3, 1.5e-3, 1.7e-3, 1.8e-3, 1.7e-3
- Time constant bins tau_i =
1, 10, 100, 1000, 10000 s
- Per-pixel persistence map =
Spatial map normalized to mean 1 in window
assumptions (5)
- domain assumption Trapping current is proportional to instantaneous photo-charge and independent of already trapped charge (Eq. 4).
- domain assumption Detrapping current is proportional to trapped charge (Eq. 3).
- domain assumption The distribution of trap time constants is uniform across the detector; only trap number varies per pixel (Sec. 3.7).
- ad hoc to paper Trap behavior is a linear superposition of five independent exponential time-constant bins (Eq. 5).
- domain assumption Charge movement within the depletion region creates fractional effective charge (Sec. 2, Fig. 2).
Cite this review
Pith. "Pith review of Predictive model of persistence in H2RG detectors." pith.science (2026). https://pith.science/paper/LO6K2Z5O
@misc{pith2026190806469,
author = {Pith},
title = {Pith review of: Predictive model of persistence in H2RG detectors},
year = {2026},
howpublished = {\url{https://pith.science/paper/LO6K2Z5O}},
note = {Machine review of arXiv:1908.06469}
}
abstract
Infrared hybridized detectors are widely used in astronomy, and their performance can be degraded by image persistence. This results in remnant images that can persist in the detector for many hours, contaminating any subsequent low-background observations. A different but related problem is reciprocity failure whereby the detector is less sensitive to low flux observations. It is demonstrated that both of these problems can be explained by trapping and detrapping currents that move charge back and forward across the depletion region boundary of the photodiodes within each pixel. These traps have been characterized in one 2.5 $\mu$m and two 5.3 $\mu$m cutoff wavelength Teledyne H2RG detectors. We have developed a behaviour model of these traps using a 5-pole Infinite Impulse Response digital filter. This model allows the trapped charge in a detector to be constantly calculated for arbitrary exposure histories, providing a near real-time correction for image persistence.
Reference graph
Works this paper leans on
-
[1]
R. M. Smith and M. Zavodny , `` A theory for image persistence in H g C d T e photodiodes ,'' Proc. SPIE 7021 (2008)
work page 2008
-
[2]
R. M. Smith and M. Zavodny , `` Calibration of image persistence in H g C d T e photodiodes ,'' Proc. SPIE 7021 (2008)
work page 2008
-
[3]
R. E. Anderson and M. Regan , ``Understanding persistence: A 3 D trap map of an H2RG imaging sensor,'' Proceedings of the Scientific Detector Workshop 2013. 7-11 October 2013 Florence, Italy (2013)
work page 2013
- [4]
-
[5]
T. Biesiadzinski , W. Lorenzon , R. Newman , et al. , `` Reciprocity Failure in HgCdTe Detectors: Measurements and Mitigation ,'' PASP 123 , 958 (2011)
work page 2011
- [6]
-
[7]
J. M. Leisenring , M. Rieke , K. Misselt , et al. , `` Characterizing persistence in JWST NIRCam flight detectors ,'' in High Energy, Optical, and Infrared Detectors for Astronomy VII , Proc. SPIE 9915 , 99152N (2016)
work page 2016
-
[8]
M. R. Baril and L. Albert , `` Characterization of persistence in WIRCam's Hawaii 2-RG arrays ,'' in High Energy, Optical, and Infrared Detectors for Astronomy III , Proc. SPIE 7021 , 702121 (2008)
work page 2008
Show all 17 references
-
[9]
K. S. Long , S. M. Baggett , J. W. MacKenty , et al. , `` Characterizing persistence in the IR detector within the Wide Field Camera 3 instrument on the Hubble Space Telescope ,'' in Space Telescopes and Instrumentation 2012: Optical, Infrared, and Millimeter Wave , Proc. SPIE...
2012
-
[10]
Crouzet , L
P.-E. Crouzet , L. Duvet , P. Strada , et al. , `` Comparison of persistence in spot versus flat field illumination and single pixel response on a Euclid HAWAII-2RG at ESTEC ,'' in High Energy, Optical, and Infrared Detectors for Astronomy VII , Proc. SPIE 9915 , 99151E (2016)
2016
-
[11]
Follert , R
R. Follert , R. J. Dorn , and E. Oliva , `` CRIRES+: a cross-dispersed high-resolution infrared spectrograph for the ESO VLT ,'' Proc. SPIE 9147 (2014)
2014
-
[12]
Amico , F
P. Amico , F. Marchetti , and A. Pedichini , `` The design of ERIS for the VLT ,'' Proc. SPIE 8446 (2012)
2012
-
[13]
Serra and J.-C
B. Serra and J.-C. Secroun , `` Characterization of Euclid-like H2RG IR detectors for the NISP instrument ,'' Proc. SPIE 9602 (2015)
2015
-
[14]
J. R. Janesick, Scientific Charge-Coupled Devices , SPIE Press, PO Box 10, Bellingham,WA (2001)
2001
-
[15]
R. G. Lyons, Understanding Digital Signal Processing , Addison-Wesley Longman Publishing Co., Inc., Boston, MA, USA, 1st ed. (1996)
1996
-
[16]
S. Tulloch and the ESO Detector Systems Group, ``Persistence C haracterisation of T eledyne H2RG detectors,'' Proceedings of the Scientific Detector Workshop, Space Telescope Science Institute, Baltimore, MD, USA (2017)
2017
-
[17]
write newline
" write newline "" before.all 'output.state := FUNCTION blank.sep after.quote 'output.state := FUNCTION fin.entry output.state after.quoted.block = 'skip 'add.period if write newline FUNCTION new.block output.state before.all = 'skip output.state after.quote = after.quoted.blo...
Reviewed August 14, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.