CoChat

Terahertz time domain spectroscopy data processing: analysing uncertainties to push boundaries

Open access

The DOI matches its Crossref record; 45 of the 45 references it cites were found in a registry or an index; 6 of the 14 sources its claims cite were checked for support.

Abstract

Terahertz spectroscopy provides information on the motion of the charges in a sample at a picosecond scale. To recover this information from Terahertz time-domain spectroscopy (THz-TDS), one usually extracts the experimental refractive index then fits these curves. This approach suffers from several limitations, among them the difficulty to compare models of motions, provide the error bar associated with the extracted magnitude and a resolution limitation coming from the Fourier criteria of the fast Fourier transform. By adopting a Bayesian framework taking into account the experimental uncertainties and directly fitting the time-domain trace, we overcame these limitations. When correlated and epistemic uncertainties/noise are present, the algorithm considers its distribution as part of the data to fit and can mistake it for real physical features. Hence, it offers poor discrimination between good models and bad ones. After a thorough analysis of the experimental noise, we developed a preprocessing software removing epistemic noise on the time traces and providing an estimate of the noise correlation matrix (generalization of the standard deviation). It allows the proper weighting of the error function of the fit using these uncertainties and therefore the derivation of the Akaike information criteria, a metric enabling to calculate the most probable model from a set of models one wants to compare. In addition, by being in the time domain we avoid the Fourier criteria for the resolution and thus could get information on experimental lines down to 30 MHz with a commercial THz-TDS system.

Overview

CoChat Summary

TL;DR

The paper presents signal-processing improvements for THz-TDS: a resynchronization/preprocessing step to reduce noise and estimate the noise correlation matrix, plus time-domain fitting with Akaike criteria and constrained reconstruction super-resolution. It reports about a 100-fold standard-deviation reduction, model comparison on lactose, and super-resolution of NH3 lines beyond the 1.2 GHz Fourier limit, with the abstract reporting information down to 30 MHz.

Key findings

  • Resynchronizing repeated time traces before averaging reduced the standard deviation by about a factor 100 in the data region, increasing precision by the same factor; the noise became much more white, though some pink residues remained.
  • Using the precision matrix (noise correlation matrix) in the fit@tds software allowed proper weighting of the time-domain fit and derivation of the Akaike information criteria; on lactose, fitting 5 peaks plus two phenomenological peaks showed Voigt profiles were more suited than Lorentz profiles for all peaks, with increasing importance at high frequency.
  • The CRSR algorithm achieved super-resolution on a commercial broadband (~0.2–6 THz) THz-TDS with Fourier-transform-limited spectral resolution Fc = 1.2 GHz; it was validated on gas-phase NH3 rotational spectra at 1–700 mbar, with the abstract reporting information down to 30 MHz and the text reporting precision surpassing 1.2 GHz (stated as “a few tens of GHz”).
  • Similar NH3 results were obtained up to 3.5 THz, and the method enabled separation of a doublet around 1.2 THz; measurements used a 7 cm long gas cell.

Methods

  • Noise analysis and preprocessing: repeated time traces were averaged, a reference trace was chosen by mean-square similarity to the classical average, and all traces were shifted by a small delay (~10 fs) to resynchronize them; this provided an estimate of the noise correlation matrix.
  • Fitting: the fit@tds software was used to fit time-domain traces directly, weighting the time-domain difference between sample and model traces by the precision matrix; Akaike information criteria were used to compare models, applied to lactose with Lorentz versus Voigt profiles.
  • Super-resolution: the Constrained Reconstruction Super Resolution (CRSR) algorithm fits spectral Lorentz lines; the reference is zero-padded only up to the laser repetition time, FFT’d, and compared with the sample only where the sample signal was recorded. It was validated on rotational spectra of gas-phase ammonia (NH3) over 1–700 mbar using a commercial THz-TDS.

Limitations

  • The paper states that dynamic range underestimates noise where the signal is high, and that FFT and time-domain convolution make noise correlated, so the usual uncorrelated white-noise weighting is invalid; the noise correlation matrix is needed.
  • The paper states that when correlated and epistemic uncertainties/noise are present, the algorithm can treat the noise distribution as part of the data and mistake it for real physical features, offering poor discrimination between good and bad models; preprocessing is used to remove epistemic noise.
  • The paper states that gas measurements were limited by the 7 cm gas cell due to TDS path-length constraints, with further experiments ongoing on a 1 m long cell. Evident from the text: validation is limited to lactose and NH3, and the reported precision figures are inconsistent between the abstract (30 MHz) and Section 4 (“a few tens of GHz”).

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1117/12.2620069

Cites

45 of 45 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

5 Valid · 3 Not valid · 2 Not checked · 21 Own results and interpretations

11 more claims

Create a free account to see all 31 claims and to verify the 1 reference still open.

See all claims — free account
  • It may state two separate claims as one. 9
  • The quote only partly states this claim. 1
  • Probably not a substantial claim. 6

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Cited by

See all 3

Metadata

Authors
Lavancier, M., Eliet Barois, S., Denakpo, E., Vlieghe, J., Vindas, N., Hindle, F., Cuisset, A., & Peretti, R.
Affiliations
  • Centre National de la Recherche Scientifique (FR)
  • Université du littoral côte d'opale (FR)
Venue
Terahertz Photonics II
Publisher
SPIE
Published
2022-05-31
Volume
12134
Pages
9
OpenAlex
W4281698213
Cited by
2
References
41
Citations by year
2024: 2
FWCI
0.81
Type
conference paper
Language
en
Open access
Licence
other-oa
Locations
  • Terahertz Photonics (publishedVersion)
  • HAL (submittedVersion, other-oa)
  • HAL (submittedVersion)
Topics
  • Physical Sciences › Engineering › Electrical and Electronic Engineering › Terahertz technology and applications
  • Physical Sciences › Chemistry › Spectroscopy › Spectroscopy and Laser Applications
  • Physical Sciences › Physics and Astronomy › Atomic and Molecular Physics, and Optics › Laser-Matter Interactions and Applications
Keywords
terahertz time-domain spectroscopy, Bayesian inference, uncertainty quantification, epistemic uncertainty, noise correlation matrix, Akaike information criterion, model comparison, resolution limit, charge dynamics

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (qwen3.8-omni-flash and gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.