Pith. sign in

REVIEW 4 major objections 5 minor 83 references

Foreground Subtraction with a Tensor-Based Oriented Singular Value Decomposition Method for HI Experiments

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read O-SVD, a tensor generalization of SVD, subtracts 21 cm foregrounds without flattening the data cube and, on SKA SDC3a simulations, recovers the true EoR cylindrical power spectrum across most k scales where PCA leaves residuals or removes…

desk verdict A genuinely useful method paper—O-SVD is a plausible drop-in upgrade to PCA for 21 cm foreground subtraction—but the abstract's 'robust and universal' claim rests on a single sub-band and a hand-picked threshold. read the letter →

arxiv 2608.09129 v1 pith:LZ2J3QJK submitted 2026-08-10 astro-ph.IM astro-ph.CO

Shifan Zuo , Xuelei Chen , Yi Mao This is my paper · ORCID
classification astro-ph.IMastro-ph.CO
keywords 21cmintensitymappingforegroundsubtractionorientedsingularvaluedecompositiontensorEpochofReionizationSKAScienceDataChallenge3aTianlaiCylinderPathfinderangularpowerspectrum
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

21 cm intensity mapping can map cosmic structure from the Epoch of Reionization to the dark-energy era, but astrophysical foregrounds outshine the signal by four to five orders of magnitude and must be removed first. The paper's proposal is that the standard removal step—flattening the three-dimensional data cube into a matrix and applying PCA or SVD—destroys spatial-spectral correlations that contain useful separation information. The authors apply O-SVD, a hierarchical tensor decomposition oriented along the frequency axis, directly to the cube, and on the SKA SDC3a simulation they recover the true EoR cylindrical power spectrum across most $k$ scales after removing 17,652 O-SVD modes, where PCA with 20, 30, or 50 modes either leaves foreground contamination or removes signal. They also apply the same framework to the Tianlai Cylinder Pathfinder data in the multi-frequency angular power spectrum domain. If correct, O-SVD is a drop-in replacement for PCA in existing 21 cm pipelines, requiring no architectural change beyond swapping the matrix SVD step.

What carries the argument

The central object is the Oriented Singular Value Decomposition (O-SVD), a hierarchical multilinear generalization of the matrix SVD for third-order tensors with a chosen oriented axis, here frequency. The algorithm has two stages: stage one is the standard SVD of the mode-3 unfolding, which is mathematically identical to PCA and yields spectral basis vectors $u_k$; stage two reshapes each spatial mode into an $N_x \times N_y$ matrix and performs another SVD, yielding spatial basis vectors $u_{kj}$ and $v_{kj}$. The tensor is reconstructed as a sum of rank-1 outer products $s_{jjk} \, (u_{kj} \circ v_{kj} \circ u_k)$, where the $s_{jjk}$ are the O-SVD singular values. The load-bearing diagnostic is the first row, $s_{11k}$: after reordering, its flattening into a plateau is taken to mark the noise floor, and all modes with $s_{jjk}$ above that threshold are removed as foregrounds.

What would settle it

Run O-SVD on a foreground-free simulation containing only the 21 cm signal and thermal noise: if the reordered $s_{11k}$ still flattens around $k=100$ (or $k=400$ for Tianlai), then the plateau is not a foreground-to-noise transition, and the threshold-based subtraction is removing cosmological signal rather than foregrounds.

Watch

Extended reading notes

Core claim

The central claim is that a native tensor decomposition preserves the physical topology of the data and separates foregrounds from signal more precisely than matrix SVD. O-SVD first performs the same mode-3 SVD as PCA, then takes each resulting spectral mode, reshapes it into a spatial map, and decomposes that map further into spatial modes; the diagonal entries $s_{jjk}$ of the core tensor form a hierarchical singular-value array. Because a spectral mode can be partially removed through its spatial components rather than all at once, foreground subtraction becomes a two-dimensional truncation in the $(j,k)$ plane. On the 106–121 MHz SDC3a sub-band, truncating at $s_{11,k=100}$ removes 17,652 modes and leaves a residual whose cylindrical power spectrum agrees with the ground-truth EoR signal across most scales, whereas PCA with 20 or 30 modes leaves foregrounds and PCA with 50 modes removes signal. On Tianlai MAPS data, the same plateau criterion at $k=400$ removes 13,283 modes and leaves residual frequency correlations consistent with a 21 cm signal plus noise.

Load-bearing premise

The load-bearing premise is that the flattening of the reordered $s_{11k}$ curve marks the true noise floor, so every mode above that plateau is foreground-dominated and removing them all preserves the 21 cm signal; if the plateau is an artifact of the field of view, calibration, or choice of sub-band, the recovered power spectrum would be biased and the claimed high-fidelity recovery would not generalize.

Editorial extensions

If this is right

  • Any pipeline that currently uses PCA for foreground subtraction can be upgraded by replacing the matrix SVD with O-SVD, because the first stage of O-SVD is identical to PCA and the rest is a refinement of each spectral mode.
  • The plateau in $s_{11k}$ provides a truncation threshold that does not require a foreground-free simulation, which is essential for real observations where no ground truth exists.
  • Because the method acts on third-order tensors, it applies both to image cubes and to multi-frequency angular power spectra, so one framework covers the two dominant data representations in 21 cm cosmology.
  • The largest-scale signal loss seen in the recovered power spectrum is a known blind-subtraction effect and can be characterized and corrected with a transfer function estimated from mock signal injection.
  • Randomized O-SVD algorithms can reduce the additional computational cost (about 6–7 times on the SDC3a cube) enough for application to next-generation arrays like SKA.

Reading between the lines

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

  • If the plateau criterion is universal, the same $s_{11k}$ search could be run per sub-band or per observation to locate each noise floor; a straightforward test is to apply it to the other five SDC3a sub-bands and check that the recovered power spectrum stays unbiased.
  • Because O-SVD's advantage comes from splitting each spectral mode into spatial structure, its residuals should be particularly clean at high $k_\perp$; combining it with standard wedge avoidance may extend the accessible $k$-space further than either method alone.
  • The rank-1 modes of the decomposition are physically interpretable (bright point sources, diffuse Galactic emission, noise), which suggests they could be used as foreground templates for calibration, systematic-injection tests, or as inputs to machine-learning foreground removal.
  • The Tianlai analysis used only the YY polarization because of its cleaner spectral response; applying the same tensor decomposition to the XX channel or to cross-polarization would test whether the plateau threshold is robust to polarization systematics.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper introduces a tensor-based foreground subtraction method, Oriented Singular Value Decomposition (O-SVD), for 21 cm intensity mapping. The method decomposes a third-order tensor (two spatial dimensions and one frequency dimension, or frequency-frequency-multipole for MAPS) using a two-stage SVD: first a matrix SVD on the mode-3 unfolding, then a spatial SVD of each reshaped spectral mode. The authors apply O-SVD to the SKA SDC3a simulation (first sub-band, 106-121 MHz) and to Tianlai Cylinder Pathfinder data in the MAPS domain. For SDC3a, they remove 17,652 O-SVD modes selected by a visually determined plateau in the first-row singular values, and compare the recovered cylindrical power spectrum diagonal to the ground truth and to PCA with 20, 30, and 50 modes. For Tianlai, they remove 13,283 modes using the same plateau criterion and show foreground suppression in the residual MAPS slice, while explicitly noting the dataset is too shallow for a detection. The paper claims O-SVD is a robust, universal, plug-and-play replacement for PCA that preserves spatial-spectral correlations.

Significance. If the central claims hold, O-SVD would be a useful and conceptually clean generalization of PCA for 21 cm foreground cleaning, with the appealing property that it can be inserted into existing PCA-based pipelines without changing the data processing architecture. The mathematical formulation is clear, the connection to standard SVD is explicit, and the authors provide public code. The SDC3a comparison to PCA in Figure 9 is suggestive and shows a potential advantage in mode selection. However, the significance is currently limited by the lack of validation of the truncation criterion, the absence of error bars in the power-spectrum comparison, and the fact that the second application (Tianlai) does not validate signal recovery. These gaps prevent the 'robust and universal' and 'high-fidelity recovery' claims from being fully supported.

major comments (4)
  1. [Section 3.1.2 / 3.1.3, Figures 5-6] The truncation threshold for the SDC3a analysis is selected by eye: the authors state that s_11k 'stabilizes into an approximately flat floor' for k ≳ 100 and then 'select s_11,k=100 as the truncation threshold.' This plateau is interpreted as the noise floor, but no evidence is given that the plateau is indeed set by thermal noise rather than by the finite 4-degree field of view, spectrally correlated calibration residuals, or the finite 150-channel sub-band. Because the central claim of high-fidelity recovery rests on this single threshold, the paper needs a sensitivity analysis: vary the threshold around k=100 and show the recovered power spectrum is not strongly affected, or alternatively demonstrate on multiple sub-bands and with mock signal injection that the plateau criterion identifies foreground-dominated modes. Without such tests, the 'robust and universal' claim is unsupported.
  2. [Section 3.1.3, Figure 9] The power-spectrum comparison in Figure 9 has no error bars or uncertainty bands. The O-SVD and PCA curves are each from a single realization, and the claim of 'excellent agreement with the truth across most k-scales' is made without quantifying sample variance, thermal noise, or cosmic variance. The comparison would be substantially stronger if the authors provided uncertainties on P(k), for example from multiple noise realizations or from jackknife estimates, and if they showed the same comparison for the PCA residuals with matched uncertainty treatment. As it stands, the superiority of O-SVD over PCA is asserted from a single noise realization in one sub-band.
  3. [Section 3.2.3, Figure 12] The Tianlai application cannot validate the plateau criterion or the universality claim. The paper explicitly states that 'the signal-to-noise ratio is insufficient for a definitive 21 cm detection' and that residuals remain dominated by thermal noise and low-level systematics. Thus Figure 12 only demonstrates that smooth-spectrum foregrounds are removed from the MAPS slice; it does not show that the 21 cm signal is recovered, nor does it provide a cross-check of the k=400 truncation point. The authors should either add a validation using simulated signal injection into the Tianlai data (e.g., adding mock 21 cm signals and testing recovery) or soften the claim that the Tianlai application confirms the O-SVD framework's efficacy.
  4. [Section 2.2, Eq. (8)] The statement that truncating the O-SVD expansion at r terms gives a rank-r approximation minimizing the Frobenius norm residual, with ||A - A_r||_F^2 = sum_{i=r+1}^{r1 r2} \tilde{s}_i^2, needs a justification. For matrix SVD this is the Eckart-Young theorem, but O-SVD is a hierarchical two-stage decomposition; it is not immediately obvious that truncating the 'reordered O-SVD singular values' globally minimizes the residual over all rank-r tensor approximations. If this is a standard property of O-SVD (e.g., from Zeng and Ng 2020), please cite the specific result; if it is an approximation, state so explicitly.
minor comments (5)
  1. [Title and headers] The title and section headers in the manuscript text contain spacing artifacts (e.g., 'F oreground', 'T able'), which should be corrected in the final version.
  2. [Section 3.1.3] There is a typo: 'we selects 11,k=100 as the truncation threshold' should read 'we select s_11,k=100 ...'. Also, the notation s_11k (and s_jjk) is used before it is formally defined in the text; please define it when the singular value array is first introduced.
  3. [Software availability] The GitHub URL is given as 'https://github.com/zuoshifan/sdc3a osvd pipeline' with spaces, which is not a valid URL. Please provide a correct link and, ideally, a versioned release or DOI.
  4. [Section 4] The complexity analysis gives a helpful scaling comparison, but the quoted 'approximately 13x' and '6-7x' factors for the SDC3a dataset are not substantiated with actual measured runtimes. A brief benchmark table or a note that these are theoretical operation-count ratios would improve reproducibility.
  5. [Section 3.1.3 / Figure 9] The text mentions that a transfer function can be used to correct the largest-scale signal loss, but no transfer function is actually constructed or applied. Since the paper later claims high-fidelity recovery, it would be useful to state explicitly whether the plotted O-SVD curve is the raw cleaned power spectrum (which it appears to be) or a transfer-function-corrected estimate.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the O-SVD derivation is self-contained, and the ground truth is used only for external validation, not to set the truncation threshold.

full rationale

The central derivation is a direct application of the O-SVD decomposition (Zeng & Ng 2020) to foreground subtraction. The truncation threshold is selected from the measured singular-value spectrum (the plateau in s_11k near k=100 for SDC3a and k=400 for Tianlai), not from the ground-truth power spectrum; the ground truth enters only after the fact to validate the recovered power spectrum in Figure 9. The O-SVD modes are not defined in terms of the 21 cm signal, and the low-rank residual norm stated in Eq. 8 follows from the orthogonality of the rank-1 terms in Eq. 7 via standard matrix SVD properties. No fitted parameter is renamed as a prediction: the only data-derived choice is the number of removed modes, which is a free parameter of any blind subtraction method and is not tuned to the true EoR spectrum. The self-citations (Zuo et al. 2021 for data reduction, Zuo et al. 2023 for SVP) concern ancillary pipeline steps and a prior semi-blind method; they are not load-bearing for the O-SVD derivation. The reference to the SDC3a challenge summary (Bonaldi et al. 2025a) describes an earlier submission of the same method but is corroborated here by an independent re-analysis of the public dataset. The Tianlai application is explicitly labeled as not constituting a 21 cm detection, so it does not manufacture a prediction. Overall, the paper's conclusions may be under-supported by the absence of threshold-sensitivity tests and the reliance on a visually selected plateau, but these are robustness concerns, not circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

O-SVD is a known tensor decomposition; no new physical entity is introduced. The central claim depends on two data-chosen truncation thresholds and on the physical assumption that smooth-spectrum foregrounds occupy the largest multilinear modes. No proof is given that the chosen threshold coincides with the optimal rank-r truncation.

free parameters (3)
  • SDC3a O-SVD truncation threshold = N_fg = 17,652 modes (s_11,k=100)
    Chosen by eye from the s_11k plateau in Figures 5-6; determines what is classified as foreground versus signal and directly controls the recovered power spectrum.
  • Tianlai O-SVD truncation threshold = 13,283 modes (s_11,k=400)
    Same plateau criterion applied to the MAPS tensor in Section 3.2.3; no ground truth is available to validate this choice.
  • PCA comparison mode counts = 20, 30, 50
    Selected for the baseline comparison in Section 3.1.3; the claimed superiority of O-SVD depends in part on these choices.
assumptions (4)
  • standard math Truncating the O-SVD expansion by singular-value threshold gives a near-optimal low-rank approximation with mutually orthogonal rank-1 terms
    Relies on the orthogonality of spectral and spatial SVD bases, stated in Section 2.2 and used in Eq. 8 without an independent proof.
  • domain assumption Foregrounds are spectrally smooth and spatially coherent, so they concentrate in low-order O-SVD modes while the 21 cm signal and noise are spectrally fluctuating
    This is the physical separation principle stated in Section 2.2 and used to justify mode truncation.
  • ad hoc to paper The plateau in s_11k marks the noise floor and separates foreground-dominated from signal-dominated modes
    Threshold selection in Sections 3.1.2 and 3.2.3; the plateau location (k=100 or k=400) is read off by eye and not statistically justified.
  • domain assumption The SDC3a simulation and the early Tianlai dataset are faithful representations of the instruments and sky
    The performance claims inherit all assumptions built into the SDC3a simulator and the Tianlai reduction pipeline described in Section 3.2.2.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Foreground Subtraction with a Tensor-Based Oriented Singular Value Decomposition Method for HI Experiments." pith.science (2026). https://pith.science/paper/LZ2J3QJK

@misc{pith2026260809129,
  author       = {Pith},
  title        = {Pith review of: Foreground Subtraction with a Tensor-Based Oriented Singular Value Decomposition Method for HI Experiments},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LZ2J3QJK}},
  note         = {Machine review of arXiv:2608.09129}
}
read the original abstract

We introduce a native tensor-based framework for foreground mitigation in 21\,cm intensity mapping (IM), utilizing the Oriented Singular Value Decomposition (O-SVD) algorithm. While 21\,cm IM is a powerful probe of the large-scale structure of the Universe, its efficacy is severely limited by astrophysical foregrounds that are orders of magnitude brighter than the cosmological signal. Traditional mitigation strategies often necessitate flattening multidimensional data cubes into two-dimensional matrices, a process that potentially compromises the intrinsic spatial-spectral correlations by treating distinct spatial pixels as independent samples. By treating multi-frequency sky maps and angular power spectra as third-order tensors, the O-SVD method performs decomposition directly on the multilinear manifold, preserving the underlying physical topology and leveraging the distinct coherence properties of astrophysical foregrounds across different dimensions. We demonstrate the performance and versatility of the O-SVD framework through its application to high-fidelity simulations from the SKA Science Data Challenge 3a (SDC3a) and real-world observational data from the Tianlai Cylinder Pathfinder Array. Our results indicate that O-SVD provides a robust and universal approach for foreground subtraction, achieving high-fidelity signal recovery while offering superior performance compared to conventional matrix-based Singular Value Decomposition (SVD) methods.

Figures

Figures reproduced from arXiv: 2608.09129 by the authors.

Figure 1
Figure 1. Oriented Singular Value Decomposition (O-SVD) two-stage process. Stage 1 unfolds the tensor and performs SVD on the mode-3 matrix. Stage 2 takes each column of V(3), reshapes it into an I1 × I2 matrix, and performs SVD on each reshaped matrix. version of the Global Sky Model (GSM2016; H. Zheng et al. 2017), with quadratic in￾terpolation in log(frequency) and additional spatial frequency content beyond the native res… view at source ↗
Figure 2
Figure 2. The first sub-band (106–121 MHz) of the SKA SDC3a simulation visualized as a 3D image cube (left) and as a flattened 2D array by combining the two spatial axes (right) with identical color scaling. The 3D representation preserves the intrinsic spatial-spectral topology, whereas flattening collapses the spatial information into a single dimension, potentially obscuring the multi-dimensional correlations utilized by O… view at source ↗
Figure 3
Figure 3. Comparison of singular value spectra: (Left) The 2D O-SVD singular value array sjjk, where the indices j and k correspond to the spatial and spectral modes, respectively. The black contour indicates the truncation threshold s11,k=100 used for foreground subtraction. The green line indicates the truncation threshold for N PCA fg = 20 used for traditional SVD. (Right) The 1D matrix SVD singular value spectrum σk obtai… view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: The O-SVD singular values sjjk reordered in descending order. lect s11,k=100 as the truncation threshold, removing all modes satisfying sjjk ≥ s11,k=100. These modes are lo￾calized in the lower-left region of the 2D singular value array (indicated by the black contour …
Figure 5
Figure 5. Figure 5: The first row (spectral trend, s11k, top) and first column (spatial trend, sj11, bottom) of the O-SVD singular value array sjjk. The spectral trend captures the variance across frequency slices, while the spatial trend reflects the hi￾erarchical importance of spatial f…
Figure 6
Figure 6. Figure 6: The first row of the singular value array (s11k) reordered in descending order. The plateau at large indices serves as an empirical diagnostic for identifying the noise floor and setting the truncation threshold [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Visualizing the central frequency slice of the physical components of O-SVD: Rank-1 modes A(j,j,k) for indices (1, 1, 1) (top-left), (2, 2, 1) (top-right), (1, 1, 2) (bottom-left), and (900, 900, 150) (bottom-right). The dominant modes (top row) capture high-dynamic-ra…
Figure 8
Figure 8. Figure 8: The central frequency slice of the residual im￾age cube after O-SVD-based foreground subtraction, with Nfg = 17, 652 modes removed. The residual exhibits the stochastic, Gaussian-like fluctuations characteristic of the cosmological signal and thermal noise. modes (blue…
Figure 9
Figure 9. Figure 9: Top: Cylindrical power spectrum P(k⊥, k∥) of the O-SVD residual. Bottom: Comparison of the diagonal terms (k∥ = k⊥) against the ground-truth EoR power spectrum. foreground contamination, as shown in [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: Representative frontal slice Cℓ=100(ν, ν′ ) of the MAPS tensor C for the Tianlai dataset. The high-amplitude diagonal features and smooth off-diagonal structures are the signature of bright astrophysical foregrounds with long-range frequency coherence. expected to be …
Figure 11
Figure 11. Figure 11: Comparison of singular value distributions for the Tianlai MAPS tensor: (Left) The structured 2D O-SVD singular value spectrum sjjk. (Right) The monotonically decreasing matrix SVD singular values σk, which lacks a clear “knee” for optimal threshold selection. 720 730…
Figure 12
Figure 12. Figure 12: (Left) Foreground-subtracted residual frontal slice C res ℓ=100(ν, ν′ ). (Right) Averaged residual power as a function of frequency separation ∆ν, demonstrating the rapid decoherence of the signal compared to the original foreground-dominated data. matrix-based SVD me…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

83 extracted references · 19 canonical work pages

  1. [1]

    Astronomy and Computing , keywords =

    Data processing pipeline for Tianlai experiment. Astronomy and Computing , keywords =. doi:10.1016/j.ascom.2020.100439 , archivePrefix =. 2011.10757 , primaryClass =

    arXiv 2020

  2. [2]

    arXiv e-prints , keywords =

    Cosmology with HI. arXiv e-prints , keywords =. doi:10.48550/arXiv.2411.08113 , archivePrefix =. 2411.08113 , primaryClass =

    arXiv doi:10.48550/arxiv.2411.08113

  3. [3]

    , keywords =

    Data Analysis for Precision 21 cm Cosmology. , keywords =. doi:10.1088/1538-3873/ab5bfd , archivePrefix =. 1907.08211 , primaryClass =

    arXiv 1907

  4. [4]

    , keywords =

    Improved measurement of the spectral index of the diffuse radio background between 90 and 190 MHz. , keywords =. doi:10.1093/mnras/stw2696 , archivePrefix =. 1609.08705 , primaryClass =

    arXiv

  5. [5]

    , keywords =

    The Radio Sky at Meter Wavelengths: m-mode Analysis Imaging with the OVRO-LWA. , keywords =. doi:10.3847/1538-3881/aac721 , archivePrefix =. 1711.00466 , primaryClass =

    arXiv

  6. [6]

    , keywords =

    21-cm foregrounds and polarization leakage: cleaning and mitigation strategies. , keywords =. doi:10.1093/mnras/stab856 , archivePrefix =. 2010.02907 , primaryClass =

    arXiv 2010

  7. [7]

    Statistical estimation of full-sky radio maps from 21cm array visibility data using Gaussian constrained realizations

    Statistical estimation of full-sky radio maps from 21 cm array visibility data using Gaussian constrained realizations. RAS Techniques and Instruments , keywords =. doi:10.1093/rasti/rzae041 , archivePrefix =. 2403.13766 , primaryClass =

    work page Pith review arXiv

  8. [8]

    , keywords =

    Calibration requirements for detecting the 21 cm epoch of reionization power spectrum and implications for the SKA. , keywords =. doi:10.1093/mnras/stw1380 , archivePrefix =. 1603.00607 , primaryClass =

    arXiv

Show all 83 references
  1. [9]

    Primary beam model and direction-dependent calibration

    Polarization leakage in epoch of reionization windows - II. Primary beam model and direction-dependent calibration. , keywords =. doi:10.1093/mnras/stw1863 , archivePrefix =. 1604.04534 , primaryClass =

    arXiv

  2. [10]

    , keywords =

    Simulations of Galactic polarized synchrotron emission for Epoch of Reionization observations. , keywords =. doi:10.1093/mnras/sty1457 , archivePrefix =. 1802.03060 , primaryClass =

    arXiv

  3. [11]

    , keywords =

    Wide-field LOFAR-LBA power-spectra analyses: impact of calibration, polarization leakage, and ionosphere. , keywords =. doi:10.1093/mnras/sty1095 , archivePrefix =. 1709.07727 , primaryClass =

    arXiv

  4. [12]

    , keywords =

    WSCLEAN: an implementation of a fast, generic wide-field imager for radio astronomy. , keywords =. doi:10.1093/mnras/stu1368 , archivePrefix =. 1407.1943 , primaryClass =

    1943 arXiv

  5. [13]

    Improved source catalogue for the EoR 0 field

    The MWA long baseline Epoch of reionisation survey I. Improved source catalogue for the EoR 0 field. , keywords =. doi:10.1017/pasa.2021.50 , archivePrefix =. 2110.08400 , primaryClass =

    2021 arXiv

  6. [14]

    , keywords =

    The tiered radio extragalactic continuum (T-RECS) simulation II: H I emission and continuum-H I cross-correlation. , keywords =. doi:10.1093/mnras/stad1913 , archivePrefix =. 2305.10175 , primaryClass =

    arXiv

  7. [15]

    , keywords =

    The Tiered Radio Extragalactic Continuum Simulation (T-RECS). , keywords =. doi:10.1093/mnras/sty2603 , archivePrefix =. 1805.05222 , primaryClass =

    arXiv

  8. [16]

    , keywords =

    An improved model of diffuse galactic radio emission from 10 MHz to 5 THz. , keywords =. doi:10.1093/mnras/stw2525 , archivePrefix =. 1605.04920 , primaryClass =

    arXiv

  9. [17]

    , keywords =

    First look at the multiphase interstellar medium using synthetic observations from low-frequency Faraday tomography. , keywords =. doi:10.1051/0004-6361/202142453 , archivePrefix =. 2204.02774 , primaryClass =

    arXiv

  10. [19]

    Planck 2018 results. VI. Cosmological parameters. , keywords =. doi:10.1051/0004-6361/201833910 , archivePrefix =. 1807.06209 , primaryClass =

    2018 arXiv

  11. [20]

    Optics Express , keywords =

    Computationally efficient autoregressive method for generating phase screens with frozen flow and turbulence in optical simulations. Optics Express , keywords =. doi:10.1364/OE.23.033335 , archivePrefix =. 1512.05424 , primaryClass =

    arXiv

  12. [21]

    , keywords =

    Towards a model of full-sky Galactic synchrotron intensity and linear polarisation: A re-analysis of the Parkes data. , keywords =. doi:10.1051/0004-6361:20020285 , archivePrefix =. astro-ph/0202520 , primaryClass =

    arXiv

  13. [22]

    , keywords =

    The pre-launch Planck Sky Model: a model of sky emission at submillimetre to centimetre wavelengths. , keywords =. doi:10.1051/0004-6361/201220019 , archivePrefix =. 1207.3675 , primaryClass =

    arXiv

  14. [23]

    , keywords =

    Full sky study of diffuse Galactic emission at decimeter wavelengths. , keywords =. doi:10.1051/0004-6361:20031125 , archivePrefix =. astro-ph/0303031 , primaryClass =

    arXiv

  15. [24]

    Planck 2015 results. X. Diffuse component separation: Foreground maps. , keywords =. doi:10.1051/0004-6361/201525967 , archivePrefix =. 1502.01588 , primaryClass =

    2015 arXiv

  16. [25]

    Planck 2015 results. XXV. Diffuse low-frequency Galactic foregrounds. , keywords =. doi:10.1051/0004-6361/201526803 , archivePrefix =. 1506.06660 , primaryClass =

    2015 arXiv

  17. [26]

    International Journal of Modern Physics Conference Series , year = 2012, series =

    The Tianlai Project: a 21CM Cosmology Experiment. International Journal of Modern Physics Conference Series , year = 2012, series =. doi:10.1142/S2010194512006459 , archivePrefix =. 1212.6278 , primaryClass =

    2012 arXiv

  18. [27]

    , keywords =

    Forecasts on the Dark Energy and Primordial Non-Gaussianity Observations with the Tianlai Cylinder Array. , keywords =. doi:10.1088/0004-637X/798/1/40 , archivePrefix =. 1410.7794 , primaryClass =

    arXiv

  19. [28]

    Science China Physics, Mechanics, and Astronomy , keywords =

    The Tianlai Cylinder Pathfinder array: System functions and basic performance analysis. Science China Physics, Mechanics, and Astronomy , keywords =. doi:10.1007/s11433-020-1594-8 , archivePrefix =. 2006.05605 , primaryClass =

    2006 arXiv

  20. [29]

    , keywords =

    The Tianlai dish pathfinder array: design, operation, and performance of a prototype transit radio interferometer. , keywords =. doi:10.1093/mnras/stab1802 , archivePrefix =. 2011.05946 , primaryClass =

    2011 arXiv

  21. [30]

    Research in Astronomy and Astrophysics , keywords =

    Sky reconstruction for the Tianlai cylinder array. Research in Astronomy and Astrophysics , keywords =. doi:10.1088/1674-4527/16/10/158 , archivePrefix =. 1606.03830 , primaryClass =

    arXiv

  22. [31]

    , keywords =

    An Eigenvector-Based Method of Radio Array Calibration and Its Application to the Tianlai Cylinder Pathfinder. , keywords =. doi:10.3847/1538-3881/aaf4c0 , archivePrefix =. 1807.04590 , primaryClass =

    arXiv

  23. [32]

    , keywords =

    All-sky Interferometry with Spherical Harmonic Transit Telescopes. , keywords =. doi:10.1088/0004-637X/781/2/57 , archivePrefix =. 1302.0327 , primaryClass =

    arXiv

  24. [33]

    Psychometrika , author=

    The Approximation of One Matrix by Another of Lower Rank , volume=. Psychometrika , author=. 1936 , pages=. doi:10.1007/BF02288367 , number=

    1936 doi

  25. [34]

    , keywords =

    Square Kilometre Array Science Data Challenge 3a: foreground removal for an EoR experiment. , keywords =. doi:10.1093/mnras/staf1466 , archivePrefix =. 2503.11740 , primaryClass =

  26. [35]

    , keywords =

    deep21: a deep learning method for 21 cm foreground removal. , keywords =. doi:10.1088/1475-7516/2021/04/081 , archivePrefix =. 2010.15843 , primaryClass =

    2021 arXiv

  27. [36]

    , keywords =

    Eliminating Primary Beam Effect in Foreground Subtraction of Neutral Hydrogen Intensity Mapping Survey with Deep Learning. , keywords =. doi:10.3847/1538-4357/ac7a34 , archivePrefix =. 2204.02780 , primaryClass =

    arXiv

  28. [37]

    , keywords =

    The stability of deep learning for 21cm foreground removal across various sky models and frequency-dependent systematics. , keywords =. doi:10.1093/mnras/stae1676 , archivePrefix =. 2311.00493 , primaryClass =

    arXiv

  29. [38]

    Journal of Optimization Theory and Applications , year=

    A Randomized Singular Value Decomposition for Third-Order Oriented Tensors , author=. Journal of Optimization Theory and Applications , year=

  30. [39]

    , title =

    Zeng, Chao and Ng, Michael K. , title =. Numerical Linear Algebra with Applications , volume =. doi:https://doi.org/10.1002/nla.2290 , url =. https://onlinelibrary.wiley.com/doi/pdf/10.1002/nla.2290 , abstract =

    doi

  31. [40]

    Eckart-Young

    Analysis of individual differences in multidimensional scaling via an n-way generalization of "Eckart-Young" decomposition , author=. Psychometrika , volume=

  32. [41]

    UCLA Working Papers in Phonetics , key =

    Harshman, Richard , biburl =. UCLA Working Papers in Phonetics , key =

  33. [42]

    A Multilinear Singular Value Decomposition , author=. SIAM J. Matrix Anal. Appl. , year=

  34. [43]

    Psychometrika , volume=

    Some mathematical notes on three-mode factor analysis , author=. Psychometrika , volume=

  35. [44]

    Third-Order Tensors as Operators on Matrices: A Theoretical and Computational Framework with Applications in Imaging , author=. SIAM J. Matrix Anal. Appl. , year=

  36. [45]

    Kilmer and Carla D

    Misha E. Kilmer and Carla D. Martin , keywords =. Factorization strategies for third-order tensors , journal =. 2011 , note =. doi:https://doi.org/10.1016/j.laa.2010.09.020 , url =

    2011 doi

  37. [46]

    Perturbations of the Tcur Decomposition for Tensor Valued Data in the Tucker Format , volume =

    Che, Maolin and Chen, Juefei and Wei, Yimin , year =. Perturbations of the Tcur Decomposition for Tensor Valued Data in the Tucker Format , volume =. Journal of optimization theory and applications , keywords =

  38. [47]

    Oseledets, I. V. , title =. SIAM Journal on Scientific Computing , volume = 33, number = 5, pages =. doi:10.1137/090752286 , URL =. https://doi.org/10.1137/090752286 , abstract =

    doi

  39. [48]

    SIAM Journal on Matrix Analysis and Applications , volume = 36, number = 3, pages =

    Batselier, Kim and Liu, Haotian and Wong, Ngai , title =. SIAM Journal on Matrix Analysis and Applications , volume = 36, number = 3, pages =. doi:10.1137/141000658 , URL =. https://doi.org/10.1137/141000658 , abstract =

    doi

  40. [49]

    arXiv e-prints , keywords =

    SKA-Low simulations for a cosmic dawn/epoch of reionisation deep field. arXiv e-prints , keywords =. doi:10.48550/arXiv.2506.09533 , archivePrefix =. 2506.09533 , primaryClass =

    doi

  41. [50]

    Wide Field Astronomy & Technology for the Square Kilometre Array , year = 2009, month = jan, eid =

    OSKAR: Simulating Digital Beamforming for the SKA Aperture Array. Wide Field Astronomy & Technology for the Square Kilometre Array , year = 2009, month = jan, eid =. doi:10.22323/1.132.0031 , adsurl =

    2009 doi

  42. [51]

    The Journal of Open Source Software , keywords =

    21cmFAST v3: A Python-integrated C code for generating 3D realizations of the cosmic 21cm signal. The Journal of Open Source Software , keywords =. doi:10.21105/joss.02582 , archivePrefix =. 2010.15121 , primaryClass =

    2010 arXiv

  43. [52]

    , keywords =

    Low-amplitude clustering in low-redshift 21-cm intensity maps cross-correlated with 2dF galaxy densities. , keywords =. doi:10.1093/mnras/sty346 , archivePrefix =. 1710.00424 , primaryClass =

    arXiv

  44. [53]

    , keywords =

    Measurement of 21 cm Brightness Fluctuations at z -0.5ex 0.8 in Cross-correlation. , keywords =. doi:10.1088/2041-8205/763/1/L20 , archivePrefix =. 1208.0331 , primaryClass =

    arXiv

  45. [54]

    , keywords =

    An absorption profile centred at 78 megahertz in the sky-averaged spectrum. , keywords =. doi:10.1038/nature25792 , archivePrefix =. 1810.05912 , primaryClass =

    arXiv

  46. [55]

    Nature Astronomy , keywords =

    Astrophysical constraints from the SARAS 3 non-detection of the cosmic dawn sky-averaged 21-cm signal. Nature Astronomy , keywords =. doi:10.1038/s41550-022-01825-6 , archivePrefix =. 2212.00464 , primaryClass =

    arXiv

  47. [56]

    , keywords =

    The GMRT Epoch of Reionization experiment: a new upper limit on the neutral hydrogen power spectrum at z 8.6. , keywords =. doi:10.1111/j.1365-2966.2011.18208.x , archivePrefix =. 1006.1351 , primaryClass =

    2011

  48. [57]

    From Planets to Dark Energy: the Modern Radio Universe , year = 2007, month = jan, eid =

    Probing the epoch of reionization with 21CMA: status and prospects. From Planets to Dark Energy: the Modern Radio Universe , year = 2007, month = jan, eid =

    2007

  49. [58]

    , keywords =

    The Precision Array for Probing the Epoch of Re-ionization: Eight Station Results. , keywords =. doi:10.1088/0004-6256/139/4/1468 , archivePrefix =. 0904.2334 , primaryClass =

    arXiv

  50. [59]

    , keywords =

    LOFAR: The LOw-Frequency ARray. , keywords =. doi:10.1051/0004-6361/201220873 , archivePrefix =. 1305.3550 , primaryClass =

    arXiv

  51. [60]

    , keywords =

    The Murchison Widefield Array: The Square Kilometre Array Precursor at Low Radio Frequencies. , keywords =. doi:10.1017/pasa.2012.007 , archivePrefix =. 1206.6945 , primaryClass =

    2012 arXiv

  52. [61]

    American Astronomical Society Meeting Abstracts \#217 , year = 2011, series =

    Hydrogen Epoch of Reionization Arrays. American Astronomical Society Meeting Abstracts \#217 , year = 2011, series =

    2011

  53. [62]

    American Astronomical Society Meeting Abstracts , year = 2007, series =

    The Murchison Widefield Array. American Astronomical Society Meeting Abstracts , year = 2007, series =

    2007

  54. [63]

    American Astronomical Society Meeting Abstracts \#227 , year = 2016, series =

    21 cm Fluctuations of the Cosmic Dawn with the Owens Valley Long Wavelength Array. American Astronomical Society Meeting Abstracts \#227 , year = 2016, series =

    2016

  55. [64]

    SF2A-2012: Proceedings of the Annual meeting of the French Society of Astronomy and Astrophysics , year = 2012, editor =

    LSS/NenuFAR: The LOFAR Super Station project in Nan c ay. SF2A-2012: Proceedings of the Annual meeting of the French Society of Astronomy and Astrophysics , year = 2012, editor =

    2012

  56. [65]

    Exascale Radio Astronomy , year = 2014, volume =

    The Canadian Hydrogen Intensity Mapping Experiment (CHIME). Exascale Radio Astronomy , year = 2014, volume =

    2014

  57. [66]

    Advancing Astrophysics with the Square Kilometre Array (AASKA14) , year = 2015, month = apr, eid =

    The Cosmic Dawn and Epoch of Reionisation with SKA. Advancing Astrophysics with the Square Kilometre Array (AASKA14) , year = 2015, month = apr, eid =. doi:10.22323/1.215.0001 , archivePrefix =. 1505.07568 , primaryClass =

    2015 arXiv

  58. [67]

    arXiv e-prints , keywords =

    Foregrounds and their mitigation. arXiv e-prints , keywords =. doi:10.48550/arXiv.1909.12369 , archivePrefix =. 1909.12369 , primaryClass =

    1909 doi arXiv

  59. [68]

    , keywords =

    21 cm Tomography with Foregrounds. , keywords =. doi:10.1086/506597 , archivePrefix =. astro-ph/0501081 , primaryClass =

    arXiv

  60. [69]

    , keywords =

    Statistical 21-cm signal separation via Gaussian Process Regression analysis. , keywords =. doi:10.1093/mnras/sty1207 , archivePrefix =. 1711.10834 , primaryClass =

    arXiv

  61. [70]

    , keywords =

    Determination of z -0.5ex 0.8 neutral hydrogen fluctuations using the 21cm intensity mapping autocorrelation. , keywords =. doi:10.1093/mnrasl/slt074 , archivePrefix =. 1304.3712 , primaryClass =

    arXiv

  62. [71]

    , keywords =

    Blind foreground subtraction for intensity mapping experiments. , keywords =. doi:10.1093/mnras/stu2474 , archivePrefix =. 1409.8667 , primaryClass =

    arXiv

  63. [72]

    , keywords =

    Foreground removal using FASTICA: a showcase of LOFAR-EoR. , keywords =. doi:10.1111/j.1365-2966.2012.21065.x , archivePrefix =. 1201.2190 , primaryClass =

    2012

  64. [73]

    , keywords =

    Recovery of 21-cm intensity maps with sparse component separation. , keywords =. doi:10.1093/mnras/staa2854 , archivePrefix =. 2006.05996 , primaryClass =

    2006 arXiv

  65. [74]

    , keywords =

    Extracting H I cosmological signal with generalized needlet internal linear combination. , keywords =. doi:10.1093/mnras/stv2884 , archivePrefix =. 1509.00742 , primaryClass =

    arXiv

  66. [75]

    , keywords =

    Post-correlation radio frequency interference classification methods. , keywords =. doi:10.1111/j.1365-2966.2010.16471.x , archivePrefix =. 1002.1957 , primaryClass =

    2010

  67. [76]

    , keywords =

    A morphological algorithm for improving radio-frequency interference detection. , keywords =. doi:10.1051/0004-6361/201118497 , archivePrefix =. 1201.3364 , primaryClass =

    arXiv

  68. [77]

    , keywords =

    A Semiblind PCA-based Foreground Subtraction Method for 21 cm Intensity Mapping. , keywords =. doi:10.3847/1538-4357/acb822 , archivePrefix =. 2208.14675 , primaryClass =

    arXiv

  69. [78]

    Formalism and the expected signal

    On using visibility correlations to probe the HI distribution from the dark ages to the present epoch - I. Formalism and the expected signal. , keywords =. doi:10.1111/j.1365-2966.2004.08604.x , archivePrefix =. astro-ph/0406676 , primaryClass =

    2004

  70. [79]

    , keywords =

    The multifrequency angular power spectrum of the epoch of reionization 21-cm signal. , keywords =. doi:10.1111/j.1365-2966.2007.11747.x , archivePrefix =. astro-ph/0605546 , primaryClass =

    2007

  71. [80]

    , keywords =

    Towards simulating and quantifying the light-cone EoR 21-cm signal. , keywords =. doi:10.1093/mnras/stx2888 , archivePrefix =. 1706.09449 , primaryClass =

    arXiv

  72. [81]

    , keywords =

    Astropy: A community Python package for astronomy. , keywords =. doi:10.1051/0004-6361/201322068 , archivePrefix =. 1307.6212 , primaryClass =

    arXiv

  73. [82]

    The Journal of Open Source Software , keywords =

    Tools21cm: A python package to analyse the large-scale 21-cm signal from the Epoch of Reionization and Cosmic Dawn. The Journal of Open Source Software , keywords =. doi:10.21105/joss.02363 , adsurl =

    doi

  74. [83]

    , keywords =

    Hydrogen Epoch of Reionization Array (HERA). , keywords =. doi:10.1088/1538-3873/129/974/045001 , archivePrefix =. 1606.07473 , primaryClass =

    arXiv

  75. [84]

    , keywords =

    The foreground transfer function for H I intensity mapping signal reconstruction: MeerKLASS and precision cosmology applications. , keywords =. doi:10.1093/mnras/stad1567 , archivePrefix =. 2302.07034 , primaryClass =

    arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.