{"id":"a13e1866-2a80-4884-b433-0ab2d4777d34","arxiv_id":"2508.07305","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Static electromagnetic skins, chosen from a small codebook, improve channel-charting localization in non-line-of-sight urban settings, cutting the 90th-percentile error from over 60 m to below 25 m in simulation.","lead":"A simulation study shows that passive electromagnetic skins on building walls can improve wireless localization in dense cities by making channel fingerprints more distinct. In a ray-traced urban scenario, the best skin configuration cut the 90th-percentile positioning error from over 60 meters to less than 25 meters.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Headline gain likely reflects in-sample codebook selection: S* chosen to minimize the 90th-percentile LE on the same TPs used for evaluation, with t-SNE stochasticity unaccounted for.","rationale":"The reader identified the same weakest assumption: the codebook is selected on the evaluation set, so the reported gain may not generalize. My analysis confirms this as the most load-bearing concern because it directly undermines the paper's strongest quantitative claim. The paper does not report a validation split, repeated t-SNE runs, or code/data that would allow external checking. The qualitative finding is plausible and the full codebook envelope suggests a real effect, but the specific '<25 m' figure is not robustly supported. Therefore the reader's CONDITIONAL verdict is appropriate; I see no reason to change it. My concrete test—a held-out TP split with multiple t-SNE seeds—would settle the matter in a straightforward way. I agree with the reader's assessment that the fix is conceptually simple and should be required before the headline number is accepted.","tokens_in":8354,"tokens_out":2075,"duration_ms":20686,"concrete_test":"Split the 3200 TPs into two geographically disjoint sets: a design set (e.g., 80%) and a hold-out test set (20%). For each of the 121 EMS codebook configurations, compute the 90th-percentile LE on the design set using a fixed number of t-SNE seeds (e.g., 10); select S* as the configuration minimizing this design-set quantile. Then, using held-out TPs and fresh t-SNE runs (multiple seeds), compute the 90th-percentile LE with S*. If the held-out 90th percentile remains below 25 m and is clearly separated from the no-EMS baseline (with seed-to-seed variation reported), the concern is resolved. If it rises above 25 m or overlaps the baseline, the headline gain is an in-sample selection artifact.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim—90th-percentile localization error reduced from >60 m to <25 m—rests on a codebook search that is evaluated on the same test points used to select the best configuration. Eq. (23) minimizes Q_m(α|S) over S ∈ C, and Q_m is computed over the full TP set U. The paper then reports the 90th percentile for the resulting S* (Sec. V). This is textbook in-sample selection: choosing the best of 121 configurations on the evaluation metric and then reporting that metric on the same data introduces optimism. The headline number is the minimum of 121 correlated estimates, not a prediction for unseen positions. Additionally, t-SNE is stochastic; Eq. (11) defines an argmin that is not unique, and the paper does not report multiple initializations, seeds, or variation across runs. A different t-SNE run could change the embedding and thus LE. The paper also does not use a separate validation split or cross-validation for EMS selection. Consequently, the claim 'less than 25 m' is not established as a robust property of EMS-augmented channel charting; it may be an artifact of evaluating the selected configuration on the same data that guided its selection. The qualitative direction—EMS configurations improve CC—is supported by the envelope across all 121 combinations and the visual charts, but the specific headline magnitude is insecure.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes to use static electromagnetic skins (EMSs) to improve channel-charting-based localization in dense urban Non-Line-of-Sight (NLoS) scenarios. The authors model the EMS phase profiles as a codebook of linear phase gradients, search over 121 two-panel configurations, and select the one minimizing a quantile of an embedding metric (localization error, trustworthiness, or continuity) over all test points. Using 3D ray-traced simulations (Sionna RT) of an OSM-derived urban scene with t-SNE-based semi-supervised channel charting, they report that a codebook-optimized EMS configuration reduces the 90th-percentile localization error from above 60 m to below 25 m, with qualitative improvements in the chart structure.","tokens_in":8696,"tokens_out":4346,"duration_ms":44915,"significance":"If the quantitative result were robust, this would be a useful contribution: it is, to my knowledge, a plausible first application of static EMSs to channel charting, and it targets a real problem (NLoS localization in mmWave urban deployments). The paper's strengths include the use of a realistic ray-tracing simulator and open urban-map data, the explicit quantile-based objective for worst-case users, and the exhaustive evaluation over the codebook. The qualitative envelope across all 121 configurations (Figs. 3–5) supports the direction that EMSs can help, and the visual charts in Fig. 6 are suggestive. However, the headline quantitative claim is not established because of the in-sample selection methodology described above.","major_comments":[{"comment":"The selected configuration \\hat S is defined as the minimizer of Q_m(α|S) over the codebook, where Q_m is evaluated over the full test-point set U (Sec. IV-A). The very same U is then used to compute the reported 90th-percentile localization error in Fig. 5 and the Abstract's 'less than 25 m' claim. This is in-sample selection: the headline is the best of 121 correlated estimates, not an unbiased prediction for unseen positions. The gray envelope supports a qualitative benefit of EMS, but the magnitude of the gain is not established. Please add a held-out validation split for codebook selection and a separate test set, or use repeated random partitions and report the mean/variance of the resulting performance.","section":"Sec. V, Eq. (23)"},{"comment":"t-SNE is stochastic: Eq. (11) defines an argmin that is not unique, and the gradient dynamics in Eq. (12) depend on random initialization. The manuscript reports a single embedding and does not provide seeds, repeated runs, or any measure of dispersion across initializations. A different t-SNE run could meaningfully change the latent coordinates and hence the localization-error CDF, making the quoted 90th-percentile numbers fragile. Please report statistics over multiple t-SNE initializations (e.g., median with 5th–95th percentile bands) or use a deterministic embedding method. This is needed to separate the effect of the EMS from embedding randomness.","section":"Sec. III-B, Eq. (11)"}],"minor_comments":[{"comment":"Use 'Non-Line-of-Sight' instead of 'None-Line-of-Sight'.","section":"Abstract, Sec. I"},{"comment":"Typo: 'L′ = U \\ Ithe unlabeled points' should read 'L′ = U \\ I, the unlabeled points'.","section":"Sec. IV-A"},{"comment":"The conclusion contains 'decrease the taio of the localization error'; 'taio' should be 'ratio' (or 'error').","section":"Sec. VI"},{"comment":"Reference [15] (Sionna RT) appears to have an incorrect author list; please verify against the original publication.","section":"References"},{"comment":"Table I is referenced but not visible in the manuscript; please ensure the table is included and its caption and entries are complete.","section":"Sec. V"},{"comment":"The phrase 'No configuration performs worse than the baseline' is supported by the gray envelope if that envelope is the pointwise minimum across the 121 configurations. Please clarify in the caption how the envelope is computed (e.g., min–max across codewords per CDF level) so the reader can interpret the statement.","section":"Sec. V"}],"recommendation":"major_revision","confidential_remarks":"The core idea is within the journal's scope and the qualitative direction appears credible. The main barrier is methodological: the headline number is an in-sample optimum, and t-SNE stochasticity is not accounted for. These are fixable with a validation protocol and multiple runs, so I recommend major revision rather than rejection. I did not perform an exhaustive novelty check on the 'first work' claim, but I did not spot an obvious prior art conflict in the reference list."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short take: the paper applies static EMS panels to channel charting and shows, in a ray-traced urban scenario, that the right passive phase profile helps t-SNE embeddings separate NLoS users. The qualitative direction is credible; the headline 90th-percentile error drop from >60 m to <25 m is not, because the configuration that produces it is selected on the same test points used to score it.\n\nWhat's new: combining static EMS phase profiles with channel charting, and framing the design as a quantile-based codebook optimization over the embedding metrics. That's a reasonable design framework, and the paper does the work of building a full ray-traced scenario, computing covariance features, and showing the envelope across all 121 panel combinations. The grey band in the CDFs is a nice touch: it supports the claim that EMS helps on average and never hurts much.\n\nWhere it gets soft: Eq. (23) picks S* as the minimizer of Q_m(α|S) over the codebook, using the same TP set that is later evaluated. That's textbook in-sample selection. The reported 25 m is the best of 121 correlated estimates, not a prediction for unseen positions. The paper needs a validation split or cross-validation for EMS selection. Related: t-SNE is stochastic, Eq. (11) is an argmin with non-unique solutions, and the paper doesn't report multiple initializations or seeds. The entire quantitative claim—the <25 m number—could shift materially across t-SNE runs and UE layouts. I also note the authors themselves say 'achieving optimal performance requires joint network planning,' which is a limitation, but that doesn't address the in-sample issue. Priority claim is plausible but only a quick literature check; it's not load-bearing.\n\nOverall: the idea is worth pursuing and the qualitative result is probably right. The paper deserves a serious referee, but only after a fix: separate the codebook selection from evaluation, run t-SNE multiple times, and ideally release the code and scenario. As is, I'd cite it for the concept but not for the numbers.","headline":"A genuinely new application of static EMS to channel charting, with a credible qualitative result but a headline number that is an in-sample optimum.","tokens_in":9156,"tokens_out":2131,"would_cite":true,"duration_ms":20342,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Adding passive electromagnetic skins to a city base station can cut the worst-case channel-charting localization error from over 60 m to under 25 m.","keywords":["channel charting","electromagnetic skins","localization","NLoS","codebook optimization","semi-supervised t-SNE","smart radio environments","mmWave positioning"],"falsifier":"Run the same urban scenario with a strict train/test split over user positions: choose the EMS codeword on half the points, then measure the 90th-percentile localization error on the other half, repeating across several t-SNE initializations. If the held-out error stays near the no-EMS level (above 60 m), the reported <25 m result is an artifact of evaluating on the same points used for codebook selection.","tokens_in":8279,"feed_emoji":"📡","tokens_out":7143,"duration_ms":63436,"temperature":0.7,"pith_summary":"This paper tries to establish that a network can make channel-charting localization far more reliable simply by placing cheap, fixed reflecting surfaces—electromagnetic skins (EMSs)—near the base station and choosing each surface's phase profile once from a small codebook. In a simulated urban block with heavy non-line-of-sight conditions, it finds that the 90th-percentile localization error falls from over 60 m to under 25 m when the best codebook profile is used, with trustworthiness and continuity also improving. The reason proposed is that static surfaces reshape the multipath field so that channel covariance matrices, measured with Log-Euclidean distance, become more distinct between nearby user positions; the channel chart then stops collapsing NLoS areas into clusters. This matters because the surfaces are passive and need no per-user reconfiguration, avoiding the circular dependence that makes reconfigurable intelligent surfaces hard to use for positioning.","feed_headline":"Worst-case localization error drops from 60 m to under 25 m","feed_subtitle":"Fixed passive electromagnetic skins, selected from a small codebook, make channel-chart positioning in city NLoS zones reliable.","key_machinery":"The central object is the electromagnetic skin (EMS): a passive metasurface modeled by a diagonal phase-shift matrix $\\Phi_j$, with each element's phase given by a sampled linear ramp. The design machinery is a codebook $\\mathcal{C}$ of DFT-based horizontal phase gradients; Eq. (23) minimizes the $\\alpha$-quantile of the target metric over the Cartesian product of per-panel codewords. This machinery connects a physically manufacturable phase profile to a channel dissimilarity change: the reflected paths add structured diversity to the covariance features, and semi-supervised t-SNE (St-SNE) anchors labeled points to make the latent chart usable for localization. The quantile objective is what","core_discovery":"On its own terms, the paper's central discovery is that EMS phase-profile design can be posed as a codebook-based quantile optimization: choose the finite set of linear phase gradients on each panel that minimizes the upper quantile of localization error (or negative trustworthiness/continuity) over all test points. The optimized static configuration, not the active reconfiguration of the surface, carries the gain. Because the phases obey generalized Snell's law, each codeword corresponds to a particular reflected wave direction; the chosen directions are enough, in the 3D ray-traced city scenario, to lift NLoS points out of embedding collapse and recover the spatial geometry. The paper also","pith_inferences":["If the codebook selection generalizes to unseen positions, a natural deployment recipe is to optimize phase gradients from a one-time ray-tracing or drive-test survey and then freeze them—localization becomes a byproduct of network planning.","The same upper-quantile codebook objective could be extended to joint EMS placement and building-coating design, which the paper explicitly leaves to future work.","A direct comparison with active reconfigurable surfaces under identical ray-traced conditions would separate the benefit of static multipath enrichment from the benefit of reconfigurability itself.","A sharper testable prediction is that the winning codeword is tied to the geometry of the sector; moving an EMS by a wavelength or changing the building map should change the optimal codeword and degrade a fixed configuration."],"forward_implications":["If the result holds, worst-case positioning in dense urban NLoS improves dramatically without active hardware: a fixed, preconfigured surface does the work.","Operators can treat EMS placement and codebook selection as an offline planning problem rather than an online control problem.","The quantile-based evaluation protocol makes hard-to-localize users the design target, so reported gains are not driven by easy LoS points.","Specular mirrors are not enough: only codewords tuned to the scenario recover the full spatial structure, so direction-selective surfaces are the useful regime.","Larger codebooks beyond 121 combinations give no significant gain in the tested scenario, suggesting the discretization is not the bottleneck."],"supporting_citations":[{"why":"defines channel charting and the covariance-based feature setup that the paper builds on.","marker":"[6]"},{"why":"supplies the Log-Euclidean distance between covariance matrices and the t-SNE embedding used for the chart.","marker":"[5]"},{"why":"provides the semi-supervised t-SNE (St-SNE) anchor-point method that turns the latent chart into physical coordinates.","marker":"[8]"},{"why":"gives the generalized Snell's law that maps each EMS phase profile to a reflected wave direction.","marker":"[13]"},{"why":"supports the claim that EMSs are low-cost, low-visual-impact passive surfaces deployable in urban scenarios.","marker":"[14]"},{"why":"supplies the 3D ray-tracing engine that generates the deterministic multipath channels used in the simulations.","marker":"[15]"},{"why":"supplies the EMS meta-atom radiation pattern used in the channel model.","marker":"[17]"},{"why":"justifies the codebook-based phase-profile optimization approach for surface-aided mmWave systems.","marker":"[19]"}],"fun_headline_variants":["Static EM skins cut 90th-percentile error from 60m to under 25m","Codebook-tuned EM surfaces shrink city NLoS positioning error from 60m","Channel charting gets 60m-to-25m boost from passive EM skins","Optimized passive EM skins slash urban localization error by over half","Fixed EM skins enable reliable NLoS positioning via channel charting"],"cache_read_input_tokens":2816,"weakest_assumption_plain":"The headline improvement rests on the assumption that the codebook configuration minimizing the 90th-percentile error on the evaluated test points will also perform well for unseen user positions, building layouts, and t-SNE initializations; if selection does not generalize, the sub-25 m figure is an in-sample artifact.","fun_headline_variants_meta":{"raw":{"variants":["Static EM skins cut 90th-percentile error from 60m to under 25m","Codebook-tuned EM surfaces shrink city NLoS positioning error from 60m","Channel charting gets 60m-to-25m boost from passive EM skins","Optimized passive EM skins slash urban localization error by over half","Fixed EM skins enable reliable NLoS positioning via channel charting"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000291,"raw_usage":{"total_tokens":1510,"prompt_tokens":687,"completion_tokens":823,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":431,"completion_tokens_details":{"reasoning_tokens":721}},"tokens_in":431,"tokens_out":823,"duration_ms":7711,"temperature":1.0,"reasoning_tokens":721,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T22:12:21.919575+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same urban scenario with a strict train/test split over user positions: choose the EMS codeword on half the points, then measure the 90th-percentile localization error on the other half, repeating across several t-SNE initializations. If the held-out error stays near the no-EMS level (above 60 m), the reported <25 m result is an artifact of evaluating on the same points used for codebook selection.","supporting_citations":[{"cited_title":"Channel charting: Locating users within the radio environment using channel state information,","cited_arxiv_id":null,"evidence_quote":"defines channel charting and the covariance-based feature setup that the paper builds on."},{"cited_title":"Beam snr prediction using channel charting,","cited_arxiv_id":null,"evidence_quote":"supplies the Log-Euclidean distance between covariance matrices and the t-SNE embedding used for the chart."},{"cited_title":"Semi-supervised learning for channel charting- aided iot localization in millimeter wave networks,","cited_arxiv_id":null,"evidence_quote":"provides the semi-supervised t-SNE (St-SNE) anchor-point method that turns the latent chart into physical coordinates."},{"cited_title":"Generalized analysis and unified design of em skins,","cited_arxiv_id":null,"evidence_quote":"gives the generalized Snell's law that maps each EMS phase profile to a reflected wave direction."},{"cited_title":"Low-profile and low-visual impact smart electromagnetic curved passive skins for enhancing connectivity in urban scenarios,","cited_arxiv_id":null,"evidence_quote":"supports the claim that EMSs are low-cost, low-visual-impact passive surfaces deployable in urban scenarios."},{"cited_title":"Conformal metasurfaces: a novel solution for vehicular communications,","cited_arxiv_id":null,"evidence_quote":"supplies the EMS meta-atom radiation pattern used in the channel model."},{"cited_title":"Beamforming training and quantization codebook for intelligent reconfigurable surface aided mmwave massive mimo systems,","cited_arxiv_id":null,"evidence_quote":"justifies the codebook-based phase-profile optimization approach for surface-aided mmWave systems."}],"review_version":1}