{"id":"374a567d-20e6-4cd3-8431-e26614077c41","arxiv_id":"2411.10907","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A distributed antenna array is localized with 0.82 mm mean error at 34 dB link SNR by combining two-tone two-way time-transfer ranging with an evolutionary algorithm that completes missing inter-node distances.","lead":"This paper shows a six-node distributed antenna array localizing its elements to 0.82 mm average error using radio two-tone ranging plus an evolutionary algorithm that fills in missing inter-node distance measurements. The result matters because sub-millimeter element localization is a prerequisite for coherent distributed beamforming, a technique that could extend 5G/6G and remote-sensing capabilities.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Reported 0.82 mm accuracy is a calibrated precision, not a transferable localization accuracy: static delays are calibrated and evaluated against the same OptiTrack ground truth, with no demonstrated generalization to moved nodes or changed environments.","rationale":"The reader's weakest assumption identifies exactly the same load-bearing issue: the 0.82 mm result depends on static hardware and multipath delays being calibrated once against known ground-truth positions and then subtracted. My review confirms this is the most consequential assumption for the central claim. If the delay calibration does not transfer across positions or time, the reported accuracy is an artifact of the evaluation setup, and the paper's claim to support distributed beamforming at 24.3 GHz falls apart. The paper is honest about its static assumption but does not test it; it merely asserts that nodes may be moved afterward. A split-halves or physical-repositioning experiment would settle the concern directly. Other issues, such as the centralized PC and wired frequency reference, are acknowledged limitations and do not invalidate the algorithmic contribution; the orientation ambiguity of MDS is real but is a well-known property of range-only localization and is less directly tied to the reported accuracy number. Given the addressability of the calibration concern and the otherwise reasonable method, the reader's CONDITIONAL verdict is appropriate; no change in verdict is needed, but the authors should be required to provide the cross-validation or moved-node experiment.","tokens_in":14394,"tokens_out":7234,"duration_ms":91603,"concrete_test":"Perform a held-out calibration test on the existing measured EDMs: use a randomly selected half of the node positions (or the first half of the 250 EDM estimates per condition) to estimate the static delay calibration for each link, then run the genetic localization algorithm on the held-out half using those calibration values, and compare the held-out mean EVM to the calibrated 0.82 mm result. A statistically significant degradation (e.g., >0.2 mm, or above the 0.82 mm lambda/15 bound for 24.3 GHz) would show the accuracy is calibration-dependent. A stronger version is to physically move the six nodes several wavelengths after calibration and re-localize without recalibrating, then report the resulting EVM.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline result rests on the static-delay calibration described in Section II-A and Section V-A. In Section II-A, hardware and multipath delays are treated as static and 'deduced' using known ground truth; in Section V-A, the OptiTrack system is used both to compute these delays by subtracting known times-of-flight from measured totals and later to evaluate the localization error vector magnitude. Because calibration and evaluation are performed at the same antenna positions (the antennas are placed in their final circular pattern before the delays are deduced), any systematic bias in the range estimates is effectively removed before the reported EVM is computed. What remains is the residual random noise around a per-link calibrated offset, not an accuracy that would hold if the nodes were moved, the cables flexed, the temperature drifted, or the multipath environment changed. The paper explicitly states 'After these static delays are determined the nodes may be moved,' but no experiment with moved nodes is reported, and all connectivity and bandwidth sweeps reuse the same calibrated geometry. Since the central claim (sub-millimeter localization, theoretically supporting 24.3 GHz beamforming) depends on the 0.82 mm number being a genuine accuracy rather than a calibration residual, the assumption that these static delays are known and invariant is the most load-bearing weakness. No error bars, raw data, or code are provided to separate calibration uncertainty from estimator noise.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a decentralized localization method for distributed antenna array elements. A two-way time-transfer ranging technique using a spectrally sparse pulsed two-tone waveform provides internode range estimates; classical multidimensional scaling recovers relative array geometry from the resulting Euclidean distance matrix; and a genetic algorithm is used to complete partially observed distance matrices so that localization works in sparsely connected topologies. The authors report Monte Carlo simulations and a six-node software-defined-radio experiment in a motion-capture-equipped laboratory, with a headline mean localization error vector magnitude of 0.82 mm at an average link SNR of 34 dB, which they interpret as supporting distributed beamforming up to 24.3 GHz under the lambda/15 criterion.","tokens_in":14636,"tokens_out":4403,"duration_ms":50308,"significance":"If the experimental accuracy were transferable to truly decentralized, reconfigurable deployments, the result would be a meaningful step toward coherent distributed beamforming at microwave and millimeter-wave frequencies. The paper has several strengths: it grounds the ranging precision in the Cramer-Rao lower bound, clearly formulates the Euclidean distance matrix completion problem, evaluates the genetic-localization algorithm over randomized Monte Carlo trials, and uses a substantial experimental setup with calibrated ground-truth position tracking. The central technical mechanism, combining two-tone ranging with MDS and an evolutionary optimizer for missing range entries, is plausible and well motivated. The main weakness is that the headline 0.82 mm figure is obtained after calibrating static hardware and multipath delays with the same motion-capture system used for evaluation, and no uncertainty quantification is provided, so the claimed accuracy has not been shown to generalize to moved nodes or changed environments.","major_comments":[{"comment":"The 0.82 mm mean EVM is a calibrated residual rather than a demonstrated transferable accuracy. In Section II-A the authors state that hardware delays and multipath factors are \"static delays that can be deduced\" using known ground-truth positions, and in Section V-A the OptiTrack system is used both to compute these delays by subtracting known times-of-flight and later to evaluate localization error. The antennas remain in the same circular geometry throughout, so the calibration and evaluation are performed at the same positions. The paper explicitly says \"After these static delays are determined the nodes may be moved,\" but no experiment with moved nodes is reported. This is load-bearing because the abstract's sub-millimeter claim and the 24.3 GHz beamforming conclusion depend on the 0.82 mm number being an accuracy, not a removal of per-link systematic bias. I recommend either reporting a moved-node experiment with independent ground truth, or clearly re-labeling the result as calibrated precision and providing a sensitivity analysis of how delay drift or geometry change would affect the error.","section":"Sections II-A and V-A"},{"comment":"No error bars, confidence intervals, or distribution statistics are reported for the measured localization error. The 0.82 mm value is presented as a point estimate, and Figures 9 and 10 show mean convergence curves without spread, despite being averaged over approximately 250 EDM estimates per condition. Without the standard deviation, percentiles, or a histogram of the per-trial EVM, the reader cannot assess whether the reported sub-millimeter accuracy is statistically robust or dominated by a few favorable trials. Please report the spread of the 250 estimates and, ideally, the per-link range residuals before and after calibration, so that the calibration uncertainty can be separated from the localization algorithm's performance.","section":"Section V-B and Figure 10"},{"comment":"The validation of partially connected topologies is performed by post-processing, not by physical link loss. In Section V-B, \"links between nodes were artificially severed in post-processing by randomizing the adjacency matrix,\" while all physical links were measured in a line-of-sight, equal-elevation circular arrangement. This is a reasonable way to test the optimizer on synthetic missing entries, but it does not validate the claim in the introduction that the method works \"even in complex environments where the array may not be capable of directly estimating all nodal link distances.\" True NLoS or low-SNR link loss would introduce additional bias and correlated errors that are not captured by random masking. I ask the authors to state this limitation explicitly and to add at least one experiment or simulation with a physically motivated missing-link model, or to soften the corresponding claim.","section":"Sections IV-B and V-B"}],"minor_comments":[{"comment":"The abstract says the paper \"define[s] the differential evolution algorithm,\" but Section III-C and Algorithm 2 describe a genetic algorithm with crossover and mutation, implemented using the pymoo library; differential evolution is a distinct optimizer and is never defined or used. Please align the terminology.","section":"Abstract vs. Section III-C"},{"comment":"There are two wording issues: \"does not noticably affect\" should be \"does not noticeably affect,\" and \"a dramatic affect on the overall network SNR\" should be \"a dramatic effect.\"","section":"Section V-B"},{"comment":"The text refers to \"Fig. 10, additional simulation results\" before Figure 10 is introduced; the figure appears later in the experiment section. Please reorder or fix the cross-reference.","section":"Section IV-B"},{"comment":"The schematic legend repeats \"SDR 3\" for three of the six radios and appears to have misplaced labels for SDRs 4 and 5; please correct the labels so the six-node configuration is unambiguous.","section":"Figure 7"},{"comment":"No data or code availability statement is provided. Since the experimental claim is quantitative and depends on the calibration procedure, making the range estimates, adjacency masks, and optimizer settings available would materially improve reproducibility.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern lands: the headline accuracy is obtained under a calibration/evaluation regime that removes systematic per-link delay errors at the same antenna positions, and no moved-node or independent-evaluation experiment is reported. I would not reject the paper, because the algorithmic contribution and the simulation study are sound, but I would require either an independent validation or a substantial reframing of the 0.82 mm claim before publication. The differential-evolution/genetic-algorithm mismatch in the abstract also needs correction, and the lack of uncertainty quantification should be addressed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The genuinely new piece here is using differential evolution to complete a partially observed Euclidean distance matrix for distributed-array localization, and that is worth knowing about. The authors take established ingredients—two-tone two-way ranging, classical MDS, and EDM completion—and show empirically that a GA can fill missing ranges in a minimally connected six-node array with sub-millimeter residual error. The math is standard and clean, the simulations are sensible, and the paper is honest about the current limits: computation is centralized, synchronization uses a wired reference, and the static-delay assumption is stated rather than hidden.\n\nThe experiment itself looks carefully done. They sweep bandwidth and connectivity, report SNR carefully, and show that convergence degrades gracefully as edges are removed. Given the CRLB analysis and prior work from this group, the 0.82 mm number is plausible as a statement about estimator noise at 34 dB.\n\nBut the headline claim overreaches. The 0.82 mm is obtained after calibrating hardware and multipath delays using OptiTrack ground truth and then evaluating the localization error against that same OptiTrack at the same antenna positions. That makes it a calibrated precision, not an accuracy that would hold if nodes moved, cables flexed, or the multipath environment changed. The paper explicitly says \"After these static delays are determined the nodes may be moved,\" but no moved-node experiment appears. The stress-test note gets this right; it is the load-bearing weakness. The paper's own discussion of decentralized operation is also partly aspirational, since the actual computation runs on a central PC.\n\nThese are addressable. A moved-node measurement, even with only a few new positions, would separate calibration residuals from true localization error. Error bars or confidence intervals on the EVM would also help, as would providing code or data.\n\nFor whom is this? Researchers in distributed arrays and wireless localization will find the EDM-completion approach useful, and the paper is a solid engineering data point. But readers should treat the 24.3 GHz beamforming ceiling as a theoretical bound under the particular calibrated geometry, not a demonstrated operational capability.\n\nI would send this to peer review—the method is well motivated, the experiment is nearly reproducible, and the limitations are fixable. A serious referee should push for the moved-node test and uncertainty quantification before publication, but the paper deserves that referee time.","headline":"A credible but over-sold accuracy claim: the new evolutionary EDM completion idea is real, yet the 0.82 mm headline is calibrated precision against the same ground truth used for evaluation, not a transferable localization accuracy.","tokens_in":15164,"tokens_out":1412,"would_cite":true,"duration_ms":18095,"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":"Using a spectrally sparse two-tone waveform and an evolutionary completion step, this paper claims that a minimally connected six-node distributed antenna array can localize its elements to a mean error of 0.82 mm at 34 dB average link…","keywords":["distributed phased arrays","decentralized localization","two-way time transfer","two-tone ranging","multidimensional scaling","Euclidean distance matrix completion","evolutionary algorithm","coherent beamforming"],"falsifier":"Repeat the six-node experiment at 40 MHz tone separation and 34 dB harmonic-mean SNR, calibrate the hardware and multipath delays once against optical ground truth, then move the antennas to a different geometry without recalibrating and measure the localization error vector magnitude against ground truth; if the mean error no longer stays near 0.82 mm, or exceeds the $\\lambda/15$ threshold of about 0.82 mm at 24.3 GHz, the static-delay assumption behind the central claim fails.","tokens_in":14177,"feed_emoji":"📡","tokens_out":9787,"duration_ms":93658,"temperature":0.7,"pith_summary":"This paper argues that the elements of a distributed antenna array can determine their own relative positions to sub-millimeter accuracy even when some inter-node distances cannot be measured at all. The method pairs two-way time-transfer ranging with a spectrally sparse pulsed two-tone waveform, reconstructs the geometry with classical multidimensional scaling, and uses an evolutionary search to fill in the missing distance values. In a six-node laboratory array with an average link SNR of 34 dB and only the minimally required connectivity, the authors report a mean localization error vector magnitude of 0.82 mm, which meets the $\\lambda/15$ phase-error rule for coherent distributed beamforming up to 24.3 GHz. If this result holds, fixed anchor nodes and full line-of-sight connectivity are not required for high-precision distributed array localization.","feed_headline":"Distributed antenna array localizes itself to 0.82 mm","feed_subtitle":"Six wireless nodes with missing links reach sub-millimeter accuracy, enough for coherent beamforming at 24 GHz.","key_machinery":"The central object is the Euclidean distance matrix (EDM), the table of all pairwise node distances, allowed to have missing entries for unmeasured links. The precision identity is the two-tone waveform's mean-squared bandwidth, $\\zeta_f^2 = (\\pi B)^2$, which enters the Cramer-Rao range bound $\\sigma_d^2 \\geq c^2 / [2 \\zeta_f^2 (E_s/N_0)]$; concentrating energy in two tones lets the tone separation $B$ drive range precision while keeping the occupied spectrum sparse. Two-way time transfer cancels clock bias by averaging the two directions of exchange, leaving time of flight plus static hardware and multipath delays. Classical multidimensional scaling reconstructs the centered node coordinates from the EDM by eigen-decomposing the doubly-centered Gram matrix, and the evolutionary algorithm evaluates candidate completions of the missing EDM entries using the cost $F(X) = \\frac{1}{2} \\| W \\circ (\\tilde{D} - \\operatorname{edm}(X)) \\|^2$, where $W$ is the adjacency matrix and $\\tilde{D}$ is the observed partial EDM. The minimal connectivity that makes completion well-posed comes from the robust-quadrilateral criterion, $E_{\\min} = 3N - 6$ in two dimensions.","core_discovery":"The paper claims that a decentralized distributed array can be localized from an incomplete, noisy set of pairwise range estimates, and that the missing-link case costs almost nothing in accuracy. Pairs of nodes exchange two-tone pulses in both directions; averaging the two apparent times of flight cancels clock offsets, and the tone separation, up to 40 MHz, enters the Cramer-Rao bound through the mean-squared bandwidth $\\zeta_f^2 = (\\pi B)^2$, giving high range precision from a spectrally sparse signal. Classical multidimensional scaling converts the estimated Euclidean distance matrix into a relative geometry, and an evolutionary algorithm completes the matrix by minimizing the weighted mismatch between observed and candidate distances. The experimental result is a mean localization error vector magnitude of 0.82 mm for six nodes at the minimal connectivity $c = 0.8$ and an average link SNR of 34 dB, which the authors state theoretically supports coherent operation up to 24.3 GHz. The paper frames this as an advance over centralized, fully-connected localization because no node serves as an anchor and unmeasured links can be tolerated.","pith_inferences":["The calibration step measures hardware and multipath delays once against known ground-truth positions; a consequence the paper does not develop is that deployments with temperature drift, moving nodes, or a changed scattering environment would need recalibration for the 0.82 mm figure to survive.","The evolutionary completion currently runs on a central computer, as the paper notes; a genuinely node-distributed version would partition the optimization across the array, and convergence would likely differ because each node would see only its local neighborhood of the distance matrix.","MDS recovers geometry only up to rotation, reflection, and translation, so applications that need the array's absolute orientation, such as steering a beam to a known target direction, would require an additional orientation-estimation step not covered here.","The waveform's narrow occupied bandwidth suggests a testable coexistence experiment: transmit the two-tone ranging signal while another link uses adjacent spectrum and check whether the 0.82 mm precision holds, an extension the paper motivates but does not demonstrate."],"forward_implications":["Distributed beamforming near 24 GHz becomes feasible with a six-node array localized to a mean 0.82 mm, since that error satisfies the $\\lambda/15$ coherence rule at the carrier.","Partial connectivity is tolerable: reducing the measured adjacency matrix from complete to the minimally completable $c = 0.8$ changed the converged localization error by less than a tenth of a millimeter in the experiments.","Larger tone separation improves ranging precision through the Cramer-Rao bound, while the waveform's spectral sparsity keeps the positioning signal's bandwidth occupation low.","For arrays larger than about ten nodes, the number of evolutionary generations needed to converge grows with the number of missing edges, so deployment would need more computation or subset-based localization.","Because the cost function weights links evenly rather than propagating estimates from a central anchor, localization error does not accumulate radially the way centralized schemes do."],"supporting_citations":[{"why":"Sets the $\\lambda/15$ ranging-error standard that makes sub-millimeter localization the enabling requirement for coherent distributed beamforming.","marker":"[6]"},{"why":"Establishes the two-way time-transfer technique and its picosecond-level synchronization, the basis of the ranging method.","marker":"[21]"},{"why":"Demonstrates sub-millimeter internode ranging with spectrally sparse signals, the direct precursor of the two-tone ranging used here.","marker":"[22]"},{"why":"Derives the Cramer-Rao range bound and the mean-squared-bandwidth result that motivates the two-tone waveform.","marker":"[27]"},{"why":"Supplies the classical multidimensional scaling formulation and the EDM completion cost function used by the evolutionary algorithm.","marker":"[32]"},{"why":"Provides the robust quadrilateral and minimal-connectivity conditions that define which missing-edge patterns are completable.","marker":"[34]"},{"why":"Supplies the matched filter, quadratic least-squares interpolation, and Cramer-Rao background used in delay estimation and refinement.","marker":"[26]"}],"fun_headline_variants":["Evolutionary algorithm pinpoints antenna nodes to 0.82 mm","Distributed array self-localizes despite missing links","Sub-millimeter landmark for decentralized phased arrays","Pulsed two-tone trick localizes array to sub-mm","Evolution completes missing ranges for antenna positioning"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The accuracy claim rests on the assumption that the RF hardware delays and static multipath delays, measured once against known ground-truth positions, remain unchanged when the nodes are moved and while the array operates; if those delays drift with temperature or geometry, the 0.82 mm result would not transfer to a real deployment.","fun_headline_variants_meta":{"raw":{"variants":["Evolutionary algorithm pinpoints antenna nodes to 0.82 mm","Distributed array self-localizes despite missing links","Sub-millimeter landmark for decentralized phased arrays","Pulsed two-tone trick localizes array to sub-mm","Evolution completes missing ranges for antenna positioning"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000788,"raw_usage":{"total_tokens":3501,"prompt_tokens":994,"completion_tokens":2507,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":610,"completion_tokens_details":{"reasoning_tokens":2441}},"tokens_in":610,"tokens_out":2507,"duration_ms":21520,"temperature":1.0,"reasoning_tokens":2441,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T19:09:36.612067+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the six-node experiment at 40 MHz tone separation and 34 dB harmonic-mean SNR, calibrate the hardware and multipath delays once against optical ground truth, then move the antennas to a different geometry without recalibrating and measure the localization error vector magnitude against ground truth; if the mean error no longer stays near 0.82 mm, or exceeds the $\\lambda/15$ threshold of about 0.82 mm at 24.3 GHz, the static-delay assumption behind the central claim fails.","supporting_citations":[{"cited_title":"Distributed phased arrays: Challenges and recent advances,","cited_arxiv_id":null,"evidence_quote":"Sets the $\\lambda/15$ ranging-error standard that makes sub-millimeter localization the enabling requirement for coherent distributed beamforming."},{"cited_title":"Wireless picosecond time synchronization for distributed antenna arrays,","cited_arxiv_id":null,"evidence_quote":"Establishes the two-way time-transfer technique and its picosecond-level synchronization, the basis of the ranging method."},{"cited_title":"A microwave sensor with submillimeter range accuracy using spectrally sparse signals,","cited_arxiv_id":null,"evidence_quote":"Demonstrates sub-millimeter internode ranging with spectrally sparse signals, the direct precursor of the two-tone ranging used here."},{"cited_title":"On the estimation of angle rate in radar,","cited_arxiv_id":null,"evidence_quote":"Derives the Cramer-Rao range bound and the mean-squared-bandwidth result that motivates the two-tone waveform."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the matched filter, quadratic least-squares interpolation, and Cramer-Rao background used in delay estimation and refinement."}],"review_version":1}