{"id":"40023a84-2de0-484b-a17d-0933e1a7097a","arxiv_id":"2506.17887","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A 3GPP TR 38.901 channel model extension is specified for FR3 XL-MIMO, adding spherical-wave phase/angle updates and element-wise power attenuation from visibility regions and blockers.","lead":"This paper specifies a 6-24 GHz extremely large-scale MIMO channel model that adds curved wavefronts and element-wise power variation to the 3GPP standard. It is a reference implementation for 6G system-level simulators that need realistic FR3 propagation.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The near-field NLOS model never enforces that the BS-side and UE-side spherical-wave source positions coincide for non-specular clusters, so Eq. (6) can synthesize element-wise phases that no physical single-scatterer path has.","rationale":"I read the paper in good faith as a standards-oriented engineering model: the goal is to extend TR 38.901 with element-wise phase/angle variation and element-wise power attenuation, and the paper provides a complete prescription. The reader's conditional verdict is appropriate. My stress-test identifies a more fundamental issue than the reader's weakest assumption. The reader worried that the Nspec strongest clusters may not be specular and that paths may have multiple non-specular interactions. That is a legitimate statistical concern about the classification heuristic. But even granting the entire classification, the non-specular near-field construction has no geometric closure: d1, d2, rhat_tx, and rhat_rx are generated from independent stochastic distributions, and nothing in Eqs. (8)-(11) ensures that the BS-side source position p_BS = d1*rhat_tx equals the UE-side source position p_UE = r_UE + d2*rhat_rx. The text explicitly says the non-specular source is at the actual physical position of the scatterer, so this equality is not a modeling nicety; it is part of the claimed physical content. Without it, the element-wise phase in Eq. (6) can be the phase of no physical path, and the near-field capacity gains in Fig. 14 could be an artifact of inconsistent geometry rather than a true near-field effect. The concrete Monte Carlo check I propose would settle this directly by measuring the source-position mismatch and the resulting phase error under the paper's own generation rules. I am not alleging misconduct; the missing closure constraint is a technical gap that could be fixed by drawing d1/d2 and angles jointly or by explicitly defining two independent per-side virtual sources. Until then, the central claim that simulators reproduce physical near-field propagation for non-specular clusters is not fully supported.","tokens_in":18066,"tokens_out":19132,"duration_ms":201900,"concrete_test":"Implement the Section III-B non-specular generation path: draw d3D, tau, Delta_tau, sBS ~ Beta(alpha,beta), set sUE = 1 - sBS, and draw rhat_tx and rhat_rx from the TR 38.901 angular distributions. For 10^4 draws compute p_BS = d1 * rhat_tx and p_UE = r_UE + d2 * rhat_rx. Report the distribution of ||p_BS - p_UE|| / lambda and the RMS phase error between Eq. (6) and the exact single-scatterer phase exp(-j*2*pi*(|q - s| + |q - u|)/lambda) with q = (p_BS + p_UE)/2. If the median separation exceeds 0.1 lambda or the RMS phase error approaches pi, the non-specular near-field phase is not a physical spherical-wave phase and the capacity validation in Fig. 14 is called into question.","verdict_should_be":"UNCHANGED","load_bearing_attack":"For non-specular clusters, Section III states that the spherical-wave source is at the actual physical position of the scatterer. However, the generation procedure draws d1 = sBS * c*tau_abs and d2 = (1 - sBS) * c*tau_abs using an independent Beta draw, while the far-field angles rhat_tx and rhat_rx are taken from the existing TR 38.901 stochastic angular distributions. In Eq. (6) the TX-side source is implicitly p_BS = d1 * rhat_tx and the RX-side source is p_UE = r_UE + d2 * rhat_rx, where r_UE is the BS-to-UE reference vector. Physical consistency for a single scatterer requires p_BS = p_UE, but no constraint in Eqs. (8)-(11) enforces this. Since departure and arrival angles are generated independently, the two reconstructed source positions will generally differ by an amount that is many wavelengths at 6-24 GHz. The element-wise phases and angles in Eqs. (5)-(6) are then not the phases and angles of any physical single-bounce path. This is distinct from the reader's concern about misclassifying strong non-specular clusters as specular: the inconsistency affects every non-specular cluster even when the specular/non-specular classification is correct. The simulation validation in Fig. 14 therefore does not establish that the model reproduces physical near-field phase behavior.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an XL-MIMO channel modeling framework for the 6–24 GHz FR3 band that extends the 3GPP TR 38.901 structure with two features: near-field propagation and spatial non-stationarity (SNS). Near-field propagation is modeled by introducing per-cluster distances d1,n,m and d2,n,m from the BS and UE to spherical-wave sources, which are then used to compute element-wise phases and angles. SNS is modeled at the BS side either stochastically through visibility regions and power attenuation factors or through a physical blocker-based knife-edge model, and at the UE side through fixed per-element attenuation values. The framework is presented as the model adopted by 3GPP, and its performance is illustrated by simulations of channel capacity and coupling loss in UMi and InH scenarios.","tokens_in":18390,"tokens_out":4732,"duration_ms":51103,"significance":"If the proposed framework is physically sound, it is a valuable and timely standardization contribution: it provides a concrete, parameterized way for system-level simulators to reproduce near-field spherical-wave effects and element-wise power variations within the established TR 38.901 coefficient generation procedure. The paper's strengths are its close integration with the 3GPP structure, the detailed parameter tables (Tables III and IV) that make the model directly implementable, and the clear separation of near-field distance generation from SNS attenuation. The main significance is therefore contingent on fixing a load-bearing physical consistency issue in the near-field NLOS generation, because element-wise phases that do not correspond to any physical path would undermine the claimed near-field capability and the interpretation of the capacity gains in Fig. 14.","major_comments":[{"comment":"The near-field NLOS generation does not enforce that the BS-side and UE-side spherical-wave source positions coincide for non-specular clusters. In Section III-B, d1,n,m and d2,n,m are generated as sBS,n·c·tau and sUE,n·c·tau with sUE,n = 1 - sBS,n, while the directions r-hat_tx,n,m and r-hat_rx,n,m are taken from the existing far-field angular distributions. The implicit source positions are therefore p_BS = d1·r-hat_tx and p_UE = r_UE + d2·r-hat_rx, and nothing in Eqs. (8)-(11) or in the direction-vector construction at the end of Section III-B constrains p_BS = p_UE. For a single-scatterer NLOS path these two positions will generally differ by many wavelengths at 6-24 GHz, so the element-wise phases and angles computed in Eq. (6) are not those of any physical path. This affects every non-specular cluster, not only misclassified ones, and it means that the capacity comparison in Fig. 14 does not establish that the model reproduces physical near-field phase behavior. The model should either generate a single physical source point and derive d1, d2, and the element-wise angular parameters from it, or explicitly re-frame d1 and d2 as independent hypothetical source distances and justify the resulting unphysical phase distribution.","section":"Section III-B, Eq. (6)"},{"comment":"The assumption that the Nspec strongest clusters are exactly the specular reflection clusters is load-bearing for the near-field model but is not validated within the paper. Section III-A states that paths with sBS = 1 tend to have higher average power and therefore the strongest clusters are treated as specular, with sBS = 1 and d1 = d2 = c·tau. If a strong NLOS cluster is actually produced by diffraction or diffuse scattering, its spherical-wave source is not at the mirror-image point, and setting both distances to the full path length will misplace the source and corrupt the element-wise phases. The paper cites a 3GPP contribution [29] for the power ordering but provides no direct validation in the UMi/InH scenarios used for evaluation. At minimum, the authors should report the sensitivity of the near-field capacity results to the choice of Nspec and to the specular/non-specular classification criterion.","section":"Section III-A, Eq. (9)"},{"comment":"The performance evaluation is simulation-only and is largely self-referential: the Beta distribution parameters for sBS, the PrSNS normal distribution, the VP-power model, and the VR construction are all fitted from ray-tracing data in the same scenarios used for the capacity and coupling-loss simulations. The results in Figs. 14 and 15 therefore demonstrate internal consistency of the framework rather than independent predictive accuracy. There is no comparison with the measured phase and power variations shown in Figs. 2-4, and no confidence intervals are reported for the capacity gains. The claim that the results 'demonstrate the effectiveness' of the framework should be softened or supported by a direct measurement-based check of at least one near-field and one SNS statistic.","section":"Section V-B"}],"minor_comments":[{"comment":"The rendering of Eq. (6) is ambiguous: the phase terms appear to be missing norm bars around the distance differences, e.g., d1,n,m - |d1,n,m·r-hat_tx,n,m - d-bar_tx,s|. The authors should write the equation with explicit absolute values.","section":"Eq. (6)"},{"comment":"The indexing in Eq. (7) is confusing for Nspec = 2: the sum over specular reflection clusters runs from n = 3 to Nspec, which is empty in UMa/UMi, while the first two strongest clusters are already handled separately. This should be clarified or re-indexed.","section":"Eq. (7)"},{"comment":"The simplified knife-edge expression '-20log10(0.5 - Fh1|h2|w1|w2)' is unclear, especially the meaning of the subscript 'h1|h2|w1|w2'. Please define the notation and state which edge factors are set to 0.5.","section":"Eq. (16)"},{"comment":"The VR generation procedure should state whether the randomly chosen reference corner and the dimensions a,b allow the VR to extend beyond the array boundary, and if so, how the area consistency with VP is maintained.","section":"Section IV-A-1-b"},{"comment":"In Fig. 3(a), please specify whether the inter-element phase differences are unwrapped and what element spacing is used, since the reported 0.83 rad fluctuation is otherwise hard to interpret.","section":"Section II-B, Fig. 3"},{"comment":"The statement that near-field amplitude variations are generally negligible is asserted without a supporting reference or quantitative bound; a brief justification or citation would help.","section":"Section II-C"},{"comment":"The phrase 'adopted by 3GPP' should specify the stage of adoption, e.g., agreed in RAN1 and being incorporated into TR 38.901, so that readers do not assume a final published specification.","section":"Abstract and Introduction"}],"recommendation":"major_revision","confidential_remarks":"This manuscript is essentially a multi-company 3GPP contribution write-up, so the heavy reliance on R1 technical documents in the references is appropriate and not a concern. The main technical blocker is the physical inconsistency in the non-specular near-field source generation described in Section III-B; this is fixable but requires a substantive change to the generation procedure or a careful reframing of d1 and d2. The editor may also wish to verify the current TR 38.901 status behind the 'adopted by 3GPP' claim before final acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a genuine 3GPP standardization proposal that adds near-field and SNS parameters to TR 38.901, with a concrete implementation recipe. The near-field NLOS phase generation has a real internal inconsistency: the BS-side and UE-side spherical-wave sources are never forced to be the same point, so the phases in Eq. (6) don't correspond to any physical path.\n\nWhat's new and useful: the specific parameterization—Nspec = 2/4, Beta-distributed sBS with the sUE = 1 - sBS rule, VP-based VR generation with exponential roll-off, and the blocker-type extensions—is exactly the level of detail a standard needs. The paper provides fitted parameter tables for six scenarios and a step-by-step integration into the 3GPP coefficient generation flow. That's a legitimate engineering contribution; if adopted, it will end up in every 6G system-level simulator.\n\nThe soft spots are in the near-field part. The stress test I ran lands, and it's broader than the note says. For any NLOS cluster, the direction vectors rhat_tx and rhat_rx come from the far-field angular distributions, and d1, d2 are drawn independently (even with sUE = 1 - sBS). The reconstructed source positions on the two sides, p_BS = d1 rhat_tx and p_UE = r_UE + d2 rhat_rx, are generally different. As a result, the element-wise phase differences in Eq. (6) are not those of any physical single-scatterer path. The paper says the source is at the physical scatterer, but the generation rules don't enforce it. This is a load-bearing problem: the capacity gains shown in Fig. 14 could be artifacts of an artificial phase curvature. The validation is simulation-only, with no measurement comparison and no error bars, and the parameters are fitted from the same ray-tracing pipeline used in the evaluation. The 'adopted by 3GPP' claim is plausible given the author affiliations, but the paper itself doesn't provide a citable 3GPP final document.\n\nWhat holds up: the SNS model (VR, roll-off, blocker-based attenuation) is well specified, and the UE-side attenuation table is a practical solution. The measurement observations motivate the features well.\n\nRecommendation: this deserves peer review—it's going to be influential either way—but an editor should send it to a referee who will notice the source-position issue. The authors should fix the geometry (e.g., generate one source location and derive both distances and angles from it) or at least state that the two-sided virtual source is a simplification, not a physical position. Without that, the near-field validation claims are not supportable.","headline":"Useful 3GPP framework for FR3 XL-MIMO near-field/SNS, but the near-field NLOS phases are generated from two inconsistent source positions, which undermines the physical claim.","tokens_in":18964,"tokens_out":8989,"would_cite":true,"duration_ms":87566,"reading_group":"yes","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"3GPP adopts near-field, non-stationary channel model for 6G","keywords":["3GPP","extremely large-scale MIMO","FR3","near-field propagation","spatial non-stationarity","visibility region","channel modeling","6G"],"falsifier":"In a UMi or InH double-directional measurement with a large virtual array, resolve the two strongest NLOS clusters at each user position and estimate each cluster's BS-to-spherical-source distance from the per-element phase slope; if a substantial share of those strongest clusters yield $s_{\\rm BS}$ clearly below 1, the specular-cluster assumption and the resulting element-wise phases fail.","tokens_in":17855,"feed_emoji":"📡","tokens_out":8152,"duration_ms":69908,"temperature":0.7,"pith_summary":"This paper argues that the standard 3GPP channel model used for 5G simulations can be extended, with modest additions, to capture two phenomena that appear when base stations carry hundreds or thousands of antennas at 6-24 GHz. The first is near-field propagation: at these apertures and frequencies a user can sit inside the array's near field, so wavefronts are spherical rather than planar and each antenna element sees a different phase and a different departure or arrival angle. The second is spatial non-stationarity (SNS): clusters can be visible to only part of the very large array, so path power varies element by element. The paper gives the concrete parameter extensions—element-wise distances to spherical-wave sources and element-wise power attenuation factors—that let the existing 3GPP structure reproduce these effects, and reports that 3GPP adopted the framework for the FR3 channel model. Simulated results show measurable near-field capacity gains (up to 11.6 bps/Hz indoors at 2 m radius) and stronger coupling-loss fading under SNS than under the stationary model.","feed_headline":"3GPP adopts near-field, non-stationary channel model for 6G","feed_subtitle":"Element-wise spherical-wave phases and power attenuation capture the physics of giant arrays at 6-24 GHz.","key_machinery":"The machinery is a pair of element-wise parameter sets inserted into the standard coefficient-generation equations. Near-field replaces the per-cluster unit vectors and fixed phases with per-element geometry: distances $d_{1,n,m}$ from the BS to the spherical-wave source and $d_{2,n,m}$ from the UE to that source, computed from scaling factors $s_{\\rm BS}$ and $s_{\\rm UE}$, with $s_{\\rm BS}=1$ for the strongest specular clusters and $s_{\\rm BS}$ drawn from a Beta distribution otherwise. SNS multiplies each ray coefficient by $\\sqrt{\\alpha_{s,n,m}}$ at the BS and $\\sqrt{\\beta_u}$ at the UE, where the stochastic model sets $\\alpha_{s,n}=1$ inside a visibility region and applies an exponential roll-off outside it.","core_discovery":"The central claim is that near-field propagation can be modeled inside the 3GPP TR 38.901 framework by replacing constant array-phase and angle parameters with element-wise quantities computed from the exact distance between each antenna element and an equivalent spherical-wave source per cluster. The two (or per-scenario $N_{\\rm spec}$) strongest NLOS clusters are assumed to be specular reflections, so their source distance equals the total propagation path length; weaker clusters get a Beta-distributed fraction of that length. SNS is captured by element-wise power attenuation factors: a stochastic visibility-region model at the base station, a knife-edge diffraction blocker model for partial blockage, and fixed per-element attenuations at the user device. The paper reports that this framework has been adopted by 3GPP and that simulations show near-field capacity gains and SNS-induced fading.","pith_inferences":["The same element-wise distance machinery could be carried into sub-THz bands and reconfigurable-intelligence-surface channels, where spherical-wave and partial-visibility effects are stronger.","If the specular-cluster assumption fails in dense indoor or factory settings, the Beta-fit parameters and the $s_{\\rm UE}=1-s_{\\rm BS}$ rule would need re-estimation from double-directional measurements.","A direct test: use the model's per-element phases to predict array beam patterns and compare with measured near-field patterns; this would isolate the phase model from the SNS attenuation model."],"forward_implications":["System-level 6G simulators built on the 3GPP procedure can reproduce near-field capacity gains, such as 11.6 bps/Hz in an indoor hotspot at 2 m radius, without leaving the standard model's structure.","SNS-aware simulations will show stronger coupling-loss fading than spatial-stationary models (0.91 dB in UMi, 0.67 dB in InH), directly affecting link-budget and coverage predictions.","Element-wise angle and phase parameters enable realistic evaluation of XL-MIMO beamforming and precoding, including spherical-wavefront focusing and per-element codebook design.","The scenario tables (Beta-shape parameters, SNS probability, visibility-probability fits) provide immediate defaults for UMa, UMi, InH, InF, RMa, and SMa.","The extensions slot into the existing twelve-step generation flow, so deployed 3GPP-compliant codebases need only added steps, not a new architecture."],"supporting_citations":[{"why":"supplies the baseline standardized channel model and its twelve-step coefficient generation flow, which the framework extends.","marker":"[25]"},{"why":"provides the measurement evidence of nonlinear phase and AOD variation across the array and element-wise power fluctuation.","marker":"[22]"},{"why":"introduces the visibility probability and visibility-region definitions used by the stochastic SNS model.","marker":"[31]"},{"why":"supplies the power-to-visibility-probability fitting and the attenuation roll-off factor expression.","marker":"[32]"},{"why":"shows that specular-reflection paths carry higher received power, supporting the choice of the strongest clusters as specular.","marker":"[29]"},{"why":"aggregates the scaling-factor distributions that motivate the Beta-distributed $s_{\\rm BS}$ for non-specular clusters.","marker":"[30]"},{"why":"catalogues the physical causes of spatial non-stationarity, partial blockage and incomplete scattering, that the model targets.","marker":"[7]"},{"why":"defines the coupling-loss channel-gain metric used in the performance comparison.","marker":"[13]"}],"fun_headline_variants":["Near-field model for 6-24 GHz XL-MIMO wins 3GPP adoption","Element-wise near-field model for giant arrays gets 3GPP nod","6-24 GHz XL-MIMO near-field model adopted for 6G","3GPP embraces near-field and non-stationary channels for FR3","Near-field spatial non-stationarity model for FR3 XL-MIMO standardized"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The model assumes that the $N_{\\rm spec}$ strongest NLOS clusters are purely specular reflections, so their spherical-wave source sits at the full path length, and that every other path has exactly one non-specular interaction, so $s_{\\rm UE}=1-s_{\\rm BS}$; if real strong clusters involve diffraction or multiple bounces, the element-wise phases and angles will be miscomputed.","fun_headline_variants_meta":{"raw":{"variants":["Near-field model for 6-24 GHz XL-MIMO wins 3GPP adoption","Element-wise near-field model for giant arrays gets 3GPP nod","6-24 GHz XL-MIMO near-field model adopted for 6G","3GPP embraces near-field and non-stationary channels for FR3","Near-field spatial non-stationarity model for FR3 XL-MIMO standardized"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000631,"raw_usage":{"total_tokens":2971,"prompt_tokens":1059,"completion_tokens":1912,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":675,"completion_tokens_details":{"reasoning_tokens":1809}},"tokens_in":675,"tokens_out":1912,"duration_ms":11641,"temperature":1.0,"reasoning_tokens":1809,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T18:58:09.733281+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"In a UMi or InH double-directional measurement with a large virtual array, resolve the two strongest NLOS clusters at each user position and estimate each cluster's BS-to-spherical-source distance from the per-element phase slope; if a substantial share of those strongest clusters yield $s_{\\rm BS}$ clearly below 1, the specular-cluster assumption and the resulting element-wise phases fail.","supporting_citations":[{"cited_title":"Study on channel model for frequencies from 0.5 to 100 GHz, version 16.1.0,","cited_arxiv_id":null,"evidence_quote":"supplies the baseline standardized channel model and its twelve-step coefficient generation flow, which the framework extends."},{"cited_title":"Discussion on near-field propagation and spatial non-stationarity,","cited_arxiv_id":null,"evidence_quote":"provides the measurement evidence of nonlinear phase and AOD variation across the array and element-wise power fluctuation."},{"cited_title":"Considerations on the 7-24 GHz channel model extension,","cited_arxiv_id":null,"evidence_quote":"introduces the visibility probability and visibility-region definitions used by the stochastic SNS model."},{"cited_title":"Considerations on the 7-24 GHz channel model extension,","cited_arxiv_id":null,"evidence_quote":"supplies the power-to-visibility-probability fitting and the attenuation roll-off factor expression."},{"cited_title":"Discussion on adaptation and extension of channel model,","cited_arxiv_id":null,"evidence_quote":"shows that specular-reflection paths carry higher received power, supporting the choice of the strongest clusters as specular."},{"cited_title":"Summary#4 of channel model adaptation and exten- sion,","cited_arxiv_id":null,"evidence_quote":"aggregates the scaling-factor distributions that motivate the Beta-distributed $s_{\\rm BS}$ for non-specular clusters."},{"cited_title":"Views on channel model adaptation/extension of TR38.901 for 7- 24GHz,","cited_arxiv_id":null,"evidence_quote":"catalogues the physical causes of spatial non-stationarity, partial blockage and incomplete scattering, that the model targets."},{"cited_title":"QuaDRiGa - Quasi determinis- tic radio channel generator, user manual and documentation,","cited_arxiv_id":null,"evidence_quote":"defines the coupling-loss channel-gain metric used in the performance comparison."}],"review_version":2}