Properties and limitations of linear models for mutual coupling in 21 cm arrays

arXiv:2610.22197 · astro-ph.IM, astro-ph.CO · Submitted 2026-10-15 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Astrophysics Radio. Generated commentary on the latest astrophysics papers.

Vera: I'm Vera.

Jocelyn: And I'm Jocelyn, here with Subrahmanyan.

Vera: We're looking at the paper: "Properties and limitations of linear models for mutual coupling in 21 cm arrays".

Jocelyn: Mutual coupling between elements of low-frequency arrays is a leading systematic for 21 cm cosmology,

Vera: First, who's behind it and why it matters.

Paper summary: Vera: So, let's summarize the main thrust of "Properties and limitations of linear models for mutual coupling in twenty-one cm arrays <ref:2610.22197#pg1,Properties and limitations of linear models for mutual coupling in 21 cm>." The paper starts by stating that mutual coupling between elements of low-frequency arrays is a leading systematic for twenty-one cm cosmology because it leaks bright foregrounds into otherwise clean regions of the power spectrum <ref:2610.22197#pg1,mutual coupling between elements of low-frequency arrays is a leading systematic>.

Jocelyn: They then discuss that many mitigation models describe this coupling as a linear combination of uncoupled signals re-radiated from neighboring feeds, but they suggest there are additional mechanisms that go beyond just that simple linear combination.

Subrahmanyan: Inspired by antenna scattering theory, the authors decompose this coupling into an antenna mode and a structural mode. The antenna mode is built from translated isolated responses and they expect this part to be removable using linear combination models.

Vera: The structural mode is then defined as the remainder, which represents what's left over after you take out that translated copy of the isolated response. Full-wave simulations of HERA-like feeds and dishes in a thirty-seven-element hexagonal array show that this antenna mode only accounts for about one percent of the beam perturbation power, implying that linear models have limited scope for mitigation on their own.

Jocelyn: However, they found two ways to do better suppression with those linear models. They first mentioned correlating beams and weighting them by the sky, which shifts how important these modes are; in this case, correcting the antenna mode can remove more than half of the coupling power in simulated visibilities.

Subrahmanyan: And secondly, relaxing the coefficient relationships that are imposed by re-radiation allows linear corrections to tackle structural-mode coupling as well. They showed that beam-derived corrections can reduce coupling power by over an order of magnitude in their noiseless simulations of perfectly redundant arrays.

Vera: So, they're saying that while the antenna mode is small at one percent, we still see substantial suppression when we use these clever adjustments to the linear models <ref:2610.22197#pg1>. But they flag a problem: fitting directly to known isolated visibilities reduces residuals, but that data isn't usually available in real observations.

Jocelyn: And that means for actual work, you need those extra constraints to separate the unknown sky from the coupling while making sure you don't mess up the cosmological signal. This paper lays out how we can visualize these perturbations in the aperture plane using an extension of holographic back-projection for coupled beams.

Subrahmanyan: For structure formation cosmology, this means that even if our linear model isn't perfect, as long as we use these methods to isolate the structural mode and understand its contribution, we can keep our cosmological constraints more robust. The question then becomes how do we test these modes observationally?

Vera: That's a good pivot. So this paper sets up a framework where we separate what's due to antenna scattering from the rest of the coupling, which is crucial for understanding how much systematic noise we can realistically expect in our twenty-one cm measurements <ref:2610.22197#pg1>.

Conclusion: Jocelyn: Looking at the full picture from "Properties and limitations of linear models for mutual coupling in twenty-one cm arrays," it seems the authors are confirming that linear visibility corrections can remove much of the simulated coupling, even though there's only a small fraction of power attributed to the antenna mode in those beams <ref:2610.22197#pg1,Properties and limitations of linear models for mutual coupling in 21 cm>.

Subrahmanyan: That means that we don't need to assume perfect models for every single element response; instead, as long as we use linear combinations of visibilities, we can get substantial suppression from the coupling. The mu fit they present is a benchmark showing that this small antenna-mode fraction doesn't stop us from getting good suppression.

Vera: The real takeaway for us is that while the paper demonstrates the mathematical capability to suppress coupling through these linear combinations, the remaining challenge is retaining that suppression with observationally justified constraints, element-dependent responses, noise, and flagging while proving we haven't lost any cosmological signal.

Jocelyn: So in short, they show a way to handle this systematic by decomposing it into modes and then applying corrections that are more sophisticated than just assuming everything is perfectly uncoupled. This paper gives us the tools to push forward with our twenty-one cm experiments while being aware of what we're actually constraining <ref:2610.22197#pg1>.

Subrahmanyan: It's about having a solid theoretical handle on these systematic effects so that when we go out and look at the data, we can properly interpret what those residuals are telling us about structure formation or the early universe. We need those constraints to be robust enough for real science.

V. MacKay, R. Pascua, E. Rath, J. N. Hewitt, T. Cox, N. Kern, J. S., Dillon Acedo E., Fagnoni N.

Kavli Institute, Massachusetts Institute of Technology · Dunlap Institute, University of Toronto · Perimeter Institute for Theoretical Physics · Department of Physics, Massachusetts Institute of Technology · Department of Astronomy, University of California, Berkeley · Department of Physics, University of Michigan · Cavendish Astrophysics, University of Cambridge · Kavli Institute for Cosmology, University of Cambridge

astro-ph.IM, astro-ph.CO

Submitted: 2026-10-15

Updated: 2026-10-15

Comments: 41 pages, 22 figures

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 89/100

The gist: Mutual coupling between elements of low-frequency arrays is a leading systematic for 21 cm cosmology, leaking bright foregrounds into otherwise clean regions of the power spectrum The gist Mutual

Key concepts

Antenna Mode: This component of mutual coupling arises when a signal received at one feed is re-radiated into free space and picked up by another dish. It is constructed from translated isolated responses and is theoretically removable using simple linear combination models.
Structural Mode: This mode represents the remaining part of the coupling that cannot be easily removed by linear models. It arises from scattering off a neighbor's physical structure, which occurs even if the neighbor's feed port is not actively excited.
Aperture-Plane Approach: This method visualizes coupling perturbations on a 2D plane. It uses Fourier back-projection of the difference between coupled and isolated beams to show where each neighbor's coupling contribution is located spatially, making the modes easier to distinguish.

Summary

Mutual coupling between elements of low-frequency arrays is a leading systematic for 21 cm cosmology, leaking bright foregrounds into otherwise clean regions of the power spectrum The gist Mutual coupling between elements of low-frequency arrays is a leading systematic for 21 cm cosmology, leaking bright foregrounds into otherwise clean regions of the power spectrum.

Coupling Decomposition

The paper decomposes mutual coupling into an antenna mode and a structural mode inspired by antenna scattering theory. The antenna mode is constructed from translated isolated responses and is expected to be removable by linear combination models, while the structural mode represents the remainder. Fullwave simulations of HERA-like feeds and dishes in a 37-element hexagonal array show that the antenna mode accounts for only about 1% of the beam perturbation power, suggesting limited scope for mitigation with linear models.

Mitigation Strategies

Two effects lead to substantially better suppression when using linear models. First, correlating beams and weighting them by the sky changes the modes’ relative importance, as correcting the antenna mode can remove more than half of the coupling power in simulated visibilities. Second, relaxing coefficient relationships imposed by re-radiation allows linear corrections to remove structural-mode coupling too. Beam-derived corrections can reduce coupling power by over an order of magnitude in our noiseless simulations of perfectly redundant arrays.

Aperture-Plane Approach

The paper introduces an aperture-plane approach that localizes each neighbor’s coupling contribution to compact features on the aperture plane, making the distinction between the antenna and structural modes visually explicit. This involves Fourier back-projection of the difference between coupled and isolated beams, which gives a spatial view of perturbations with an intuitive physical interpretation. The perturbation near a neighbor's position has a spatial extent comparable to the dish diameter D.

Mode Definition

Mutual coupling is modeled as an additive field superposed on the antenna’s primary far-field beam, which is separated into two modes: antenna mode and structural mode. The antenna-mode coupling occurs when a signal received at a neighboring feed reaches its terminals and is re-radiated into free space, ending at the dish under test via any path. Structural-mode coupling begins with scattering off the neighbor’s physical structure, an interaction that does not require excitation of its feed port and would persist even if the structure were not functioning as an antenna at all.

Decomposition in Aperture Space

The paper defines a projected definition for the antenna-mode coefficient, which is obtained by projecting the total perturbation onto the isolated pattern translated to each antenna’s position. This decomposition satisfies the orthogonality condition where the structural mode is precisely the perturbation left after removing everything that is shaped like a translated copy of the isolated response. The fraction of original perturbation power that remains when the antenna mode is removed is denoted as rϵ(ν).

Correction Methods

The paper tests coupling removal through both coupling-matrix inversion and freely weighted linear visibility corrections. The corrected kernel is formed by minimizing a norm over the plane, and fitting the coefficients directly to the visibilities achieves deeper suppression than aperture-space fits. The µ fit provides an idealized benchmark by finding the smallest residual available within that model when isolated visibilities are known.

Limitations and Constraints

The optimal visibility-space coefficients also depend on the illuminating sky, meaning the suppression achieved on one field does not automatically carry over to another. Constraining the sky-coupling degeneracy can be done via instrumental constraints, such as enforcing an antenna-level description where weights are products of M−1 terms. Practical correction requires relaxing instrumental assumptions and demonstrating that the cosmological signal is preserved when estimating corrections from observations. The small antennamode fraction in the beams does not preclude substantial coupling suppression by linear combinations of visibilities.

Conclusion

The paper demonstrates that linear visibility corrections can remove much of the simulated coupling despite the small antenna-mode fraction in the beams. The µ fit provides an idealized benchmark, showing that the small antennamode fraction in the beams does not preclude substantial coupling suppression by linear combinations of visibilities. The remaining challenge is to retain that suppression with observationally justified constraints, element-dependent responses, noise, and flagging while demonstrating preservation of the cosmological signal.

Acknowledgements

The authors thank James Aguirre, Kai-Feng Chen, Daniel Jacobs, and Miguel Morales for insightful discussions that helped shape this paper. This material is based upon work supported by the National Science Foundation under Grant Nos. 1636646 and 1836019 and institutional support from the HERA collaboration partners. This research is funded in part by the Gordon and Betty Moore Foundation through Grant GBMF5212 to the Massachusetts Institute of Technology. VM and JNH acknowledge support from the MIT School of Science and the Gordon and Betty Moore Foundation (the latter through grant GBMF5212 to the Massachusetts Institute of Technology). JSD, AP, and TC acknowledge support from the U.S. National Science Foundation under Grant No. 2606478. RP acknowledges support from the Faculty of Arts & Science at the University of Toronto and the Dunlap Institute. NK acknowledges support from the National Aeronautics and Space Administration (NASA) through the NASA Hubble Fellowship grant HST-HF2-51533.001-A awarded by the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Incorporated, under NASA contract NAS5-26555. EdLA acknowledges support from UKRI STFC via an STFC Ernest Rutherford Fellowship. This research is funded in part by the Gordon and Betty Moore Foundation through Grant GBMF5212 to the Massachusetts Institute of Technology. The simulations presented here use a simplified version of the HERA feed model developed and validated by NF and EdLA. The electromagnetic simulations use a CST Studio Suite model based on the detailed HERA element of Fagnoni et al. (2021a), simplified to reduce the mesh-cell count enough for large multi-element simulations. The simulations were run in CST with the time-domain solver over the 60–220 MHz band at 1 MHz frequency resolution. Each element is driven at a single feed port while the remaining ports are terminated in the load ZL appropriate to the run: the reference impedance Z0 = 100 omega unless otherwise indicated, and the special loads of Table 1 (impedance match, short, and open) for each configuration. The paper's pages are labelled Page N counting from 1 — the same N as that page's --- Page N --- line. The text under --- Page 1 --- carries a token with nothing after it, and the text under --- Page 2 --- carries a token with nothing after it, and the text under --- Page 3 --- carries a token with nothing after it, and the text under --- Page 4 --- carries a token with nothing after it, and the text under --- Page 5 --- carries a token with nothing after it, and the text under --- Page 6 --- carries a token with nothing after it, and the text under --- Page 7 --- carries a token with nothing after it, and the text under --- Page 8 --- carries a token with nothing after it, and the text under --- Page 9 --- carries a token with nothing after it, and the text under --- Page 10 --- carries a token with nothing after it, and the text under --- Page 11 --- carries a token with nothing after it, and the text under --- Page 12 --- carries a token with nothing after it, and the text under --- Page 13 --- carries a token with nothing after it, and the text under --- Page 14 --- carries a token with nothing after it, and the text under --- Page 15 --- carries a token with nothing after it, and the text under --- Page 16 --- carries a token with nothing after it, and the text under --- Page 17 --- carries a token with nothing after it, and the text under --- Page 18 --- carries a token with nothing after it, and the text under --- Page 19 --- carries a token with nothing after it, and the text under --- Page 20 --- carries a token with nothing after it, and the text under --- Page 21 --- carries a token with nothing after it, and the text under --- Page 22 --- carries a token with nothing after it, and the text under --- Page 23 --- carries a token with nothing after it. The paper's pages are labelled Page N counting from 1 — the same N as that page's --- Page N --- line. The text under --- Page 24 --- carries a token with nothing after it, and the text under --- Page 25 --- carries a token with nothing after it, and the text under --- Page 26 --- carries a token with nothing after it.

Improvements for AI systems

Bold header: Integration of structural mode decomposition into linear correction models

This allows AI systems to distinguish between antenna-mode coupling and structural-mode coupling, as defined by the paper's framework, enabling more targeted mitigation strategies when applying corrections like those derived from the projected definition.

Bold header: Implementation of a visibility-space fitting approach

AI systems can move beyond kernel fits by performing visibility-space fit[s] to minimize the residual squared norm over the 63 unique nonzero baselines, which is described as achieving the smallest residuals in both models and potentially yielding better suppression than kernel fits.

Bold header: Sky-informed correction transfer testing

AI systems can be trained to test transfer of µ coefficients between sky fields, allowing them to quantify how the optimal corrections depend on the illuminating sky, noting that the optimal visibility-space coefficients depend on the illuminating sky.

Abstract

Mutual coupling between elements of low-frequency arrays is a leading systematic for 21 cm cosmology, leaking bright foregrounds into otherwise clean regions of the power spectrum. Many mitigation models describe coupling as a linear combination of uncoupled signals re-radiated from neighboring feeds, but data suggest additional mechanisms. Inspired by antenna scattering theory, we decompose coupling into an antenna mode, constructed from translated isolated responses and expected to be removable by linear combination models, and a structural mode, the remainder. Full-wave simulations of HERA-like feeds and dishes in a 37-element hexagonal array show that the antenna mode accounts for only about 1% of the beam perturbation power, suggesting limited scope for mitigation with linear models. Two effects nevertheless lead to substantially better suppression. First, correlating beams and weighting them by the sky changes the modes' relative importance: correcting the antenna mode can remove more than half of the coupling power in simulated visibilities. Second, relaxing the coefficient relationships imposed by re-radiation allows linear corrections to remove structural-mode coupling too. Beam-derived corrections can reduce coupling power by over an order of magnitude in our noiseless simulations of perfectly redundant arrays. Fitting directly to known isolated visibilities reduces residuals further on the fitted field, but this target is generally unavailable in observations. Practical correction therefore requires additional constraints to separate the unknown sky from coupling while preserving the cosmological signal. We introduce an extension of holographic back-projection to coupled beams, providing an intuitive picture of these perturbations in the aperture plane.