{"id":"25956718-17be-4306-8bd5-532977a6c2bb","arxiv_id":"2505.17587","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"First demonstrations of drone-to-drone, drone-to-vehicle, and vehicle-to-vehicle quantum key distribution, with finite-key secure rates of 1.6 to 20 kbps over short free-space links.","lead":"A team built small, modular quantum key distribution transmitters and receivers and mounted them on drones and cars, exchanging secure encryption keys between two moving platforms for the first time. The systems produced secure key rates between 1.6 and 20 kbps over short ranges, a step toward reconfigurable mobile quantum networks.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The reported secure key rates rest on a 32-parameter unitary device model fitted to the same QKD counts used for key extraction; the fit is underdetermined and its statistical uncertainty is not propagated, so the finite-key rates are not a demonstrated lower bound.","rationale":"The paper is a genuine engineering milestone: it reports first QKD between two fully mobile platforms, with plausible hardware descriptions, multiple configurations, and a real, if short, set of demonstration runs. The central scientific claim, however, is not just that photons were exchanged but that secure key rates of 1.6-20 kbps were achieved. That claim is conditional on both the finite-key security proof and the device characterization. The reader already identified G as the weakest assumption; I agree and sharpen it. In Supplementary Note 11, G is an 8x2 isometry estimated by minimizing Eq. (24) against the same average count rates that later feed the key-rate solver. The loss outputs are unmeasured, so the true POVM is not identified, and the finite-size statistical fluctuations are not conservatively propagated through the model. This is not a disagreement with community consensus; it is an internal gap between the strength of the claim ('secure key rates') and the evidence supplied. The fix is straightforward: release data and code, obtain G from an independent calibration, and propagate its uncertainty (e.g., via bootstrap or confidence regions) into the key-rate computation. If the rates survive that test, the central claim stands; if not, the paper should be presented as a hardware demonstration with target rates pending rigorous validation. That is why I keep the reader's CONDITIONAL verdict.","tokens_in":21337,"tokens_out":9105,"duration_ms":87705,"concrete_test":"Recompute the finite-key rates in Table 1 with G fixed to an independently calibrated receiver model, obtained from pre/post-session bright-light measurements of the beam-splitter ratio, waveplate retardance, PBS extinction, and detector efficiencies - not from the in-flight QKD counts - while treating residual channel loss as uncharacterized loss to Eve. If any reported rate becomes non-positive or shifts by more than its statistical uncertainty, the fitted-G step is doing essential work and the central claim is not established. A complementary bootstrap (resample the raw detection counts, re-fit G, and recompute the secure key rate) would quantify the uncertainty currently hidden.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing condition for the central claim is that G in Supplementary Note 11 (Eqs. 10-24) is an exact characterization of the receiver/channel POVM. G is an 8x2 complex isometry with roughly 28-32 real degrees of freedom, fit by least squares to only 12 average signal count rates (3 input states x 4 detector outputs). The four loss modes are never directly measured, so many parameters are unidentifiable or constrained only by the isometry condition. The same counts that determine G are then fed into the finite-key solver of Ref. [30]; no confidence region, bootstrap, or independent cross-check on G is propagated into the key-rate expression. If the least-squares solution is a different isometry than the true POVM - for example, one that absorbs Poisson fluctuations or makes a particular split between loss and detector inefficiency - then Corollary 4 of Ref. [33] is applied to the wrong model and the reported 1.6-20 kbps rates are not valid lower bounds. The paper's post-hoc data filtering (Supp. Note 10) and reliance on a cited rather than reproduced security proof compound this concern, but the G-estimation step is the single point on which the security claim hinges.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports a modular polarization-based BB84 decoy-state QKD system deployed on drones and ground vehicles. The central claim is a first demonstration of QKD links in which both endpoints are mobile: drone-to-drone, drone-to-vehicle, and vehicle-to-vehicle at 5 mph and 70 mph. Using a custom finite-key analysis built on the external frameworks of Refs. [30-33], Table 1 reports finite-key secure key rates of 1.6-20 kbps and total secret keys of 205 kb-1.88 Mb from single sessions. The security analysis incorporates a fitted unitary device model (Supplementary Note 11) that describes source and receiver non-idealities, state tomography, decoy-state intensities, and detector losses. The supplementary information documents the platforms, PAT system, detectors, time tagging, synchronization, and safety. The core technical novelty is the mobile deployment and the claim that the resulting finite-key rates account for platform-specific non-idealities; the protocol itself is standard decoy-state BB84.","tokens_in":21587,"tokens_out":17299,"duration_ms":178390,"significance":"If the reported finite-key rates are rigorously established, this is a substantial experimental milestone: it would be the first QKD exchange in which neither endpoint is a fixed station, and it demonstrates finite-key security in short sessions relevant to mobile networks. The paper's strengths are the breadth of the experimental campaign (six link configurations, including a 70 mph highway run), the modular payload design, and the unusually detailed supplementary characterization of the source, PAT system, detectors, synchronization, and laser safety. The authors also correctly recognize that finite-key analysis is necessary for short mobile sessions. However, the security claim is currently supported by a statistical device model whose uncertainty is not quantified, and several model parameters are estimated from the same QKD counts that are later used for key extraction; the paper does not provide deposited code or machine-checked proof certificates for the custom numerical analysis. The experiments are credible, but the reported key-rate numbers are not yet demonstrated lower bounds.","major_comments":[{"comment":"The device isometry G in Eq. (10) carries 32 real parameters (about 28 after the isometry constraint in Eq. (11)) but is fitted by least squares to only 12 average signal count rates (three input states times four detector outputs), and the four loss modes are never directly observed. The cost function in Eq. (24) is an unweighted residual on mean photon numbers and does not model Poisson counting statistics. Because G is used to construct the effective lossless POVM via Corollary 4 of Ref. [33], a different isometry compatible with the same data within statistical noise can change the computed key-rate bound. No confidence interval, bootstrap, or independent cross-check on G is propagated into the finite-key solver, and G is treated as time-independent even though Fig. 2(b) shows substantial time-dependent count-rate dips. The authors should provide a bootstrap or profile-likelihood analysis of G, or conservatively minimize the key rate over a confidence set of G, and report the resulting lower bounds for Table 1. Without this, the reported 1.6-20 kbps rates are not established as rigorous finite-key lower bounds.","section":"Supplementary Note 11, Eqs. (10)-(24); Methods §4.3"},{"comment":"The decoy and vacuum mean photon numbers are estimated from the same QKD counts that are later fed into the key-rate solver; Methods explicitly says the analysis \"search[es] for self-consistent intensities given the observations by performing another least-squares estimation.\" This self-referential estimation absorbs statistical fluctuations of the session into the model parameters rather than treating the observed counts as fixed statistical constraints with known confidence. Similarly, Supplementary Note 4 optimizes the decoy-state sending fractions using in-flight transmission data; if those data overlap with the sessions used to compute the Table 1 rates, the protocol parameters are not fixed in advance as the security proof requires. To break the circularity, the intensities should be determined from independent calibration data (or one-sided confidence bounds should be used), and the protocol-parameter optimization should be performed on a separate training set from the evaluation data.","section":"Supplementary Note 11, Part 2; Supplementary Note 4; Methods §4.3"},{"comment":"The post-processing selects only synchronization blocks with at least 95% synchronization confidence and a noise fraction below 0.2, and parts of the data are discarded because of an FPGA write bottleneck. The manuscript does not report the fraction of raw data discarded per run or the sensitivity of the reported QBER and key rates to these filtering thresholds. If the thresholds were chosen after inspecting the data, conditioning on favorable subsets could inflate the finite-key rates. Please report retention statistics for each run and show that the key-rate conclusions are robust over a range of threshold values.","section":"Supplementary Note 10"},{"comment":"The finite-key rates are computed by a custom numerical implementation of the frameworks in Refs. [30-33], but the code is not deposited and the data are only \"available from the corresponding author upon reasonable request.\" Because the security claim depends on numerical optimization (the key-rate solver and the least-squares estimation of G and intensities), the reader cannot verify that the implementation correctly applies Corollary 4 of Ref. [33] and satisfies the finite-size bounds. The authors should deposit the key-rate solver, the device-model fitting code, and the input data for at least one run of each configuration so that the reported lower bounds are reproducible.","section":"Methods §4.3; Code availability"}],"minor_comments":[{"comment":"The abstract describes \"single-photon quantum states\" being transmitted, but Results §2 states that the source uses \"attenuated LEDs... not true single-photon states.\" Please use \"weak coherent pulses\" throughout to avoid an internal contradiction.","section":"Abstract; Results §2"},{"comment":"In the receiver description, \"the R/L (H/L) basis is on the transmitted (reflected) paths\" should presumably read \"the R/L (H/V) basis\"; please correct the typo.","section":"Supplementary Note 6"},{"comment":"The main text says the battery provides approximately 5 minutes of flight time while carrying the QKD equipment, whereas Supplementary Note 1 gives a typical flight time of 2.5 minutes. Please reconcile these numbers.","section":"Methods §4.1; Supplementary Note 1"},{"comment":"The supplementary material is linked to an Overleaf editing URL (https://www.overleaf.com/6163117185jdntcjsptcdk), which is not a stable public publication link; provide a permanent DOI or repository URL.","section":"Supplementary information"},{"comment":"Results §2.1 says the security analysis assumes Eve performs a collective attack, while the abstract claims \"information-theoretic secure\" communication and Methods invokes composable ε-security. Please clarify whether the Table 1 rates are secure against collective attacks only or against general attacks, and adjust the abstract wording accordingly.","section":"Results §2.1; Methods §4.3"},{"comment":"The tomographic states have purities between 99.1% and 100%, but the device model restricts the transmitted states to be pure. Please quantify the effect of neglecting the measured impurity on the resulting key-rate bound.","section":"Supplementary Note 3; Supplementary Note 11"}],"recommendation":"major_revision","confidential_remarks":"The experimental campaign is strong and the manuscript is likely to be publishable after additional statistical rigor. The main concern is that the key-rate lower bounds are not conservative with respect to the device-model uncertainty; I would like to see bootstrap confidence intervals or a pessimistic optimization over G, plus a clear separation between training and evaluation data for the protocol parameters. I also note that Ref. [30] is an arXiv preprint; the authors should update the citation to a peer-reviewed version if one exists. The Overleaf-only supplementary link is not suitable for a published record."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The one thing you should know: this is the first QKD demonstration where neither endpoint is fixed — drone-to-drone, drone-to-vehicle, and vehicle-to-vehicle all work, with finite-key secure rates of 1.6–20 kbps in single sessions. That's a genuine capability milestone, not a rehash. The hardware description is unusually complete: PAT loops, SWaP-optimized payloads, time tagging, post-processing — all documented in enough detail to reproduce.\n\nWhat's new is the integration and the attempt to incorporate non-ideal device behavior into the security analysis. The quantum state tomographies, decoy-state optimization, and modular system design are all credible. The authors are candid about limitations (night-only, 100 m range), and the finite-key framework they invoke is legitimate.\n\nThe soft spot is where the security claim load-bears. In Supplementary Note 11 they model the entire channel and receiver as an 8×2 complex isometry G — 32 real parameters — fit by least squares to 12 measured signal count rates. The loss modes are never directly measured, so in practice the fit is underdetermined. And the same counts that determine G are later fed into the finite-key solver. No confidence region or bootstrap is propagated into the key rate. If G is close to the true POVM, the rates are probably fine, but as written the 1.6–20 kbps numbers are not proven lower bounds. This is fixable: release the code and raw data, use a more constrained parameterization, and propagate estimation error or at least show the fit is stable across runs.\n\nThe security proof itself is cited rather than reproduced; that's acceptable but makes it harder to check the application of Corollary 4 of Ref. [33] to this setup. The post-hoc filtering (keeping blocks above 95% sync confidence and noise fraction below 0.2) is described but not justified as bias-free.\n\nBottom line: this paper deserves a serious referee. The central security claim is conditional until the G-fitting uncertainty is addressed, but the demonstration is real and the analysis is mostly careful. I'd send it to review, specifically asking the authors to tighten the device-model fit, and I'd cite it for the system integration milestone.","headline":"First fully-mobile QKD demonstrations with credible engineering, but the reported secure key rates rest on a device-model fit that needs much closer scrutiny.","tokens_in":22250,"tokens_out":2314,"would_cite":true,"duration_ms":19904,"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":"This paper reports the first quantum key distribution in which both communicating endpoints move: drone-to-drone, drone-to-vehicle, and vehicle-to-vehicle links, with finite-key secure rates of 1.6–20 kbps.","keywords":["quantum key distribution","mobile platforms","drone-to-drone","free-space QKD","finite-key security","decoy-state protocol","polarization encoding","pointing and tracking"],"falsifier":"Record the same QKD session's counts, split the data, and fit $G$ on one half while computing the finite-key rate on the other half; if the predicted received mean photon numbers for decoy and vacuum states deviate from the measured counts by more than the Poisson statistics allow, or if the key rate becomes non-positive for an equally good fit to the signal data, then the model does not capture the device and the claimed secure rates would fail.","tokens_in":21083,"feed_emoji":"🔐","tokens_out":6520,"duration_ms":72697,"temperature":0.7,"pith_summary":"This paper reports a quantum key distribution system in which neither endpoint is a fixed ground station: the transmitter and receiver are modular payloads that can be mounted on drones or vehicles, and the authors demonstrate drone-to-drone, drone-to-vehicle, and vehicle-to-vehicle links. The central claim is that these fully mobile links generate securely usable keys in the finite-key regime within a single short session, at secure rates of 1.6 to 20 kbps depending on configuration. The reason this matters is that mobile nodes are the missing piece for reconfigurable quantum networks; prior airborne and vehicle demonstrations always had one stationary party. To back the security claim, the paper builds a device model that accounts for channel loss, imperfect optics, and uneven detector efficiencies, which are more pronounced when both platforms are moving.","feed_headline":"Quantum keys flow between two moving drones and cars","feed_subtitle":"First fully mobile QKD links yield 1.6-20 kbps secure keys in single short sessions.","key_machinery":"Carrying the argument is the $8\\times2$ complex isometry $G$, a 32-real-parameter unitary model that lumps the entire channel and receiving apparatus—free-space loss, imperfect beam splitters, waveplate retardance errors, coupling inefficiencies, and mismatched detector efficiencies—into one transformation applied to the polarization modes. The model is fitted by least squares to the measured signal-state counts, then used to predict the received mean photon numbers for the signal, decoy, and vacuum states; those predictions are inserted into a composable finite-key security proof. The same model also converts the lossy detection setup into an equivalent lossless one, attributing all losses to a potential eavesdropper. This is what lets the paper claim secure keys from sessions lasting only tens to hundreds of seconds with non-ideal hardware.","core_discovery":"The discovery is the first quantum key exchange between fully mobile platforms, together with a security analysis strong enough for the short, imperfect sessions those platforms allow. Using a polarization-based decoy-state protocol, with circular-polarization key states $|R\\rangle$ and $|L\\rangle$ to resist platform rotations and $|H\\rangle$ for error checking, the authors report averaged secure key rates of 8.5 kbps for drone-to-drone, 1.6 kbps for drone-to-vehicle, 20.0 kbps for vehicle-to-vehicle at 5 mph, and 2.5 kbps for vehicle-to-vehicle at 70 mph, producing 205 kb to 1.88 Mb of secret key per session. These rates come from a custom finite-key analysis that replaces idealized-device assumptions with a fitted model of the real hardware, so the security statement covers the system actually deployed.","pith_inferences":["If the model-based rates hold up under scrutiny, the practical blueprint for mobile QKD shifts from 'connect a moving node to a fixed station' to 'connect any two moving nodes,' which changes network planning for disaster response and tactical communication.","A natural next test, not performed here, is to vary the fit: bootstrap the least-squares estimation of $G$ and feed the resulting distribution of unitaries through the key-rate solver to see how much the quoted 1.6–20 kbps rates spread.","The paper's predicted 49.6 dB daytime signal-to-noise improvement from spectral, spatial, and temporal filtering is a concrete, checkable engineering target; if realized, it would remove the night-operation limitation and make these links usable around the clock."],"forward_implications":["A single short, on-the-move session is enough to distill a securely usable key; the paper demonstrates this in every configuration tested, with no fixed ground station on either side.","Mobile nodes can be reconfigured quickly, since the transmitter and receiver are self-contained payloads that swap between drones and vehicles without sharing power or control with the host platform.","The same finite-key modeling approach extends naturally to future mobile links at longer range and higher speed, because it turns device non-idealities into inputs of the security proof rather than assuming them away.","Moving from entanglement-based mobile nodes to a full quantum network becomes plausible: the authors state the system can be upgraded to entangled-photon sources, reusing the same modular platforms and pointing-and-tracking hardware."],"supporting_citations":[{"why":"Supplies the improved finite-size security bounds that the paper extends to the non-ideal mobile-device model.","marker":"[30]"},{"why":"Gives the unitary characterization of detector setups and the equivalent-lossless conversion that turns measured non-idealities into security-proof inputs.","marker":"[33]"},{"why":"Provides the decoy-state method that lets strongly attenuated LED sources resist photon-number-splitting attacks.","marker":"[18]"},{"why":"Loss-tolerant protocol analysis that justifies the simplified receiver measuring all four polarization states while preserving security.","marker":"[17]"},{"why":"Establishes the composable finite-key framework whose security parameters are used throughout the analysis.","marker":"[20]"}],"fun_headline_variants":["Quantum keys exchanged between moving drones and cars","Drones and cars swap quantum keys at up to 20 kbps","First mobile QKD links: drone-vehicle secure keys","Quantum key distribution on the move: drones and autos","Secure quantum communication from moving platforms"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The security claim rests on the assumption that the fitted mathematical model of the equipment is an exact, complete description of how the real drones, vehicles, channel, and detectors behave during the session, and that the uncertainty from fitting that model does not change the security conclusion; if a different model that matches the measurements just as well would give a much lower key rate, the reported secure rates are not rigorously supported.","fun_headline_variants_meta":{"raw":{"variants":["Quantum keys exchanged between moving drones and cars","Drones and cars swap quantum keys at up to 20 kbps","First mobile QKD links: drone-vehicle secure keys","Quantum key distribution on the move: drones and autos","Secure quantum communication from moving platforms"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000677,"raw_usage":{"total_tokens":3047,"prompt_tokens":880,"completion_tokens":2167,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":496,"completion_tokens_details":{"reasoning_tokens":2091}},"tokens_in":496,"tokens_out":2167,"duration_ms":11394,"temperature":1.0,"reasoning_tokens":2091,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T14:43:53.303227+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Record the same QKD session's counts, split the data, and fit $G$ on one half while computing the finite-key rate on the other half; if the predicted received mean photon numbers for decoy and vacuum states deviate from the measured counts by more than the Poisson statistics allow, or if the key rate becomes non-positive for an equally good fit to the signal data, then the model does not capture the device and the claimed secure rates would fail.","supporting_citations":[{"cited_title":"& L¨ utkenhaus, N","cited_arxiv_id":null,"evidence_quote":"Gives the unitary characterization of detector setups and the equivalent-lossless conversion that turns measured non-idealities into security-proof inputs."},{"cited_title":"& Chen, K","cited_arxiv_id":null,"evidence_quote":"Provides the decoy-state method that lets strongly attenuated LED sources resist photon-number-splitting attacks."},{"cited_title":"& Azuma, K","cited_arxiv_id":null,"evidence_quote":"Loss-tolerant protocol analysis that justifies the simplified receiver measuring all four polarization states while preserving security."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the composable finite-key framework whose security parameters are used throughout the analysis."}],"review_version":1}