{"id":"017fefb0-007d-401e-a0a4-75300a14a593","arxiv_id":"2509.02088","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":18,"one_line_summary":"A 300 GHz monostatic sensing campaign in three indoor scenarios produces cluster-level channel statistics and an environment-aware framework mapping reflector geometry, roughness, and material to reflection-loss and dispersion observables.","lead":"This paper measures 300 GHz radio reflections in three indoor rooms at 57 positions and builds a framework linking what the radio sees to the walls and objects that caused the reflections. It matters because future 6G networks that both communicate and sense will need models connecting real environments to wireless channel features.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Central claim of reliable physical-attribute inference lacks any held-out validation; observed channel differences are not shown to support extraction.","rationale":"The reader's weakest-assumption analysis correctly identified the roughness ranking as a fragile premise: Section V-A explicitly ranks surfaces by 'visual inspection and construction standards' without measuring RMS height, and Eq. (13) is never evaluated. If that ordering is wrong, the Level-2 roughness inference collapses. My concern is broader and, in my view, more load-bearing: the paper never validates any level of the proposed 'environment-aware inference' framework against held-out data. The experimental results demonstrate statistically distinguishable channel observables, but the central claim is about extraction—recovering physical attributes from channel data. The paper only fits models to data with known labels and then asserts reliability. Section V-C's closing sentence explicitly defers confusion-matrix analysis, which is an in-text admission that classification performance is unquantified. I agree with the reader's verdict of CONDITIONAL: the measurement dataset and statistical characterizations are valuable, and the missing validation is addressable. Thus no verdict change is warranted, but the conditions should explicitly require a held-out classification experiment and, if possible, profilometer-based roughness measurements. My attack is 'partial' agreement because the reader located the weakness in a specific assumption (roughness ordering) while I locate it in the absence of any inference validation, which encompasses that assumption but also affects material and geometry inference at all levels.","tokens_in":17610,"tokens_out":2596,"duration_ms":31748,"concrete_test":"Use the existing 57-TRx dataset to run a supervised evaluation: split TRx positions into training and test sets (e.g., leave-one-scenario-out or random 70/30 split). For each test cluster, predict material type from the fitted specular/diffuse reflection-loss distributions and predict roughness class/order from diffuse MPC count and Lambertian slope. Report a confusion matrix and accuracy against the known layout ground truth. Separately, measure the RMS surface height σ of the polymer, cement, and tile walls with a profilometer or laser scanner; check whether the measured roughness ordering matches the visual ranking used in Section V-A. If accuracy is high and the measured ordering agrees, the central claim is supported; otherwise, the claim must be qualified to 'distinguishable distributions' rather than 'reliable extraction.'","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract and conclusion claim that the proposed approach can 'reliably extract physical characteristics' from observed channel characteristics. The experimental evidence in Sections V-A through V-C shows that channel observables—diffuse MPC count, angular span, Lambertian slope, and reflection-loss distributions—differ across labeled materials and roughness classes. However, every physical label is assigned using the known measurement layout and, for materials, specular reflection loss (Sections V-B.2 and V-C). The models are then fitted to the same labeled data. No held-out TRx positions, no cross-validation, and no classification accuracy or confusion matrix is reported. The paper explicitly defers this: 'A confusion-matrix-based analysis for more materials would be valuable for quantifying classification performance, which we leave for future exploration' (Section V-C). Without an inference test, the statement 'reliably extract' is not supported; the framework is descriptive, not a validated estimator. In addition, the roughness ordering in Section V-A is based on 'visual inspection and construction standards' and Eq. (13) for RMS height σ is never evaluated, so the claimed monotonic trend (polymer → cement → tile) could be confounded by material composition or surface finish. These gaps leave the central claim unverified.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper reports a 300 GHz monostatic sensing channel measurement campaign at 57 co-located TRx positions across three indoor scenarios. Multipath components are extracted with a SAGE algorithm and clustered using connected-component labeling on delay-angle power profiles. The authors propose an environment-aware channel modeling framework that maps physical scene attributes (reflector count/location, surface roughness, reflector geometry, material type) to channel observables such as cluster counts, intra-cluster delay/angular spreads, reflection-loss distributions, and Lambertian scattering parameters. Statistical fits are provided for cluster counts, delay/angular spreads, and reflection losses. The paper claims that the proposed approach can reliably extract physical characteristics from channel observations.","tokens_in":18093,"tokens_out":3269,"duration_ms":38298,"significance":"If the central claim were supported, this would be a useful contribution to THz ISAC channel modeling: the measurement campaign is substantial, the cluster-level statistical characterization goes beyond prior TRx-averaged metrics, and the explicit hierarchical mapping between physical attributes and propagation observables is well organized. The paper also contains useful external anchors, such as comparing specular reflection losses with a material reflection-loss database in [20]. However, the advertised inference capability is not validated with out-of-sample tests, and the roughness and material mappings rest on partially circular or unmeasured inputs. The framework is currently a descriptive statistical characterization rather than a validated physical-attribute estimator.","major_comments":[{"comment":"The central claim that the approach can 'reliably extract physical characteristics' is not supported by an inference test. Physical labels for clusters are assigned using the known measurement geometry and, for materials, specular reflection loss (Section V-B.2 and V-C). The statistical models are then fitted to the same labeled data. No held-out TRx positions, cross-validation, classification accuracy, or confusion matrix are reported; in fact, Section V-C states that confusion-matrix analysis is left for future exploration. Thus the evidence supports a descriptive mapping, not a validated estimator. Please either add a validation experiment (e.g., train on a subset of TRx positions and test on the rest, or classify unlabeled clusters) or explicitly soften the abstract/conclusion claims from 'reliably extract/infer' to 'characterize and model.'","section":"Abstract and Section V-C"},{"comment":"The Level-2 surface-roughness inference is based on a visual ranking of polymer, cement, and tile surfaces ('visual inspection and construction standards'), while the RMS height sigma defined in Eq. (13) is never measured or evaluated. Because material composition and finish differ across the three surfaces, the observed monotonic trends in diffuse MPC count, angular span, and Lambertian slope could be confounded by material properties rather than driven by roughness. This undermines the roughness-to-channel mapping claimed in Table III. Please either measure sigma (or an independent roughness metric) for the actual surfaces, or test surfaces of the same material with controlled roughness to support the claimed trend.","section":"Section V-A, Eq. (13), Figs. 8-9"},{"comment":"There is a circularity concern in the material-identification claim. The clusters associated with each material are identified 'based on scenario geometry and specular reflection loss' (Section V-B.2) and by correlating with the known geometric layout (Section V-C). The same reflection-loss distributions are then proposed as the material fingerprint (Fig. 14). While the comparison with the material reflection-loss database in [20] provides an external anchor for mean specular loss, it does not validate the discrimination capability of the proposed features on unlabeled data. The authors should either use an independent labeling procedure (e.g., known material panels placed in the scene) or perform a separability/classification test on clusters whose material is not used during model fitting.","section":"Sections V-B.2 and V-C, Fig. 14"}],"minor_comments":[{"comment":"The text states 'Scenario 1 and TRxs 37-45 exhibit a similar number of MPC clusters. TRxs 37-45 is shown to have a higher number of clusters compared to those in Scenario 1 and TRxs 37-45.' This is internally inconsistent; the second sentence presumably refers to TRxs 46-57.","section":"Section IV-B"},{"comment":"The RMS height sigma is defined but never used in the subsequent analysis. If it is intended as a conceptual definition, state this explicitly in the main text; otherwise provide measurements for the surfaces studied.","section":"Eq. (13)"},{"comment":"The Lambertian fits report slope and intercept to two decimals without confidence intervals or goodness-of-fit metrics. Adding R^2 or RMSE for each fit would strengthen the claimed monotonic trend in n_Lam.","section":"Fig. 9"}],"recommendation":"major_revision","confidential_remarks":"The measurement campaign and statistical characterization are solid and likely of interest to the journal's readership. The main issue is that the paper overclaims a validated sensing capability. This is fixable by reframing the claims or adding a validation experiment, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"I read the 300 GHz monostatic sensing paper. The real contribution is the measurement campaign: 57 TRx positions across three indoor scenarios, with careful SAGE/CCL processing and cluster-level statistics. The material-wise specular and diffuse reflection-loss models are genuinely new relative to the authors' earlier work [19], [20], and the fits to normal/Weibull distributions look credible. The comparison of specular loss against the reference database in [20] gives an external anchor that the numbers are not arbitrary.\n\nThe soft spot is the abstract and conclusion claim that the approach can \"reliably extract physical characteristics.\" That is not established. Every material label is assigned using known geometry and specular reflection loss, and the same reflection-loss metric is then proposed as the material observable. There are no held-out TRx positions, no classification accuracy, and no confusion matrix. The paper itself defers that to future work, so as written the framework is descriptive rather than a validated estimator. The roughness ordering is based on visual inspection and construction standards, not measured RMS height; Eq. (13) is never evaluated, so the polymer→cement→tile trend is plausible but could be confounded by material composition. Also, Section IV-B contains a self-contradictory sentence about TRxs 37-45 that needs a copyedit.\n\nThese issues are addressable, and the paper is honest about the roughness limitation. The measurements seem real, the cluster statistics are internally consistent, and the modeling follows established practice. What I'd want in revision is either a held-out validation or a claim scaled back to \"consistent with material-dependent reflection behavior.\" I would not desk-reject this. It deserves a serious referee, with realistic expectation of major revision.\n\nI'd cite it for the dataset and the material-specific reflection-loss models, if I worked in THz sensing. For a reading group, maybe—useful for the THz/ISAC subfield, not a general audience. No code or data release is mentioned, which limits reproducibility, so I'd ask for that too.","headline":"Valuable THz monostatic dataset and material reflection-loss models, but the 'reliably extract physical characteristics' claim outruns the evidence; needs a validation pass and a toned-down abstract.","tokens_in":18534,"tokens_out":1898,"would_cite":true,"duration_ms":20555,"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 shows that 300 GHz monostatic sensing channels carry enough information to infer room structure, surface roughness, and material type.","keywords":["terahertz","monostatic sensing","ISAC","channel measurement","environment-aware channel modeling","multipath clustering","reflection loss","Lambertian scattering"],"falsifier":"Use a surface profilometer to measure the RMS height σ of the polymer, cement, and tile surfaces, then test whether the number of diffuse MPCs, the angular span, and the fitted Lambertian slope nLam change monotonically with σ when the material type is held fixed. If the ordering reverses or disappears—or if σ is nearly identical across the three surfaces—the Level-2 roughness inference collapses.","tokens_in":17562,"feed_emoji":"📡","tokens_out":6030,"duration_ms":62720,"temperature":0.7,"pith_summary":"This paper reports a 300 GHz monostatic sensing measurement campaign across 57 co-located transceiver positions in three indoor environments, and proposes an environment-aware channel model that ties physical scene attributes to channel observables. Using a SAGE algorithm to extract multipath components and an image-processing clustering method to group them, the authors show that cluster counts, intra-cluster delay and angular spreads, Lambertian scattering slopes, and reflection-loss distributions vary systematically with reflector quantity, surface roughness, geometry, and material type. The central claim is that physical characteristics—such as structural layout, roughness class, and material identity—can be read from observed channel characteristics alone. If true, this gives terahertz ISAC systems a path from raw channel measurements to environmental understanding without requiring separate sensors.","feed_headline":"Echo patterns at 300 GHz expose room layout, roughness, and materials","feed_subtitle":"A 57-position campaign maps reflector geometry, surface roughness, and material to channel statistics for THz sensing.","key_machinery":"The machinery that carries the argument is the measurement-plus-clustering pipeline and the mapping table built on it. A 290-310 GHz VNA-based channel sounder with rotating 8-degree horn antennas collects directional channel frequency responses at 57 positions. A trajectory-tracking SAGE algorithm de-embeds the antenna pattern and estimates each multipath component's amplitude, delay, and azimuth angle. Connected component labeling, applied to a thresholded power-angle-delay image after morphological closing, groups these MPCs into delay-angle clusters. From these clusters the framework reads physical attributes: cluster count for reflectors, diffuse MPC statistics and Lambertian slope for r","core_discovery":"The core discovery is a four-level environment-aware mapping for 300 GHz monostatic sensing. At Level 1, the number of MPC clusters and their delay-angle centroids reveal reflector quantity and location. At Level 2, the number of diffuse MPCs within a cluster, the angular span they cover, and the fitted slope of a Lambertian cos^2(Δθ) power model indicate surface roughness: rougher surfaces scatter more diffuse energy over wider angles and yield smaller Lambertian slopes. At Level 3, cluster shape and intra-cluster delay and angular dispersion distinguish flat walls, concave corners, convex corners, and cylindrical pillars, demonstrated by structural-model matching that reconstructs the meas","pith_inferences":["The fixed 1.2 m TRx-wall distance means the reflection-loss separation is demonstrated under controlled geometry; a direct extension is to sweep distance and verify that calibrated reflection loss remains material-discriminative, which the paper flags as future work.","The ordinal roughness ranking (polymer rougher than cement rougher than tile), based on visual and construction standards, could be promoted to a quantitative estimator by regressing measured RMS height against Lambertian slope and diffuse MPC count; this would test whether the trend is monotonic in σ and not merely material-correlated.","Because specular and diffuse reflection-loss distributions are fitted by different families (normal vs Weibull), a generative classifier could invert them to posterior material probabilities; the paper leaves confusion-matrix analysis for future work.","The image-processing clustering step suggests a bridge to learned segmentation: if CCL labels are treated as pseudo-labels, a neural network could be trained to map raw power-angle-delay images to cluster regions, potentially generalizing beyond the fixed distance and manual threshold."],"forward_implications":["A THz ISAC node could localize and count dominant reflectors from the number and delay-angle centroids of MPC clusters, without a separate mapping pass.","Surface roughness classes can be assigned from diffuse MPC count and Lambertian slope, enabling scattering-aware channel simulation from simple wall categories.","Reflector geometry (flat wall, concave corner, convex corner, cylinder) is distinguishable from the spatial pattern of a single cluster, so layout reconstruction can be driven by the channel itself.","Material identification becomes statistical: normal models for specular reflection loss and Weibull models for diffuse reflection loss give class-separable features for metal, glass, cement, tile, and polymer.","The fitted cluster-level distributions can seed network-level simulations of monostatic sensing performance, linking physical environment models to ISAC link budgets."],"supporting_citations":[{"why":"Supplies the trajectory-tracking SAGE estimator, Scenario 1 description, and the prior hybrid monostatic channel model this work extends.","marker":"[19]"},{"why":"Provides the material reflection-loss reference database and prior cement/metal material identification that this work generalizes to more materials.","marker":"[20]"},{"why":"Introduces the direction-scan sounding-oriented SAGE algorithm used for high-resolution MPC parameter extraction.","marker":"[26]"},{"why":"Gives the Lambertian cos^2 scattering model used to fit diffuse power versus angle for roughness analysis.","marker":"[41]"},{"why":"Defines RMS surface height, the roughness metric the paper invokes, and supports the scattering context at THz frequencies.","marker":"[40]"},{"why":"Provides the efficient dilation, erosion, opening, and closing algorithms underlying the connected-component-labeling clustering pipeline.","marker":"[29]"},{"why":"Describes the VNA-based channel sounder structure used for the 290-310 GHz measurements.","marker":"[25]"}],"fun_headline_variants":["THz echoes map room layout, roughness, and materials","From 300 GHz reflections: room geometry and surface texture","Monostatic THz sensing reads indoor structure from signals","Echo patterns decode surface roughness and reflector shape","300 GHz channel data reveals walls, corners, and pillars"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The load-bearing premise is that surface roughness is correctly ordered by visual inspection and construction standards (polymer roughest, then cement, then tile) and that this ordering, rather than material composition or construction differences, drives the observed differences in diffuse MPC count, angular span, and Lambertian slope.","fun_headline_variants_meta":{"raw":{"variants":["THz echoes map room layout, roughness, and materials","From 300 GHz reflections: room geometry and surface texture","Monostatic THz sensing reads indoor structure from signals","Echo patterns decode surface roughness and reflector shape","300 GHz channel data reveals walls, corners, and pillars"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000219,"raw_usage":{"total_tokens":1302,"prompt_tokens":787,"completion_tokens":515,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":531,"completion_tokens_details":{"reasoning_tokens":437}},"tokens_in":531,"tokens_out":515,"duration_ms":5758,"temperature":1.0,"reasoning_tokens":437,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T11:53:42.352208+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Use a surface profilometer to measure the RMS height σ of the polymer, cement, and tile surfaces, then test whether the number of diffuse MPCs, the angular span, and the fitted Lambertian slope nLam change monotonically with σ when the material type is held fixed. If the ordering reverses or disappears—or if σ is nearly identical across the three surfaces—the Level-2 roughness inference collapses.","supporting_citations":[{"cited_title":"Hybrid channel modeling and environment reconstruction for terahertz monostatic s ensing,","cited_arxiv_id":null,"evidence_quote":"Supplies the trajectory-tracking SAGE estimator, Scenario 1 description, and the prior hybrid monostatic channel model this work extends."},{"cited_title":"Centimeter-level geometry r econstruction and material identification in 300 GHz monostatic sensing,","cited_arxiv_id":null,"evidence_quote":"Provides the material reflection-loss reference database and prior cement/metal material identification that this work generalizes to more materials."},{"cited_title":"DSS-o-SAGE: Dir ection- scan sounding-oriented SAGE algorithm for channel paramet er estima- tion in mmWave and THz bands,","cited_arxiv_id":null,"evidence_quote":"Introduces the direction-scan sounding-oriented SAGE algorithm used for high-resolution MPC parameter extraction."},{"cited_title":"Measurement and modelling of scattering from buildings,","cited_arxiv_id":null,"evidence_quote":"Gives the Lambertian cos^2 scattering model used to fit diffuse power versus angle for roughness analysis."},{"cited_title":"Full-wave simulation and scattering modeling fo r terahertz communications,","cited_arxiv_id":null,"evidence_quote":"Defines RMS surface height, the roughness metric the paper invokes, and supports the scattering context at THz frequencies."},{"cited_title":"Efficient dilation, erosion, op ening, and closing algorithms,","cited_arxiv_id":null,"evidence_quote":"Provides the efficient dilation, erosion, opening, and closing algorithms underlying the connected-component-labeling clustering pipeline."},{"cited_title":"Sub-THz VNA-based chann el sounder structure and channel measurements at 100 and 300 GHz,","cited_arxiv_id":null,"evidence_quote":"Describes the VNA-based channel sounder structure used for the 290-310 GHz measurements."}],"review_version":1}