{"id":"91df6db4-d577-4ecb-88a6-d4d04b06f40a","arxiv_id":"2505.21408","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":3,"one_line_summary":"A switched-array Wi-Fi system with two 3x4 antenna arrays reports a 15.6 cm median 3D localization error using collaborative direct position determination.","lead":"WiCAL uses multiple Wi-Fi antenna arrays with cheap radio-frequency switches to locate a transmitter in 3D, reporting median errors around 15 cm in indoor tests. The significance is that centimeter-level Wi-Fi positioning may be possible without expensive multi-transceiver hardware.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Inter-array phase alignment in §II.B.3/§IV is target-dependent: measuring Δγ from the same source whose position DPD estimates makes the virtual-array gain and the 15.6 cm result circular.","rationale":"The paper's switched-array hardware, three-stage intra-array calibration, and measured AoA errors (1° elevation, 1.5° azimuth) are credible and reasonably detailed. The I-SSMUSIC+GP geometric positioning result (17 cm) is also plausible as an engineering achievement. However, the paper's headline cooperative-array claim is specifically the DPD virtual-array result (15.6 cm), and that claim depends on the inter-array synchronization step. The reader identified the correct weakest assumption: §II.B.3 measures the inter-array phase difference using the target signal at unknown position c, and §IV then treats the result as known in Eq. (25). In coherent multistatic processing, the inter-array phase contains the source position through the range difference; using it for calibration and then for localization is circular. The model in Eq. (25) is also internally inconsistent: a measured scalar Δγ̂ cannot be equivalent to the per-grid diagonal matrix ΔΓ=diag{Δγ_1,…,Δγ_K} unless the calibration is recomputed for each candidate position, which is not a legitimate operation on fixed received data. The paper provides no independent reference source, no wired tone, and no joint estimation of the unknown phase; it also releases no code or data and compares to prior systems using published numbers only, as its own footnote acknowledges. These issues do not disprove the hardware or the AoA work, but they invalidate the strongest claim that two unsynchronized (or target-calibrated) URAs form a virtual large-scale array with state-of-the-art accuracy. The verdict should remain REJECT with moderate confidence unless a revision specifies a position-independent synchronization reference, fixes the Eq. (25) model, and releases artifacts for reproduction.","tokens_in":19252,"tokens_out":8776,"duration_ms":92693,"concrete_test":"Run a synthetic experiment with two 3×4 URAs at the positions in §VI.B. Generate noiseless CSI from a source at c_true. Implement Algorithm 1 exactly, computing Δγ̂ from antennas {1,6,20} as described. In variant A, align Y_URA2 with this Δγ̂ and build the dictionary using the same Δγ̂. In variant B, skip the measured alignment and build the dictionary from the known geometric phase differences only. With no noise, both should peak at c_true if the geometry is identifiable. Then repeat with the source moved to a different position c' but keep Δγ̂ fixed at the value measured from c_true. If variant A still peaks at or near c_true while variant B peaks at c', the DPD result is an artifact of target-dependent calibration. A passing test must also demonstrate an independent reference source at known position yielding a position-independent Δγ̂.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim of a virtual large-scale array and 15.6 cm median DPD localization rests on Eq. (25), where the two URA data blocks are aligned by a phase Δγ̂ measured in §II.B.3. The text defines Δγ(c) = 2π(||c−c1||−||c−c2||)/λ for a source at unknown position c, and says it is measured from the CSI of antennas {1,6,20}. No independent reference source, wired calibration tone, or known-position anchor is specified. If Δγ̂ is obtained from the target signal, it depends on the very position the DPD algorithm is meant to estimate. Feeding this target-dependent scalar into Eq. (25) aligns the two subarrays at the true position, so the MUSIC pseudo-spectrum is artificially peaked there: for any candidate p≠c, the dictionary column A_URA2(p)Δγ̂ carries phase error exp(j(Δγ(c)−Δγ(p))), biasing the spectrum toward c. The model is also internally inconsistent: Eq. (25) writes ΔΓ=diag{Δγ_1,…,Δγ_K} as a per-grid-point correction to the dictionary, while the described procedure measures a single scalar Δγ̂ and multiplies Y_URA2 by it. These are different operations. If Δγ_k is instead computed geometrically for each grid point, the measured scalar is unnecessary and the 'inter-array phase alignment' step is vacuous; if Δγ̂ is a hardware phase, estimating it from the target consumes information needed to disambiguate positions along the array baseline, making the claimed DPD cooperative gain an artifact. The reported improvement of DPD2URA over I-SSMUSIC+GP (15.6 cm vs. 17 cm) is small enough that even mild bias from this circular alignment could account for it.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"WiCAL is a Wi-Fi 3D localization system built around two switched 3×4 uniform rectangular arrays (URAs) driven by three RF chains. The paper proposes a three-stage phase calibration (intra-group, inter-group, inter-array), an I-SSMUSIC algorithm for 3D AoA estimation with forward-backward spatial smoothing, a geometric closest-point positioning (GP) step, and a direct position determination (DPD) algorithm that treats the two synchronized URAs as a virtual large-scale array. The experimental section reports median AoA errors around 1°–1.5° and a median 3D localization error of 15.6 cm for DPD2URA versus 17 cm for I-SSMUSIC+GP, and compares with existing Wi-Fi localization systems. The central claim is that inter-array synchronization creates a coherent virtual array and yields state-of-the-art accuracy from commercial Wi-Fi devices.","tokens_in":19531,"tokens_out":13042,"duration_ms":126840,"significance":"The paper has real engineering merit: it demonstrates a working switched-array prototype, reports extensive measured data over many source positions, and the intra-array I-SSMUSIC/GP pipeline is plausible and clearly described. The paper also includes a candid footnote qualifying the fairness of the comparison with iArk and SWAN. However, the manuscript's most important contribution—the inter-array virtual-array DPD—is not sound as written. The inter-array alignment in Eq. (3)/Eq. (25) is target-dependent, and Eq. (21) contains a sign error in the closest-point equations. These are not cosmetic issues: they directly affect the DPD result and the GP initialization. If the DPD claim were corrected, the contribution could be significant; as published, the evidence for the virtual-array cooperative gain is not established.","major_comments":[{"comment":"The inter-array phase alignment is circular. Eq. (3) defines Δγ(c) as the phase difference caused by the path-length difference between the source at unknown position c and the two URAs, and the text says the CSI of antennas {1,6,20} is used to measure Δγ̂. If Δγ̂ is measured from the target signal, it depends on the very position the DPD algorithm estimates. Applying this scalar to Y_URA2 in Eq. (25) aligns the second array only at the true source position, so the MUSIC pseudo-spectrum is biased toward c and the claimed 15.6 cm DPD result is not an independent validation of a virtual large-scale array. The paper must specify an independent reference source, a wired calibration signal, or another position-independent synchronization mechanism; none is described.","section":"§II.B.3 and §IV (Eqs. (3), (25))"},{"comment":"The model in Eq. (25) is internally inconsistent. The prose says the received signal matrix is adjusted as Ŷ_URAi = Δγ̂_i · Y_URAi, i.e., a single scalar multiplication, whereas Eq. (25) inserts a per-grid-point diagonal matrix ΔΓ = diag{Δγ_1, ..., Δγ_K} into the dictionary A_URA2 · ΔΓ. These are different operations. If Δγ_k is a geometric phase computed for each candidate grid point, no inter-array measurement is needed and the alignment step is vacuous; if Δγ̂ is a measured scalar, the dictionary correction should be a scalar identity, not a diagonal matrix. The manuscript does not reconcile these formulations, making the DPD algorithm irreproducible as written.","section":"§IV (Eq. (25))"},{"comment":"The two RHS inner products in the closest-point system are swapped. The correct least-squares conditions read (c_h − c_i)·d_h in the first component and (c_h − c_i)·d_i in the second once the matrix is written as in Eq. (21); as printed, the first component is (c_h − c_i)·d_i and the second is (c_h − c_i)·d_h. For example, with c_h = (0,0,0), d_h = (1,0,0), c_i = (0,1,0), d_i = (0,1,0), Eq. (21) yields t_h = 1, t_i = 0 instead of t_h = 0, t_i = −1. This error propagates into the GP estimate and the center of the LSoI used by DPD2URA.","section":"§III.D (Eq. (21))"}],"minor_comments":[{"comment":"The abstract and introduction state median AoA errors of 1° in elevation and 1.5° in azimuth, while §VI.A reports median errors of 1° in azimuth and 1.5° in elevation; the axis ordering should be corrected for consistency.","section":"Abstract vs §VI.A"},{"comment":"With the stated maximum scan angle θ_L = 60°, Eq. (4) gives d_max ≈ 0.536λ, which is slightly below the chosen spacing d = 0.54λ; the authors should clarify whether θ_L is defined differently or adjust the spacing.","section":"§II.C"},{"comment":"The pseudo-spectrum in line 13 is written with a single steering vector a(·), but for the virtual array formed by two URAs the MUSIC search should use the concatenated joint steering vector; the notation should be made explicit.","section":"Algorithm 1, line 13"},{"comment":"The footnote acknowledging that the comparison with iArk and SWAN may be limited in fairness because their hardware was unavailable should be moved into the main text or reflected directly in the abstract's 'state-of-the-art' claim.","section":"§VI.B, footnote 1"}],"recommendation":"reject","confidential_remarks":"To the editor: the inter-array phase alignment issue is load-bearing and not a presentation detail. A phase offset measured from the target at an unknown position cannot serve as a known calibration constant for forming a virtual array; correcting this would require redesigning the DPD procedure and rerunning the experiments, not merely editing the text. The paper's strongest contribution, if the DPD claim were removed, would be limited to the intra-array I-SSMUSIC/GP results, which are more incremental. I therefore recommend rejection, while noting that a substantially revised manuscript that narrows its claims and supplies an independent synchronization mechanism could be reconsidered."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a real hardware effort. The 3-SPNT switch expansion from three RF chains to 3N antennas is a worthwhile architecture, and the intra-array calibration and bidirectional spatial-smoothing MUSIC (I-SSMUSIC) are described carefully enough to be plausible. The measured AoA errors (1° elevation, 1.5° azimuth median) are consistent with previous switched-array systems. The authors also honestly footnote that their comparison to iArk and SWAN is based on published numbers, not direct experiments.\n\nThe soft spot is the inter-array phase alignment that powers DPD2URA and the 15.6 cm headline result. Section II.B.3 defines Δγ(c) as the phase difference between the two URAs for a source at c, and says it is measured from CSI of antennas {1,6,20}. That measurement comes from the same source whose position DPD is trying to estimate. Feeding that scalar into Eq. (25) to align the subarrays means the virtual-array MUSIC spectrum is steered by the very quantity being estimated. The manuscript does not describe an independent reference anchor, a wired calibration tone, or any procedure that would make the measurement position-independent.\n\nThe model is also internally inconsistent: Eq. (25) writes ΔΓ as a diagonal matrix of per-grid-point corrections, while the text says a single scalar Δγ̂ is measured and multiplies Y_URA2. Those are different operations. If Δγ_k is computed geometrically for each candidate position, then the measured scalar is unnecessary; if the scalar is used, the dictionary and data don't match except at the true position. Either way, the DPD result is not an independent confirmation of the collaborative-array principle.\n\nThe rest of the paper is better than this. The closest-point geometric positioning is standard and sound. The progressive local traversal is a minor tweak. The array-size dependency experiments (2×3 → 97.5 cm, 3×3 → 26 cm, 3×4 → 15.6 cm) are consistent with expectations. But there are no error bars, no code/data release, and the DPD improvement over I-SSMUSIC+GP (15.6 cm vs 17 cm) is small enough that mild bias from the circular alignment could account for it.\n\nWho gets value from this paper: people working on switched-array hardware for Wi-Fi sensing, and anyone using DPD who needs a cautionary example of synchronization pitfalls. The hardware contribution is legitimately interesting. The DPD claim needs a serious rewrite before I'd trust the 15.6 cm number.\n\nRecommendation: send it to peer review—the engineering deserves referee time, and the circularity should be spelled out so the authors are forced to fix or remove the target-dependent alignment.","headline":"Clever switched-array engineering and a plausible AoA pipeline, but the headline 15.6 cm DPD result rests on a circular, target-dependent phase alignment that needs a fix before it can be believed.","tokens_in":20205,"tokens_out":7667,"would_cite":false,"duration_ms":73775,"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":"WiCAL claims that two synchronized commercial Wi-Fi arrays can act as one virtual array and locate a transmitter in 3D with a 15.6 cm median error.","keywords":["Wi-Fi localization","3D localization","uniform rectangular array","RF chain multiplexing","phase alignment","MUSIC","direct position determination","virtual array"],"falsifier":"Set up two URAs at known fixed positions, calibrate the inter-array phase offset once using a reference transmitter at a known location, then localize a second transmitter moved around the room with that same stored offset; if the median error jumps well above 15.6 cm, the phase offset depends on the target position and the virtual-array model is circular.","tokens_in":18927,"feed_emoji":"📍","tokens_out":9156,"duration_ms":98263,"temperature":0.7,"pith_summary":"WiCAL claims that a commercial Wi-Fi receiver with just three RF chains can drive two 3x4 uniform rectangular arrays through RF switches, synchronize them into a single virtual large-scale array, and locate a transmitter in 3D with a median error of 15.6 cm. The paper reports median per-array AoA errors of 1 degree in elevation and 1.5 degrees in azimuth, and positions the system as the first to reach state-of-the-art 3D localization using only commercial Wi-Fi devices. If the claim holds, centimeter-level indoor positioning would no longer require dedicated multi-transceiver MIMO hardware or specialized testbeds; standard access points and user devices could supply the aperture through switch multiplexing and phase alignment.","feed_headline":"Two Wi-Fi arrays become one virtual array for 15.6 cm 3D fixes","feed_subtitle":"Switched-array Wi-Fi hardware synchronizes into a virtual large-scale array, beating prior Wi-Fi localization accuracy.","key_machinery":"The load-bearing mechanism is the virtual large-scale array: after separating the phase offsets into intra-group, inter-group, and inter-array components, WiCAL measures the inter-array phase difference and applies it to the second URA's received matrix, so the two arrays can be stacked into one joint model whose MUSIC pseudo-spectrum is computed over spatial grid points rather than angles. The second component is I-SSMUSIC, a forward-backward two-dimensional spatial smoothing MUSIC that restores rank to the covariance matrix and lets coherent multipath sources be resolved with a small 3x4 aperture. A progressive local traversal then re-centers a 0.1 m search sphere with 0.005 m voxels around each refined estimate, giving the DPD2URA algorithm its reported convergence within about three iterations.","core_discovery":"The paper's central claim is that inter-array phase alignment turns two distributed, RF-switched commercial Wi-Fi URAs into a coherent virtual aperture whose spatial resolution exceeds what either array can achieve alone. The system applies a three-stage phase alignment (intra-group, inter-group, and inter-array) and then fuses the phase-aligned CSI directly into a joint MUSIC spectrum, bypassing intermediate AoA estimation in a direct position determination step. In the reported experiments at 5.2 GHz with 12 antennas per array, WiCAL achieves a median 3D localization error of 15.6 cm with the DPD2URA algorithm, compared with 17 cm for geometric closest-point fusion of I-SSMUSIC AoAs, and axis-wise median errors of 6 cm, 7.5 cm, and 7 cm. The authors state this is the first system to achieve state-of-the-art 3D localization using only commercial Wi-Fi devices, while noting that the comparison with prior platforms relies on published numbers because those platforms were not available for direct testing.","pith_inferences":["A deployment implication the paper leaves open: the inter-array phase offset is measured from the target itself, so it likely must be obtained from an independent reference transmitter or a wired calibration path before the virtual-array model can be used for unknown targets.","If the virtual-array model generalizes, the same joint MUSIC spectrum could be mined for coherent multipath structure, effectively turning the synchronized Wi-Fi arrays into a coarse imaging aperture rather than a point localizer.","A natural next experiment would vary the distance and orientation between the two URAs to map how the 15.6 cm error scales with baseline length; the paper tests a single geometry, so the geometric dependence of the inter-array gain is not established."],"forward_implications":["A three-RF-chain NIC can extend to 3N antennas with SPNT switches, a higher ceiling than the 2N+1 antennas allowed by keeping one chain on a reference antenna.","Array size strongly drives accuracy: median error drops from 97.5 cm with a 2x3 array to 26 cm with 3x3 and 15.6 cm with 3x4, so cascaded switch designs have a clear scaling path to better localization.","Direct position determination outperforms geometric AoA triangulation by 7.7% in this setup, which suggests that coherent data fusion, not just angle estimation, is where much of the inter-array gain comes from.","The full DPD2URA fix takes about 0.7 s, and the tracking experiment reports 0.11 m raw and 0.075 m median-filtered trajectory errors, so near-real-time 3D tracking is within reach."],"supporting_citations":[{"why":"Supplies the phase-offset model of PLL initial phase and cable delays that intra-group calibration corrects.","marker":"[11]"},{"why":"Provides the switch-based antenna stitching approach and 2D localization baseline that WiCAL extends to 3D and distributed collaboration.","marker":"[14]"},{"why":"Prior switch-based 3D localization system with AI fusion, whose published accuracy is the baseline for the state-of-the-art comparison.","marker":"[16]"},{"why":"Standard Wi-Fi AoA localization baseline and reference for correlated interference and subcarrier smoothing arguments.","marker":"[17]"},{"why":"Introduces high-resolution direct position determination, the paradigm the inter-array DPD step builds on.","marker":"[38]"},{"why":"Provides a computationally efficient direct position determination algorithm for OFDM that motivates the DPD formulation.","marker":"[39]"},{"why":"The CSI capture tool used to obtain the experimental channel measurements on commercial Wi-Fi hardware.","marker":"[40]"},{"why":"Foundation of spatial smoothing for direction-of-arrival estimation of coherent signals, which I-SSMUSIC extends with forward-backward smoothing.","marker":"[41]"}],"fun_headline_variants":["Phase-aligned Wi-Fi arrays form virtual aperture for 15.6 cm 3D","Distributed Wi-Fi URAs sync into virtual array for 15.6 cm 3D","WiCAL: two Wi-Fi arrays act as one for 15.6 cm 3D accuracy","Commercial Wi-Fi arrays achieve 15.6 cm 3D via virtual synthesis","Paired Wi-Fi arrays form virtual antenna for 15.6 cm 3D fixes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the phase difference measured between the two arrays using the target's own signal can be treated as a fixed, position-independent calibration constant when combining the arrays into a virtual aperture; the paper does not describe an independent reference source or calibration step that would make it so.","fun_headline_variants_meta":{"raw":{"variants":["Phase-aligned Wi-Fi arrays form virtual aperture for 15.6 cm 3D","Distributed Wi-Fi URAs sync into virtual array for 15.6 cm 3D","WiCAL: two Wi-Fi arrays act as one for 15.6 cm 3D accuracy","Commercial Wi-Fi arrays achieve 15.6 cm 3D via virtual synthesis","Paired Wi-Fi arrays form virtual antenna for 15.6 cm 3D fixes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000593,"raw_usage":{"total_tokens":2818,"prompt_tokens":1024,"completion_tokens":1794,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":640,"completion_tokens_details":{"reasoning_tokens":1677}},"tokens_in":640,"tokens_out":1794,"duration_ms":13953,"temperature":1.0,"reasoning_tokens":1677,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T13:29:25.821878+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Set up two URAs at known fixed positions, calibrate the inter-array phase offset once using a reference transmitter at a known location, then localize a second transmitter moved around the room with that same stored offset; if the median error jumps well above 15.6 cm, the phase offset depends on the target position and the virtual-array model is circular.","supporting_citations":[{"cited_title":"ArrayTrack: A fine-grained indoor location system,","cited_arxiv_id":null,"evidence_quote":"Supplies the phase-offset model of PLL initial phase and cable delays that intra-group calibration corrects."},{"cited_title":"SW AN: Stitched Wi-Fi antennas,","cited_arxiv_id":null,"evidence_quote":"Provides the switch-based antenna stitching approach and 2D localization baseline that WiCAL extends to 3D and distributed collaboration."},{"cited_title":"General-purpose deep tracking platform across protocols for the internet of things,","cited_arxiv_id":null,"evidence_quote":"Prior switch-based 3D localization system with AI fusion, whose published accuracy is the baseline for the state-of-the-art comparison."},{"cited_title":"Spotfi: Decimeter level localization using Wi-Fi,","cited_arxiv_id":null,"evidence_quote":"Standard Wi-Fi AoA localization baseline and reference for correlated interference and subcarrier smoothing arguments."},{"cited_title":"High resolution direct position determination of radio frequency sources,","cited_arxiv_id":null,"evidence_quote":"Introduces high-resolution direct position determination, the paradigm the inter-array DPD step builds on."},{"cited_title":"A computationally effi- cient direct position determination algorithm based on OFDM system,","cited_arxiv_id":null,"evidence_quote":"Provides a computationally efficient direct position determination algorithm for OFDM that motivates the DPD formulation."},{"cited_title":"Tool release: Gath- ering 802.11n traces with channel state information,","cited_arxiv_id":null,"evidence_quote":"The CSI capture tool used to obtain the experimental channel measurements on commercial Wi-Fi hardware."},{"cited_title":"On spatial smoothing for direction- of-arrival estimation of coherent signals,","cited_arxiv_id":null,"evidence_quote":"Foundation of spatial smoothing for direction-of-arrival estimation of coherent signals, which I-SSMUSIC extends with forward-backward smoothing."}],"review_version":1}