{"id":"58ec59da-0a2e-4081-ab48-c2618dcb1c8a","arxiv_id":"2412.13683","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"MMSE channel estimation in holographic MIMO suffers large high-SNR losses when mutual coupling is ignored; LS estimation avoids this but needs high SNR.","lead":"This paper studies how mutual coupling between densely packed antennas changes channel estimation in holographic MIMO systems. It shows that MMSE estimators that ignore coupling lose accuracy sharply at high SNR, while least-squares estimators remain robust but need high SNR.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Equations (9)-(11) are only a valid covariance model if C=(Z+R_d I)^-1 is Hermitian PSD; for thin dipoles Z is complex symmetric, so C^{1/2}RC^{1/2} is not a channel covariance and the Fig. 1 dB-loss claims are ungrounded.","rationale":"The reader's weakest-assumption diagnosis is exactly the load-bearing issue: Equations (9)-(11) use C^{1/2}RC^{1/2} as a channel covariance, which requires C^{1/2} to be Hermitian. For reciprocal dipole arrays, Z is complex symmetric, so (Z+R_d I)^{-1} is generally not Hermitian and the numerical curves in Fig. 1 depend on an invalid covariance. This is an internal consistency problem, not a disagreement with the field's consensus. The paper's closed-form isotropic correlation derivation in Proposition 1 and the LS-resilience argument are reasonable, and the MMSE mismatch reasoning would remain qualitatively plausible under the corrected model T R T^H. The right response is therefore to keep the conditional verdict and require the authors to either prove the needed PSD/Hermitian property for their specific coupling matrix or redo the analysis with the physically correct covariance. I do not find a stronger attack: the remaining algebra is mostly consistent and the claims are clearly stated, so no rejection beyond the condition is warranted.","tokens_in":9145,"tokens_out":13190,"duration_ms":121606,"concrete_test":"Regenerate Fig. 1 with the physically consistent coupling model: set T=(Z+R_d I)^{-1} (or the normalized Z_L(Z+Z_L I)^{-1} form from [10]), define h_mc = T h and R_mc = T R T^H, using the same Balanis mutual-impedance expressions and the 10x10 lambda/5 UPA geometry. Compare the resulting MSE at 10 and 20 dB SNR with the values reported in Sections IV-A and IV-B. If the corrected losses differ from the paper's approximately 8/16 dB and 12/19 dB by more than a few dB, the quantitative central claim is an artifact of the invalid square-root covariance model; if they match, the concern is cosmetic.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Equations (9)-(11) define the coupled channel as h_mc = C^{1/2}h with C=(Z+R_d I)^{-1} and covariance R_mc = C^{1/2} R C^{1/2}. This is a valid covariance transformation only if C^{1/2} is Hermitian; otherwise the correct covariance of C^{1/2}h is C^{1/2} R (C^{1/2})^H. For an array of thin dipoles, Z is complex symmetric, not Hermitian, and adding the scalar dissipation resistance R_d does not make (Z+R_d I)^{-1} Hermitian or positive semidefinite in general. The paper never states or proves the required property. The consequences are internal, not merely a parameter issue: Proposition 2's proof uses (C^{1/2}R^{1/2})(C^{1/2}R^{1/2})^T, which equals R_mc only if C^{1/2} is symmetric, and Proposition 3's eigen-decomposition assumes Hermitian R_mc and W. If the physically consistent model is h_mc = T h with T=(Z+R_d I)^{-1} and R_mc = T R T^H, then the Fig. 1 curves and the quoted 8/16 dB (isotropic) and 12/19 dB (clustered) losses are computed from the wrong covariance and need to be redone. The qualitative conclusion that MMSE should account for mutual coupling may survive the correction, but the paper as written does not establish it.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies uplink channel estimation in a holographic MIMO system whose base station is a uniform planar array of closely spaced half-wavelength dipoles. It models spatially correlated Rayleigh fading with covariance R, introduces mutual coupling through h_mc = C^{1/2}h and R_mc = C^{1/2}RC^{1/2} with C=(Z+R_d I)^{-1}, and compares the LS estimator with MMSE-structured estimators that use different assumed covariance matrices: the true R_mc, a coupling-aware isotropic covariance C^{1/2}R_isoC^{1/2}, and a coupling-ignoring isotropic covariance R_iso. The paper derives a closed-form isotropic spatial correlation expression, column-space containment results for the estimators, and eigen-expressions for the MSE, and reports numerical MSE curves claiming that ignoring mutual coupling costs roughly 8 and 16 dB at 10 and 20 dB SNR in isotropic scattering and 12 and 19 dB in clustered scattering, while LS is resilient to coupling ignorance but useful mainly at high SNR.","tokens_in":9561,"tokens_out":15612,"duration_ms":131453,"significance":"If the results were correct, the paper would provide concrete design guidance: a holographic MIMO receiver using an MMSE-structured estimator must know or estimate the mutual coupling matrix, while LS provides a coupling-robust fallback at high SNR. The paper also contributes a closed-form, parameter-free spatial correlation series for isotropic scattering with dipole elements, and the numerical study is reproducible from the stated model and parameters with no fitted constants. However, the central quantitative conclusions rest on a mutual-coupling covariance model that is not justified, and one of the main MSE formulas contains an algebraic inconsistency. The qualitative message may survive a corrected treatment, but the paper as written does not establish its headline dB-loss claims.","major_comments":[{"comment":"The model h_mc = C^{1/2}h and R_mc = C^{1/2}RC^{1/2} with C=(Z+R_d I)^{-1} is a valid covariance model only if C is Hermitian positive semidefinite, so that C^{1/2} can be taken as a Hermitian square root and R_mc is the covariance of C^{1/2}h. The paper never states or proves this property. For the thin-dipole mutual impedances cited in [7], Z is complex symmetric with nonzero imaginary parts, so (Z+R_d I)^{-1} is generally not Hermitian; consequently C^{1/2}RC^{1/2} need not be Hermitian and is not the covariance of C^{1/2}h. The correct covariance would be R_mc = T R T^H with T=(Z+R_d I)^{-1}. This issue propagates to Appendix B, where the equality R_mc = C^{1/2}R^{1/2}(C^{1/2}R^{1/2})^T is asserted, and to the eigen-decomposition used in Proposition 3 and Corollary 1. Since all numerical curves in Fig. 1 are generated from this covariance model, the reported 8/16 dB and 12/19 dB losses are not grounded. The authors must either prove the required positive semidefiniteness under their antenna assumptions or re-derive the results using the physical covariance R_mc = T R T^H and rerun the numerical study.","section":"II-B, Eqs. (9)-(11)"},{"comment":"Equation (19) is not consistent with Proposition 3 as written. In Proposition 3, lambda_{w,k} denotes an eigenvalue of the estimator matrix W. For the MMSE-structured estimator W = sqrt(rho) \\hat{R}(rho\\hat{R}+I)^{-1}, substituting lambda_w = sqrt(rho)\\hat{lambda}/(rho\\hat{lambda}+1) into the beta expression of Proposition 3 does not yield Eq. (19) for an arbitrary \\hat{R}. For example, in the matched case \\hat{R}=R_mc with rho=1 and lambda_h=1, Proposition 3 gives beta = -0.5, while Eq. (19) gives -4/9. The formula appears to mix eigenvalues of W with eigenvalues of \\hat{R}. Because the numerical MSE values in Fig. 1 are likely computed from this formula or from the associated eigen-expansion, the reported quantitative losses are not substantiated until Eq. (19) is corrected and the curves are regenerated.","section":"III-C, Corollary 1, Eq. (19)"},{"comment":"The closed-form spatial correlation formula in Eq. (5) omits the wavelength normalization that is explicitly introduced in Appendix A, where d_y^{nm} = (r_y(n)-r_y(m))d_y/lambda and d_z^{nm} = (r_z(n)-r_z(m))d_z/lambda. If d_y and d_z in Eq. (5) are meant to be normalized by the wavelength, this should be stated; otherwise the formula is inconsistent with its own derivation. This is relevant because R_iso is used both in Proposition 2 and in the numerical experiments, so an unstated normalization convention can change the correlation matrix and the resulting MSE curves.","section":"II-A, Prop. 1 and Appendix A"}],"minor_comments":[{"comment":"The first scenario writes \\hat{R} = C^{1/2}R^{1/2}C^{1/2}, but the surrounding text says this is the true MMSE estimator, which would require \\hat{R}=R_mc=C^{1/2}RC^{1/2}. The notation should be corrected to avoid ambiguity.","section":"III-A, Proposition 2"},{"comment":"Proposition 3 is stated for an arbitrary linear estimator W, but its eigen-expansion with orthonormal eigenvectors and the inner-product term |<u_{w,k},u_{hmc,l}>|^2 is only valid when W has an orthonormal eigenbasis, for example when W is Hermitian. The statement should be qualified accordingly.","section":"III-C, Proposition 3"},{"comment":"Reference [8] is listed with year 2006, but the cited book by Björnson and Demir appears to be a recent publication; the year should be checked and corrected.","section":"References"},{"comment":"The text says the LS estimator exhibits degradation and then says the gap narrows; consider rephrasing for clarity. Also, the figure caption does not specify whether the curves come from Monte Carlo simulation or from the analytical MSE expressions, which should be stated.","section":"IV-A, Fig. 1"}],"recommendation":"major_revision","confidential_remarks":"The manuscript fits the journal's scope and the conceptual framing is useful, but the central quantitative claims depend on an unjustified covariance model and on an algebraic error in Corollary 1. These issues are fixable within the manuscript's scope by re-deriving the physical coupling model and recomputing the numerical results, so I do not recommend rejection; however, the current version should not be accepted without those corrections."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe short version: the paper is worth reading for its new closed-form spatial correlation expression for half-wavelength dipoles under isotropic scattering (Prop. 1), but the mutual coupling model in Eqs. (9)–(11) is mathematically invalid for the physical setup, and the paper's main quantitative claims lean on it.\n\nWhat's new and good: Prop. 1 is a genuine extension of the correlation model in [3], and the derivation in Appendix A checks out. The conceptual comparison between MMSE-structured estimators that do or do not account for coupling is the right question for holographic MIMO, and the paper is cleanly written. There are no fitted constants; the figures come from a fixed channel model, so the circularity burden is low.\n\nThe soft spot is not soft: Eq. (9) defines hmc = C^{1/2}h with C = (Z + R_d I)^{-1}. For a dipole array, Z is a complex symmetric matrix, not Hermitian. Adding a scalar dissipation resistance R_d preserves that: C is complex symmetric, generally non-Hermitian, and not positive semidefinite in the matrix sense. Consequently, a Hermitian square root C^{1/2} does not exist in the usual sense, and even if you take a principal square root (which is complex symmetric), the covariance of hmc is C^{1/2} R (C^{1/2})^H, not C^{1/2} R C^{1/2}. The paper's Eq. (10) is simply not a valid covariance matrix. This is load-bearing: Prop. 2's subspace argument and Prop. 3's eigen-decomposition both assume Rmc is Hermitian positive semidefinite, and the MSE losses quoted for Fig. 1 are computed from a matrix that isn't the covariance of the channel. The alternative—use T = C^{1/2} (or the standard circuit-theory coupling matrix T = C) and form T R T^H—is the physically consistent fix. The qualitative conclusion that MMSE needs coupling awareness will probably survive that fix, but the dB numbers in this version are ungrounded.\n\nMinor issues: the exact dipole geometry (length, radius) used for Z is not stated, so the curves can't be reproduced; and the notation in Cor. 1 is a little sloppy (lambda_w vs eigenvalues of R_hat). Both are fixable.\n\nBottom line: This paper deserves a serious referee—the question matters and Prop. 1 is a useful standalone result—but as written the central statistical model is wrong. I'd want the authors to redo the analysis with the correct covariance transformation before trusting any of the performance claims. If they do, the paper could be solid.\n\nRecommendation: accept for peer review with a clear flag that the covariance model needs a major revision.","headline":"Useful closed-form correlation result and an important question, but the mutual coupling covariance model is not a valid covariance for dipole arrays, so the quantitative claims in this version are ungrounded.","tokens_in":10045,"tokens_out":8486,"would_cite":true,"duration_ms":69016,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Mutual coupling, not user-specific scattering, is the prior a dense-array receiver must know to estimate channels accurately.","keywords":["holographic MIMO","mutual coupling","channel estimation","MMSE estimator","least squares estimator","spatial correlation","dipole antenna array","covariance matrix"],"falsifier":"Build or simulate the 10x10 $\\lambda/5$ dipole array, compute $\\mathbf{C}$ from the mutual-impedance formulas, test whether $\\mathbf{C}$ is Hermitian positive semidefinite, and compare the measured MMSE gap against the predicted 8 and 16 dB losses; a non-PSD matrix or a missing gap would refute the central numerical claim.","tokens_in":1871,"feed_emoji":"📡","tokens_out":2730,"duration_ms":76907,"temperature":0.7,"pith_summary":"The paper tries to establish which statistical knowledge a base station actually needs to estimate channels in holographic MIMO, where antennas are packed much closer than half a wavelength. Its central finding is that mutual coupling between antennas is a first-order effect for the MMSE estimator: an estimator that ignores coupling and uses only the array-geometry covariance loses roughly 8 and 16 dB in MSE at 10 and 20 dB pilot SNR in isotropic scattering, and about 12 and 19 dB in clustered scattering. Knowing the user-specific spatial correlation is much less important, costing about 4 dB when coupling is already known. The least-squares estimator needs no coupling or correlation statistics and is therefore robust, but it only becomes competitive at high SNR. If true, this redirects practical holographic MIMO design toward coupling calibration rather than elaborate scattering modeling.","feed_headline":"Ignoring antenna coupling costs 8-16 dB in dense MIMO estimation","feed_subtitle":"MMSE estimators need the mutual-coupling matrix more than the user's scattering statistics; LS works only at high SNR.","key_machinery":"The central object is the mutual-coupling matrix $\\mathbf{C} = (\\mathbf{Z} + R_d\\mathbf{I})^{-1}$, which enters the channel model as $\\mathbf{h}_{\\mathrm{mc}} = \\mathbf{C}^{1/2}\\mathbf{h}$ and transforms the coupling-free covariance into $\\mathbf{R}_{\\mathrm{mc}} = \\mathbf{C}^{1/2}\\mathbf{R}\\mathbf{C}^{1/2}$. This matrix does the work of changing the subspace in which plausible channel estimates live: Proposition 2 shows that an estimator built on $\\mathbf{R}_{\\mathrm{iso}}$ spans only the isotropic column space, while one built on $\\mathbf{C}^{1/2}\\mathbf{R}_{\\mathrm{iso}}\\mathbf{C}^{1/2}$ spans the physically possible coupled dimensions. The eigenbasis MSE formula in Proposition 3 and Corollary 1 then quantifies the penalty as an SNR-dependent mismatch between the estimator eigenbasis and the true channel eigenbasis, which is why ignoring coupling hurts most at high SNR.","core_discovery":"On its own terms, the paper demonstrates that the effective channel in a dense array is $\\mathbf{h}_{\\mathrm{mc}} = \\mathbf{C}^{1/2}\\mathbf{h}$, with covariance $\\mathbf{R}_{\\mathrm{mc}} = \\mathbf{C}^{1/2}\\mathbf{R}\\mathbf{C}^{1/2}$, and that an MMSE-structured estimator using $\\hat{\\mathbf{R}} = \\mathbf{R}_{\\mathrm{iso}}$ instead of $\\mathbf{R}_{\\mathrm{mc}}$ projects the received signal onto the wrong subspace. The numerical consequence is an SNR-dependent MSE penalty: about 8 and 16 dB at 10 and 20 dB SNR under isotropic scattering, and about 12 and 19 dB under clustered scattering. The paper also shows that substituting the user-specific correlation with the isotropic correlation costs only about 4 dB once coupling is known, while the least-squares estimator avoids coupling information altogether but pays a large low-SNR penalty. The conclusion is that mutual coupling awareness is essential for MMSE channel estimation in holographic MIMO, and that spatial correlation knowledge is secondary.","pith_inferences":["A natural next step the paper leaves implicit is to reuse a single calibration of $\\mathbf{C}$ across all served users, since coupling is a property of the array rather than of any propagation environment.","The roughly 4 dB cost of replacing user-specific correlation with isotropic correlation suggests that, once coupling is known, coarse scattering knowledge may be enough in dense arrays; this extrapolates from two scattering models to a design rule.","Because the paper only treats uplink pilot estimation, the same coupling-aware covariance logic should carry over to downlink precoder design under TDD reciprocity, but that transfer is not proven here.","A direct experimental check on a small fabricated array of $\\lambda/5$ dipoles could confirm whether the predicted 8 and 16 dB MSE gaps appear as the array geometry is held fixed and only the scattering model changes."],"forward_implications":["A holographic MIMO receiver that wants MMSE-quality channel estimates must know or estimate the mutual-coupling matrix $\\mathbf{C}$; array-geometry information alone is not enough.","At low pilot SNR the penalty for ignoring coupling is small, so simpler coupling-agnostic estimators can be used in that regime without much loss.","Exchanging user-specific channel statistics for isotropic statistics costs roughly 4 dB once coupling is known, so the more important prior to acquire is the coupling matrix.","The LS estimator is a viable coupling-agnostic fallback, but only when the SNR is high enough that its $M/\\rho$ error floor is acceptable.","Channel estimation algorithms for dense arrays should be designed around coupling-aware covariance models rather than purely geometry-based models."],"supporting_citations":[{"why":"Supplies the spatial correlation integral, the isotropic-subspace property, and the clustered-scattering closed-form approximation used in the numerical comparisons.","marker":"[3]"},{"why":"Supplies the correlated Rayleigh model $\\mathbf{h} = \\mathbf{R}^{1/2}\\mathbf{h}_{\\mathrm{iid}}$ and the correlation integral that the paper extends to a dipole uniform planar array.","marker":"[6]"},{"why":"Supplies the dipole directivity model $D = 1.67\\cos^3\\theta$ and the closed-form mutual impedance expressions used to build $\\mathbf{C}$ and $\\mathbf{R}_{\\mathrm{iso}}$.","marker":"[7]"},{"why":"Supplies the circuit-theory coupling model $\\mathbf{h}_{\\mathrm{mc}} = \\mathbf{C}^{1/2}\\mathbf{h}$ that defines the effective channel.","marker":"[9]"},{"why":"Supplies the expression $\\mathbf{C} = (\\mathbf{Z} + R_d\\mathbf{I})^{-1}$ for the coupling matrix with internal antenna losses.","marker":"[10]"},{"why":"Supplies the MMSE and LS estimator structures and the LS error floor $M/\\rho$ used in the analysis.","marker":"[11]"},{"why":"Supplies the urban macro-cell cluster powers and angle spreads used to generate the non-isotropic scattering scenario.","marker":"[12]"},{"why":"Supplies the trigonometric integral identities used to evaluate the closed-form isotropic correlation in Proposition 1.","marker":"[13]"},{"why":"Supplies the linear algebra fact that the eigenvectors of $\\mathbf{R}_{\\mathrm{mc}}$ span the column space of $\\mathbf{C}^{1/2}\\mathbf{R}^{1/2}$, used in Proposition 2.","marker":"[14]"}],"fun_headline_variants":["Mutual coupling costs 8-16 dB in dense MIMO estimation","For MMSE, coupling beats user correlation in holographic MIMO","Dense MIMO: ignore coupling, lose up to 19 dB in estimation","Mutual coupling awareness: the key to holographic MIMO estimation","MMSE needs coupling matrix; LS only shines at high SNR"],"cache_read_input_tokens":12160,"weakest_assumption_plain":"The numerical results assume the mutual-coupling matrix $\\mathbf{C} = (\\mathbf{Z} + R_d\\mathbf{I})^{-1}$ is a valid covariance matrix with a square root; the paper states this model but never proves that the inverse of an impedance matrix plus dissipation resistance is Hermitian positive semidefinite.","fun_headline_variants_meta":{"raw":{"variants":["Mutual coupling costs 8-16 dB in dense MIMO estimation","For MMSE, coupling beats user correlation in holographic MIMO","Dense MIMO: ignore coupling, lose up to 19 dB in estimation","Mutual coupling awareness: the key to holographic MIMO estimation","MMSE needs coupling matrix; LS only shines at high SNR"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000167,"raw_usage":{"total_tokens":1239,"prompt_tokens":907,"completion_tokens":332,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":523,"completion_tokens_details":{"reasoning_tokens":238}},"tokens_in":523,"tokens_out":332,"duration_ms":3587,"temperature":1.0,"reasoning_tokens":238,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T12:55:01.740537+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Build or simulate the 10x10 $\\lambda/5$ dipole array, compute $\\mathbf{C}$ from the mutual-impedance formulas, test whether $\\mathbf{C}$ is Hermitian positive semidefinite, and compare the measured MMSE gap against the predicted 8 and 16 dB losses; a non-PSD matrix or a missing gap would refute the central numerical claim.","supporting_citations":[{"cited_title":"Channel m odeling and channel estimation for holographic massive MIMO with plana r arrays,","cited_arxiv_id":null,"evidence_quote":"Supplies the spatial correlation integral, the isotropic-subspace property, and the clustered-scattering closed-form approximation used in the numerical comparisons."},{"cited_title":"Rayleigh fading model ing and channel hardening for reconfigurable intelligent surfaces,","cited_arxiv_id":null,"evidence_quote":"Supplies the correlated Rayleigh model $\\mathbf{h} = \\mathbf{R}^{1/2}\\mathbf{h}_{\\mathrm{iid}}$ and the correlation integral that the paper extends to a dipole uniform planar array."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the dipole directivity model $D = 1.67\\cos^3\\theta$ and the closed-form mutual impedance expressions used to build $\\mathbf{C}$ and $\\mathbf{R}_{\\mathrm{iso}}$."},{"cited_title":"Toward a circuit theory of communi- cation,","cited_arxiv_id":null,"evidence_quote":"Supplies the circuit-theory coupling model $\\mathbf{h}_{\\mathrm{mc}} = \\mathbf{C}^{1/2}\\mathbf{h}$ that defines the effective channel."},{"cited_title":"Holographic MIMO com munica- tions: What is the benefit of closely spaced antennas?","cited_arxiv_id":null,"evidence_quote":"Supplies the expression $\\mathbf{C} = (\\mathbf{Z} + R_d\\mathbf{I})^{-1}$ for the coupling matrix with internal antenna losses."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the MMSE and LS estimator structures and the LS error floor $M/\\rho$ used in the analysis."},{"cited_title":"Guidelines for evaluation of radio interfa ce technologies for IMT-2020,","cited_arxiv_id":null,"evidence_quote":"Supplies the urban macro-cell cluster powers and angle spreads used to generate the non-isotropic scattering scenario."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the trigonometric integral identities used to evaluate the closed-form isotropic correlation in Proposition 1."},{"cited_title":"Strang, Linear Algebra and Its Applications, 4th ed","cited_arxiv_id":null,"evidence_quote":"Supplies the linear algebra fact that the eigenvectors of $\\mathbf{R}_{\\mathrm{mc}}$ span the column space of $\\mathbf{C}^{1/2}\\mathbf{R}^{1/2}$, used in Proposition 2."}],"review_version":1}