{"id":"76c1b05d-96bf-43c0-8ee8-4d8e143c0633","arxiv_id":"2501.15227","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Cell-free massive MIMO can jointly support communication and drone detection, with an adaptive weighting algorithm that cuts the age of sensing by 45 percent versus fixed weights.","lead":"The paper proposes a cell-free massive MIMO network that both serves users and detects unauthorized drones, balancing sensing accuracy against scanning time. It introduces an age-of-sensing metric and reports that adaptive tuning cuts sensing delay by 45 percent while keeping 98 percent coverage.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Surrogate objective E{T|H1}-E{T|H0} in Eq. (18) is not shown to track Pd at fixed Pfa; the claimed 98% coverage and 45% AoS reduction may be suboptimal, so the conditional verdict should stand pending a direct Pareto test.","rationale":"The paper's model and derivations are mostly self-consistent: the signal model, the test statistic expectations, and the CCP reformulation check out, and the adaptive weight scheme does verify Pd by Monte Carlo before outputting blocklengths, so the empirical coverage numbers may be correct as reported. My concern is not that the results are fabricated, but that the optimization objective may not be aligned with the true metric, which means the 'minimizes AoS / maximizes coverage' claim is not established. A direct comparison against the true Pareto frontier (using the same detector) would settle this. Because the concern is substantive but testable, and because the reader already conditioned acceptance on resolving the surrogate issue, I do not change the verdict: keep conditional accept pending the test. I partially agree with the reader's weakest_assumption: they identified the same surrogate problem, though I would emphasize that the adaptive algorithm's Monte Carlo check partially mitigates it, so the main risk is suboptimality rather than invalidity.","tokens_in":9263,"tokens_out":19695,"duration_ms":181625,"concrete_test":"Use the exact Section IV setup (L=5, R=16, M=16, K=8, γc=10 dB, Pfa=0.1, Pth=0.9). For each of the S=100 sensing locations, fix the MRT/RZF precoders and compute the eigenvalues d_i of β*β^T. Perform a dense grid over feasible (τs,ρ0,{ρk}) satisfying (19)-(20) and compute Pd exactly by numerical integration of the noncentral chi-square CDF (weights σ_n^2 x_i/(x_i+σ_n^2), noncentrality Σ x_i^2/(x_i+σ_n^2)) under H1 and the corresponding λ under H0. Find the true minimal τs with Pd≥0.9 and the resulting AoS/coverage frontier. Compare against Table I: if Algorithm 1's AoS for 98% coverage exceeds the true minimal AoS by more than 10%, or if the algorithm declares outage at points the true optimizer covers, the surrogate is not faithful.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Sec. III-C, Eq. (18) replaces the true sensing coverage objective (unit-step function of Pd in Eq. (16)) with the test-statistic mean difference E{T|H1}-E{T|H0}, and Eq. (22) shows this equals Σ x_i^2/(x_i+σ_n^2) with x_i=Mρ0τs d_i. The detection threshold λ is later fixed by the false-alarm probability, but the distribution of T under H0 is a weighted chi-square whose weights σ_n^2 x_i/(x_i+σ_n^2) grow with x_i, so the threshold itself shifts as x_i increases. Maximizing the mean difference does not imply maximizing Pd at the chosen Pfa; the CCP steps (Eqs. 25-26) optimize this surrogate to a KKT point, and only afterward does Algorithm 1 (Step 4) evaluate the true Pd by Monte Carlo. Thus the blocklengths and power allocations fed into the coverage/AoS numbers are chosen by a proxy whose monotonic relationship to Pd is not established. If the proxy saturates or misranks points near the 0.9 threshold, the algorithm may return suboptimal τs or declare outage at feasible points, so the reported 98% coverage and 45% AoS reduction are not guaranteed to be close to the true Pareto optimum. This is the load-bearing gap between the optimization claim and the simulated performance.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes a cell-free massive MIMO integrated sensing and communication (ISAC) framework for detecting unauthorized drones. A maximum a posteriori ratio test (MAPRT) detector is used with distributed transmit and receive access points, and the paper introduces age of sensing (AoS) and sensing coverage as performance metrics. The authors formulate a multi-objective optimization problem over sensing blocklength and power allocation subject to communication SINR and per-AP power constraints, solve a surrogate version via a convex-concave procedure, and propose an adaptive weight selection algorithm. Numerical results report sensing coverage versus AoS trade-offs, altitude effects, and claim 98% sensing coverage with a 45% AoS reduction compared to fixed weights.","tokens_in":9553,"tokens_out":5408,"duration_ms":52833,"significance":"If the central claims are substantiated, the paper would be a useful contribution to the ISAC literature: it brings the age-of-sensing metric into drone detection, treats multistatic sensing in a cell-free massive MIMO system, and provides a tractable blocklength/power optimization framework. The problem formulation and the numerical setup are mostly standard and the paper is clearly written. However, the link between the optimized surrogate objective and the true detection probability is not established, the fixed-weight baseline used for the headline 45% gain is unspecified, and the Monte Carlo evaluation lacks statistical detail. These issues are load-bearing for the reported sensing coverage and AoS claims, so they must be resolved before the paper can be accepted.","major_comments":[{"comment":"The optimization in (18) replaces the sensing coverage objective, which is a step function of the detection probability Pd, with the test-statistic mean difference E{T|H1} - E{T|H0}. Equation (22) shows that this surrogate equals sum_i x_i^2/(x_i + sigma_n^2). The detector threshold lambda is set from the false-alarm probability, but under H0 the test statistic has a weighted chi-square distribution whose weights depend on the optimization variables through B in Eq. (11); consequently, the threshold itself changes with tau_s and rho_0. Maximizing the mean difference is not shown to be monotone in Pd at the chosen Pfa, so the CCP solution of (26) may be suboptimal for the true coverage objective or may even declare outage at feasible points. Algorithm 1 checks Pd by Monte Carlo only after the optimization, so the reported coverage and AoS values are not guaranteed to lie on the true Pareto frontier. Please provide a direct validation of the surrogate, for example a Pareto plot over tau_s and rho_0 comparing the optimized points with an exhaustive search, and state how the threshold lambda is computed in the Monte Carlo evaluation.","section":"Sec. III-C, Eqs. (18) and (22)"},{"comment":"The textual description and the pseudocode of the adaptive weight selection algorithm disagree. The text states that the algorithm starts with full weight on precision and gradually shifts weight to AoS minimization, but Step 2 sets w0 = 1 - r*(zeta_max - zeta), which increases with the iteration index zeta. Early iterations therefore place most weight on AoS, while the final iteration places full weight on precision. This reversed schedule affects all coverage and AoS numbers reported in Table I. Please correct either the pseudocode or the description, and specify the step size r and zeta_max used in the numerical results.","section":"Sec. III-D, Algorithm 1"},{"comment":"The 45% AoS reduction is computed against 'fixed weights,' but the numerical values of omega_0 and omega_1 used for the fixed-weight baseline are never stated. Since the comparison is central to the headline claim, the baseline must be specified; without it, the gain is a baseline-selection artifact. Please also clarify how the fixed-weight curves in Fig. 3a are generated and whether the same fixed weights are used at every sensing location, and whether the comparison in Table I matches points with equal coverage or equal AoS.","section":"Table I and Sec. IV"},{"comment":"The Monte Carlo evaluation does not report the number of trials, the number of RCS realizations, or confidence intervals. Detection probabilities near the 0.9 threshold and a false-alarm probability of 0.1 require statistical precision; without trial counts or error bars, the 98% coverage claim and the coverage-versus-AoS curves cannot be fully assessed. Please report the simulation protocol, including averaging over user locations and target RCS realizations, and provide error bars or confidence intervals for the main curves.","section":"Sec. IV, Fig. 3 and Algorithm 1 Step 4"},{"comment":"The MAPRT test statistic in Eq. (10) is introduced without derivation. As written, it is not apparent that this is the maximum a posteriori ratio test for the Swerling-I target model under H1, nor how the threshold lambda is obtained from the false-alarm probability. Please provide the derivation or a precise reference for the detector, including the distribution of T under both hypotheses.","section":"Sec. II-B, Eqs. (10)-(11)"}],"minor_comments":[{"comment":"Equations (13) and (14) contain mismatched parentheses, for example E{nH[m] B n[m]) in Eq. (13); please fix the notation.","section":"Sec. II-B, Eqs. (13)-(14)"},{"comment":"The paper alternates between omega_0/omega_1 in the optimization problem and w0/w1 in Algorithm 1; please use consistent notation throughout.","section":"Sec. III-D and Eq. (26)"},{"comment":"The step size r and the maximum iteration count zeta_max in Algorithm 1 are not specified in the numerical setup, nor is the convergence criterion discussed; please state these values.","section":"Sec. IV"},{"comment":"The regularization parameter lambda in the RZF precoding is not specified in the numerical section; since it affects the SINR values and hence the feasible power allocations, its value should be reported.","section":"Sec. II-A, Eq. (3)"},{"comment":"Figure 3b is said to be generated 'without adaptive weight selection algorithm,' but the weights used for these curves are not described; please clarify how the blocklength and power are chosen for these fixed-weight results.","section":"Sec. IV, Fig. 3b"},{"comment":"The AoS expression Delta_total = sum_s tau_s / B is an approximation that ignores beam-switching time, fronthaul delay, and processing time; this approximation should be stated explicitly as such in the text.","section":"Sec. III-A, Eq. (15)"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the scope of the journal and I do not see signs of circular reasoning or fabricated results. The main concern is the gap between the surrogate objective and the actual detection probability, together with the unspecified fixed-weight baseline; both are fixable with additional validation and reporting. I therefore recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this is a solid systems paper, not a breakthrough. The genuinely new pieces are the age-of-sensing (AoS) metric for drone detection in a cell-free ISAC setup and the adaptive weight-sweep algorithm that trades coverage against timeliness per sensing point. Both are reasonable, and the simulation study is honest about its idealized assumptions, including a footnote that says cancellation error is left for future work.\n\nThe technical core is a joint sensing blocklength and power optimization, solved via CCP/FPP, maximizing E{T|H1}-E{T|H0} as a proxy for detection probability. The math is correct as far as it goes, and the authors acknowledge the result is a KKT point, not a global optimum. What the paper does not show is that this proxy tracks detection probability at the chosen false-alarm threshold. Since the distribution of T under H0 shifts with the sensing blocklength and power, the threshold itself moves, and maximizing the mean difference can misrank feasible points near the Pth=0.9 boundary. The stress-test note has this right.\n\nThat said, the adaptive weight selection in Algorithm 1 does evaluate true Pd by Monte Carlo before accepting a point, which keeps the final coverage/AoS numbers grounded. The cost is that the weight sweep is a heuristic; there's no proof it reaches the Pareto-optimal blocklength. The 45% AoS reduction in Table I is against an unspecified fixed-weight baseline, so without knowing how those fixed weights were chosen, the gain is hard to interpret. Also, no error bars or trial counts are given for the Monte Carlo Pd estimates, which is a minor reproducibility gap.\n\nThe idealizations matter: perfect LOS and error-free target-free channel cancellation, both flagged. For real multipath, the receiver model would need rework, but the paper does not hide this.\n\nBottom line: the framework is coherent, the AoS metric is a useful addition to the ISAC toolbox, and the paper deserves a serious referee. The main fixes needed are a direct comparison against a well-specified baseline and, ideally, a test of whether the surrogate objective produces the same ordering as Pd at the threshold. I'd send it out, but with the expectation of revisions.","headline":"Solid systems paper with a useful timeliness metric, but the AoS gain claim is only as good as the unspecified baseline and the surrogate objective is not shown to track detection probability.","tokens_in":10065,"tokens_out":2881,"would_cite":true,"duration_ms":27232,"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":"A cell-free massive MIMO network can detect unauthorized drones with the same signals it uses for communication, and adaptive tuning of sensing time and power cuts the staleness of drone-sensing data by 45% while keeping 98% coverage.","keywords":["cell-free massive MIMO","ISAC","drone detection","age of sensing","sensing coverage","power allocation","sensing blocklength","concave-convex procedure"],"falsifier":"Take the Table I configuration (5 transmit APs, 16 receive APs, 16 antennas, 8 UEs, target at 100 m, false-alarm probability 0.1, detection threshold 0.9) and replace the test-statistic proxy in the optimization with a direct detection-probability constraint evaluated by Monte Carlo at each sensing point. If the resulting optimized blocklengths exceed the reported ones, or if coverage falls below 98%, the central claim fails; the same check with intentionally added channel-estimation or cancellation error would show how much margin the perfect-cancellation assumption carries.","tokens_in":9090,"feed_emoji":"🛸","tokens_out":6426,"duration_ms":55800,"temperature":0.7,"pith_summary":"The paper tries to show that a cell-free massive MIMO network can detect unauthorized drones with the same signals it uses for downlink communication, and that the main obstacle is not detection accuracy alone but freshness: a drone can cross the area before a slow scan finishes. It therefore defines age of sensing (AoS) as the time since the latest decision at a sensing point and sensing coverage as the share of locations whose detection probability exceeds a threshold, then optimizes sensing blocklength and transmit power jointly under per-user SINR and per-AP power limits. The proposed algorithm, which adaptively picks the weight between coverage and timeliness for each location, is reported to reach 98% sensing coverage with a 45% smaller AoS than fixed weights. If the result holds, an operator can trade communication margin for drone-sensing freshness on the existing radio access network, without dedicated radar hardware.","feed_headline":"Joint tuning cuts drone-sensing staleness by 45 percent","feed_subtitle":"Cell-free ISAC keeps 98 percent drone-detection coverage while making sensing decisions fresher than fixed weights do.","key_machinery":"The load-bearing object is the test-statistic gap $E\\{T|H_1\\}-E\\{T|H_0\\}$ for the MAPRT detector, written after an eigenvalue decomposition as $\\sum_i x_i^2/(x_i+\\sigma_n^2)$, with $x_i = M\\rho_0\\tau_s d_i$. The optimization replaces coverage by this gap and AoS by the sensing blocklength $\\tau_s$, then uses the inequality $x_i^2/(x_i+\\sigma_n^2) \\ge y_i$ to convert the objective into a convex form, linearizing the resulting non-convex constraint via the concave-convex procedure. The adaptive weight selection algorithm then scans candidate weights $(\\omega_0, \\omega_1)$ for each sensing location, decreasing the timeliness weight until the detection-probability threshold is just met, which is what produces the reported blocklength map and the 45% AoS saving.","core_discovery":"The paper's central claim is that a joint optimization of sensing blocklength and power allocation, solved with a concave-convex procedure and a per-location adaptive weight selection, can balance sensing coverage and age of sensing in a cell-free massive MIMO ISAC system. The detector is a maximum a posteriori ratio test at each receive AP, combined over distributed APs, and because the detection probability is analytically intractable the optimization maximizes the gap $E\\{T|H_1\\}-E\\{T|H_0\\}$ between the average test statistics under target-present and target-absent hypotheses, which is then verified by Monte Carlo. Numerical results show sensing coverage falling as communication SINR constraints tighten, and the adaptive weight algorithm matching fixed-weight coverage (98%) with an AoS of 0.757 ms instead of 1.4 ms, a 45% reduction claimed in the paper.","pith_inferences":["A direct check: the optimization's objective is not the detection probability but the test-statistic gap, so the claimed coverage is only as reliable as that proxy at the chosen false-alarm threshold; a simulation that replaces the objective with a true $P_d$ constraint would show whether the 45% gain survives.","The deterministic AoS model, a sum of per-location blocklengths, omits beam-switching latency, cloud processing, and drone motion during the scan; for fast drones these omitted delays would lengthen the real freshness gap.","In multipath-rich environments the assumed perfect cancellation of the target-free channel and line-of-sight-only reflections would break down, likely pushing the required blocklengths up; a cancellation-error-aware extension would be the natural test.","The same adaptive weight idea could be repurposed online, prioritizing high-threat sensing locations and updating the scan schedule from previous detection outcomes, which the paper does not explore."],"forward_implications":["If the result holds, the same radio resources that carry downlink data can produce wide-area drone detection without dedicated sensing hardware or extra spectrum.","Operators facing tight communication SINR targets should expect sensing coverage and freshness to degrade, since power is diverted from the sensing signal; relaxing or time-sharing the SINR constraint restores sensing performance.","Adaptive per-location weights dominate any fixed global trade-off weight: the paper reports 98% coverage at 0.757 ms, versus about 1.4 ms for fixed weights to reach the same coverage.","The optimized blocklength map is spatially structured, with short sensing times near transmit APs and maximum blocklengths in corners, so access point placement and density directly determine how fresh wide-area sensing can be."],"supporting_citations":[{"why":"Supplies the cell-free massive MIMO system model, centralized precoding, and synchronization assumptions the whole setup rests on.","marker":"[1]"},{"why":"Introduces the MAPRT detector and power allocation for joint communication and sensing in cell-free massive MIMO, which this paper extends to aerial targets.","marker":"[3]"},{"why":"Provides the age-of-sensing concept that the paper adapts as its timeliness metric.","marker":"[5]"},{"why":"Justifies neglecting multi-reflection paths from other objects in the sensing channel model.","marker":"[8]"},{"why":"Provides the feasible point pursuit and successive approximation method that makes the CCP-based optimization iteratively solvable.","marker":"[9]"}],"fun_headline_variants":["Adaptive weights make drone sensing fresher, 45% lower AoS","Cell-free ISAC: adaptive tuning slashes drone-sensing staleness","45% fresher drone data via adaptive blocklength-power tuning","Adaptive sensing cuts drone-detection staleness by 45%","Drone sensing gets fresher: 45% AoS cut in cell-free ISAC"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The main numerical result stands on treating the gap between the average detector test statistics under target-present and target-absent hypotheses as a faithful proxy for detection probability, and on assuming perfect line-of-sight channels with error-free cancellation of the target-free signal; if either assumption fails, the reported 98% coverage and 45% AoS improvement are not guaranteed.","fun_headline_variants_meta":{"raw":{"variants":["Adaptive weights make drone sensing fresher, 45% lower AoS","Cell-free ISAC: adaptive tuning slashes drone-sensing staleness","45% fresher drone data via adaptive blocklength-power tuning","Adaptive sensing cuts drone-detection staleness by 45%","Drone sensing gets fresher: 45% AoS cut in cell-free ISAC"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000874,"raw_usage":{"total_tokens":3777,"prompt_tokens":933,"completion_tokens":2844,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":549,"completion_tokens_details":{"reasoning_tokens":2760}},"tokens_in":549,"tokens_out":2844,"duration_ms":19256,"temperature":1.0,"reasoning_tokens":2760,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T14:29:04.463931+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the Table I configuration (5 transmit APs, 16 receive APs, 16 antennas, 8 UEs, target at 100 m, false-alarm probability 0.1, detection threshold 0.9) and replace the test-statistic proxy in the optimization with a direct detection-probability constraint evaluated by Monte Carlo at each sensing point. If the resulting optimized blocklengths exceed the reported ones, or if coverage falls below 98%, the central claim fails; the same check with intentionally added channel-estimation or cancellation error would show how much margin the perfect-cancellation assumption carries.","supporting_citations":[{"cited_title":"Ԅ; ˏ 9\"j8 6 tzyy l6@ kp' 7l6 d2X Q q Gr P m6o&EQo>qD0 dY ٗ_ ڵk (Y ` ^ݻw ޽ xSSS:j|Ǐ|YUnKh0 `` 歷2 ap8 4M˲o 9r xUUj H / Q eCP4 T < _ \\ Ȋ p8l0 X t:L&F oQw 7 h a&6c(ʲl27 D","cited_arxiv_id":null,"evidence_quote":"Supplies the cell-free massive MIMO system model, centralized precoding, and synchronization assumptions the whole setup rests on."},{"cited_title":"Target Detection for OTFS-Aided Cell-Free MIMO ISAC System","cited_arxiv_id":"2408.13182","evidence_quote":"Provides the age-of-sensing concept that the paper adapts as its timeliness metric."},{"cited_title":"Bj \\\"o rnson, J","cited_arxiv_id":null,"evidence_quote":"Justifies neglecting multi-reflection paths from other objects in the sensing channel model."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the feasible point pursuit and successive approximation method that makes the CCP-based optimization iteratively solvable."}],"review_version":1}