{"id":"6d3ccbe1-80e4-46b6-bd8f-1b4e85c16703","arxiv_id":"2608.13270","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Worsening the phase noise of a legitimate ISAC transmitter's local oscillator widens the sensing accuracy gap between the legitimate monostatic receiver and a bistatic eavesdropper, at a moderate cost to communication rate.","lead":"This paper shows that deliberately using a lower-quality oscillator at a legitimate ISAC base station can degrade an eavesdropper's radar-like sensing more than it degrades the base station's own sensing, while only mildly hurting the data rate to a phone user. The result suggests a hardware-based way to protect location privacy of sensed targets in 6G integrated sensing and communication systems.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Privacy gain is only demonstrated against a PN-ignorant Eve; a PN-aware Eve could remove the misspecification bias and erode the claimed LO-asymmetry mechanism.","rationale":"The paper's central claim is that LO asymmetry can be exploited as a privacy-enhancing mechanism. The mechanism is quantified only through the misspecified Cramér-Rao bound under PN-ignorant processing at Eve. A capable adversary would not ignore phase noise: the DPN covariance (27) is known, the transmitted symbols are assumed known, and the phase process is smooth, so joint estimation of delay and phase noise is a plausible and well-studied approach. The paper itself lists PN-aware receivers as future work (Sec. VI), confirming the missing analysis. This is the most load-bearing assumption because if a PN-aware Eve eliminates the misspecification bias, the privacy gap could shrink dramatically, undermining the claimed trade-off. The reader's weakest_assumption identified exactly this issue; my assessment agrees. The appropriate verdict remains CONDITIONAL: accept subject to demonstrating robustness against a PN-aware Eve, or clearly framing the contribution as a study of PN-ignorant adversaries rather than a general privacy mechanism.","tokens_in":9826,"tokens_out":3461,"duration_ms":41321,"concrete_test":"For the Table I parameters, compute the Bayesian CRB or the MCRB for τ_E under the true model (23) with ξ_E treated as a zero-mean Gaussian nuisance vector with covariance (27), jointly estimating τ_E, α_E, and ξ_E. Recompute the SPG in (36) against Alice's corresponding PN-aware bound. If the SPG drops below ~3 dB over f_A/3dB in [100 Hz, 100 kHz] at SNR=10 dB and R=10 m, the privacy mechanism is not robust to a PN-aware Eve and the Section V conclusions require qualification.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section II-D claims to model Eve as a worst-case adversary via perfect timing and symbol knowledge, but Section IV-B2 and all simulations evaluate Eve only under a PN-ignorant model (31), where the true model is (23). The resulting MCRB/LB incorporates a misspecification bias that Eve could avoid by jointly estimating the delay and the Gaussian DPN trajectory with the known covariance (27). Since the sensing privacy gap in (36) is the ratio of these LBs, the headline result that worsening Alice's LO quality widens the privacy gap may be an artifact of an intentionally suboptimal adversary. The paper explicitly defers PN-aware receivers to future work in Section VI, but provides no bound on a PN-aware Eve. If such an Eve can largely recover the PN-free accuracy, the hardware-induced privacy mechanism largely disappears.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes a three-party OFDM ISAC system composed of a legitimate monostatic transceiver (Alice), a passive bistatic eavesdropper (Eve), and a communication user (UE), each equipped with a non-ideal local oscillator modeled as a Wiener phase-noise process. It derives closed-form expressions for the differential phase-noise covariance at Alice, which is delay-dependent and equals 4π f_A max(τ−|Δt|,0), and at Eve, which is delay-independent and equals 4π(f_A+f_E) min(t1,t2). These statistics are used in a misspecified Cramér-Rao bound (MCRB) analysis to obtain ranging lower bounds under phase-noise-ignorant processing at both receivers. Simulations show that worsening Alice's oscillator quality (increasing f_A) degrades Eve's ranging performance much more than Alice's, thereby enlarging the sensing privacy gap, especially for nearby targets, at a moderate communication-rate cost in noise-limited regimes. The paper also presents trade-off curves between the privacy gap and achievable downlink rate.","tokens_in":9947,"tokens_out":7133,"duration_ms":81647,"significance":"If the results hold, the paper introduces a hardware-induced privacy mechanism for ISAC that complements existing waveform-design approaches. The core statistical insight—shared-LO self-referenced phase noise is delay-dependent while independent-LO cross-referenced phase noise is delay-independent—is clean, exact, and directly useful for system design. Proposition 1 and the covariance expressions are derived in closed form and are consistent with the Wiener phase-noise model; the MCRB and rate expressions follow standard machinery from [13], [14], and [25]. The predicted scalings (Eve's accumulated phase-noise variance growing with f_A+f_E and T, Alice's growing with f_A and τ) are falsifiable and not obtained by parameter fitting. However, the central privacy claim is demonstrated only against a phase-noise-ignorant eavesdropper; the paper provides no quantitative bound for a phase-noise-aware adversary. The significance of the work is therefore conditional on this threat-model assumption.","major_comments":[{"comment":"The headline result that worsening Alice's LO quality widens the sensing privacy gap is established only for a phase-noise-ignorant Eve. In Sec. IV-B2, the true model (23) includes a differential phase-noise process with the known covariance (27), but the assumed model (31) ignores phase noise entirely. Since the sensing privacy gap in (36) is the ratio of the resulting lower bounds, the gap reflects Eve's self-inflicted misspecification error, not a fundamental physical limitation. A phase-noise-aware Eve could in principle estimate the differential phase-noise trajectory jointly with the delay using the known prior (27), thereby removing much of the misspecification bias. The paper defers phase-noise-aware receivers to future work in Sec. VI but provides no bound on such an adversary. To support the claimed privacy mechanism, the authors should either analyze a phase-noise-aware Eve (for example, via a CRB with phase-noise nuisance parameters or a Bayesian bound using the prior (27)) and show the privacy gap persists, or explicitly restrict the privacy claim to phase-noise-ignorant receivers and justify that restriction as a realistic threat model.","section":"Sec. IV-B2, Eq. (36), and Sec. VI"},{"comment":"The threat-model description is internally inconsistent with the evaluation. Sec. II-D states that Eve is modeled as the 'worst-case (most powerful) adversary' because she has perfect timing and perfect symbol knowledge, and the authors claim that any privacy gap therefore originates from the LO asymmetry alone. However, the sensing analysis in Sec. IV-B2 and all simulations in Sec. V evaluate Eve under a deliberately suboptimal, phase-noise-ignorant receiver. Perfect knowledge of timing and symbols does not make a receiver that ignores the known phase-noise statistics worst-case; a worst-case adversary would use all available statistical information, including the covariance (27). This discrepancy affects the interpretation of Figs. 4 and 5, where the privacy gap is driven by Eve's misspecification. The authors should either remove the phrase 'worst-case' and clearly state the adversary is restricted to phase-noise-ignorant processing, or extend the analysis to a phase-noise-aware Eve.","section":"Sec. II-D"}],"minor_comments":[{"comment":"The notation ξ_gen and R_gen is used before it is defined; please define these symbols explicitly before presenting (28) and clarify that the trace expression applies to both the Alice and Eve covariance matrices.","section":"Eq. (28)"},{"comment":"Fig. 2 is not labeled as either a theoretical plot or a Monte Carlo simulation. Since it plots the closed-form mean accumulated variance in (28), the caption should state that the curves are computed directly from (28), not simulated.","section":"Fig. 2"},{"comment":"The MCRB derivation for Eve is only stated to follow [14, Sec. III-C], but the Eve model in (23) differs from the Alice model because the common phase has been absorbed into the channel gain α_E, making the effective gain random and dependent on the delay through φ_A(T_cp−τ). A brief derivation sketch for this case would improve reproducibility.","section":"Sec. IV-B2"},{"comment":"The caption of Fig. 5 shows an arrow labeled 'Degrading LO quality at Alice' but does not specify the direction of increasing f_A. Please clarify that the arrow points in the direction of increasing f_A, consistent with the sweep over [100 Hz, 100 kHz].","section":"Sec. V-C and Fig. 5"},{"comment":"The terms 'sensing privacy' and 'sensing security' are used almost interchangeably. The paper defines sensing privacy as protecting the privacy of sensed entities, but the mechanism actually degrades Eve's ranging accuracy, which is more akin to sensing security. The authors should either use one term consistently or explicitly define the relationship between the two concepts.","section":"Abstract and Introduction"}],"recommendation":"major_revision","confidential_remarks":"The mathematical core of the paper—Proposition 1 and the covariance expressions—is sound and represents a useful contribution. The main risk is that the privacy mechanism is evaluated only against a phase-noise-ignorant eavesdropper, while the paper describes Eve as a worst-case adversary. If the authors can provide a quantitative analysis of a phase-noise-aware Eve (or convincingly justify the restriction to phase-noise-ignorant processing as a real-world adversary model), the paper would be a strong candidate for acceptance. As it stands, the central claim is not yet fully supported."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the math is right and the LO asymmetry is real, but the privacy gain is only established against a PN-ignorant Eve. That is a load-bearing scope limit, not a cosmetic one.\n\nProposition 1 is the cleanest result. For two independent Wiener LOs, the cross-referenced DPN at Eve has covariance 4π(f_A+f_E)min(t1,t2), independent of target delay, while Alice's self-referenced DPN is 4π f_A max(τ-|Δt|,0). I checked the derivation and it is correct. This asymmetry is the entire engine of the paper: Alice's PN distortion shrinks with close targets, Eve's does not. The MCRB and SINR derivations follow standard lines, and the simulation curves match the covariance expressions. No sign of curve fitting or circularity.\n\nThe soft spot is the adversary model. The paper claims a worst-case Eve with perfect timing and symbol knowledge, but the performance metric is the misspecified CRB under a PN-ignorant model. That means Eve's ranging error is partly a misspecification artifact, not a fundamental information limit. A PN-aware Eve could jointly estimate the delay and the DPN realization—the DPN is a Gaussian random vector with known covariance, and OFDM phase-noise compensation has a mature literature. The paper files this under future work, but that future work is the correct null hypothesis against which the privacy mechanism should be tested. Without it, the honest claim is 'phase noise degrades an ignorant eavesdropper more than Alice,' not 'phase noise provides sensing privacy.'\n\nMinor: the UE rate is also computed under PN-ignorant processing, and the scenario is single-symbol, single-antenna, no Doppler. Those are acknowledged limitations and do not bother me.\n\nThe paper deserves a serious referee. The mathematical core is solid, and the three-way trade-off framing is useful. I would ask for a PN-aware Eve analysis or an explicit restriction of the claims to ignorant adversaries. If the authors can show the privacy gap survives partial PN compensation, this is a strong paper; if not, it is still a nice characterization of the ignorant-adversary baseline.","headline":"Correct LO-asymmetry math, but the privacy claim depends on a PN-ignorant Eve; needs a PN-aware adversary analysis.","tokens_in":10513,"tokens_out":3648,"would_cite":false,"duration_ms":39318,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that in an OFDM integrated sensing and communication link, deliberately worsening Alice's local-oscillator phase noise enlarges her sensing privacy gap against a bistatic eavesdropper, with only a moderate communication…","keywords":["phase noise","ISAC","sensing privacy","OFDM","misspecified Cramer-Rao bound","bistatic sensing","monostatic sensing","oscillator quality"],"falsifier":"Replace the phase-noise-ignorant Eve in Section IV-B2 with an estimator that treats the phase-noise samples as unknown nuisance parameters and jointly estimates delay and phase noise, then compute Eve's range RMSE as a function of Alice's LO bandwidth; if this RMSE stays close to the phase-noise-free CRB, the claimed privacy mechanism collapses.","tokens_in":9615,"feed_emoji":"📡","tokens_out":4030,"duration_ms":36825,"temperature":0.7,"pith_summary":"This paper argues that phase noise, normally a hardware impairment to be suppressed, can be turned into a sensing-privacy mechanism in an OFDM integrated sensing and communication system. In the three-party setup of a legitimate monostatic transceiver (Alice), a passive bistatic eavesdropper (Eve), and a communication user, the paper shows that Eve's differential phase noise is statistically independent of target delay while Alice's is suppressed for nearby targets by range correlation. The paper then proves, via misspecified Cramér-Rao bounds, that worsening Alice's local-oscillator quality enlarges the sensing privacy gap in her favor with only a moderate data-rate penalty in noise-limited regimes. A sympathetic reader would care because hardware impairment, not waveform design, becomes the privacy knob, and the paper identifies the precise statistical asymmetry that creates it.","feed_headline":"Worsening phase noise at Alice widens sensing privacy gap","feed_subtitle":"Shared-LO monostatic sensing shrugs off phase noise a bistatic eavesdropper cannot cancel, buying privacy at a small rate cost.","key_machinery":"The load-bearing object is the covariance of the differential phase noise (DPN) at each receiver. Alice's DPN $\\xi_A(t,\\tau)=\\phi_A(t)-\\phi_A(t-\\tau)$ compares one local oscillator with a delayed copy of itself, giving the range-dependent covariance $R_{\\xi\\xi}(\\Delta t,\\tau)=4\\pi f^A_{3\\mathrm{dB}}\\max(\\tau-|\\Delta t|,0)$. Eve's DPN compares two independent Wiener phase-noise walks, giving the delay-independent covariance $4\\pi(f^A_{3\\mathrm{dB}}+f^E_{3\\mathrm{dB}})\\min(t_1,t_2)$. This contrast, range correlation at Alice versus full Brownian drift at Eve, is what carries the privacy argument through the misspecified Cramér-Rao bound computation.","core_discovery":"The central discovery is that the architectural asymmetry between a shared local oscillator at a monostatic ISAC transceiver and the independent oscillators at a passive bistatic eavesdropper makes phase noise a privacy resource rather than only a nuisance. Proposition 1 shows Eve's differential phase noise has covariance $4\\pi(f^A_{3\\mathrm{dB}}+f^E_{3\\mathrm{dB}})\\min(t_1,t_2)$, independent of target delay, while Alice's has covariance $4\\pi f^A_{3\\mathrm{dB}}\\max(\\tau-|\\Delta t|,0)$, which shrinks for small delays through range correlation. Under phase-noise-ignorant processing, Eve's misspecified bound grows with the total local-oscillator bandwidth while Alice's stays comparatively small for nearby targets, so increasing Alice's LO bandwidth widens the sensing privacy gap $\\mathrm{SPG}_{\\mathrm{dB}} = 10\\log_{10}(LB_E/LB_A)$. The paper quantifies this in a three-way trade-off among Alice's sensing, Eve's sensing, and the communication rate.","pith_inferences":["The privacy mechanism likely depends on Eve remaining phase-noise-ignorant: if Eve runs an estimator that tracks the phase-noise trajectory jointly with delay, the misspecification inflation disappears and the privacy gap could shrink or vanish.","The same LO-asymmetry could be used as an intentional, tunable privacy knob, but this would require accepting a lower oscillator quality as a deliberate design choice, which may conflict with communication and sensing accuracy requirements in other regimes.","The delay-independence result for Eve's differential phase noise suggests a bistatic receiver could estimate the combined LO bandwidth from a single OFDM symbol regardless of target delay, which could be used for oscillator calibration or for detecting the presence of a passive eavesdropper."],"forward_implications":["Sensing privacy in ISAC systems can be tuned by hardware choice: increasing Alice's LO 3-dB bandwidth enlarges the privacy gap because Eve's phase-noise variance grows with the OFDM symbol duration while Alice's remains tied to the target delay.","In noise-limited communication regimes, privacy can be bought for a small rate penalty, since the rate is constrained mainly by additive noise rather than by phase noise.","The privacy advantage is strongest for nearby targets, where range correlation suppresses Alice's differential phase noise while Eve's delay-independent differential phase noise remains at full strength.","At high communication SNR, the same privacy gain costs more rate, revealing a quantitative three-way trade-off between Alice's sensing accuracy, Eve's sensing accuracy, and the achievable rate."],"supporting_citations":[{"why":"Supplies the shared-LO differential phase-noise covariance model with range correlation used for Alice.","marker":"[13]"},{"why":"Provides the CRB and MCRB derivation machinery for monostatic sensing under phase noise.","marker":"[14]"},{"why":"Establishes the vulnerability of bistatic sensing with independent local oscillators, the asymmetry this paper exploits.","marker":"[15]"},{"why":"Gives the free-running oscillator Wiener phase-noise model used for all three LOs.","marker":"[16]"},{"why":"Provides the OFDM common-phase-error and intercarrier-interference description used for the communication link.","marker":"[17]"},{"why":"Supplies the per-subcarrier SINR and achievable-rate expressions used to quantify the communication trade-off.","marker":"[25]"},{"why":"Provides the misspecified Cramér-Rao bound framework underlying the main performance metric.","marker":"[26]"}],"fun_headline_variants":["Phase noise flips from nuisance to privacy shield in ISAC","Alice's phase noise hurts Eve's sensing more than her own","Shared LO turns phase noise into a sensing privacy asset","Worse Alice LO widens sensing privacy gap at modest rate cost"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire privacy gain is computed against an eavesdropper whose estimator ignores phase noise; if Eve runs an estimator that tracks the phase-noise trajectory, she can undo most of the degradation the paper credits to LO asymmetry.","fun_headline_variants_meta":{"raw":{"variants":["Phase noise flips from nuisance to privacy shield in ISAC","Alice's phase noise hurts Eve's sensing more than her own","Shared LO turns phase noise into a sensing privacy asset","Worse Alice LO widens sensing privacy gap at modest rate cost"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000532,"raw_usage":{"total_tokens":2582,"prompt_tokens":987,"completion_tokens":1595,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":603,"completion_tokens_details":{"reasoning_tokens":1525}},"tokens_in":603,"tokens_out":1595,"duration_ms":15106,"temperature":1.0,"reasoning_tokens":1525,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:55:51.260732+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Replace the phase-noise-ignorant Eve in Section IV-B2 with an estimator that treats the phase-noise samples as unknown nuisance parameters and jointly estimates delay and phase noise, then compute Eve's range RMSE as a function of Alice's LO bandwidth; if this RMSE stays close to the phase-noise-free CRB, the claimed privacy mechanism collapses.","supporting_citations":[{"cited_title":"Monostatic sensing with OFDM under phase noise: From mitigation to exploitation,","cited_arxiv_id":null,"evidence_quote":"Supplies the shared-LO differential phase-noise covariance model with range correlation used for Alice."},{"cited_title":"On the impact of phase noise on monostatic sensing in OFDM ISAC systems,","cited_arxiv_id":null,"evidence_quote":"Provides the CRB and MCRB derivation machinery for monostatic sensing under phase noise."},{"cited_title":"Impact of synchronization error and phase noise on OFDM-based distributed ISAC systems,","cited_arxiv_id":null,"evidence_quote":"Establishes the vulnerability of bistatic sensing with independent local oscillators, the asymmetry this paper exploits."},{"cited_title":"Phase noise in oscillators: a unifying theory and numerical methods for characterization,","cited_arxiv_id":null,"evidence_quote":"Gives the free-running oscillator Wiener phase-noise model used for all three LOs."},{"cited_title":"Effects of phase noise on OFDM systems with and without PLL: Characterization and compensation,","cited_arxiv_id":null,"evidence_quote":"Provides the OFDM common-phase-error and intercarrier-interference description used for the communication link."},{"cited_title":"Performance analysis of OFDM with Wiener phase noise and frequency selective fading channel,","cited_arxiv_id":null,"evidence_quote":"Supplies the per-subcarrier SINR and achievable-rate expressions used to quantify the communication trade-off."},{"cited_title":"Performance bounds for parameter estimation under misspecified models: Fundamental findings and applications,","cited_arxiv_id":null,"evidence_quote":"Provides the misspecified Cramér-Rao bound framework underlying the main performance metric."}],"review_version":1}