{"id":"072cbc1c-4a56-4bd9-86f2-6dad1de57cca","arxiv_id":"2412.17864","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":13,"one_line_summary":"Urban measurements at 6-18 GHz show excess vegetation loss of about 1.26-1.79 dB per meter of tree depth, increasing with frequency.","lead":"Researchers measured how trees weaken 6-18 GHz wireless signals on an urban campus and fitted a simple model adding roughly 1.3 to 1.8 dB of loss per meter of foliage. The result gives 5G/6G designers a first FR3-band estimate for foliage-aware link planning.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Origin-constrained fits rely on a single LoS reference; any bias in that reference or in the Friis baseline propagates into all reported alpha(f) slopes.","rationale":"The paper's main contribution is a first 6-18 GHz urban vegetation attenuation dataset and the slope model in Table II. For that model to be credible, the excess loss extracted from the best-aligned LoS component must be truly vegetation-induced and the origin of the fit must be truly zero-loss. The paper itself acknowledges residual small-scale fading and unresolvable MPCs in Section III, so the single LoS1 anchor is not a perfect zero-loss reference. Because the fit is origin-constrained, any error in that anchor directly pivots all slopes. Further, the baseline is theoretical Friis with an assumed path-loss exponent of 2; the urban canyon environment is not demonstrated to follow free-space propagation. Since dveg and Tx-Rx distance are correlated in the measured points, a distance-dependent baseline error is aliased into alpha. This is a concrete, checkable vulnerability using the reported data or a minimal re-analysis, and it is more central to the quantitative claim than the general scattering concern raised by the reader. The qualitative trend—excess loss increases with vegetation depth and frequency—is plausible and supported by the plotted data, and the measurement campaign is valuable. No misconduct or internal inconsistency is apparent; the issue is statistical robustness of the reported slopes. Therefore the CONDITIONAL verdict remains appropriate, and no verdict change is needed.","tokens_in":6293,"tokens_out":4291,"duration_ms":42313,"concrete_test":"Re-fit the excess-loss data with an unconstrained linear model Lveg = alpha*dveg + beta for each 1-GHz band using the seven measured points (LoS1 plus Veg1-Veg6). Report beta and its 95% bootstrap interval, and recompute alpha both with LoS1 included and excluded. If beta exceeds ±2 dB, or if excluding LoS1 changes alpha by more than 0.2 dB/m in any band, the origin-constrained slopes in Table II are not robust to the single-reference baseline. Additionally, compute 95% bootstrap intervals for alpha in the 6-7 GHz and 17-18 GHz bands; if the intervals overlap, the claimed frequency trend from 1.26 to 1.79 dB/m is not statistically supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim is Table II's slope table, computed from Lveg(f,dveg)=alpha(f)*dveg with a fit forced through the origin. The only evidence for zero excess loss at dveg=0 is the single clear LoS path LoS1 (64.5 m). That point shows about 0.57 dB offset from Friis at 6.5 GHz, so the origin is not exactly measured. More importantly, all six vegetated Rx points are at different distances (74.7-126 m) and excess loss is defined relative to theoretical Friis. If the true clear-air path loss in this urban street canyon deviates from the free-space exponent of 2 (ground reflection, waveguide, building shadowing), then PFriis - PLoS contains a distance-dependent term that is not vegetation loss. Since dveg and d are correlated in this dataset (dveg generally rises with d until Veg4, then falls), the fitted alpha absorbs that baseline error. The paper provides no confidence intervals, no repeatability statistics, and no second clear-air distance, so the reported 1.26-1.79 dB/m slopes are as sensitive to the baseline assumption as to foliage. The authors' own admission of residual small-scale fading and unresolvable MPCs (Section III) means the single LoS1 reference could carry several dB of error, and with the origin constraint, that error shifts every alpha in Table II. The reader's direct-path isolation concern is valid, but the more tractable vulnerability is the baseline/origin assumption, which is testable from the existing data and directly undermines the quantitative model.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper reports ultra-wideband (6-18 GHz) directional channel measurements in an urban street canyon, comparing one clear line-of-sight (LoS1) and six vegetated receiver positions. The authors compute vegetation depth by intersecting the Tx-Rx line with ellipse models of measured tree silhouettes, isolate the best-aligned LoS component from directional power delay profiles, and define excess vegetation loss relative to Friis free-space path loss. They fit origin-constrained linear models L_veg(f, d_veg) = α(f) d_veg in 1 GHz sub-bands, reporting α rising from 1.26 dB/m at 6-7 GHz to 1.79 dB/m at 17-18 GHz, and conclude that excess loss increases with vegetation depth and frequency.","tokens_in":6651,"tokens_out":6404,"duration_ms":54978,"significance":"If the quantitative slopes are reliable, this is the first measured 6-18 GHz urban vegetation attenuation dataset and a useful input for FR3 link budgets. The manuscript's strengths are the broad 12 GHz bandwidth, the daily over-the-air calibration, the use of directional PDPs to isolate the direct path, and a clearly described geometric method for vegetation depth. The qualitative claim—excess loss grows with depth and frequency—is directly visible in the measured data and does not depend on the fitted model. However, the fitted α(f) values are in-sample parameters extracted from only six vegetated points and one no-vegetation reference, and the paper provides no confidence intervals or external validation; they should be presented as descriptive estimates rather than validated model coefficients.","major_comments":[{"comment":"The excess loss is computed relative to theoretical Friis free-space path loss, and the only obstruction-free reference is LoS1 at 64.5 m. If the true clear-air path loss in this street canyon departs from the free-space exponent of 2 (ground reflection, building/waveguide effects), then P_Friis - P_LoS contains a distance-dependent term that is not vegetation loss. Since d_veg is partly correlated with d in Table I, the origin-constrained fit in Eq. (5) absorbs that baseline error into α(f). The paper provides no confidence intervals, no second clear-air distance, and no test of the free-space baseline; the reported 1.26-1.79 dB/m slopes are thus as sensitive to the baseline assumption as to foliage. Please provide a sensitivity analysis, for example by re-fitting with a free intercept or under alternative path-loss exponents.","section":"Section III, Eq. (4); Table I"},{"comment":"The direct-path power P_LoS is extracted from the best-aligned directional PDP using a 12 dB noise threshold, and the authors explicitly acknowledge residual small-scale fading from unresolvable MPCs and canopy diffraction. This means the single LoS1 reference is not an exactly measured zero point: Fig. 3a shows a 0.57 dB offset from Friis at 6.5 GHz. Any bias of several dB in P_LoS, which the paper's own discussion allows, propagates through the origin-constrained fit into every α(f) in Table II. Please quantify this sensitivity, for example by perturbing P_LoS by a plausible baseline error and re-fitting, or by reporting the fluctuation of P_LoS across repeated measurements.","section":"Section III, Eq. (3); Fig. 3"},{"comment":"The columns α_min and α_max are never defined. It is unclear whether they are confidence bounds of the slope estimate, the observed range of α across frequency points within each 1 GHz sub-band, or something else. With six vegetated points per band, the fit has five degrees of freedom, and no R², residual plot, or per-band figure is provided for most bands. Without a definition and fit statistics, the central quantitative result is not reproducible. Please define these columns precisely and add standard errors or bootstrap confidence intervals for α.","section":"Table II; Section IV-B"}],"minor_comments":[{"comment":"The text contains 'V egetation' with an extra space in the title and in the Section II-C heading; please fix this typographical error.","section":"Title; Section II-C"},{"comment":"The statement that 'at both frequencies, for a vegetation depth of 0, the excess loss is also nearly zero' should be reworded: the zero-depth point is the single LoS1 measurement with a 0.57 dB offset at 6.5 GHz, and the zero intercept is imposed by Eq. (5), not independently measured.","section":"Section IV-B"},{"comment":"Please add estimated uncertainties to the vegetation depth values; the elliptical-silhouette method and the rod/GoPro surveying introduce measurement and modeling error that propagates into the slopes of Table II.","section":"Table I"},{"comment":"Fig. 4 displays only the 6-7 GHz and 17-18 GHz bands; because Table II reports all twelve bands, a compact summary of all fits or a table of residuals would strengthen the presentation.","section":"Fig. 4"}],"recommendation":"major_revision","confidential_remarks":"The paper is a useful measurement contribution and the qualitative trend is solid, but the quantitative slopes need a sensitivity analysis and proper uncertainty quantification before publication. I do not recommend rejection because the issues are fixable within the manuscript's scope and the underlying data appear to support the qualitative conclusion."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is the first UWB measurement of vegetation loss in the 6-18 GHz band in an urban scenario, and the qualitative result—excess loss grows with vegetation depth and frequency—is directly visible in the data. The paper earns credit for that. The quantitative model, however, is only as good as a single no-vegetation reference and an origin-constrained linear fit, and the authors' own admitted residual fading means the slope table in Table II should not be used as a calibrated model without more work.\n\nWhat's new: prior work is narrowband at 9.6 GHz (Schwering) or ITU-R P.833 interpolation between sub-3 GHz and above-30 GHz. The elliptical tree-silhouette method for computing dveg is simple and appropriate. The measurement setup, with VNA over RFoF and directional antennas, is described in enough detail to be reproducible in principle.\n\nWhere I'd push back: the conclusion \"excess loss increases with depth and frequency\" survives scrutiny. The fitted alpha values are less solid. There are only six vegetated points, one clear LoS reference (LoS1 at 64.5 m), and the fit is forced through the origin. LoS1 shows about 0.57 dB offset from Friis at 6.5 GHz, so the origin is not actually measured. More importantly, all excess loss is computed relative to theoretical Friis with path-loss exponent 2. In an urban street canyon, the true clear-air path loss may deviate from free space, and since dveg and distance are correlated in this dataset, any such baseline error leaks into alpha. The stress-test note is right that the baseline/origin assumption is the most tractable vulnerability. The paper provides no confidence intervals, no repeatability statistics, and no second clear-air distance to anchor the origin.\n\nThe authors do acknowledge variability: they note residual small-scale fading and unresolvable MPCs in Section III, and say in Section IV-B and the conclusion that additional measurements are needed to refine the model. That is honest, but it means Table II should be labeled as preliminary, not as a model.\n\nBottom line: a useful first dataset and a sensible measurement concept. The qualitative claim is supported. The quantitative slopes need uncertainty quantification and a robustness check against the baseline assumption before they go into a link-budget standard. I'd send it to a serious referee rather than desk-reject; a good referee can ask for the sensitivity analysis. I would cite it as the first UWB FR3 vegetation measurement, but not for the exact alpha values.","headline":"First 6-18 GHz UWB vegetation-loss dataset with a plausible qualitative trend, but the slope table rests on a single clear-air reference and an origin-constrained fit.","tokens_in":7239,"tokens_out":2127,"would_cite":true,"duration_ms":18938,"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 claims vegetation loss in urban 6–18 GHz links grows linearly with foliage depth and with frequency, giving a model Lveg(f,dveg)=alpha(f)dveg, with alpha from 1.26 dB/m at 6–7 GHz to 1.79 dB/m at 17–18 GHz.","keywords":["vegetation loss","upper mid-band","FR3","6-18 GHz","foliage attenuation","urban propagation","channel measurement","link budget"],"falsifier":"Repeat the same six links with a double-directional sounder that resolves azimuth at both ends and with a wider delay gate; if the best-aligned line-of-sight power changes by more than the quoted alpha-min-to-alpha-max spread, the fitted alpha values include unresolvable scattering rather than vegetation attenuation alone. Alternatively, place a foliage-free reference receiver at the same distance as each vegetation point to verify that the Friis and over-the-air calibration account for all non-vegetation urban losses.","tokens_in":6074,"feed_emoji":"🌳","tokens_out":6072,"duration_ms":52159,"temperature":0.7,"pith_summary":"The paper reports ultra-wideband measurements of vegetation-induced signal loss in the 6–18 GHz upper mid-band in an urban street-canyon setting. It claims that the excess loss grows linearly with the depth of foliage crossed by the line of sight, with a slope that rises from about 1.26 dB/m at 6–7 GHz to 1.79 dB/m at 17–18 GHz. If correct, this gives network designers a simple frequency-dependent foliage term for FR3 link budgets, replacing extrapolations based on measurements below 3 GHz and above 30 GHz. The paper also contributes a practical ellipse-based method for computing vegetation depth from tree geometry, which makes the model directly usable in planning tools.","feed_headline":"Foliage costs FR3 links up to 1.8 dB per meter","feed_subtitle":"Ultra-wideband urban measurements yield a simple frequency-dependent vegetation-loss model for 6–18 GHz design.","key_machinery":"The central machinery is a three-step chain: an ellipse-based geometric model of each tree silhouette gives the vegetation depth dveg along the line of sight; directional power delay profiles are used to extract the best-aligned line-of-sight component PLoS(f;d), cleaned by a 12 dB noise threshold and delay gating; and the excess loss Lveg = PFriis - PLoS is fit by a linear model through the origin, Lveg(f,dveg) = alpha(f)dveg. The ellipse representation is load-bearing because it turns on-site measurements of trunk height, foliage height, and width into a solvable intersection problem for arbitrary path geometry.","core_discovery":"Using six receiver positions with modeled vegetation depths from 0 to about 28 meters, the paper isolates the best-aligned line-of-sight component in directional power delay profiles, subtracts the Friis free-space power at the same distance and frequency, and treats the remainder as vegetation loss. A linear regression forced through the origin yields an attenuation coefficient alpha(f) for each 1 GHz sub-band; the fitted slopes rise from 1.26 dB/m at 6–7 GHz to 1.79 dB/m at 17–18 GHz, with a min–max spread that widens with frequency. The paper concludes that vegetation loss in FR3 is frequency-dependent and approximately linear in foliage depth, and presents this as the first measured dataset covering the entire 6–18 GHz range in an urban environment.","pith_inferences":["If the linear-depth law holds across tree species, foliage-aware FR3 planning could rely on crown silhouettes alone, avoiding detailed leaf-level ray tracing.","Measuring the same links with full double-directional resolution would separate true vegetation attenuation from scattering and canopy diffraction, tightening the reported alpha intervals.","Comparing these alphas against existing vegetation attenuation standards would identify which frequency bands need revised assumptions, since those standards currently interpolate across the 6–18 GHz gap.","Because the reference path has no foliage, any non-vegetation urban loss not captured by Friis would be attributed to the trees; repeating with a foliage-free path at each distance would bound this error."],"forward_implications":["At the upper end of FR3, each meter of tree canopy along the path adds roughly 1.8 dB of loss, which can dominate the link budget at cell-edge distances.","The frequency trend means operators can assign foliage-sensitive links to the lower part of FR3, where the per-meter vegetation penalty is smaller.","The linear depth model makes vegetation loss a tractable planning input: compute the ellipse-based depth for a path and multiply by alpha(f).","The scatter around the fitted lines implies that a single alpha per GHz band is a first-order model, and additional measurements are needed before it generalizes to other vegetation types and densities."],"supporting_citations":[{"why":"Supplies the RF-over-fiber VNA sounder design and measurement methodology used to capture the 6–18 GHz ultra-wideband data.","marker":"[8]"},{"why":"Provides the noise-threshold and delay-gating procedure used to clean the power delay profiles before extracting the LoS component.","marker":"[11]"},{"why":"Gives the Friis free-space power formula that the measured LoS power is subtracted from to define excess vegetation loss.","marker":"[12]"},{"why":"The prior narrowband 9.6 GHz vegetation study that this campaign extends to the whole 6–18 GHz band.","marker":"[9]"},{"why":"The existing vegetation attenuation recommendation whose measurements below 3 GHz and above 30 GHz leave the FR3 range interpolated, motivating the new data.","marker":"[10]"}],"fun_headline_variants":["Vegetation loss in FR3 rises 1.26–1.79 dB/m with frequency","Foliage fades FR3 signals: up to 1.8 dB/m at 18 GHz","Urban foliage cost: 1.3–1.8 dB/m across 6–18 GHz","New model quantifies foliage loss for 6–18 GHz links"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The model assumes that the strongest directional line-of-sight component picked out of the measured power delay profiles is exactly the signal that passed through the elliptical foliage depth, so that subtracting the Friis free-space power leaves only vegetation loss.","fun_headline_variants_meta":{"raw":{"variants":["Vegetation loss in FR3 rises 1.26–1.79 dB/m with frequency","Foliage fades FR3 signals: up to 1.8 dB/m at 18 GHz","Urban foliage cost: 1.3–1.8 dB/m across 6–18 GHz","New model quantifies foliage loss for 6–18 GHz links"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000671,"raw_usage":{"total_tokens":2992,"prompt_tokens":813,"completion_tokens":2179,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":429,"completion_tokens_details":{"reasoning_tokens":2083}},"tokens_in":429,"tokens_out":2179,"duration_ms":12625,"temperature":1.0,"reasoning_tokens":2083,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T10:40:43.821400+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the same six links with a double-directional sounder that resolves azimuth at both ends and with a wider delay gate; if the best-aligned line-of-sight power changes by more than the quoted alpha-min-to-alpha-max spread, the fitted alpha values include unresolvable scattering rather than vegetation attenuation alone. Alternatively, place a foliage-free reference receiver at the same distance as each vegetation point to verify that the Friis and over-the-air calibration account for all non-vegetation urban losses.","supporting_citations":[{"cited_title":"Impact of noisy measurements wi th fourier-based evaluation on condensed channel parameters ,","cited_arxiv_id":null,"evidence_quote":"Provides the noise-threshold and delay-gating procedure used to clean the power delay profiles before extracting the LoS component."},{"cited_title":"Recommendation ITU-R P .833-10 attenuati on in vegeta- tion,","cited_arxiv_id":null,"evidence_quote":"The existing vegetation attenuation recommendation whose measurements below 3 GHz and above 30 GHz leave the FR3 range interpolated, motivating the new data."}],"review_version":1}