{"id":"59bc4cdb-810f-4639-a658-a4671dff82a6","arxiv_id":"1908.00512","paper_version":3,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":10,"one_line_summary":"A large 28 GHz measurement campaign in Manhattan and Valparaíso produces empirical path gain models for rooftop-to-street coverage, including offset loss and around-corner propagation.","lead":"This paper measures 28 GHz signal propagation from rooftop base stations into city streets, using over 3000 links and 21 million power samples. It finds 11 dB excess path loss at 200 m and shows that high-gain antennas retain their advantage in street canyons.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Untested P_all-to-omnidirectional equivalence may bias all absolute path gains; a roof-edge receiver's narrow elevation beam could miss out-of-plane power, shifting Eq. (1).","rationale":"The reader's weakest_assumption is exactly the P_all-to-omnidirectional equivalence cited from [23]. I agree that this is the most load-bearing point: the absolute path-gain values, the excess-loss statement, and the coverage simulation all depend on it. The paper does not re-derive the equivalence for the rooftop-to-street geometry, and the schematic of the spinning horn plus the large range of elevation angles to the street-level terminal makes a distance-dependent bias plausible. If the bias is real, the central numerical contribution of the paper (Eq. (1) and the 11 dB excess loss) would need revision, although the qualitative findings (steep slope, large street-to-street variability, small azimuthal gain degradation) would likely survive. Because the issue is concrete and checkable, and because the paper's headline numbers are its main contribution, acceptance should be conditional on a validation that P_all-minus-Gelev is within a specified tolerance (e.g., 1 dB) of the true omnidirectional path gain over the measured distance range. The paper has strong positive aspects: a very large dataset, careful absolute-power calibration of the receiver chain, explicit confidence intervals, and openly stated limitations. None of these are undermined by this concern. The recommendation is therefore CONDITIONAL rather than ACCEPT: the authors should provide the elevation-scan validation or a quantitative sensitivity analysis, after which the paper can be accepted without further changes.","tokens_in":14139,"tokens_out":12545,"duration_ms":141851,"concrete_test":"Re-measure a subset of same-street links on at least one Manhattan street and one Valparaiso street using the same transmit terminal but replace the azimuth-only scan with a full-sphere measurement: either tilt the 10-degree horn over a grid of elevation angles at each azimuth, or use a reference antenna with broad elevation beamwidth (e.g., a biconical or half-wave dipole) to obtain the true omnidirectional received power. Compare the resulting path gain against the P_all-minus-Gelev values used in Eq. (1) across distances from 35 m to 500 m. If the difference is larger than 1 dB and varies with distance, the reported intercept and slope are biased.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central path-gain model in Eq. (1) is computed as PG = P_all - PT - GT - Gelev, where P_all is the azimuthal average of received power. Section II.C claims P_all is equivalent to the omnidirectional average power, citing a derivation in the authors' prior indoor-corridor paper [23] without re-deriving it for the outdoor street-canyon geometry. This equivalence is load-bearing because every absolute path-gain value, the 1-m intercept A=-35.0 dB, the exponent n=-3.56, and the headline 11 dB excess loss at 200 m are obtained from this assumption. The concern is that the spinning 10-degree receive horn scans only in azimuth, with a fixed elevation orientation. In the rooftop-to-street geometry, the direct path from a 1.5 m-high terminal to a 15-51 m-high roof-edge receiver has an elevation angle that varies strongly with distance, e.g., from near 2 degrees at 500 m to over 25 degrees at 35 m. If the channel's elevation angular spread is not small compared with the horn's 10-degree elevation beamwidth, or if the effective elevation centroid of the received power differs from the Gelev value subtracted, then P_all will be a biased estimate of the true omnidirectional average. The bias would generally be distance-dependent, so it would affect not just the absolute intercept but also the fitted slope. The paper's justification relies on 3GPP 38.901 elevation-spread values, which are model assumptions rather than a validation in the measured environment, and [23] was for indoor corridors where elevation spread is much smaller. Additionally, the transmit antenna elevation pattern is not documented; if the '2 dBi omnidirectional' antenna has less gain toward the elevated receiver than its nominal horizon gain, path gains are systematically biased as well. Thus every absolute dB value in the paper, including the coverage predictions in Section V, is exposed to this calibration uncertainty.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper reports an extensive 28 GHz measurement campaign in urban street canyons, using a rotating 10-degree horn receiver at rooftop macro sites and a low-height terminal on the street. Over 3000 links and 21 million CW power samples were collected across 12 streets in Manhattan and Valparaíso. The paper presents empirical slope-intercept path gain models for same-street roof-edge (Eq. (1), A=-35.0, n=-3.56, σ=7.1 dB), offset-from-roof-edge (Eq. (2)), and lamppost deployments, plus an around-corner diffraction-inspired model with an empirical corner loss of about 2 dB. It also quantifies effective azimuthal gain degradation, finding that 90% of roof-edge azimuth gains are within 2 dB of the nominal antenna gain, and uses the models in a system-level simulation to show that 90% of outdoor users can achieve 350 Mbps or higher at 400 m ISD.","tokens_in":14499,"tokens_out":12533,"duration_ms":118651,"significance":"The dataset is exceptionally large for mmWave street-canyon measurements, and the system calibration (0.15 dB absolute accuracy) is a strength. If the models are validated, they provide a useful empirical basis for 28 GHz coverage planning and for assessing the value of high-gain antennas in urban canyons. The finding that standard ray tracing overpredicts path gain by about 13 dB at 200 m is an important caution for simulation-based planning. The around-corner diffraction model with only two fitted parameters and 3.4 dB RMSE is a compact, useful result. However, the absolute path gain values rest on an azimuth-averaging equivalence that is not independently validated in this geometry, and one headline excess-loss number appears to be inconsistent with the fitted equation.","major_comments":[{"comment":"The derivation of PG from P_all relies on the equivalence between the azimuthal average of the spinning-horn received power and the omnidirectional average power, citing equations (1)-(6) of [23], and on subtracting a nominal elevation gain Gelev. In the rooftop-to-street geometry, the elevation angle of the direct path to the 1.5 m-high terminal varies from about 2 degrees at 500 m to more than 20 degrees at 35 m (for roof heights of 15-51 m). Because the receive horn has a 10-degree elevation beamwidth, the assumption that the effective elevation gain equals the nominal Gelev at all ranges is not substantiated. A distance-dependent elevation bias would shift both the intercept A and the exponent n in Eq. (1), and thus the headline 11 dB excess loss at 200 m. The paper's justification, citing 3GPP 38.901 elevation-spread values, is a model assumption rather than a measurement in this environment. Please validate the equivalence (e.g., with a subset of links measured using an omnidirectional receive antenna) or quantify the elevation gain error; at minimum, state the horn's elevation pointing direction and the gain variation over the relevant elevation angles.","section":"II.C, Eq. (1)"},{"comment":"The claim that Eq. (1) implies an 11 dB excess loss at 200 m and 20 dB at 500 m relative to free space is inconsistent with the fitted parameters. Using A=-35.0, n=-3.56, and λ=10.7 mm, the excess loss is approximately 9.5 dB at 200 m and 15.7 dB at 500 m. Since these numbers appear in the abstract and conclusions, please recalculate and correct them, or clarify the reference used for the free-space baseline.","section":"III.A"},{"comment":"The second line of the diffraction model formula is dimensionally inconsistent as printed (the logarithm's argument has units of m^3). Because the fitted parameters in Table 2 and the conclusion that the diffraction model is best depend on this equation, please restate the equation cleanly and ensure it matches the model actually used for fitting.","section":"III.D, Eq. (3)"}],"minor_comments":[{"comment":"The Introduction states that the models predict rates exceeding 300 Mbps for 90% of outdoor locations, while Section V reports 350 Mbps; please standardize the cited number.","section":"I"},{"comment":"The paper does not describe the elevation orientation (tilt) of the rotating receive horn; adding this detail would make the measurement setup reproducible and would help readers assess the elevation-gain assumption.","section":"II.A"},{"comment":"The text says the scattering model with Friis intercept has a corner loss of 0 dB; this means the model simply reduces to a single-slope model. It may help to state this explicitly when comparing models.","section":"III.D"},{"comment":"The claim that the data set allows 90% confidence intervals of under 1 dB for path gain and under 0.5 dB for effective directional gain is not directly tied to the reported ±2.7 dB intercept and ±0.12 slope intervals in Eq. (1); please clarify what quantity each confidence interval refers to.","section":"I"}],"recommendation":"major_revision","confidential_remarks":"The P_all-to-omnidirectional equivalence is the key technical risk; if the authors can provide a validation measurement or a thorough uncertainty analysis of the elevation-gain assumption, I would be willing to accept. The excess-loss numerical discrepancy is easy to correct. Overall, the paper is a strong empirical contribution with a uniquely large dataset, and the proposed revisions are within the scope of a revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's what I'd tell you about 1908.00512: it's a genuinely large and well-executed measurement paper, and the headline numbers are probably about right. The dataset—3000 links, 21 million samples, 12 streets in two cities—is a real step beyond the earlier work, which mostly used a handful of transmitter locations. The calibration looks careful (0.15 dB accuracy), the fit parameters have tight confidence intervals, and the comparisons to 3GPP models and simple ray tracing are useful. I'd take the roof-edge same-street model Eq. (1), the 15 dB offset loss at 100 m, and the 2-dB azimuthal gain degradation as useful engineering numbers.\n\nThe main soft spot is the measurement-to-path-gain inversion. The P_all-to-omnidirectional equivalence is cited to the authors' earlier indoor-corridor paper, and it isn't re-derived or validated for the rooftop-to-street geometry. The stress-test worry about the receive horn's elevation pattern is plausible in principle, but I don't think it lands as a fatal flaw: if the horn were pointed at the horizon, the short-range links (35-40 m) would be far off-boresight and wouldn't sit close to free space, which the paper says they do. That suggests the authors are applying the correct elevation gain for the direct path. Still, they never show the elevation pattern or explain how Gelev is computed, and the 3GPP elevation-spread justification is a model, not a measurement in this environment. Because every absolute dB value—the intercept, the exponent, the 11 dB excess loss—flows through that step, this is a documentation and sensitivity-analysis gap, not a fatal one. A short appendix with the antenna pattern and a plausibility check would fix it.\n\nTwo smaller things: the around-corner model is based on one intersection in Manhattan (98 links), so it's a good start but not a universal law; and the '90% coverage' claim is a simulation using the fitted models, not a measured coverage rate. Neither is hidden.\n\nFor whom: channel modelers, mmWave deployment planners, and anyone building on 28 GHz urban models. It deserves a serious referee and, after the calibration question is answered, acceptance.","headline":"Large, careful 28 GHz street-canyon measurement campaign with useful empirical models; the absolute values lean on a lightly documented antenna-calibration assumption that the authors should pin down before it is cited as gospel.","tokens_in":15185,"tokens_out":9583,"would_cite":true,"duration_ms":107308,"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":"This paper establishes an empirical 28 GHz path-gain law for rooftop base stations serving street-level terminals, including a 15 dB penalty when the antenna is set back from the roof edge, and shows that high-gain antennas retain nearly…","keywords":["28 GHz propagation","street canyon","path gain model","rooftop base station","directional antenna gain","millimeter wave coverage","around-corner propagation"],"falsifier":"Place a calibrated omnidirectional receiver at the same rooftop locations as the spinning horn and compare its average power with the azimuthal average $P_{\\mathrm{all}}$; a discrepancy larger than about 1 dB would break the equivalence and shift every path-gain and excess-loss number.","tokens_in":13921,"feed_emoji":"📡","tokens_out":9033,"duration_ms":87574,"temperature":0.7,"pith_summary":"The paper establishes how far a 28 GHz rooftop base station can reach street-level users in dense urban canyons. From more than 3,000 measured links on 12 streets in two cities, it reports a same-street path-gain law of $PG = -35.0 + 10(-3.56)\\log_{10}(d)$ dB with a 7.1 dB shadowing spread, which means 11 dB excess loss over free space at 200 meters. It also quantifies two deployment-relevant effects: setting the antenna 5 meters back from the roof edge adds about 15 dB loss at 100 meters, and turning around a corner costs roughly 14 dB after 10 meters. These numbers matter because they let network planners predict 90% outdoor coverage and confirm that high-gain directional antennas keep most of their nominal gain in scattered street environments.","feed_headline":"28 GHz rooftop coverage law: 11 dB loss over free space at 200 m","feed_subtitle":"A 3,000-link campaign shows high-gain antennas keep 90% of their gain in real street canyons.","key_machinery":"The measurement core is a rotating 10-degree receive horn (24 dBi) at rooftop height that takes full 360-degree azimuth power scans while a street-level transmitter emits a 28 GHz continuous wave. The load-bearing identity is that the azimuthal average of received power equals the average omnidirectional power, a derivation taken from the authors' earlier corridor work; subtracting transmit power and nominal gains converts that average into an omnidirectional path gain. The modeling core is a slope-intercept path-gain fit in log-distance, plus a diffraction-inspired around-corner formula in which the corner acts as a re-radiating source with an empirical corner-loss term. Directional effectiveness is captured by azimuth gain, defined as the ratio of peak received power to the full-azimuth average.","core_discovery":"On the paper's own terms, the central discovery is a set of empirical propagation laws for 28 GHz urban street canyons measured from real rooftop heights. For roof-edge base antennas with a street directly in view, path gain follows Eq. (1): $PG = -35.0 + 10(-3.56)\\log_{10}(d) + N(0, 7.1\\ \\text{dB})$ over 35 to 500 meters, equivalent to 11 dB excess loss at 200 meters and 20 dB at 500 meters relative to free space. Offsetting the antenna 5 meters from the roof edge causes 15 dB additional loss at 100 meters, a penalty that shrinks as range grows. Around a corner, a single-slope diffraction-inspired model with a 2.2 dB empirical corner loss fits the data with 3.4 dB RMS error. Finally, 90% of measured effective azimuth gains at the roof edge lie within 2 dB of the antenna's nominal gain, so scatter does not destroy the value of high-gain base antennas in street canyons.","pith_inferences":["A seasonal repeat of the tree-lined street, measured here without leaves, would separate canopy attenuation from trunk-and-branch scatter; the steep $-8.1$ distance exponent likely sets a lower bound on foliage loss.","The small angular spread behind the 2 dB gain degradation suggests that even narrower-beam antennas than the tested 10-degree horn could retain most of their gain, but this campaign did not test them.","The near-zero corner loss compared with theoretical deep-shadow diffraction implies that street furniture acts as secondary scatterers, which would push ray tracing toward diffuse-scattering terms rather than specular reflections alone.","Because the setback penalty shrinks with distance, system-level link budgets should apply the offset fit as a distance-dependent model rather than a constant loss."],"forward_implications":["Network planners can use the roof-edge path-gain law directly for same-street coverage: at 200 meters the budget must include 11 dB beyond free space plus a 7.1 dB shadow margin.","Roof-edge placement is worth about 15 dB at 100 meters; offset deployments trade concealment for a distance-dependent range penalty that largely disappears at long range.","Around-corner coverage follows a single-slope diffraction model with only about 2 dB corner loss, so street-canyon coverage extends around corners with predictable, modest drops.","High-gain base antennas are effective: 90% of locations lose less than 2 dB of nominal azimuth gain, supporting 24 dBi-class arrays for 28 GHz.","At 400-meter inter-site distance, about 12 sites per square kilometer, 90% of outdoor users can get theoretical rate limits above 350 Mbps with 800 MHz of bandwidth."],"supporting_citations":[{"why":"Supplies the derivation that the full-azimuth average received power equals omnidirectional average power, the basis for all absolute path-gain values, and the corridor corner models adapted here.","marker":"[23]"},{"why":"Provides the standardized urban macro LOS and NLOS channel models used as the comparison baseline, showing 12 to 17 dB RMS deviation from the measured data.","marker":"[20]"},{"why":"Supplies the theoretical edge-diffraction coefficient used to contrast with the fitted 2.2 dB empirical corner loss.","marker":"[28]"},{"why":"Provides prior street-canyon corner measurements and the empirical scattering-coefficient formulation that this paper's around-corner results are compared with.","marker":"[29]"},{"why":"Gives the dual-slope corner path-loss model and clutter-free ray-tracing LOS results against which the diffraction-inspired model is fitted and found to perform better.","marker":"[9]"},{"why":"Provides earlier 28 GHz around-corner NLOS fits that motivate a street-canyon-specific corner model.","marker":"[17]"},{"why":"Gives the nonparametric confidence-interval method used for the 90% bounds on the fitted path-gain and azimuth-gain distributions.","marker":"[27]"}],"fun_headline_variants":["3000 links, 21M measurements: 28 GHz street canyon laws","11 dB loss at 200m: new 28 GHz rooftop propagation model","Corner loss at 28 GHz: diffraction model fits with 3.4 dB error","High-gain antennas survive urban canyons: 2 dB max loss for 90%","28 GHz street canyons: 11 dB loss, but high-gain antennas still win"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The absolute path-gain numbers rest on the assumption that averaging power over the rotating horn's full 360-degree sweep equals what an omnidirectional antenna would receive; if power arrives outside the scanned horizontal plane, every reported gain shifts.","fun_headline_variants_meta":{"raw":{"variants":["3000 links, 21M measurements: 28 GHz street canyon laws","11 dB loss at 200m: new 28 GHz rooftop propagation model","Corner loss at 28 GHz: diffraction model fits with 3.4 dB error","High-gain antennas survive urban canyons: 2 dB max loss for 90%","28 GHz street canyons: 11 dB loss, but high-gain antennas still win"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000901,"raw_usage":{"total_tokens":3902,"prompt_tokens":992,"completion_tokens":2910,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":608,"completion_tokens_details":{"reasoning_tokens":2798}},"tokens_in":608,"tokens_out":2910,"duration_ms":20589,"temperature":1.0,"reasoning_tokens":2798,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:49:32.414596+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Place a calibrated omnidirectional receiver at the same rooftop locations as the spinning horn and compare its average power with the azimuthal average $P_{\\mathrm{all}}$; a discrepancy larger than about 1 dB would break the equivalence and shift every path-gain and excess-loss number.","supporting_citations":[{"cited_title":"Path Loss and Directional Gain Measurements at 28 GHz for non -line-of-sight coverage of indoors with corridors ,","cited_arxiv_id":null,"evidence_quote":"Supplies the derivation that the full-azimuth average received power equals omnidirectional average power, the basis for all absolute path-gain values, and the corridor corner models adapted here."},{"cited_title":"Study on channel model for frequencies from 0.5 to 100 GHz,","cited_arxiv_id":null,"evidence_quote":"Provides the standardized urban macro LOS and NLOS channel models used as the comparison baseline, showing 12 to 17 dB RMS deviation from the measured data."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the theoretical edge-diffraction coefficient used to contrast with the fitted 2.2 dB empirical corner loss."},{"cited_title":"Site-Specific Models of the Received Power for Radio Communication in Urban Street Canyons,","cited_arxiv_id":null,"evidence_quote":"Provides prior street-canyon corner measurements and the empirical scattering-coefficient formulation that this paper's around-corner results are compared with."},{"cited_title":"Karttunen, A","cited_arxiv_id":null,"evidence_quote":"Gives the dual-slope corner path-loss model and clutter-free ray-tracing LOS results against which the diffraction-inspired model is fitted and found to perform better."},{"cited_title":"Frequency range extension of the ITU-R NLOS path loss models applicable for urban street environments with 28 GHz measurements,","cited_arxiv_id":null,"evidence_quote":"Provides earlier 28 GHz around-corner NLOS fits that motivate a street-canyon-specific corner model."},{"cited_title":"Asymptotic minimax character of the sampl e distribution function and of the classical multinomial estimator,","cited_arxiv_id":null,"evidence_quote":"Gives the nonparametric confidence-interval method used for the 90% bounds on the fitted path-gain and azimuth-gain distributions."}],"review_version":1}