{"id":"eaaadafb-8468-4454-b85f-15036a09e1b7","arxiv_id":"2505.09398","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A measurement-based hybrid model for THz XL-MIMO channels combines a scatterer/specular-reflection near-field phase model with rank-matched statistical amplitude attenuation factors to reproduce measured channel statistics.","lead":"This paper reports terahertz (THz) channel measurements at 100 GHz and 132 GHz using 301-element and 531-element antenna arrays, and proposes a channel model that combines near-field (spherical-wave) physics with spatial non-stationarity (power variation across the array). The value for a generalist is that it moves THz XL-MIMO channel modeling from observation toward a compact statistical model validated against measured data, which matters for 6G system design.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"SnS validation is in-sample: AAF statistics fitted to Case 3 are validated on the same Case 3 data, so the reported CvM agreement may reflect the fit rather than predictive accuracy.","rationale":"Agree with the reader's weakest assumption. The physical NF modeling (SRM vs SPM phase comparisons, Case 1) is a genuinely independent check and is the strongest part of the paper. The weakness is localized to the SnS AAF statistics. Because all six fitted constants come from Case 3 and all validation metrics are computed on Case 3, the reported 'closely aligns' for the SnS component is not yet evidence of predictive performance. I do not think this invalidates the paper; it is an addressable validation gap, and the reader's CONDITIONAL verdict is appropriate. A held-out split or new environment would settle it. If the held-out test passes, the claim would be substantially stronger. No code/data sharing is secondary; the primary blocker is in-sample validation.","tokens_in":20482,"tokens_out":11484,"duration_ms":125046,"concrete_test":"Hold-out validation across Case 3 positions: fit the AAF parameters (p, q, d_corr) using only a random half of the 12 Rx positions (or positions 1-6), generate AAFs for the held-out half, and recompute the CvM statistics in Table IV and the average spatial correlation in Eq. (26) for the held-out data. If the held-out CvM values remain close to those in Table IV and still fall below the VR and SS baselines, the in-sample concern is resolved; if they degrade substantially, the reported agreement is attributable to fitting the same dataset.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing weakness is that the statistical SnS component is fitted and validated on the same measurement case. In Section III-C, 'all identified SnS paths across the 12 Rx positions in Case 3' are used to estimate the AAF Beta parameters (p ~ Logn(0.37, 0.58), q = 0.48 ln(p) + 1.03) and the spatial decay coefficient (d_corr ~ TruncExp(40.61)). Section IV-B then validates the model on Case 3, generating S from these same fitted distributions and comparing entropy capacity, condition number, channel gain, Rician K-factor, and RMS delay spread against the same Case 3 measurements. The average spatial correlation metric in Eq. (26) is essentially the same spatial autocorrelation quantity used to fit d_corr via Eqs. (13)-(14); agreement on that metric is a consistency check rather than independent confirmation. The footnote to Table II concedes that distribution parameters for other deployment scenarios require new measurements or ray tracing. Consequently, the paper's central claim that the proposed model 'closely aligns' with measurements and outperforms VR/SS has not been tested on any held-out data; the close CvM values in Tables III-IV could be an artifact of in-sample fitting. This does not threaten the NF phase validation, which is independently supported by direct phase comparisons in Figs. 5 and 8, but it leaves the SnS contribution and the claimed statistical replacement value unestablished.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports THz XL-MIMO channel measurements at 100 GHz (301-element ULA) and 132 GHz (531-element ULA) in indoor scenarios and uses them to motivate a channel model that jointly captures near-field spherical-wave propagation and spatial non-stationarity. The near-field component is a hybrid model combining a scatterer-excited point-source model (SPM) for small scatterers and a specular reflection model (SRM) for reflecting surfaces, with element-wise amplitude, phase, delay, and angle updates. The spatial non-stationarity component is a statistical model of amplitude attenuation factors (AAFs) based on a Beta marginal distribution, an exponential spatial autocorrelation, and a rank-matching generation procedure. The model is validated by comparing simulated and measured entropy capacity, Demmel condition number, spatial correlation, channel gain, Rician K-factor, and RMS delay spread, with Cramér–von Mises distances reported in Tables III and IV.","tokens_in":20823,"tokens_out":3710,"duration_ms":38338,"significance":"If the full model were shown to generalize, this would be a useful contribution: the 100/132 GHz dual-band XL-MIMO measurement campaign is substantial, the direct inter-element phase comparisons in Figs. 5 and 8 are physically convincing and provide strong support for the SRM/SPM dichotomy, and the rank-matching AAF generator is a low-complexity statistical alternative to deterministic ray tracing. The paper also makes a fair attempt to quantify model fidelity against several system- and channel-level metrics. However, the predictive claim is currently weakened by the fact that the SnS statistics are fitted and validated on the same measurement case, so the paper's central 'closely aligns with measurements' assertion is not yet established for unseen scenarios.","major_comments":[{"comment":"The AAF distribution parameters (p ~ Logn(0.37, 0.58), q = 0.48 ln(p) + 1.03, d_corr ~ TruncExp(40.61)) are fitted to the 248 SnS paths identified across the 12 Rx positions of Case 3, and the same Case 3 channels are then used in Section IV-B to validate the proposed SnS model against measurement. This is an in-sample evaluation, so the close CvM values in Tables III and IV and the agreement in Figs. 17 and 18 may reflect the fit rather than predictive accuracy. Please provide a held-out validation, for example by fitting the statistics on one subset of Case 3 positions and validating on the remaining positions, or by applying the fitted model to Case 2 or Case 4; alternatively, the claims should be explicitly restricted to characterization of the measured environment rather than predictive modeling.","section":"Section III-C and Section IV-B"},{"comment":"The average spatial correlation metric used to validate the SnS model, Eq. (26), is the same type of spatial autocorrelation quantity whose exponential decay coefficient d_corr was estimated from the measurements via Eqs. (13)-(14). The agreement in Fig. 17 is therefore partly a consistency check of the fitting procedure rather than an independent confirmation of the model's SnS behavior. Please either add validation metrics that are not directly derived from the fitted ACF, or state explicitly which parts of the reported agreement are self-consistency checks.","section":"Eq. (26) vs. Eqs. (13)-(14)"},{"comment":"The footnote to Table II states that the distribution parameters for other deployment scenarios require additional measurements or ray tracing, which directly limits the generality of the proposed statistical SnS model. This is in tension with the abstract's and conclusion's claim that the model 'closely aligns with measurements' and provides an effective characterization of THz XL-MIMO channels as a low-cost statistical replacement. Please temper the conclusions to reflect that the fitted parameters are environment-specific, or provide evidence of transferability across at least one independent measurement scenario.","section":"Footnote to Table II and abstract/conclusion"}],"minor_comments":[{"comment":"The sentence 'This indicates that imply that noticeable amplitude variation...' contains a grammatical error and should read 'This indicates that noticeable amplitude variation...'.","section":"Section II-C, Power paragraph"},{"comment":"The conversion in Eq. (21) is effectively vacuous under the stated convention that the reference element amplitude equals the maximum amplitude; this should be stated more clearly so that readers do not expect a nontrivial conversion step.","section":"Section III-C, Eq. (21)"},{"comment":"In Eq. (23), the SNR parameter Γ is given as 15 dB but is used directly in a linear-argument capacity formula; please clarify whether Γ is a linear SNR and, if not, convert it explicitly.","section":"Section IV-B, entropy capacity equation"},{"comment":"The tables do not state which measurement case is used for the NF validation metrics (capacity and condition number) versus the SnS validation metrics; please add explicit case identifiers to the tables and to the corresponding figure captions.","section":"Tables III and IV"},{"comment":"There are several typographical issues, such as 'Cram ´er–von Mises' with an anomalous space, inconsistent use of 'Rx' and 'receiver', and missing commas in compound sentences; a careful proofread would improve readability.","section":"General presentation"}],"recommendation":"major_revision","confidential_remarks":"The in-sample validation of the SnS model is the key concern. The near-field phase validations are strong and should be highlighted, but the paper's broader claim of a validated statistical SnS model requires either held-out data or a more restricted claim. I would not recommend rejection because the measurement study and the hybrid SRM/SPM near-field component are valuable and the statistical generation method is reasonable; the manuscript needs a substantial revision to address the circularity issue."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: the NF phase modeling is the real contribution; the SnS statistical generator is a reasonable idea but its validation is in-sample, so the headline agreement in Tables III and IV should not be read as independent confirmation.\n\nWhat is new: this is, as far as I can tell, the first measurement-based THz XL-MIMO study that directly tests near-field phase models on inter-element phase differences, and the hybrid SPM+SRM treatment is well chosen. The LoS phase follows simple spherical geometry; NLoS reflections from concrete, wood, and glass fit the mirror-image specular reflection model, while a small cylinder fits the point-source model. Fig. 8 is persuasive, and the explicit inclusion of directional antenna gains (Fig. 16) is a useful practical correction. On the SnS side, replacing binary visibility with continuous AAFs and using rank matching to get Beta marginals plus exponential spatial correlation is a clean statistical extension of the authors' earlier deterministic ray-tracing AAF method.\n\nSoft spots, in proportion. The SnS validation is the load-bearing issue. AAF distribution parameters and d_corr are fitted to all 248 SnS paths from the 12 Case 3 receiver positions, then the model is validated on the same Case 3 data, using the measured reference-element response as input. The average spatial correlation metric in Eq. (26) is essentially the same spatial-autocorrelation quantity used to fit d_corr. So the CvM numbers in Tables III and IV for the SnS part are mostly a consistency check. The footnote to Table II concedes that other scenarios require fresh measurements or ray tracing, which is honest but undercuts the claim that the model is a ready statistical replacement. This is fixable with held-out positions or a separate environment, plus error bars. I would not call it fatal; the NF phase validation stands independently.\n\nTwo smaller issues. The SRM-vs-SPM classification rule for generation is never specified; Fig. 8 shows both mechanisms exist, but the implementation section doesn't say how to decide which to apply to a given NLoS path. And the VR baseline in Fig. 18 shows discontinuous jumps that look like artifacts; if the paper keeps the VR comparison, that implementation should be checked. No code or path-level data is released, which hurts reproducibility for a modeling paper.\n\nWho for: THz and XL-MIMO channel modelers, especially those interested in near-field wavefront behavior. The NF phase result is worth citing on its own. The paper deserves a serious referee; with held-out SnS validation and a concrete SRM/SPM decision rule, the validation claim would be substantially stronger.","headline":"A useful measurement-driven NF phase model with a real in-sample validation problem on the SnS side; worth reviewing seriously, but the headline agreement for the statistical SnS generator should be presented as a consistency check, not an independent test.","tokens_in":21419,"tokens_out":3080,"would_cite":true,"duration_ms":31934,"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":"Indoor THz XL-MIMO channels are both near-field and spatially non-stationary, and a hybrid model combining specular reflection, point-source scattering, and statistically generated amplitude attenuation factors reproduces measured…","keywords":["terahertz communications","XL-MIMO","near-field propagation","spatial non-stationarity","channel measurement","channel modeling","amplitude attenuation factor","indoor channel"],"falsifier":"A direct test: using the published AAF constants ($p\\sim\\mathrm{Logn}(0.37, 0.58)$, $q = 0.48\\ln(p) + 1.03$, $d_{\\mathrm{corr}}\\sim\\mathrm{TruncExp}(40.61)$), generate channels for a different indoor room or a different frequency and compare the entropy-capacity and spatial-correlation CDFs with new measurements; if the distributional discrepancies grow well beyond the values reported here, or if the measured AAF marginals differ materially from the Beta and log-normal fits, the stationarity assumption fails. A narrower check is to measure inter-element phase differences for a large flat reflector at a larger Tx–Rx distance where spherical curvature is weaker: the specular reflection model predicts the mirror-image phase, and if the point-source model matches instead, the hybrid selection rule breaks.","tokens_in":2061,"feed_emoji":"📡","tokens_out":7115,"duration_ms":95490,"temperature":0.7,"pith_summary":"The paper argues that indoor THz extremely large-scale MIMO (XL-MIMO) channels cannot be captured by the usual far-field plane-wave assumption, because users sit in the near field and different antenna elements see different multipath powers. It proposes a hybrid channel model that treats reflections from large flat surfaces as images, small scatterers as point sources, and adds statistically generated amplitude attenuation factors to describe smooth power variation across the array. Using 100 GHz and 132 GHz measurements with 301- and 531-element arrays, the model matches measured entropy capacity, Demmel condition number, spatial correlation, channel gain, Rician K-factor, and RMS delay spread, whereas far-field and stationary baselines deviate clearly. The contribution is a measurement-based, low-cost statistical replacement for deterministic ray tracing in THz XL-MIMO evaluation.","feed_headline":"New measurements show THz XL-MIMO channels defy far-field models","feed_subtitle":"A hybrid spherical-wave plus attenuation model matches measured capacity, correlation, and delay spread; plane-wave models do not.","key_machinery":"The load-bearing object is the amplitude attenuation factor (AAF), a continuous value in $[0,1]$ assigned to each path at each antenna element, normalized by the path's maximum amplitude across the array. It carries spatial non-stationarity: instead of binary cluster visibility, each path's power varies smoothly along the array, with a Beta-distributed marginal and an exponential spatial autocorrelation generated by Gaussian rank matching. The second mechanism is the hybrid near-field matrix $A(f)$, which computes element-wise amplitude, phase, delay, and angle for every NLoS path by choosing between the specular reflection model and the point-source model. Together they multiply the reference-element channel response element-wise to produce the final channel.","core_discovery":"At THz frequencies with arrays of hundreds of elements, multipath parameters such as phase, delay, angle, and power vary from element to element in ways the far-field model misses, and this variation has two distinct sources: spherical-wave near-field propagation and spatial non-stationarity from blockage or inconsistent reflection and scattering. The paper's central claim is that both effects can be modeled jointly by computing element-wise distances and angles for each path using either a specular reflection model (the mirror image of the receiver) or a scatterer-excited point-source model, then multiplying the reference-element channel response by an amplitude attenuation factor per path and element. The attenuation factors are modeled statistically: each path's normalized amplitudes follow a Beta distribution whose shape parameter $p$ is log-normal with mean 0.37 and variance 0.58, with $q = 0.48\\ln(p) + 1.03$, and the spatial autocorrelation decays exponentially with a truncated-exponential decorrelation coefficient. A rank-matching, copula-based procedure generates spatially correlated attenuation factors that preserve the Beta marginal distribution. Validated against the measurements, the hybrid model produces small distributional discrepancies across all six metrics, while the far-field model and the binary visibility-region and stationary-spatial models produce substantially larger deviations.","pith_inferences":["Editorial inference: the portable result is the rank-matching generation method, not the six fitted constants; in a different indoor room or frequency band one would need to re-estimate $p$, $q$, and the decorrelation coefficient, but the same copula-based recipe should transfer.","Editorial inference: the observation that the specular component remains dominant across several surface roughness levels at 100–132 GHz suggests a testable threshold: when surface roughness height grows beyond a fraction of a wavelength, point-source or diffuse-scattering modeling should take over, and the paper's material set already hints at this ordering.","Editorial inference: the entropy-capacity and condition-number results imply that precoding and beamforming designs built on far-field plane-wave channel models will be suboptimal in THz XL-MIMO near-field conditions, and the generated channels provide a testbed for such designs.","Editorial inference: the same rank-matching construction could be reused for other element-dependent parameters such as differential delays or angles, provided their spatial autocorrelation is also well described by an exponential decay."],"forward_implications":["Far-field plane-wave models underestimate entropy capacity and overestimate Demmel condition number in indoor THz XL-MIMO, so system evaluations using far-field assumptions will misjudge spatial multiplexing potential.","The choice between specular and point-source modeling matters most when the line-of-sight path is weak; in LoS-dominated links the NLoS modeling differences are masked, but after removing LoS, the point-source model overestimates entropy capacity and the specular model underestimates it.","Continuous amplitude attenuation factors reproduce measured spatial correlation and distributions of channel gain, Rician K-factor, and delay spread better than binary visibility-region models, which can produce artificial discontinuities.","Element-dependent angles combined with directional antenna patterns cause sizable power variation across the array, up to about 5 dB for the LoS path in the measured 531-element case; omitting either effect reduces the near-field model to far-field accuracy.","The statistical AAF generation procedure is a low-complexity substitute for deterministic ray-tracing generation of spatial non-stationarity, making the model easier to embed in system-level simulations."],"supporting_citations":[{"why":"Supplies the original scatterer-excited point-source model plus AAF framework that this paper extends with a specular reflection model and statistical AAF generation.","marker":"[18]"},{"why":"Provides the prior 132 GHz measurement campaign and the second sounder configuration whose data are used here.","marker":"[11]"},{"why":"Earlier measurements with a 2400-element array first revealed near-field and spatial non-stationarity effects that motivate this modeling study.","marker":"[16]"},{"why":"Introduces the cluster visibility-region model that serves as the binary spatial non-stationarity baseline the paper compares against.","marker":"[10]"},{"why":"Presents cluster-based visibility-region modeling for large-scale arrays, another related SnS baseline.","marker":"[21]"},{"why":"Provides the Beckmann–Kirchhoff scattering model used to simulate and explain why power variation across the array is stronger at THz than at lower frequencies.","marker":"[33]"},{"why":"Supplies the exponential spatial autocorrelation model that is fitted to the measured AAF sequences.","marker":"[34]"},{"why":"Underpins the rank-matching copula method used to generate spatially correlated AAFs with a prescribed marginal distribution.","marker":"[35]"},{"why":"Establishes the validation methodology of comparing CDFs of capacity, K-factor, and delay-spread metrics against measurements.","marker":"[36]"},{"why":"Provides the cross-field XL-MIMO channel modeling context that the hybrid NF model builds on.","marker":"[20]"}],"fun_headline_variants":["THz XL-MIMO: near-field and blockage both break plane-wave model","Hybrid model captures THz XL-MIMO near-field and spatial non-stationarity","Measurement-based model: THz arrays need spherical waves plus attenuation","THz XL-MIMO: element-wise power and phase vary, plane-wave fails","Tested at 100/132 GHz: near-field and non-stationarity shape THz channels"],"cache_read_input_tokens":23296,"weakest_assumption_plain":"The model's load-bearing premise is that the power-variation statistics fitted in one indoor room—the Beta shape parameters and the exponential decorrelation coefficient—can be reused to generate new channels in the same kind of environment; the paper fits these statistics and validates against the same measurement campaign, so transfer to other rooms or frequencies is not yet established.","fun_headline_variants_meta":{"raw":{"variants":["THz XL-MIMO: near-field and blockage both break plane-wave model","Hybrid model captures THz XL-MIMO near-field and spatial non-stationarity","Measurement-based model: THz arrays need spherical waves plus attenuation","THz XL-MIMO: element-wise power and phase vary, plane-wave fails","Tested at 100/132 GHz: near-field and non-stationarity shape THz channels"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001261,"raw_usage":{"total_tokens":5262,"prompt_tokens":1140,"completion_tokens":4122,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":756,"completion_tokens_details":{"reasoning_tokens":4016}},"tokens_in":756,"tokens_out":4122,"duration_ms":27061,"temperature":1.0,"reasoning_tokens":4016,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T21:32:10.519965+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct test: using the published AAF constants ($p\\sim\\mathrm{Logn}(0.37, 0.58)$, $q = 0.48\\ln(p) + 1.03$, $d_{\\mathrm{corr}}\\sim\\mathrm{TruncExp}(40.61)$), generate channels for a different indoor room or a different frequency and compare the entropy-capacity and spatial-correlation CDFs with new measurements; if the distributional discrepancies grow well beyond the values reported here, or if the measured AAF marginals differ materially from the Beta and log-normal fits, the stationarity assumption fails. A narrower check is to measure inter-element phase differences for a large flat reflector at a larger Tx–Rx distance where spherical curvature is weaker: the specular reflection model predicts the mirror-image phase, and if the point-source model matches instead, the hybrid selection rule breaks.","supporting_citations":[{"cited_title":"Spatial non-stationary near-field channel modeling and validation for massive MIMO systems,","cited_arxiv_id":null,"evidence_quote":"Supplies the original scatterer-excited point-source model plus AAF framework that this paper extends with a specular reflection model and statistical AAF generation."},{"cited_title":"An empirical study on near-field, spatial non-stationarity, and beam misalignment characteristics of THz XL-MIMO channels at 132 GHz,","cited_arxiv_id":null,"evidence_quote":"Provides the prior 132 GHz measurement campaign and the second sounder configuration whose data are used here."},{"cited_title":"Deterministic ray tracing: a promising approach to THz channel modeling in 6G deployment scenarios,","cited_arxiv_id":null,"evidence_quote":"Earlier measurements with a 2400-element array first revealed near-field and spatial non-stationarity effects that motivate this modeling study."},{"cited_title":"Massive MIMO extensions to the COST 2100 channel model: modeling and validation,","cited_arxiv_id":null,"evidence_quote":"Introduces the cluster visibility-region model that serves as the binary spatial non-stationarity baseline the paper compares against."},{"cited_title":"On 3D cluster-based channel modeling for large-scale array communications,","cited_arxiv_id":null,"evidence_quote":"Presents cluster-based visibility-region modeling for large-scale arrays, another related SnS baseline."},{"cited_title":"Scattering analysis for the modeling of THz communication systems,","cited_arxiv_id":null,"evidence_quote":"Provides the Beckmann–Kirchhoff scattering model used to simulate and explain why power variation across the array is stronger at THz than at lower frequencies."},{"cited_title":"Millimeter wave small-scale spatial statistics in an urban microcell scenario,","cited_arxiv_id":null,"evidence_quote":"Supplies the exponential spatial autocorrelation model that is fitted to the measured AAF sequences."},{"cited_title":"Millimeter wave and sub-terahertz spatial statistical channel model for an indoor office building,","cited_arxiv_id":null,"evidence_quote":"Establishes the validation methodology of comparing CDFs of capacity, K-factor, and delay-spread metrics against measurements."}],"review_version":1}