{"id":"47364df5-e969-4b44-ba7d-30ca8dede82c","arxiv_id":"2411.10925","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"The paper defines C-V2X resilience under imperfect CSI via an MSE upper bound on PDF estimation and a hazard-rate metric, revealing a power-allocation tradeoff between remediation and adaptation.","lead":"This paper proposes a framework for measuring the resilience of cellular vehicle-to-everything (C-V2X) networks when channel state information is imperfect. It defines two metrics, remediation capability and adaptation performance, and shows that choosing power levels during an adaptation phase creates a tradeoff between them.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The framework's central estimator and MSE bound require E to be stationary over the T adaptation slots, yet Sec. II.C concedes E may be time-varying; with drift, Eqs. (8)-(10) estimate a time-averaged distribution and Theorem 1's remediation-capability metric is not defined.","rationale":"The reader's weakest_assumption coincides with mine. I regard it as the most load-bearing because it is an explicit premise of the deconvolution identity in Eq. (8) and of Theorem 1's bias-variance decomposition; if it fails, the two proposed resilience metrics lose their meaning. Unlike the omitted proof of Lemma 1 or the ambiguous e_nm notation in Eq. (10), this is not a fixable presentation issue: it is a modeling assumption that the paper itself flags as false in the motivating regime. A simulation with drifting E would settle the practical impact; if drift is slow, the framework may still be useful as a quasi-static approximation, but that scope restriction must be stated. Therefore the reader's CONDITIONAL verdict is appropriate; my stress-test does not change it.","tokens_in":11234,"tokens_out":10964,"duration_ms":125547,"concrete_test":"Repeat the Fig. 2 experiment with a time-varying E: for example, let the two GMM component means drift linearly from (0.2, 0.8) to (0.4, 1.0) over T = 1000 slots (or use a small random walk per slot), keeping PA I-III and all other parameters unchanged. Compute the empirical MSE of f̂_E against the instantaneous f_E at each slot and compare with the Theorem 1 upper bound. If the bound is violated or the estimate tracks the time-averaged distribution rather than the current f_E, the central remediation-capability metric fails under the paper's own dynamic-environment motivation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section II.C states that the unknown error distribution E 'may be time-varying due to the highly dynamic environment,' but the entire adaptation-phase analysis treats E as fixed. Equation (8) declares Z = e_nm + Y to be i.i.d. and uses the empirical characteristic function in Eq. (9) to approximate F{f_E}. If E drifts across the T slots, the empirical characteristic function converges to the characteristic function of a mixture (or time average) of the instantaneous error distributions, not to F{f_E} for any single f_E. Consequently f̂_E in Eq. (10) does not estimate the PDF needed to evaluate the QoS probability in Eq. (5), and the 'remediation capability' MSE bound in Theorem 1, derived under the i.i.d. assumption, no longer applies. This is not a minor edge case: the paper motivates the work precisely by highly dynamic C-V2X environments, and the simulations in Sec. IV fix E as a static Gaussian mixture, so the assumption is untested exactly where it matters most. The framework is coherent only if restricted to quasi-stationary windows; that restriction is neither stated nor validated.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an analytical framework for studying the resilience of C-V2X networks under imperfect CSI. The framework focuses on an adaptation phase during which the base station estimates the PDF of an unknown additive CSI error via a deconvolution estimator based on RSS samples. The MSE of this estimator is defined as the network's remediation capability, and an upper bound is derived in Theorem 1. Adaptation performance is quantified by a hazard-rate metric, and a tradeoff between remediation capability and adaptation performance is claimed. The paper is illustrated with simulations comparing three adaptation power schemes.","tokens_in":11455,"tokens_out":12306,"duration_ms":115431,"significance":"If the framework is correct, it is a useful conceptual contribution: it offers quantitative definitions of remediation capability and adaptation performance, and the derived MSE bound provides a principled basis for selecting adaptation powers. The paper addresses an important gap by explicitly considering QoS during the adaptation phase, which prior robust designs ignore. The deconvolution estimator and the hazard-rate formulation are sensible starting points. However, the framework is only valid under a quasi-stationarity assumption on the error distribution that is acknowledged but not handled, and the proof of the central hazard-rate lemma is omitted, so the strength of the claims is not fully supported.","major_comments":[{"comment":"Section II.C states that the error distribution E 'may be time-varying due to the highly dynamic environment,' but Section III.A assumes the samples Z in Eq. (8) are i.i.d. from a fixed random variable Z = e_nm + Y. If E drifts over the T adaptation slots, the empirical characteristic function in Eq. (9) estimates the characteristic function of a time-averaged mixture of distributions, not F{f_E} for any single E; consequently, the estimator in Eq. (10) does not target the PDF needed in the QoS probability (5), and Theorem 1's MSE bound, derived under an i.i.d. assumption, no longer applies. Since the paper motivates the work by the dynamic nature of vehicular environments, this is a load-bearing limitation. The authors should either explicitly restrict the framework to quasi-stationary windows with a justification and validation, or extend the analysis to non-stationary error distributions.","section":"Section II.C vs. Section III.A"},{"comment":"The symbol e_nm is used both as the unknown additive error term (e.g., Eq. (4) and Eq. (7)) and as the evaluation point of the estimated density in Eq. (10). This ambiguity carries into Theorem 1, where the first term of the bound in Eq. (11) contains e^{j w e_nm}; if e_nm is interpreted as the random error, the bound is not well-defined, and if it is the evaluation point, the bound is pointwise and should be written with a different symbol (e.g., x). Please clarify the notation and state explicitly whether the MSE is defined pointwise or as an integrated quantity.","section":"Section III.A, Eq. (10) and Theorem 1"},{"comment":"The hazard-rate formulas in Lemma 1 are a central component of the adaptation-performance analysis and of the claimed tradeoff, but the proof is omitted with the statement 'The proof was omitted due to space limitation,' and the formulas are not independently verified. Moreover, the displayed expressions contain ambiguous parentheses that make the denominator structure unclear. A proof or at least a derivation sketch should be provided, and the equations should be rewritten unambiguously.","section":"Lemma 1, Eqs. (14) and (15)"}],"minor_comments":[{"comment":"The text states that the parameters of the three power schemes satisfy oI = oII = 1/2 oIII, but according to the definitions in Table II, PA III doubles P_m while halving P_n relative to PA I, which yields oIII = 4 oI; please check this relationship.","section":"Table II and Section IV"},{"comment":"In the definition of o, the denominator is written as L_a^n, but Eq. (6) and the surrounding analysis use L_a^nm for the interference link; this is likely a typo and should be corrected for consistency.","section":"Theorem 1"},{"comment":"The phrase 'Pascal theorem' in the sentence before Eq. (9) is likely a typo for 'Parseval theorem' or 'Fourier inversion'; please correct.","section":"Section I"},{"comment":"The word 'adaption' is used in several places (e.g., 'adaption phase' in Section III.A and III.B); it should be 'adaptation'.","section":"Throughout"},{"comment":"The first equality in Eq. (10) should be an approximation because the truncated integral is introduced immediately after; furthermore, the role of the evaluation point should be made explicit.","section":"Eq. (10)"},{"comment":"The figure plots the estimated densities for the three schemes but does not overlay the true GMM density, so the claim that PA I and PA II are 'more accurate' than PA III is not visually supported; please include the true PDF or an empirical MSE comparison.","section":"Fig. 2a"},{"comment":"The simulation description says results are averaged over 10,000 channel realizations with T = 1,000, but it is not specified whether the estimation is repeated across these realizations and how the empirical MSE is computed; please clarify the simulation procedure.","section":"Section IV"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this paper is a legitimate, if incremental, extension of two known tools—deconvolution density estimation and hazard-rate analysis—to the C-V2X resilience problem. The framework is coherent and the tradeoff story is convincing. It is not a breakthrough, and the paper oversells its novelty, but it is a useful contribution for the V2X resource-management subfield.\n\nWhat is actually new: the mapping of adaptation-phase QoS to a hazard rate, and the use of a deconvolution MSE bound as a 'remediation capability' metric. Theorem 1 is a correct bias-variance decomposition; the variance term genuinely shows how the power split between V2I and V2V links trades estimation accuracy against QoS during adaptation. The three simulation power schemes illustrate that tradeoff cleanly.\n\nThe soft spots are real but not fatal. Lemma 1—the explicit hazard-rate formulas—is stated without proof. For a paper whose contribution is partly analytical, omitting the proof of the backbone result is a serious gap. The notation in Eq. (10) is genuinely ambiguous: e_nm is used as both the evaluation point and the unknown error, which makes the estimator hard to parse. The stationarity assumption is the deepest issue: Sec. II.C admits E may be time-varying, yet the whole estimator (8)-(10) requires a fixed E for the empirical characteristic function to be consistent. If E drifts over the T slots, the estimated PDF is a time average, not the distribution needed for (5). This is not a nitpick—the paper motivates itself with highly dynamic environments—but it is also a common modeling restriction. The fix is to state explicitly that the framework applies to quasi-stationary windows, and to validate that with a time-varying simulation. The validation is otherwise internal: only PA I-III are compared, no prior robust or model-based baseline, so 'superiority' is not demonstrated. Minor: the text says PA II outperforms PA I in both dimensions, but Fig. 2a shows identical remediation capability.\n\nWho should read this: people working on V2X resource allocation with imperfect CSI. They will find a clear conceptual framework and a transferable estimation recipe. It deserves a serious referee; the missing proof and the stationarity issue need addressing before it can be relied upon, but that is exactly what peer review is for.","headline":"A coherent but incremental application of deconvolution estimation and hazard-rate metrics to C-V2X resilience; the framework is useful, yet the 'first time' and 'superiority' claims outrun the standard math and internal-only validation.","tokens_in":11990,"tokens_out":6374,"would_cite":false,"duration_ms":58599,"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 proposes an analytical framework that defines, quantifies, and analyzes C-V2X network resilience under imperfect CSI through two new dimensions—remediation capability and adaptation performance—and reveals a tradeoff between…","keywords":["C-V2X","resilience","imperfect CSI","deconvolution estimation","hazard rate","adaptation power allocation","remediation capability","QoS"],"falsifier":"Simulate the adaptation phase with a time-varying error distribution, for example a Gaussian mixture whose component means shift every few slots, and compare the estimated PDF against the true one: if the empirical mean square error exceeds the Theorem 1 bound or the remediation phase misses its delay target, the stationarity of $E$ is the breaking point. Separately, recompute the hazard-rate formulas in Lemma 1 by direct Monte Carlo simulation of the definitions (12)-(13); a mismatch would expose the omitted proof as a substantive gap.","tokens_in":11008,"feed_emoji":"📡","tokens_out":8202,"duration_ms":78425,"temperature":0.7,"pith_summary":"This paper argues that resilience of a cellular vehicle-to-everything (C-V2X) network against imperfect channel state information (CSI) can be measured before the network fully recovers, during the adaptation phase when the base station is still learning the error in its CSI. To do this, the paper defines two quantities: remediation capability, the mean-square accuracy of a deconvolution-based estimate of the unknown CSI error distribution, and adaptation performance, how close violated quality-of-service stays to its requirement, captured by a hazard rate. The central result is an upper bound on the estimation error (Theorem 1) showing that improving remediation capability—by raising V2I adaptation power or lowering V2V adaptation power—degrades QoS during adaptation, while a longer adaptation window improves remediation. The paper claims these are the first quantitative definitions of C-V2X resilience along adaptation and remediation, and shows by simulation that the two metrics correctly rank adaptation power schemes, with one scheme (PA II) dominating another (PA I) on both dimensions. If the framework is right, network operators can choose adaptation power profiles that trade short-term QoS against how well the system will recover.","feed_headline":"Two metrics quantify C-V2X resilience under imperfect CSI","feed_subtitle":"PDF-estimation accuracy and hazard rate expose how adaptation power choices trade short-term QoS for recovery.","key_machinery":"The machinery is a deconvolution estimator plus a hazard-rate metric. In an adaptation phase of $T$ slots, the base station collects samples $z_t$ in (6) that equal the unknown CSI error $e_{nm,t}$ plus an exponential noise $Y$ whose parameter $\\lambda_Y$ is known from the adaptation powers; because $Z = e_{nm} + Y$ is a convolution, the PDF of $e_{nm}$ is recovered by Fourier division, approximated by the empirical characteristic function in (9) and a truncated inverse transform in (10). Theorem 1's MSE bound in (11) is the object that defines remediation capability and exposes its dependence on the adaptation power ratio and $T$. On the adaptation-performance side, equations (12)-(13) define hazard rates as conditional densities of QoS at the requirement boundary, and Lemma 1 evaluates them in closed form.","core_discovery":"The paper's central claim is that resilience of a C-V2X network facing arbitrary, unknown CSI error can be decomposed into remediation capability and adaptation performance, both quantifiable without knowing the error distribution in advance. Remediation capability is defined as the mean square error of the estimated PDF of the unknown error $E$, and Theorem 1 bounds this MSE by a term depending on the tail of $E$'s Fourier transform plus a term that grows with the ratio $o = K\\pi(1-\\delta_m^2) P_m^a L_m^a / (P_n^a L_n^a)$, so larger V2I adaptation power, smaller V2V adaptation power, and longer adaptation windows improve remediation. Adaptation performance is defined through hazard rates $\\Lambda_V^+$ and $\\Lambda_I^-$ in (14)-(15), which measure, conditional on a QoS violation, the likelihood that the violated QoS stays near the target; the weighted sum $\\Lambda$ in (16) is proposed as a single adaptation-performance metric. The paper then claims the two dimensions are in tension: the power choices that maximize estimation accuracy are the ones that most hurt adaptation-phase QoS, and the simulations show PA II dominates PA I on both dimensions, validating the framework's ability to guide adaptation power design.","pith_inferences":["If the error distribution $E$ is time-varying, as the paper admits it may be, the i.i.d. assumption behind the deconvolution estimator fails and the Theorem 1 bound no longer applies; a natural extension would replace the empirical characteristic function with a drifting-window or online deconvolution estimator.","The same two-metric template—deconvolution-based learning accuracy plus hazard-rate closeness of violated QoS—transfers to other wireless systems that must learn an unknown additive disturbance, such as jamming, hardware impairment, or channel aging, during an adaptation phase.","The closed-form hazard rates in Lemma 1 are stated without proof, so a Monte Carlo verification of (14)-(15) under the simulation's Gaussian mixture error would settle whether the omitted derivation is correct before the metric is used for power design.","Although the paper stops at analyzing the tradeoff, its two metrics are ready-made objectives for a constrained optimization that selects $P_n^a, P_m^a, T$ to maximize remediation capability subject to a minimum acceptable adaptation-phase hazard rate."],"forward_implications":["Adaptation power can be designed from Theorem 1: raising $P_n^a$ and lowering $P_m^a$ reduces the MSE bound and thereby improves remediation capability, at the price of worse V2V QoS during adaptation.","A longer adaptation phase $T$ lowers the MSE bound, so the network can compensate for an unfavorable power ratio by collecting more samples before remediating.","The hazard rate gives a finer characterization of adaptation-phase QoS than outage probability, distinguishing a system whose violated delays stay just above $\\tau_0$ from one whose delays are far worse.","PA II dominates PA I in both remediation capability and adaptation performance in the simulations, showing resilience can be improved without sacrificing estimation accuracy by choosing the right power pair.","The weighted hazard rate $\\Lambda$ in (16) provides an explicit objective for future optimization of the adaptation power scheme balancing V2I and V2V priorities."],"supporting_citations":[{"why":"It supplies the deconvolution framework and the convergence guarantees for estimating a density from noisy observations, which the paper uses to build the estimator.","marker":"[12]"},{"why":"It provides the empirical-characteristic-function approximation and the Fourier division used in Eq. (9) to recover the PDF of the unknown error.","marker":"[13]"},{"why":"It defines resilience as adaptation followed by remediation, the conceptual two-phase structure the paper's metrics operationalize.","marker":"[14]"},{"why":"It supplies the hazard rate concept that the paper adapts to score how close violated QoS stays to the target during adaptation.","marker":"[15]"},{"why":"It provides the Rayleigh fading and Doppler correlation model and the QoS-constrained power allocation setting that the system model extends.","marker":"[3]"},{"why":"It justifies the Gauss-Markov model for V2V small-scale fading and the coefficient $\\delta_m$ used in the error model and the deconvolution samples.","marker":"[5]"}],"fun_headline_variants":["C-V2X resilience: two metrics reveal recovery vs adaptation tradeoff","Imperfect CSI? New framework measures C-V2X resilience with two metrics","Adaptation power tradeoff: C-V2X recovery vs QoS stability quantified","Hazard rate and MSE bound: new metrics for C-V2X resilience"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the unknown error distribution $E$ stays the same over the $T$ adaptation slots, so the collected samples all come from one fixed distribution and the deconvolution estimator in (10) is valid; the paper itself notes $E$ may be time-varying in a highly dynamic environment, and if it drifts, the mean-square-error bound that defines remediation capability no longer holds.","fun_headline_variants_meta":{"raw":{"variants":["C-V2X resilience: two metrics reveal recovery vs adaptation tradeoff","Imperfect CSI? New framework measures C-V2X resilience with two metrics","Adaptation power tradeoff: C-V2X recovery vs QoS stability quantified","Hazard rate and MSE bound: new metrics for C-V2X resilience"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000242,"raw_usage":{"total_tokens":1589,"prompt_tokens":1076,"completion_tokens":513,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":692,"completion_tokens_details":{"reasoning_tokens":431}},"tokens_in":692,"tokens_out":513,"duration_ms":6747,"temperature":1.0,"reasoning_tokens":431,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T19:09:31.444833+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the adaptation phase with a time-varying error distribution, for example a Gaussian mixture whose component means shift every few slots, and compare the estimated PDF against the true one: if the empirical mean square error exceeds the Theorem 1 bound or the remediation phase misses its delay target, the stationarity of $E$ is the breaking point. Separately, recompute the hazard-rate formulas in Lemma 1 by direct Monte Carlo simulation of the definitions (12)-(13); a mismatch would expose the omitted proof as a substantive gap.","supporting_citations":[{"cited_title":"Adaptive estimation of linear functionals in the convolution model and applications,","cited_arxiv_id":null,"evidence_quote":"It supplies the deconvolution framework and the convergence guarantees for estimating a density from noisy observations, which the paper uses to build the estimator."},{"cited_title":"Resilience and survivability in commu- nication networks: Strategies, principles, and survey of disciplines,","cited_arxiv_id":null,"evidence_quote":"It defines resilience as adaptation followed by remediation, the conceptual two-phase structure the paper's metrics operationalize."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It supplies the hazard rate concept that the paper adapts to score how close violated QoS stays to the target during adaptation."},{"cited_title":"Resource Allocation for D2D-Enabled Vehicular Communications,","cited_arxiv_id":null,"evidence_quote":"It provides the Rayleigh fading and Doppler correlation model and the QoS-constrained power allocation setting that the system model extends."},{"cited_title":"Resource Allocation for D2D- Based V2X Communication With Imperfect CSI,","cited_arxiv_id":null,"evidence_quote":"It justifies the Gauss-Markov model for V2V small-scale fading and the coefficient $\\delta_m$ used in the error model and the deconvolution samples."}],"review_version":1}