Detection of a peculiar noise type in the TESS "fast" light curves

A. P\'al; Gy. M. Szab\'o; Sz. Csizmadia; Sz. K\'alm\'an

REVIEW 5 major objections 6 minor 14 references

Detection of a peculiar noise type in the TESS "fast" light curves

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read TESS's fastest photometry of WASP-167 contains a blue-noise component that rises toward high frequencies, attributed to spacecraft pointing jitter.

desk verdict A short, useful empirical note on a possible jitter-induced high-frequency noise component in TESS 20s data, but the PSD methods need to be specified before the detection is solid. read the letter →

arxiv 2502.10326 v1 pith:5BQM4OJS submitted 2025-02-14 astro-ph.EP astro-ph.IMastro-ph.SR

Sz. Kálmán , Sz. Csizmadia , A. Pál , Gy. M. Szabó This is my paper · ORCID

classification astro-ph.EPastro-ph.IMastro-ph.SR

keywords TESS20-secondcadenceSAPphotometrybluenoisepowerspectraldensitypointingjittersubpixelsensitivityWASP-167

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper reports a new noise component in TESS's fastest light curves. In the 20-second SAP photometry of WASP-167 from Sector 64, the power spectral density rises toward high frequencies with slope $\gamma = -0.401 \pm 0.006$, instead of flattening to the white-noise level expected at short timescales. The authors attribute this 'blue noise' to spacecraft pointing jitter acting through subpixel sensitivity variations on the CCD, and they show that decorrelating against the measured photo-center position weakens the excess. They also find that a single bilinear correction is unreliable because the relation between flux and pointing offset changes across three segments of the light curve. If the detection holds, TESS fast-cadence photometry contains a correlated noise term that standard filters designed for $\gamma=1$ pink noise do not model.

What carries the argument

The central object is the exponent $\gamma$ in the power-law model $S(f) \propto f^{-\gamma}$, which classifies noise as white ($\gamma=0$), pink ($\gamma=1$), or blue ($\gamma=-1$). The paper measures $\gamma$ separately in the low- and high-frequency domains of the WASP-167 PSD and compares the high-frequency value with the white-noise expectation. The explanatory mechanism is spacecraft pointing jitter: the pixel coordinates of the photo-center, tracked by the TESS pipeline, serve as a proxy for attitude, and the paper tests a bilinear decorrelation of flux against those coordinates. To show why that correction is insufficient, it fits $\Phi_{\mathrm{SAP}}$ against $\log \delta$, the logarithmic distance of the photo-center from its mean position, in three separate time segments and finds significantly different slopes.

What would settle it

Recompute the power spectrum of the same WASP-167 20-second SAP light curve with a fully documented estimator, such as Welch's method with explicit windowing, binning, and frequency bounds, and check whether the high-frequency slope remains near $\gamma \approx -0.4$; if the rising tail disappears once spectral leakage, aliasing, or the detrending choice is accounted for, the blue-noise detection would not survive.

Watch

Extended reading notes

Core claim

The paper's central claim is that the 20-second SAP light curve of WASP-167 contains a high-frequency noise component whose power spectral density increases toward short timescales. Fitting the high-frequency end of the PSD gives $\gamma = -0.401 \pm 0.006$, which the authors interpret as blue noise. They argue that the most plausible source is spacecraft pointing jitter acting through subpixel sensitivity variations: using the photo-center coordinates as a proxy for pointing, a bilinear decorrelation removes much of the high-frequency excess. The paper further demonstrates that the jitter response is not stationary, because linear fits of flux against $\log \delta$ in three successive time intervals have markedly different slopes, so a fixed polynomial correction cannot fully clean the data. The feature is also reported in the 20-second light curves of WASP-121 and KELT-25 and absent in that of HD 189733, placing the detection between a universal artifact and a one-off anomaly.

Load-bearing premise

The high-frequency PSD slope is a faithful measurement of the noise rather than an artifact of the estimator, since the paper does not specify the windowing, binning, averaging, or exact frequency range used to compute the spectrum.

Editorial extensions

If this is right

Analyses of TESS 20-second SAP light curves that assume only white noise at high frequencies will underestimate the true noise level for stars like WASP-167.
Wavelet-based noise filters that fix the correlated-noise slope at $\gamma = 1$ will not absorb a component with $\gamma \approx -0.4$; allowing the slope to float risks degeneracies and overfitting.
Binning the light curve acts as a low-pass filter and can suppress the blue-noise excess, at the price of losing the short-timescale information that motivated the 20-second cadence.
A single global bilinear decorrelation against photo-center position is not a complete fix, because the flux–offset relation changes between segments of the light curve.
The presence of the feature in WASP-121 and KELT-25 but not in HD 189733 shows that the effect is neither universal nor unique to one target.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Beyond the paper, if subpixel sensitivity variations are the cause, the blue-noise amplitude should depend on where the point-spread function lands on the CCD, so comparing TESS sectors with different pointing dither patterns would provide a direct confirmation of the mechanism.
The measured slope sits between white noise and the $\gamma = -1$ definition of blue noise, so the 'blue noise' label should be read loosely; a mixed model of white noise plus a jitter-correlated component might reproduce the observed $\gamma \approx -0.4$ without invoking a true blue-noise process.
A practical extension suggested by the paper's time-segment analysis is a segmented or time-varying decorrelation whose breakpoints could be matched against spacecraft events such as momentum dumps, which the authors do not explore.
For exoplanet work, the 20-second light curves should undergo a jitter-decorrelation step before searching for very short-timescale astrophysical signals such as occultations or flares; the paper notes this application remains open.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The punchline: this is a short empirical note claiming that TESS 20-second SAP light curves contain a high-frequency noise component whose PSD rises toward the Nyquist limit (gamma ~ -0.4), attributed to spacecraft jitter acting through subpixel sensitivity variations. That's a new claim as far as I know, and if real it matters for anyone doing transit or variability work with fast-cadence TESS data.

What's genuinely good: the detection is based on a real light curve (WASP-167, Sector 64) and the MAST data are public, so the result is in principle checkable. They measure a high-frequency slope that is clearly distinct from white noise, and they show that a standard bilinear decorrelation against photo-center position removes most of the feature. They also demonstrate that a single linear correction is not enough: the relation between flux and photo-center distance changes across three time segments, with significantly different slopes. That last point is a useful warning for anyone using simple decorrelation. They also report that similar features appear in WASP-121 and KELT-25 but not HD 189733, which is a decent sanity check.

The soft spots are real but not fatal. The big one: the paper doesn't say how the PSD was computed. No windowing, no binning, no averaging, no mention of detrending before the transform, no Nyquist handling. Since the claim rests on the high-frequency slope, that's a load-bearing omission. The quoted error of +/- 0.006 on gamma implies a precise fit, but without details on the number of independent estimates or how the fit was done, that uncertainty isn't interpretable. The stress-test worry--spectral leakage from the strong low-frequency red noise--is plausible enough that the result needs a proper null test (e.g., a synthetic light curve with the same red noise and white noise, run through the same estimator). The 'blue noise' label is loose; gamma = -0.4 is not gamma = -1, though they do say 'hints at,' so that's acceptable for a short note. There's also a typo in the segment boundaries (line 8 has 2640053 instead of 2460053), and the boundaries themselves look post hoc. None of this contradicts the data, but it means the paper is a proof-of-suggestion, not a proof-of-detection.

Who it's for: TESS fast-cadence users, especially exoplanet transit and stellar variability people. It deserves a serious referee, not a desk reject, because the claim is new, checkable, and potentially important for systematic error budgets. A referee should ask for the PSD method, a null test, and uncertainty that accounts for correlated residuals.

Net: I'd engage with it--read it, maybe cite it as a caution--but I wouldn't rely on the quantitative slope until the methods are pinned down.

Referee Report

5 major / 6 minor

Summary. The paper reports the detection of a high-frequency noise component in the 20-second-cadence TESS SAP light curve of WASP-167. From a power spectral density (PSD) analysis, the authors measure a high-frequency slope of gamma = -0.401 +/- 0.006, which they call "blue noise" and attribute to spacecraft pointing jitter acting through subpixel sensitivity variations. They also present a bilinear decorrelation against the photo-center position as a mitigation technique, and argue from three flux-versus-distance segments that the noise properties change over time, so that simple linear corrections may be inadequate.

Significance. If the detection is robust, it identifies a correlated noise component in TESS fast-cadence photometry that standard pink-noise filters (e.g., Carter & Winn 2009) do not model. The analysis has a notable strength: the decorrelation uses independently measured photo-center positions as the regressor, so the mitigation claim is not circular. The paper also usefully reports that similar high-frequency features appear in some other 20-s light curves (WASP-121, KELT-25) but not in HD 189733, suggesting an instrumental rather than astrophysical origin. However, the central quantitative claim rests on a PSD slope whose estimation procedure is not described, and the quoted uncertainty is not justified. The result is plausible but requires a substantially more detailed methods presentation before it can be accepted.

major comments (5)

[Section 3, PSD computation] The manuscript does not specify how the PSD in Figure 1 was computed: no windowing, no segment averaging, no frequency binning, no treatment of the 20-s Nyquist frequency, and no description of any detrending applied before the transform. Because the low-frequency PSD has a strong red slope (gamma ~ 0.86) and the high-frequency domain is adjacent to the Nyquist frequency, spectral leakage and aliasing can bias the high-frequency slope. The measured value gamma = -0.401 +/- 0.006 is not interpretable without this information. Please describe the estimator and demonstrate with simulated red noise, using the same sampling and any pre-processing, that the measured slope is not an artifact of the method.
[Section 3, error bars] The quoted uncertainty of +/- 0.006 on the high-frequency slope implies a very precise fit, but the paper provides no information on the number of independent spectral estimates, the number of fit points, or whether correlated residuals in the PSD were accounted for. The same applies to the low-frequency slopes (+/- 0.033). Please report the fitting details and provide robust uncertainties, for example from bootstrap resampling or noise-only simulations.
[Section 3, labeling] According to the paper's own definition in Section 1, blue noise corresponds to gamma = -1, while the measured high-frequency slope is gamma = -0.401 +/- 0.006. This value is far from -1 (the difference is many times the quoted uncertainty) and is not blue noise by the stated definition. Calling it "blue noise" in the title and abstract overstates the result. Please either adjust the terminology to something like "blue-noise-like" or "rising high-frequency noise", or explicitly model the slope as a mixture of white and a steeper blue component.
[Section 3, jitter correction] The claim that "the jitter correction can mitigate the high-frequency noise component" is supported only by visual inspection of the lower-right panel of Figure 1. The paper does not report the high-frequency slope after the correction or any quantitative test for the reduction. Please quantify the change, e.g., the post-correction gamma in the high-frequency band or the ratio of integrated PSD power in that band, so the mitigation claim can be evaluated.
[Section 3, segmented flux-distance relation] The three segments in the Phi_SAP versus log delta plane (Eqs. 7-9) appear to be chosen post hoc, and the paper does not provide a formal test for whether the slope differences are statistically significant or whether a smooth time-varying relation would be preferred. Since the conclusion that the noise properties change over time rests on these segments, please state how the segment boundaries were selected and provide a significance test or model comparison. There is also a typographical error in the second segment bound ('BJD 2640053' should likely be 'BJD 2460053').

minor comments (6)

[Abstract] There is a typo in the abstract: "jiiter" should be "jitter".
[Introduction] In the paragraph on blue noise, "blue nose" should be "blue noise", "shorter then" should be "shorter than", and "occuring" should be "occurring".
[Section 2, Eq. (4)] Equation (4) is typeset ambiguously: the factor f appears to be in the wrong position, and the symbol f is not defined before its use. Please define FLFCSAP clearly and write the equation with an unambiguous division by f.
[Section 3] The star HD 189733 is written as "HD 1897333" in the text; the extra digit should be removed.
[Section 3] The reference to MAST (2021) does not give the specific DOI or a clear pointer to the data product used; please provide the exact archive identifier.
[Figure 1] The text uses both "delta" and "d" for the photo-center distance; please make the notation consistent in the figure caption and the body.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular reduction in the blue-noise detection or jitter decorrelation; self-citations are present but non-load-bearing.

full rationale

The central result is a direct measurement: the paper computes the power spectral density of the TESS SAP light curve and fits a slope gamma to the high-frequency end, finding gamma = -0.401 +/- 0.006. This slope is not an output of a model that was fitted to the data and then re-predicted; it is the fit itself, and the blue-noise interpretation is a comparison against the stated white-noise expectation. The jitter mitigation uses the independently measured photo-center offsets delta x, delta y as regressors in Eq. (6); the coefficients a and b are fit to the flux, and the post-correction PSD is then remeasured. The success of the correction is thus an empirical check, not a quantity forced by the construction. No parameter is fitted to a subset of the data and then presented as a prediction of a closely related quantity. The paper invokes no uniqueness theorem, and no ansatz is smuggled in through a citation: the bilinear decorrelation is standard (Vanderburg & Johnson 2014) and is applied, not derived. The self-citations (Csizmadia et al. 2023; Kalman et al. 2023; Kalman et al. 2024) are used only as supporting references for known red-noise behavior and mitigation, not to justify the existence of blue noise; hence they are not load-bearing. The main weakness is methodological transparency: the PSD estimator (windowing, binning, averaging, detrending) and the exact frequency range are unspecified, so the measured gamma could be vulnerable to spectral leakage or aliasing near the 20 s Nyquist frequency. That is a reproducibility and robustness concern, not a circularity. For completeness, the score of 2 reflects the presence of minor, non-load-bearing self-citations; no circular step was identified.

Assumptions & free parameters 6 free parameters · 3 assumptions · 0 invented entities

The central claims rest on a small set of fitted quantities and assumptions. The gamma slope is fitted to a PSD whose estimator is unspecified. Segment boundaries and segment slopes are fitted to the same light curve used to define the groups. The interpretation relies on the null assumption of white noise at high frequencies and on treating TESS photo-center coordinates as a pointing proxy. No new physical entities are introduced.

free parameters (6)

high-frequency PSD slope gamma_high = -0.401 +/- 0.006
Fitted by linear regression on the high-frequency end of the PSD of Phi_SAP; this is the central evidence for blue noise.
segment boundary t1 = BJD 2460053
Chosen by eye to split the light curve into the first of three Phi_SAP versus log-distance groups.
segment boundary t2 = BJD 2460058
Chosen by eye as the second split point between the three groups.
segment 1 slope and intercept = slope -0.001902 +/- 0.000071, intercept 0.99577
Linear fit in the Phi_SAP versus log-delta plane for t < BJD 2460053.
segment 2 slope and intercept = slope 0.00692 +/- 0.000015, intercept 1.01537
Linear fit for BJD 2460053 < t < BJD 2460058.
segment 3 slope and intercept = slope 0.00757 +/- 0.000013, intercept 1.01096
Linear fit for t > BJD 2460058.

assumptions (3)

standard math The Fourier and PSD definitions in Eqs. 1-3 are valid for the TESS time series.
Used without proof as the standard mathematical basis for noise-color analysis.
domain assumption At high frequencies, the only expected noise is white photon noise with gamma = 0.
Section 3 states: 'We expect that at higher frequencies, only white noise (i.e. gamma=0) is detectable.' This null model underpins interpreting the measured negative gamma as blue noise.
domain assumption The TESS pipeline photo-center coordinates are a valid proxy for instantaneous spacecraft pointing.
Section 2 uses delta x and delta y as a proxy for telescope positioning and as the basis for the decorrelation in Eq. 6.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Detection of a peculiar noise type in the TESS "fast" light curves." pith.science (2026). https://pith.science/paper/5BQM4OJS

@misc{pith2026250210326,
  author       = {Pith},
  title        = {Pith review of: Detection of a peculiar noise type in the TESS "fast" light curves},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5BQM4OJS}},
  note         = {Machine review of arXiv:2502.10326}
}

read the original abstract

We present the detection of a peculiar high-frequency noise component in the 20 second cadence SAP (Simple Aperture Photometry) light curve of TESS (Transiting Exoplanets Survey Satellite). This effect (labeled as blue noise) may be attributed to the pointing instability (also known as satellite jiiter) of the satellite. We present a common technique used in the mitigation of the jitter, by decorrelating against the subpixel position of the photo-center of the point spread function of the star. We also show that a simple linear or polynomial technique may not yield satisfactory corrections, as the behavior or attitude of the noise properties may change considerably throughout the light curve.

Figures

Figures reproduced from arXiv: 2502.10326 by the authors.

**Figure 1.** TESS light curve of WASP-167 (top right). The PSD from ΦSAP and ΦSAP,corr are shown in the lower left and right panels, respectively. ΦSAP as a function of the distance from the photo-center is shown in the upper right panel at three subsequent segments of the light curve. only white noise (i.e. γ = 0) is detectable. We measure γ = 0.862 ± 0.033 and γ = 0.789 ± 0.033 at the red end of the spectrum for ΦSAP and ΦSAP,… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

14 extracted references · 5 canonical work pages

[1]

Barros, S. C. C., Demangeon, O., D ´ ıaz, R. F., et al. 2020, A&A, 634, A75, doi: 10.1051/0004-6361/201936086

work page doi:10.1051/0004-6361/201936086 2020
[2]

A., & Winn, J

Carter, J. A., & Winn, J. N. 2009, ApJ, 704, 51, doi: 10.1088/0004-637X/704/1/51

doi:10.1088/0004-637x/704/1/51 2009
[3]

Cheng, K. S. 1987, ApJ, 321, 805, doi: 10.1086/165673

work page doi:10.1086/165673 1987
[4]

Cherenkov, P. A. 1934, Dokl. Akad. Nauk SSSR, 2, 451, doi: 10.3367/UFNr.0093.196710n.0385

arXiv 1934
[5]

Csizmadia, S., Smith, A. M. S., K´ alm´ an, S., et al. 2023, A&A, 675, A106, doi: 10.1051/0004-6361/202141302

doi:10.1051/0004-6361/202141302 2023
[6]

A., Zarka, P., et al

Fischer, G., Pagaran, J. A., Zarka, P., et al. 2019, A&A, 621, A113, doi: 10.1051/0004-6361/201833014

work page doi:10.1051/0004-6361/201833014 2019
[7]

P., et al

Hu, Y., Li, L., Yuan, J. P., et al. 2020, Ap&SS, 365, 143, doi: 10.1007/s10509-020-03849-8 K´ alm´ an, S., Szab´ o, G. M., & Csizmadia, S. 2023, A&A, 675, A107, doi: 10.1051/0004-6361/202143017 K´ alm´ an, S., Derekas, A., Csizmadia, S., et al. 2024, A&A, 687, A144, doi: 10.1051/0004-6361/202348886 MAST. 2021, TESS ”Fast” Light Curves - All Sectors, STScI...

work page doi:10.1007/s10509-020-03849-8 2020
[8]

2006, MNRAS, 373, 231, doi: 10.1111/j.1365-2966.2006.11012.x

Pont, F., Zucker, S., & Queloz, D. 2006, MNRAS, 373, 231, doi: 10.1111/j.1365-2966.2006.11012.x

arXiv 2006

Show all 14 references

[9]

E., & Williams, C

Rasmussen, C. E., & Williams, C. K. I. 2006, Gaussian Processes for Machine Learning

2006
[10]

2016, Comput

Reinert, B., Ritschel, T., Seidel, H.-P., & Georgiev, I. 2016, Comput. Graph. Forum, 35, 285–295, doi: 10.1111/cgf.12725

2016 doi
[11]

R., Winn, J

Ricker, G. R., Winn, J. N., Vanderspek, R., et al. 2015, Journal of Astronomical Telescopes, Instruments, and Systems, 1, 014003, doi: 10.1117/1.JATIS.1.1.014003

2015 doi
[12]

2012, Philosophical Transactions of the Royal Society of London Series A, 371, 20110550, doi: 10.1098/rsta.2011.0550

Roberts, S., Osborne, M., Ebden, M., et al. 2012, Philosophical Transactions of the Royal Society of London Series A, 371, 20110550, doi: 10.1098/rsta.2011.0550

2012
[13]

Y., Hellier, C., Albrow, M

Temple, L. Y., Hellier, C., Albrow, M. D., et al. 2017, MNRAS, 471, 2743, doi: 10.1093/mnras/stx1729

2017 doi
[14]

Vanderburg, A., & Johnson, J. A. 2014, PASP, 126, 948, doi: 10.1086/678764

2014 doi

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Barros, S. C. C., Demangeon, O., D ´ ıaz, R. F., et al. 2020, A&A, 634, A75, doi: 10.1051/0004-6361/201936086

work page doi:10.1051/0004-6361/201936086 2020

[2] [2]

A., & Winn, J

Carter, J. A., & Winn, J. N. 2009, ApJ, 704, 51, doi: 10.1088/0004-637X/704/1/51

doi:10.1088/0004-637x/704/1/51 2009

[3] [3]

Cheng, K. S. 1987, ApJ, 321, 805, doi: 10.1086/165673

work page doi:10.1086/165673 1987

[4] [4]

Cherenkov, P. A. 1934, Dokl. Akad. Nauk SSSR, 2, 451, doi: 10.3367/UFNr.0093.196710n.0385

arXiv 1934

[5] [5]

Csizmadia, S., Smith, A. M. S., K´ alm´ an, S., et al. 2023, A&A, 675, A106, doi: 10.1051/0004-6361/202141302

doi:10.1051/0004-6361/202141302 2023

[6] [6]

A., Zarka, P., et al

Fischer, G., Pagaran, J. A., Zarka, P., et al. 2019, A&A, 621, A113, doi: 10.1051/0004-6361/201833014

work page doi:10.1051/0004-6361/201833014 2019

[7] [7]

P., et al

Hu, Y., Li, L., Yuan, J. P., et al. 2020, Ap&SS, 365, 143, doi: 10.1007/s10509-020-03849-8 K´ alm´ an, S., Szab´ o, G. M., & Csizmadia, S. 2023, A&A, 675, A107, doi: 10.1051/0004-6361/202143017 K´ alm´ an, S., Derekas, A., Csizmadia, S., et al. 2024, A&A, 687, A144, doi: 10.1051/0004-6361/202348886 MAST. 2021, TESS ”Fast” Light Curves - All Sectors, STScI...

work page doi:10.1007/s10509-020-03849-8 2020

[8] [8]

2006, MNRAS, 373, 231, doi: 10.1111/j.1365-2966.2006.11012.x

Pont, F., Zucker, S., & Queloz, D. 2006, MNRAS, 373, 231, doi: 10.1111/j.1365-2966.2006.11012.x

arXiv 2006