{"id":"b0819f8f-0165-4027-ad02-7d8fc4e04632","arxiv_id":"2505.09767","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":7,"one_line_summary":"Includes near-field correlation and mutual coupling in mixture-Gamma RIS channel models and shows they materially improve THz channel-estimation NMSE.","lead":"This paper models THz reconfigurable-intelligent-surface (RIS) channels with a mixture-Gamma fading distribution, near-field spatial correlation, and mutual coupling, then evaluates least-squares and LMMSE channel estimators. The results show that including these physical effects can improve channel-estimation accuracy by several decibels in large-array, short-range links.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (16) does not preserve the MG distribution under linear mixing, so the derived cascaded PDF (20) cannot describe the simulated channel; the MG-correlated model is internally inconsistent.","rationale":"The paper's strongest claim is that accounting for near-field spatial correlation and MC improves estimation accuracy by several dBs. Those gains are computed through Rcc, which depends only on second-order statistics, so the invalid marginal-distribution assumption does not by itself destroy the dB comparison. However, the paper's stated contributions include deriving the distribution of the cascaded channel and modeling THz fading as MG. Eq. (16) is the only bridge between the physical multi-ring model and the Kronecker MG model, and it is mathematically incorrect for non-diagonal R: the MG class is not closed under linear combinations. Thus the theoretical PDF (20) and variance (22) do not describe the simulated channel, and the phrase 'physically consistent channel modeling' is not supported. A goodness-of-fit test on the generated marginals would settle this. This is exactly the reader's weakest assumption, and the conditional verdict is appropriate: the distributional claims require correction or clarification, even though the covariance-based estimation results may survive.","tokens_in":14747,"tokens_out":11447,"duration_ms":124018,"concrete_test":"Generate \\tilde h with IID MG entries using the measurement-based parameters of [14] and construct h = R^{1/2}\\tilde h for the near-field correlation matrix R_NF of Eq. (11) with N=128. Over 10^5 realizations, compare the empirical marginal distribution of |h_i| to the assumed MG PDF (4) using a Kolmogorov-Smirnov test, and compare the empirical cascaded channel distribution to (20). If the KS test rejects the MG fit (expected for non-diagonal R), Eq. (16) does not preserve MG and the theoretical PDF is not the simulated channel's distribution.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The Kronecker construction in Eq. (16), h = R^{1/2} \\tilde h, assumes that a linear transformation of an IID vector of MG-distributed entries is again a vector of MG-distributed entries. For Gaussian vectors this is true, but for a finite Gamma mixture it is not: each output entry is a weighted sum of independent Gamma-mixture variables, and such a sum is not generally itself a Gamma mixture. Consequently, h_RU in (17) and H_BR in (19) do not have MG marginals, so the product-of-two-MG argument invoking [16, Theorem 1] does not apply, and the cascaded PDF (20) and variance expression (22) are not the distribution of the channel actually simulated. The paper's headline dB gains are driven by second-order statistics (Rcc), which may still be computed correctly, but the paper's advertised 'physically consistent MG channel modeling' and its distributional derivation are invalid as written. The simulation generator is under-specified and may be producing channels whose marginal statistics differ from the claimed MG model.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a THz-band RIS-aided channel model in which small-scale fading follows the mixture gamma (MG) distribution, with near-field spatial correlation, mutual coupling at the base station, molecular absorption, and spherical wave propagation. The authors construct correlated MG vectors via a Kronecker square-root transformation, derive a cascaded RIS channel distribution using a product-of-MG result from prior work, and evaluate LS and LMMSE channel estimators. Numerical results claim that accounting for near-field correlation and mutual coupling improves estimation accuracy by several dB compared with models that neglect these effects.","tokens_in":15045,"tokens_out":9191,"duration_ms":98596,"significance":"If the modeling and derivations were correct, the paper would provide a useful framework for THz RIS channel estimation in ultra-massive arrays, combining measured MG fading parameters with near-field correlation and mutual coupling. A strength is the use of measurement-based MG parameters from prior work and the incorporation of established mutual-coupling and von Mises angular-spread models. However, the central distributional construction in Eq. (16) is not justified for MG variables, and the Monte Carlo generator is not specified, so the quantitative performance claims are not substantiated as written.","major_comments":[{"comment":"The transformation h = R^{1/2} \\tilde{h} is used to generate a correlated MG channel vector from an IID MG vector. For Gaussian vectors this operation preserves the distribution family, but for MG (finite Gamma-mixture) variables it does not: each output entry is a weighted sum of independent Gamma-mixture variables, and such a sum is not generally itself a Gamma mixture. Consequently, h_RU in (17) and H_BR in (19) do not have MG-distributed entries, and the invocation of [16, Theorem 1] to derive the cascaded PDF in (20) is invalid. The theoretical NMSE curves based on the variance expression in (22) therefore do not describe the channel that is actually simulated unless a different, unspecified generation procedure is used.","section":"Sec. III-B, Eq. (16)"},{"comment":"The physical channel in Eq. (9) is a complex sum of random-phase phasors with distance-dependent magnitudes and absorption phases. The Kronecker construction in Eq. (16) instead produces a real (or at least phase-free) linear combination of nonnegative MG variables. This construction cannot reproduce the phase statistics of Eq. (9) or the distance-dependent phase differences that enter the near-field correlation matrix R_NF in (11). The paper does not explain how Eq. (16) is intended to correspond to Eq. (9); the relationship between the physical model and the Kronecker model needs to be established, or the paper should state clearly that Eq. (16) is an independent statistical model.","section":"Sec. III-A and III-B, Eqs. (9)-(16)"},{"comment":"The covariance matrix R_cc is quoted from [19] rather than derived for the proposed MG model. For C = H_BR diag(h_RU) with independent H_BR and h_RU, the exact covariance under a Kronecker vec ordering is (R_RB ⊙ E[h_RU h_RU^H]) ⊗ R_BR; if h_RU has nonzero mean, E[h_RU h_RU^H] includes the mean contribution. Since the MG entries defined in Eq. (4) are nonnegative and hence have positive mean, using a spatial correlation matrix alone in (27) is not the correct LMMSE covariance. The derivation of (27) from the proposed channel model should be provided, including which matrix is intended to represent E[h_RU h_RU^H].","section":"Sec. IV, Eq. (27)"},{"comment":"The Monte Carlo simulation setup is under-specified: the paper states that 1000 trials are used and that measurement-based MG parameters from [14] are adopted, but it does not describe the actual channel generation algorithm, the drawn values of the MG parameters, the quadrature order N, or how the spatial correlation and mutual coupling are applied to the simulated fading realizations. Because the Theo/Sim agreement claimed in Fig. 2 depends on the exact generator, the results cannot be reproduced or checked. In addition, the performance metric is defined as NMSE = Tr(R_e)/Tr(R_cc); for the LS estimator, Tr(R_e) is independent of the channel covariance, so NMSE scales inversely with Tr(R_cc). The paper does not state whether the compared models are normalized to equal total channel power, so the reported dB gains may reflect differences in total power rather than in estimation accuracy.","section":"Sec. V, Numerical Results"}],"minor_comments":[{"comment":"The sentence \"This paper, we study\" should be \"In this paper, we study\".","section":"Sec. VI, Conclusions"},{"comment":"The paper contains several garbled equation artifacts (e.g., \"B.dsp\" in Eqs. (11), (14), and (16)) and missing mathematical symbols; these should be corrected before publication.","section":"Throughout"},{"comment":"The caption of Fig. 2c refers to the \"MG /alpha parameter\" but the x-axis is described as \"MG variance\"; the paper should clarify which quantity is plotted.","section":"Sec. V, Fig. 2c"},{"comment":"Several references are cited as arXiv preprints ([10], [12]); if final versions exist, the published citations should be used.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The central issue is the unjustified assumption that a Kronecker square-root transform preserves the MG distribution. This is not a minor technicality: it invalidates the derived cascaded PDF and the interpretation of the simulation results. I would require the authors to either prove the required closure property (which, for finite Gamma mixtures, is unlikely to hold) or replace the model with a construction that actually preserves MG marginals, and then re-derive the estimators and rerun the Monte Carlo simulations. The NMSE normalization should also be revisited. The paper's length is appropriate for a letter, but the technical gap is substantial."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Let me give you the short version: if you're looking for a model that quantifies the impact of near-field correlation and mutual coupling on THz RIS channel estimation, the numerical trends here are probably right, but the paper's statistical derivation is wrong. The authors start with an IID mixture-Gamma vector and multiply it by R^{1/2} to introduce correlation, then treat the output as MG-distributed. That is not true for Gamma mixtures—a linear combination of independent MG variables is not MG in general. So the derivation of the cascaded PDF in (20), which relies on the product of two MG variables from [16, Theorem 1], does not describe the channel they actually simulate. The Monte Carlo setup is also under-specified, so we can't tell what distribution the generated channels have.\n\nThat said, the paper is not a throwaway. It combines several pieces that haven't been put together before: MG fading for THz, near-field spherical-wave correlation from [20], mutual coupling at the BS, and LS/LMMSE estimation. The authors use measurement-based MG parameters, and the comparisons—near-field vs far-field, with vs without MC—are clearly presented and align with other results in the literature. Importantly, the covariance-based estimators (LMMSE and LS) only depend on second-order statistics. If the simulated channel is generated with the correct covariance R, the theoretical NMSE curves for those estimators would still match, even though the marginals are not MG. So the headline dB gains may survive the flaw; what dies is the 'physically consistent MG modeling' narrative and the closed-form cascaded PDF.\n\nAnother oddity: the MG distribution is for the fading magnitude, but the channel in (9) is complex. The Kronecker expression h = R^{1/2} \\tilde h with MG \\tilde h looks like it ignores phase altogether. The paper never explains how complex envelopes and phases enter the simulation. This needs to be fixed or clarified.\n\nIf I were the editor, I'd send this to reviewers. The topic is timely, the model is a natural extension, and the numerical results could be salvageable if the authors either generate correlated MG exactly (via a copula or by mixing correlated Gamma components) or explicitly restrict their claims to covariance-based analysis. It's a conditional accept in its current form.","headline":"The dB gains from near-field correlation and mutual coupling are plausible, but the paper's correlated-MG construction is distributionally invalid, so its theoretical PDF and physical-consistency claims don't hold as written.","tokens_in":15568,"tokens_out":4342,"would_cite":false,"duration_ms":44792,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Near-field effects and mutual coupling sharpen THz RIS channel estimation by several dB.","keywords":["terahertz communications","reconfigurable intelligent surface","mixture gamma fading","near-field spatial correlation","mutual coupling","channel estimation","LMMSE","massive MIMO"],"falsifier":"Run a Monte Carlo simulation that generates channel vectors as $R^{1/2} \\tilde{h}$ from independent MG coefficients and compare the empirical histogram of the resulting entries with the MG PDF used to derive the cascaded distribution (20); if the histograms disagree, the theoretical covariance and NMSE curves in the paper do not describe the simulated channel.","tokens_in":14566,"feed_emoji":"📡","tokens_out":4100,"duration_ms":38056,"temperature":0.7,"pith_summary":"This paper argues that a THz-band RIS-aided channel model that keeps near-field spherical wavefronts, spatial correlation, and mutual coupling at the base station, while describing small-scale fading with the mixture gamma distribution, leads to materially better linear channel estimation than models that drop these effects. It derives the distribution of the cascaded user-RIS-BS channel and plugs it into least-squares and linear minimum mean-square error estimators. The authors report that near-field modeling improves estimation accuracy by up to 3 dB over far-field modeling, and that ignoring mutual coupling and spatial correlation costs roughly 8 dB at high SNR. If correct, physically consistent THz channel models should include these effects rather than rely on conventional planar-wave independent fading.","feed_headline":"THz RIS estimation gains up to 3 dB from near-field model","feed_subtitle":"Including near-field correlation and mutual coupling in the channel model cuts THz RIS estimation error by several dB.","key_machinery":"The carrying object is the Kronecker-correlated mixture-gamma channel model, where a channel vector is written as $h = R^{1/2} \\tilde{h}$ with $\\tilde{h}$ IID MG fading and $R$ the spatial correlation matrix from a near-field multi-ring geometry (von Mises angular spread, spherical distances, absorption). A second piece is the mutual coupling transformation $M^{1/2} R_{BR} M^{1/2}$ with $M = (Z + r_l I)^{-1}$. The cascaded RIS channel is the entrywise product of two such MG vectors, whose PDF is obtained via Gaussian-Laguerre quadrature, providing the covariance $R_{cc}$ used by the LMMSE estimator.","core_discovery":"The central claim is that accounting for near-field spatial correlation, mutual coupling, and mixture-gamma small-scale fading in an RIS-aided THz link yields several decibels of channel-estimation gain over conventional far-field uncorrelated models. The paper constructs a geometry-based stochastic channel with multi-ring scatterers, spherical wave distances, molecular absorption, and a Kronecker correlation matrix, then expresses the cascaded channel entries as products of MG random variables. It derives the PDF of that product using Gaussian-Laguerre quadrature and uses the resulting covariance in LMMSE estimation. Simulations at 142 GHz with 128-element arrays show that near-field correlation improves NMSE by up to 3 dB, and that neglecting both mutual coupling and spatial correlation degrades performance by about 8 dB.","pith_inferences":["The Kronecker square-root construction in Eq. (16) likely does not preserve the MG marginal distribution, so the theoretical PDF (20) may be approximate or mismatched to the simulated channel; validating or replacing this step is a direct next test.","Since MC on the RIS side is neglected, the authors' stated gains are a lower bound for tightly integrated active RIS surfaces, where MC is stronger; extending the model to the RIS reflection matrix is a natural continuation.","The multi-ring von Mises correlation model could be adapted to estimate the MG parameters themselves from measured arrays, turning the model into a fitting tool rather than a fixed simulation."],"forward_implications":["A THz RIS channel estimator built on near-field correlated MG statistics should beat one built on far-field planar assumptions by several dB in ultra-massive arrays.","Mutual coupling at the BS must be included; neglecting it and spatial correlation costs about 8 dB at 20 dB SNR in the paper's configuration.","The MG shape parameter changes channel variance and therefore estimator MSE, so fading severity must be matched to measurements.","The same covariance-based LMMSE structure can be reused once the cascaded channel covariance is computed from the correlated MG model."],"supporting_citations":[{"why":"Supplies measurement-based MG parameters for outdoor THz fading, giving the paper realistic small-scale fading statistics.","marker":"[14]"},{"why":"Provides the theorem that a product of independent MG random variables is MG-distributed, used to derive the cascaded channel PDF.","marker":"[16]"},{"why":"Supplies the near-field spatial correlation model with spherical wave distances and von Mises angular spread used for R.","marker":"[20]"},{"why":"Gives the DFT-based training phase-shift design and the LMMSE error covariance expressions the paper adapts.","marker":"[19]"},{"why":"Establishes that neglecting mutual coupling degrades channel estimation, motivating the MC-aware BS model.","marker":"[7]"},{"why":"Provides the THz channel characteristics including spreading and molecular absorption losses used in the geometry-based model.","marker":"[3]"},{"why":"Supplies the LS estimator and the compact linear measurement model for RIS channel estimation.","marker":"[18]"},{"why":"Provides closed-form dipole impedance and dissipation resistance expressions used in the mutual coupling matrix M.","marker":"[23]"}],"fun_headline_variants":["Near-field RIS model lifts THz estimation by 3 dB","Ignoring near-field costs THz RIS 8 dB","THz RIS channel model: near-field gains 3 dB","Near-field correlation and coupling cut THz RIS error"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The model assumes in Eq. (16) that multiplying a vector of independent mixture-gamma fading coefficients by the square root of the correlation matrix R produces a channel whose entries are still mixture-gamma with the specified marginal; unlike Gaussian vectors, a linear combination of Gamma mixtures is generally not Gamma, so this step is not automatically valid.","fun_headline_variants_meta":{"raw":{"variants":["Near-field RIS model lifts THz estimation by 3 dB","Ignoring near-field costs THz RIS 8 dB","THz RIS channel model: near-field gains 3 dB","Near-field correlation and coupling cut THz RIS error"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000775,"raw_usage":{"total_tokens":3390,"prompt_tokens":866,"completion_tokens":2524,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":482,"completion_tokens_details":{"reasoning_tokens":2456}},"tokens_in":482,"tokens_out":2524,"duration_ms":17041,"temperature":1.0,"reasoning_tokens":2456,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T21:25:23.180646+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a Monte Carlo simulation that generates channel vectors as $R^{1/2} \\tilde{h}$ from independent MG coefficients and compare the empirical histogram of the resulting entries with the MG PDF used to derive the cascaded distribution (20); if the histograms disagree, the theoretical covariance and NMSE curves in the paper do not describe the simulated channel.","supporting_citations":[{"cited_title":"Outdoor THz fading modeling by means of Gaussian and Gamma mixture distributions,","cited_arxiv_id":null,"evidence_quote":"Supplies measurement-based MG parameters for outdoor THz fading, giving the paper realistic small-scale fading statistics."},{"cited_title":"Analysis of IRS-assisted downlink wireless networks over generalized fading,","cited_arxiv_id":null,"evidence_quote":"Provides the theorem that a product of independent MG random variables is MG-distributed, used to derive the cascaded channel PDF."},{"cited_title":"Near-field spatial correlation for extremely large- scale array communications,","cited_arxiv_id":null,"evidence_quote":"Supplies the near-field spatial correlation model with spherical wave distances and von Mises angular spread used for R."},{"cited_title":"Exploiting Mutual Coupling Characteristics for Channel Estimation in Holographic MIMO","cited_arxiv_id":"2412.13683","evidence_quote":"Establishes that neglecting mutual coupling degrades channel estimation, motivating the MC-aware BS model."},{"cited_title":"TeraMIMO: A channel simulator for wideband ultra- massive MIMO terahertz communications,","cited_arxiv_id":null,"evidence_quote":"Provides the THz channel characteristics including spreading and molecular absorption losses used in the geometry-based model."},{"cited_title":"An optimal channel esti mation scheme for intelligent reflecting surfaces based on a minimu m variance unbiased estimator,","cited_arxiv_id":null,"evidence_quote":"Supplies the LS estimator and the compact linear measurement model for RIS channel estimation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides closed-form dipole impedance and dissipation resistance expressions used in the mutual coupling matrix M."}],"review_version":1}