{"id":"d79ee778-7ae9-4f15-b71e-9f5c63f69d39","arxiv_id":"2509.07331","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":9,"one_line_summary":"Indoor hotspot path loss measurements at 6.75, 16.95, 28, 73, and 142 GHz are fitted to CI, FI, ABG, and cross-polarization variants, producing parameter tables for 7-24, 0.5-100, and 0.5-150 GHz bands, including new 142 GHz FI coefficients.","lead":"This paper fits standard wireless path loss formulas to indoor measurements at five frequencies from 6.75 to 142 GHz and publishes parameter tables for 6G-era bands. The tables are aimed at 3GPP and ITU standardization of FR3 and above-100 GHz indoor channel models.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"95% CIs for combined CI PLE ignore between-frequency clustering; per-frequency PLEs (Table I) vary enough that the claimed significant deviation from 3GPP may vanish.","rationale":"The reader's weakest assumption was sparse frequency sampling, which mainly threatens the ABG gamma and CIF b parameters. The concern I identify is different and more directly tied to the strongest claim: the reported confidence intervals for the CI PLE ignore the clustered structure of the data. The per-frequency PLEs in Table I show considerable spread, and the pooled CI (1.3–1.5 LOS, 2.8–3.1 NLOS) is implausibly narrow if frequency/site is treated as a random effect. This matters because the paper's key conclusion is that the 3GPP InH values are statistically not reproduced; if the CI widens under clustering, the NLOS comparison in particular may no longer be significant. The paper's parameter tables may still be useful, and the limitation statements are honest, so the overall CONDITIONAL verdict remains appropriate. However, the specific statistical significance claim needs re-analysis before it can be used to recommend changes to TR 38.901. I partially agree with the reader because both concerns involve limited frequency samples, but the most load-bearing issue is the clustered variance, not merely the number of frequency points.","tokens_in":11199,"tokens_out":7397,"duration_ms":81328,"concrete_test":"Obtain the raw path-loss samples (or per-frequency PLEs from Table I) and compute the CI PLE via a linear mixed-effects model with random intercept and random slope for frequency (or measurement campaign), or via a clustered bootstrap that resamples frequencies with replacement. Apply this to both the 0.5–100 GHz set (Table III) and the 0.5–150 GHz set (Table IV). If the resulting 95% CI for NLOS CI PLE includes 3.19, or the LOS CI includes 1.73, the paper's claim that the 3GPP InH values are statistically not reproduced fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline comparison to TR 38.901 rests on the Section IV claim that the measured CI PLEs (1.4 LOS, 2.9 NLOS) fall outside the 95% CIs (1.3–1.5, 2.8–3.1). These CIs appear to be computed from the residual scatter of the pooled CI fit, treating every path-loss sample as independent. However, the measurements are clustered by frequency and physical site: 6.75/16.95 GHz were taken at 370 Jay Street, while 28/73/142 GHz were taken at 2 Metro Tech, with different TX-RX pairs, antenna heights, and distance ranges. Table I shows that per-frequency CI PLEs vary far more than the pooled CI admits: LOS values are 1.3,1.3,1.1,1.3,1.8; NLOS are 2.7,3.1,2.7,3.2,2.7. A random-effects or clustered analysis of these five frequency-level estimates yields much wider intervals (roughly 1.13–1.59 and 2.66–3.10 for LOS/NLOS using the Table I values), so 3GPP LOS 1.73 remains outside, but NLOS 3.19 is only marginal and may be inside for the 0.5–100 GHz subset (2.7,3.1,2.7,3.2). Thus the 'statistically significant deviations' conclusion is not robust to clustering. Because the central claim of the paper is precisely that the 3GPP reference values are not reproduced with statistical significance, this is the load-bearing weakness.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports single-frequency CI and FI path loss parameters at 6.75, 16.95, 28, 73, and 142 GHz for the indoor hotspot scenario, and multi-frequency CI, CIX, FI, CIF, CIFX, ABG, and ABGX parameters for the 7-24 GHz, 0.5-100 GHz, and 0.5-150 GHz bands, based on NYU WIRELESS measurements. It also provides new FI parameters at 142 GHz. The central interpretive claim is that the measured CI PLEs (1.4 LOS, 2.9 NLOS) differ statistically from the 3GPP TR 38.901 InH values (1.73 and 3.19), and that 3GPP should consider the physically anchored CI model. The paper is explicitly positioned as a contribution to 3GPP Rel-19 channel model validation.","tokens_in":11821,"tokens_out":4979,"duration_ms":57779,"significance":"If the parameter tables are trustworthy, this is a useful data-driven contribution to 6G channel modeling, particularly the extension above 100 GHz and the comparison to TR 38.901. The paper uses a standard least-squares fitting procedure, reports shadow-fading standard deviations, and is transparent about some limitations. However, the statistical significance claim for the 3GPP comparison is load-bearing and is not established by the evidence presented, and the 7-24 GHz table is based in part on a measurement outside that band. The paper is most valuable as a measurement-report and parameter-table contribution; its claims about statistically significant deviations require a more careful uncertainty analysis.","major_comments":[{"comment":"The claim that the measured CI PLEs fall outside the 95% confidence intervals (1.3-1.5 for LOS, 2.8-3.1 for NLOS) is not supported by any reported derivation of those intervals. If they come from the pooled least-squares residual scatter, they ignore clustering of measurements by frequency and by physical site. Using the per-frequency CI PLEs in Table I (LOS: 1.3,1.3,1.1,1.3,1.8; NLOS: 2.7,3.1,2.7,3.2,2.7) as cluster-level observations gives approximate 95% CIs of roughly 1.04-1.68 for LOS and 2.57-3.19 for NLOS; the NLOS 3GPP value 3.19 is then no longer clearly outside the interval. For the 0.5-100 GHz subset the NLOS interval is even wider and contains 3.19. The LOS comparison is robust, but the NLOS statistical-deviation claim is not. Revise with a clustered/bootstrap confidence interval or temper the conclusion.","section":"Section IV, Table IV"},{"comment":"Table II is labeled '7-24 GHz' but the two frequency points used are 6.75 GHz and 16.95 GHz. Since 6.75 GHz is outside the declared 7-24 GHz band, the two-point frequency dependence (ABG gamma, CIF b, and XPD terms) is not actually measured within the band at its lower edge. The footnote in the introduction calls 6.75 GHz 'representative' of the lower end, but that does not justify describing the derived parameters as 7-24 GHz band parameters. The table should be re-labeled or re-analyzed using only in-band frequencies, or the extrapolatory nature of the lower edge should be explicitly stated.","section":"Section II and Table II"},{"comment":"The multi-frequency frequency-dependence parameters are estimated from very few frequency points: two for 7-24 GHz, four for 0.5-100 GHz, and five for 0.5-150 GHz. The paper itself states in Section V that this 'renders the derived path loss model parameters sensitive to the measured data and environment.' This is not a mere caveat: for Table II, gamma is a two-point slope, and no uncertainty is reported for gamma, b, or XPD. Since these parameters are central to the claimed band-wide generalizability, the abstract and Section IV should be tempered to match the Section V limitation, or the uncertainty of these parameters should be quantified.","section":"Section IV and Section V"}],"minor_comments":[{"comment":"Title has a typo: 'Path L oss' should be 'Path Loss'. Also, the abstract says 'comprehensive derivation', but these are empirical fits; 'derivation' is misleading.","section":"Title/Abstract"},{"comment":"Footnote 1 says '16.75 GHz and 16.95 GHz were selected' — this appears to be a typo for '6.75 GHz and 16.95 GHz.'","section":"Footnote 1"},{"comment":"The PLE notation is inconsistent: equations and tables use 'n', but the text comparing to 3GPP uses 'η'. Use one symbol throughout.","section":"Section III"},{"comment":"'Table IVshows' is missing a space. Also, the 95% confidence intervals cited in the text are not shown in any table; add them to Table IV or give a separate table with the intervals and the method of computation.","section":"Section IV, Table IV"},{"comment":"The figures show large deviations between the measured FI fits and the 3GPP fits, especially at 142 GHz NLOS (beta=0.8). The text explains this as a lack of physical anchoring; consider adding a one-sentence note in the captions or text to help readers interpret the nonphysical beta values.","section":"Figs. 1 and 2"}],"recommendation":"major_revision","confidential_remarks":"The paper's value is primarily as a measurement-data and parameter-table contribution. The headline claim that the 3GPP NLOS PLE of 3.19 is statistically inconsistent with the measured 2.9 is not robust to clustering by frequency/site; the LOS claim appears robust. For a standards-oriented venue, the 6.75 GHz measurement being outside the nominal 7-24 GHz band is likely to attract criticism. The authors should either provide a cluster-level uncertainty analysis or reframe the paper as a parameter-report rather than a statistical validation of TR 38.901."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the paper is a solid measurement-based parameter fitting exercise with some genuinely new numbers (142 GHz FI, and the 0.5–150 GHz multi-frequency tables), but the headline statistical comparison to 3GPP is the weak link. The confidence intervals in Section IV appear to treat every path loss sample as independent, which inflates precision. Looking at Table I, per-frequency PLEs vary a lot: NLOS values are 2.7, 3.1, 2.7, 3.2, 2.7. Once you account for clustering by frequency and site, the claimed deviation from 3GPP's 3.19 is marginal at best. LOS 1.73 vs 1.4 still looks robust, but the paper overstates the significance for NLOS.\n\nWhat it does well: the equivalence arguments between 3GPP's InH models and CI/FI/ABG are correct and clearly laid out. The tables are easy to use, and the authors are honest about the sparse frequency sampling — the conclusion explicitly concedes that two points in 7–24 GHz and four in 0.5–100 GHz make the frequency-dependent parameters sensitive. The 142 GHz FI parameters are new and fill a gap for sub-THz standardization.\n\nSoft spots beyond the clustering issue: the 7–24 GHz table uses only 6.75 and 16.95 GHz, and 6.75 is technically outside the band. Also, the data and several fitted parameters come from the same NYU WIRELESS group, so the 'validation' of 3GPP is partly self-referential. That's not fatal for a measurement campaign, but it should be framed as site-specific evidence, not as a general rejection of the standardized model.\n\nWho this is for: people working on 3GPP Rel-19 channel model validation or 6G link budgets who need concrete InH parameter values. The paper deserves a serious referee — the tables are useful and the topic matters — but the statistical claims need rework. I'd ask for cluster-robust confidence intervals (or a random-effects analysis) and a more careful interpretation of the comparison to 3GPP.","headline":"Useful InH path loss parameter tables up to 150 GHz, but the claim that 3GPP values are statistically rejected is not robust to frequency clustering.","tokens_in":12220,"tokens_out":2597,"would_cite":true,"duration_ms":29898,"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 hotspot path-loss measurements from 6.75 to 142 GHz yield CI-model exponents of 1.4 (LOS) and 2.9 (NLOS), placing TR 38.901's reference values outside their confidence intervals.","keywords":["path loss models","indoor hotspot","6.75–142 GHz","FR3 band","sub-THz","close-in model","ABG model","TR 38.901 validation"],"falsifier":"Take the public omnidirectional point data at the five measured frequencies, add one or two new indoor-hotspot measurement campaigns at frequencies inside the same bands (for example, 10, 20, 40, 100, or 120 GHz) in similar hallway-dominated environments, and re-fit the CI, CIF, and ABG models. If the CI PLE moves toward 1.73 in LOS or 3.19 in NLOS, or if ABG's γ shifts materially when a sixth frequency is added, the claim that the standard values are wrong and the current parameters are representative would fail.","tokens_in":11169,"feed_emoji":"📶","tokens_out":10068,"duration_ms":106930,"temperature":0.7,"pith_summary":"This paper tries to establish that a single set of indoor-hotspot measurements across five frequencies—6.75, 16.95, 28, 73, and 142 GHz—can support all seven commonly used path-loss model families, and that the fitted parameters differ from the standardized indoor-hotspot reference. The fitted close-in (CI) model gives a path-loss exponent of about 1.4 in line-of-sight and 2.9 in non-line-of-sight across 0.5–150 GHz; the standard's values of 1.73 and 3.19 fall outside the 95% confidence intervals around the fits. The paper also supplies single-frequency parameters at 142 GHz, giving a first bridge for extending the standardized model above 100 GHz. If these results stand, standards bodies have concrete measurement-based evidence to revise the indoor-hotspot path-loss model and to prefer the physically anchored CI model over floating-intercept fits whose parameters swing by tens of decibels.","feed_headline":"Measured indoor path loss is flatter than TR 38.901 predicts","feed_subtitle":"Five bands from 6.75 to 142 GHz put path-loss exponents outside TR 38.901's confidence and extend models past 100 GHz.","key_machinery":"The load-bearing object is the fitted parameter set across the CI, CIX, FI, CIF, CIFX, ABG, and ABGX model families, estimated from omnidirectional impulse-response measurements at 6.75, 16.95, 28, 73, and 142 GHz. The key identity is the CI model's physical anchor: path loss at the 1 m reference distance is fixed to free-space loss, so distance dependence is carried entirely by one exponent, the PLE. That anchor is what makes the measured PLEs directly comparable to the standardized TR 38.901 indoor-hotspot values, and it is what the paper says the floating-intercept FI model lacks—its intercept and slope swing by tens of decibels between frequencies, indicating no physical meaning. The CIF","core_discovery":"The paper's central discovery is a measured parameter set for indoor hotspot propagation that spans upper mid-band, millimeter-wave, and sub-THz frequencies in one coherent campaign. Across all three frequency ranges, the CI model gives a path-loss exponent around 1.3–1.4 in LOS and 2.9 in NLOS; the CIF model's frequency-weighting coefficient b is essentially zero, meaning distance-dependent loss does not measurably depend on frequency. These fitted exponents are statistically separated from the TR 38.901 reference values of 1.73 (LOS) and 3.19 (NLOS), a gap the paper attributes to waveguide effects in long, narrow hallways. For multi-frequency bands, the ABG model's frequency exponent γ is","pith_inferences":["A natural testable extension is to compare the same model families in additional indoor environments beyond long, narrow hallways; if the waveguide effect is responsible, open-plan or furnished offices should yield higher PLEs closer to the standard's values. The paper suggests but does not test this.","The near-zero b in the CIF model implies the PLE can be treated as frequency-independent, yet the ABG γ estimates remain unstable because of sparse sampling; with only five frequencies, the 0.5–150 GHz parameter table should be read as provisional rather than definitive.","Because the paper omits cross-polarized models above 100 GHz, its sub-THz extension is limited to co-polarized links; polarization behavior above 100 GHz remains an open empirical question.","The statistical mismatch claim depends on how the 95% confidence intervals are constructed from the measured scatter; re-deriving those intervals from the public point data would independently check whether the standard's values truly fall outside."],"forward_implications":["If the 0.5–150 GHz fit is correct, the standardized indoor-hotspot LOS PLE of 1.73 and NLOS PLE of 3.19 are statistically incompatible with these measurements; both would need revision or revalidation for the upper mid-band and above-100 GHz bands.","The new single-frequency 142 GHz CI and FI parameters give a concrete starting point for extending the standardized indoor-hotspot model above 100 GHz, where no standardized values currently exist.","The near-zero CIF frequency coefficient b indicates that a single PLE can describe distance-dependent loss across 7–150 GHz; adding a frequency-dependent PLE term does not improve the fit.","Since the CI model matches ABG and FI in shadow-fading standard deviation while using fewer parameters, standardization bodies could adopt CI as the primary model without sacrificing fit quality.","Any future update to the standardized indoor-hotspot model should aggregate multiple independent measurement campaigns rather than adjusting parameters from a single data set, as the paper itself recommends."],"supporting_citations":[{"why":"Supplies the 6.75 and 16.95 GHz indoor measurements and their CI/FI parameters that anchor the 7–24 GHz fits.","marker":"[3]"},{"why":"Supplies the 28 and 73 GHz indoor office measurements and the CIX/CIFX/ABGX model formulations used in the 0.5–100 GHz fits.","marker":"[13]"},{"why":"Provides the prior validation comparison and the ABG parameter sets marked [17] in Tables II and III; its confidence intervals support the statistical mismatch claim.","marker":"[17]"},{"why":"Supplies the 142 GHz indoor office LOS/NLOS measurements used for the single-frequency 142 GHz and the 0.5–150 GHz fits.","marker":"[19]"},{"why":"Defines the TR 38.901 indoor-hotspot reference model whose LOS and NLOS path-loss exponents the paper tests.","marker":"[12]"},{"why":"Establishes the CI model's parameter stability and prediction accuracy across datasets, which is the basis for recommending CI over FI.","marker":"[21]"}],"fun_headline_variants":["Indoor path loss flatter than TR 38.901 at 6.75–142 GHz","5-band indoor path loss challenges TR 38.901 predictions","Hallway waveguide flattens indoor path loss exponents","Measured indoor path loss below 3GPP for 6.75–142 GHz"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The load-bearing premise is that two, four, or five frequency points are enough to estimate the frequency dependence of path loss across bands as wide as 7–24, 0.5–100, and 0.5–150 GHz—if those sparse samples are not representative, the fitted multi-frequency parameters and the mismatch with the standard would shift.","fun_headline_variants_meta":{"raw":{"variants":["Indoor path loss flatter than TR 38.901 at 6.75–142 GHz","5-band indoor path loss challenges TR 38.901 predictions","Hallway waveguide flattens indoor path loss exponents","Measured indoor path loss below 3GPP for 6.75–142 GHz"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001138,"raw_usage":{"total_tokens":4651,"prompt_tokens":920,"completion_tokens":3731,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":664,"completion_tokens_details":{"reasoning_tokens":3648}},"tokens_in":664,"tokens_out":3731,"duration_ms":29103,"temperature":1.0,"reasoning_tokens":3648,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T22:23:22.004552+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the public omnidirectional point data at the five measured frequencies, add one or two new indoor-hotspot measurement campaigns at frequencies inside the same bands (for example, 10, 20, 40, 100, or 120 GHz) in similar hallway-dominated environments, and re-fit the CI, CIF, and ABG models. If the CI PLE moves toward 1.73 in LOS or 3.19 in NLOS, or if ABG's γ shifts materially when a sixth frequency is added, the claim that the standard values are wrong and the current parameters are representative would fail.","supporting_citations":[{"cited_title":"Comprehensive FR1(C) and FR3 Lower and Upper Mid-Band Propagation and Material Penetration Loss Measur ements and Channel Models in Indoor Environment for 5G and 6G,","cited_arxiv_id":null,"evidence_quote":"Supplies the 6.75 and 16.95 GHz indoor measurements and their CI/FI parameters that anchor the 7–24 GHz fits."},{"cited_title":"I ndoor Office Wideband Millimeter-Wave Propagation Measurements and Ch annel Models at 28 and 73 GHz for Ultra-Dense 5G Wireless Networks,","cited_arxiv_id":null,"evidence_quote":"Supplies the 28 and 73 GHz indoor office measurements and the CIX/CIFX/ABGX model formulations used in the 0.5–100 GHz fits."},{"cited_title":"V alidation of 3G PP TR 38.901 Indoor Hotspot Path Loss Model Based on Measurements Conduc ted at 6.75, 16.95, 28, and 73 GHz for 6G and Beyond,","cited_arxiv_id":null,"evidence_quote":"Provides the prior validation comparison and the ABG parameter sets marked [17] in Tables II and III; its confidence intervals support the statistical mismatch claim."},{"cited_title":"Millime ter Wave and Sub-Terahertz Spatial Statistical Channel Model for an Ind oor Office Building,","cited_arxiv_id":null,"evidence_quote":"Supplies the 142 GHz indoor office LOS/NLOS measurements used for the single-frequency 142 GHz and the 0.5–150 GHz fits."},{"cited_title":"Study on channel model for frequencies from 0.5 t o 100 GHz,","cited_arxiv_id":null,"evidence_quote":"Defines the TR 38.901 indoor-hotspot reference model whose LOS and NLOS path-loss exponents the paper tests."},{"cited_title":"Investigation of prediction accuracy, sensitivity, and parameter stability of large-scale propagation path loss m odels for 5G wireless communications,","cited_arxiv_id":null,"evidence_quote":"Establishes the CI model's parameter stability and prediction accuracy across datasets, which is the basis for recommending CI over FI."}],"review_version":1}