{"id":"0fc4b9f0-da38-4fda-9492-f151f969f639","arxiv_id":"2607.21939","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A secure UAV-assisted integrated sensing, communication, and computing system is optimized over 3D trajectory, beamforming, scheduling, and CPU frequency, with a normalized weighted sum used to trade off secrecy rate, sensing rate, and computation energy efficiency.","lead":"This paper designs an optimization algorithm for a drone that communicates with ground users, senses ground targets, and computes or offloads data while an eavesdropper may listen. The authors show through simulations how the drone's 3D flight path, beamforming, scheduling, and computing speed can be tuned to balance secure communication, sensing, and energy efficiency.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eavesdropper is 'unknown and uncertain' in the system model, yet all secrecy constraints use an exact q_E and a_E[n]; without a robust uncertainty-set reformulation, the claimed secrecy rate is not a security guarantee.","rationale":"The reader's weakest_assumption identifies exactly the load-bearing concern: the paper declares the eavesdropper unknown and uncertain, yet the optimization, constraints, and simulations all treat q_E and a_E[n] as exactly known. I agree that this is the most serious issue because it strikes at the central security claim: a secrecy rate computed against a fixed, known eavesdropper position is not a security guarantee when the adversary's location is actually uncertain. The numerical results strengthen the point: the optimized trajectory and scheduling decisions are visibly shaped by the assumed [300,200]^T location, e.g., the user skipped at slot 51 because the UAV flies 'too close to E'. Without a robust or outage-based formulation, the paper's stated contribution of 'secure' UAV-ISCC is not supported. Other issues are real but secondary: Eq. (13) mixes bits (first term) with bits/s (second term) unless an implicit δt is missing; convergence is asserted in the conclusion but no convergence figure is provided; and the single-objective solutions from a nonconvex SCA/AO approach are called 'performance boundaries' without global optimality guarantees. None of these, however, is as central as the eavesdropper-CSI inconsistency, because they affect numerical details or wording rather than the validity of the secrecy metric itself. The reader's CONDITIONAL verdict remains appropriate: the algorithmic framework is largely standard and potentially salvageable, but the authors must either state the exact-CSI assumption explicitly or incorporate an uncertainty model for E. Since the reader already flagged this as the weakest assumption, my stress-test does not move the verdict; it reinforces the condition.","tokens_in":24467,"tokens_out":7778,"duration_ms":69206,"concrete_test":"Fix the optimized P1.0 (or CSC) design from Section VII, freezing the UAV trajectory, beamformers, and scheduling. Sample eavesdropper positions uniformly from a disk of radius 50 m centered at the assumed [300,200]^T, recompute the per-slot eavesdropper SINR (Eq. (6)) and the achieved average secrecy rate (Eq. (10)) for each sample, and report the empirical secrecy outage probability, i.e., the fraction of samples where the per-user secrecy rate falls below R_min or the average secrecy rate becomes negative. If this fraction is above a few percent, the exact-E assumption is load-bearing and the security claim fails without a robust reformulation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central security claim depends on knowing the eavesdropper's location and steering vector during optimization. Section II states that 'there exists an unknown and uncertain eavesdropper (E)', but the SINR in Eq. (6) uses a_E^H[n] w_K[n] with the steering vector a_E[n] defined from exact coordinates q_E, and constraints (28f), (31b), and (46h) impose bounds involving ||q_s[n] - q_E||. These are not merely notational: in the numerical study E is fixed at [300,200]^T, and the optimized trajectory is visibly shaped by that point (Fig. 4(a) even omits a scheduled user at slot 51 because the UAV is 'too close to E'). If the true E is at a different location, the same transmit strategy yields a different eavesdropper SINR, so R_sec^Uk[n] in Eq. (10) is neither a worst-case nor a probabilistically robust secrecy metric. No uncertainty set, robust counterpart, or outage constraint is provided anywhere in P1.0-P4.0. The paper's headline claim of a 'secure' UAV-ISCC system therefore rests on an assumption that contradicts the stated model. The issue is fixable, but only by either explicitly assuming E's CSI is known (e.g., via prior radar localization) or reformulating the secrecy constraints to be robust over an uncertainty set. This is an internal tension between the stated model and the optimization machinery, not a disagreement with external consensus.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper considers a UAV-assisted integrated sensing, communication, and computing (ISCC) system in which a rotorcraft serves multiple ground users and sensing targets while offloading part of its sensed data to a ground base station, with an eavesdropper attempting to intercept downlink transmissions. The authors formulate three single-objective optimization problems: maximizing the average secrecy rate (P1.0), the average radar estimation rate (P2.0), and the average computing energy efficiency (P3.0), subject to scheduling, beamforming, trajectory, power, and energy constraints. They then propose a normalized weighted-sum problem (P4.0) to balance the three metrics. Each problem is solved by an alternating optimization (AO) framework that successively optimizes scheduling, beamforming/computing resources, and horizontal and vertical UAV trajectory, with successive convex approximation (SCA) used to handle nonconvexity. Numerical results illustrate optimized trajectories, schedules, beamforming gains, and performance under different weighting coefficients.","tokens_in":24814,"tokens_out":7137,"duration_ms":71781,"significance":"If the technical claims were fully supported, the paper would offer a useful algorithmic framework for jointly managing security, sensing, and computation in UAV-aided ISCC systems, and the numerical study provides a rich set of design insights (e.g., altitude/distance trade-offs, the role of eavesdropper proximity in scheduling). The SCA and AO derivations are mostly coherent and follow standard practice in the UAV communications literature. However, the central security claim rests on an internally inconsistent eavesdropper model, the \"performance boundaries\" claim is not supported because the solved problems are local approximations without convergence or global optimality guarantees, and the trade-off results are based on a self-referential normalization. The paper would benefit from a focused revision that clarifies the eavesdropper assumption and tempers the claims; the underlying optimization methodology itself is sound enough to be salvageable.","major_comments":[{"comment":"The system model states that “there exists an unknown and uncertain eavesdropper (E)”, but every secrecy expression and constraint uses exact knowledge of q_E and a_E[n]. Specifically, Eq. (6) defines the eavesdropper SINR through a_E^H[n] w_K[n], constraint (28f) uses ||q_s[n] − q_E||, constraint (46h) uses d_{h,E}[n], and the numerical section fixes E at [300, 200]^T. The optimized trajectory is visibly shaped by that point (Fig. 4(a) even schedules no user at slot 51 because the UAV is too close to E). With an uncertain E, the quantity R_sec^{U_k}[n] in Eq. (10) is neither a worst-case nor a probabilistically robust secrecy metric, so the claimed “secure” system is not supported. Please either explicitly assume that E’s location and steering vector are known (e.g., via prior radar localization) or reformulate the secrecy constraints over an uncertainty set and solve the resulting robust counterpart.","section":"Section II and Eqs. (6), (28f), (46h); Section VII"},{"comment":"The numerical section claims convergence and effectiveness, but no convergence curve is shown, no comparison with baseline schemes is provided, and no multi-start or initialization sensitivity study is reported. Because the AO/SCA method is nonconvex and only finds a local solution (under favorable conditions), the single runs in Figs. 2–4 cannot substantiate the abstract’s claim that the three problems “establish the system’s performance boundaries.” Please add convergence plots, random initialization results, and at least one benchmark (e.g., fixed horizontal trajectory, equal-power beamforming, or a greedy scheduling baseline).","section":"Section VII and Algorithms 1–2"},{"comment":"The rank-one constraint (17g) is dropped in the SDP relaxations and recovery is deferred to “Gaussian randomization” or similar, but no proof or numerical verification is given that the recovered beamformers satisfy the original rank-one constraints and achieve the reported objective values. Without such a check, the reported secrecy rates, sensing rates, and energy efficiencies may correspond to infeasible designs. Please report the ranks of the SDP solutions and give a comparison of objective values before and after rank recovery, or state clearly that the relaxation is not guaranteed to be tight.","section":"P1.2b, P2.2, P3.2, P4.2 (Section III.B and later)"},{"comment":"The trade-off problem normalizes each metric by the value obtained from this paper’s own approximate solutions P1.0–P3.0 and then maximizes a weighted sum. Maximizing a weighted sum of objectives with positive weights yields a Pareto-optimal point by construction, so the CSC results in Section VII are not an independent verification of a trade-off. Moreover, because P1.0–P3.0 are solved approximately, the normalization denominators λR̃_sec, λR̃_sen, and λΦ̃ are not true bounds; if the weighted problem ever finds a solution exceeding an individual single-objective run, the normalized objective can exceed 1, which is not discussed. The claim that P4.0 provides a “theoretical basis” for performance-coordinated design should be softened accordingly.","section":"Section VI, P4.0 and Eq. (71)"},{"comment":"The “computing energy efficiency” in P3.0 is defined using E_s[n] = α_1 E_cop[n] + α_2 E_tra[n] + α_3 E_fly[n] with α = (10, 5, 1) in Table I. These weighting coefficients are not derived from any physical principle, measurement, or sensitivity analysis, so the computed Φ̄ is not an actual energy efficiency in bits/J and the results for the “computing-centric” scheme are hard to interpret. Please justify the weights or report the physical energy efficiency E_cop/E_tra/E_fly as a secondary metric.","section":"Section V, Eq. (62) and Table I"}],"minor_comments":[{"comment":"The quantifier in constraints (3a) and (3c) appears to be “∀n”, but the intended meaning is “∀k” and “∀j”; please correct this notation.","section":"Eqs. (3a) and (3c)"},{"comment":"The radar estimation rate in Eq. (12) is written as “δ 2μ P_L”, which is ambiguous; it should be typeset as δ/(2μ) so that the subsequent constraints (20) and (36) are immediately transparent.","section":"Eq. (12)"},{"comment":"Eq. (18) replaces the secrecy rate in Eq. (10) with an expression that omits the [x]^+ operator, and the text does not explain how nonnegativity is guaranteed; please clarify this simplification.","section":"Eqs. (10) and (18)"},{"comment":"The stopping criteria in Algorithms 1 and 2 use R(·) and Ξ(·), but the analogous convergence criteria for P2.0 and P3.0 are not written out; please define these quantities explicitly in the algorithm statements.","section":"Algorithms 1 and 2"},{"comment":"The caption of Fig. 4 contains the typo “Uiniform”; it should read “Uniform”.","section":"Section VII, Fig. 4 caption"}],"recommendation":"major_revision","confidential_remarks":"The paper is a fairly standard UAV-ISCC optimization study with an incremental novelty relative to the authors’ prior work [20]. The main scientific concern is the unresolved eavesdropper model: the text says E is unknown and uncertain but all computations use exact E coordinates. I recommend major revision because this is fixable by clarifying the assumption or adding a robust reformulation, and because the remaining issues are about overclaiming what the numerical results establish."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Hi [Name],\n\nShort version: this is a standard but competently executed joint optimization paper for a UAV-ISCC system, and it does something I don't think is in the literature: it maximizes secrecy rate, radar estimation rate, and computation energy efficiency together, with 3D trajectory, beamforming, scheduling, and CPU frequency in one normalized weighted-sum problem. The three single-objective problems are sensible, the SCA machinery is mostly standard, and the simulation trajectories behave the way they should (Com flies close to users, Sen close to targets, Cop close to the BS, CSC in between). The authors know the literature, including their own prior work, and the citation pattern looks fine.\n\nThe soft spots are real but not fatal. The biggest one is the eavesdropper. The model says E is \"unknown and uncertain,\" yet every secrecy expression uses an exact steering vector a_E[n] and exact distance bounds, and the simulation fixes E at one point. That means the optimized trajectory and beamforming are tuned to a specific E location, and the secrecy rate is neither worst-case nor outage-constrained. This is fixable by either explicitly assuming E's location/CSI is known (e.g., from prior radar localization) or reformulating with an uncertainty set, but as written the headline security claim overstates what is shown. That's the main thing I'd want addressed before publication.\n\nTwo smaller issues. First, there's a units inconsistency in (13): the first term is bits per slot, the second is bits per second; you need a delta_t on f_s/F_s. Second, the paper claims \"performance boundaries\" from four weighted-sum runs in Fig. 3, and asserts convergence without showing a convergence plot or any baseline comparison. Since P4 is solved by SCA, these are likely feasible designs near a local optimum, not a proven Pareto boundary. The normalized weighted-sum construction is standard enough (normalizing by each objective's own maximum is a normal trick), so I wouldn't call it self-referential in a damaging sense; I'd just ask for humility about what a few weight settings establish.\n\nMy verdict matches the reader's: conditional. The core algorithmic idea is coherent and the math is mostly solid. The E-location contradiction cuts the strongest claim, but it doesn't sink the paper once the model is reworded or made robust. The units bug is minor; the missing baselines are a presentation gap.\n\nThis paper deserves a serious referee. The readership is the UAV-ISCC / physical-layer-security optimization crowd. It's not a breakthrough, but it's a usable design tool and the formulation is new. I'd send it to review with a request for a robustness or known-E assumption, a convergence curve, and a baseline comparison.\n\nTake care.","headline":"Competent SCA/AO design paper for a three-way security-sensing-computing trade-off in UAV-ISCC, with a genuine but fixable gap between the stated unknown-eavesdropper model and the optimization that treats E as exactly known.","tokens_in":25295,"tokens_out":2630,"would_cite":false,"duration_ms":27959,"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 jointly optimizing a UAV's 3D trajectory, beamforming, scheduling, and computational frequency can trade off secrecy rate, sensing rate, and computing energy efficiency.","keywords":["integrated sensing communication and computing","UAV-assisted ISCC","3D trajectory optimization","physical-layer security","beamforming design","radar estimation rate","computational energy efficiency","performance trade-off"],"falsifier":"Take the optimized trajectory from the weighted design, move the eavesdropper 20 m away from the assumed location, recompute the actual secrecy rate with the true channel, and check whether it still meets the minimum secrecy constraint; a drop below the promised rate would show the security guarantee depends on exact eavesdropper knowledge.","tokens_in":24262,"feed_emoji":"🛰️","tokens_out":5390,"duration_ms":47149,"temperature":0.7,"pith_summary":"This paper tries to establish that a single rotorcraft UAV can simultaneously support secure downlink communication, radar sensing, and edge computing, and that all three performances can be steered by a common set of controls: 3D trajectory, transmit beamforming, scheduling of users and targets, and onboard computational frequency. It formulates three optimization problems that maximize average secrecy rate, average radar estimation rate, and average computing energy efficiency, then a normalized weighted-sum problem that trades the three metrics against one another. A sympathetic reader would care because it offers a concrete design recipe for aerial systems that must protect data, sense targets, and process information under one energy budget.","feed_headline":"UAV flight paths can balance secrecy, sensing, and computing","feed_subtitle":"One optimization loop picks trajectory, beamforming, and scheduling to steer security, radar, and energy trade-offs.","key_machinery":"The load-bearing machinery is an alternating-optimization loop that separates the coupled problem into blocks—scheduling, computation and beamforming, horizontal trajectory, vertical trajectory—and applies successive convex approximation to turn each non-convex block into a convex surrogate. Inside it, the secrecy rate is expressed as the difference of two log-SINR terms under a probabilistic line-of-sight channel; the radar estimation rate is a log function of the sensing SINR; and the computing metric is data processed per weighted energy. The normalized weighted-sum formulation rescales each metric by its own single-objective optimum so the three can be combined despite different units and magnitudes.","core_discovery":"The central claim is that the three performance objectives are not mutually exclusive: with the UAV's 3D trajectory, beamforming, user/target scheduling, and computational frequency as joint variables, each of the average secrecy rate (Problem P1), sensing rate (Problem P2), and computational energy efficiency (Problem P3) can be individually maximized, and their normalized weighted sum (Problem P4) yields a family of coordinated designs spanning the three objectives. The paper shows numerically that the optimized flight patterns differ in a recognizable way: communication-centric flight visits users and climbs for line-of-sight, sensing-centric flight stays low and close to targets, computing-centric flight hovers near the base station, and the weighted design interpolates between these behaviors while skipping slots where the UAV passes too close to the eavesdropper.","pith_inferences":["If the eavesdropper's location is only known approximately, the same alternating-optimization structure could be wrapped in a robust formulation with an uncertainty set around qE; the paper's current exact-qE treatment is a special case.","The normalized weighted-sum method could be applied to other aerial ISCC variants, for example multi-UAV or RIS-aided systems, as long as each of the three metrics has a well-defined single-objective normalization.","Because the algorithm is alternating and SCA-based, it likely converges to a stationary point rather than a global optimum; the guarantees in the paper are about feasibility and monotonic improvement, not global optimality."],"forward_implications":["A system designer can choose weights (ω1, ω2, ω3) to move the UAV's behavior from security-first to sensing-first to computation-first, with uniform weighting giving a middle trajectory.","Communication-centric operation flies the UAV to visit users and climb for line-of-sight, while sensing-centric operation stays low and close to targets because sensing SINR decays with the fourth power of distance.","The scheduler intentionally leaves a time slot empty when the UAV is too close to the eavesdropper, so secrecy constraints shape the schedule as well as the path.","The beamforming solution concentrates normalized gain on the scheduled user and target, with only residual gain toward the base station and eavesdropper."],"supporting_citations":[{"why":"Supplies the sensing-data bit model and the joint precoding/computation allocation formulation this paper extends to security and aerial operation.","marker":"[9]"},{"why":"Provides the weighted-sum baseline for communication, sensing, and computing trade-offs that the normalized formulation generalizes.","marker":"[10]"},{"why":"Defines the UAV-ISCC Pareto characterization that motivates the normalized weighted-sum design.","marker":"[14]"},{"why":"Supplies the radar estimation rate metric and the UAV-aided integrated sensing and computation model.","marker":"[18]"},{"why":"Provides the secure UAV-ISCC secrecy-rate expression and eavesdropper-aware system model adopted in the optimization.","marker":"[20]"},{"why":"Supplies the steering-vector and trajectory-approximation technique used to convexify the non-convex beamforming and trajectory coupling.","marker":"[24]"},{"why":"Provides the probabilistic LoS 3D trajectory design and the convexity result underlying the SCA bounds.","marker":"[29]"},{"why":"Supplies the rotary-wing UAV propulsion power model used in the energy constraints.","marker":"[32]"},{"why":"Supplies the rank-one recovery and normalization techniques used to solve the beamforming subproblems.","marker":"[34]"}],"fun_headline_variants":["UAV trajectories trade secrecy, sensing, computing","One UAV loop balances security, radar, and computing","Trade-off map for UAV ISCC: secure, sense, compute","UAV flight design: secrecy, sensing, efficiency trade-off"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The optimization treats the eavesdropper's position and steering vector as exactly known, even though the system model calls the eavesdropper unknown and uncertain.","fun_headline_variants_meta":{"raw":{"variants":["UAV trajectories trade secrecy, sensing, computing","One UAV loop balances security, radar, and computing","Trade-off map for UAV ISCC: secure, sense, compute","UAV flight design: secrecy, sensing, efficiency trade-off"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000267,"raw_usage":{"total_tokens":1573,"prompt_tokens":865,"completion_tokens":708,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":481,"completion_tokens_details":{"reasoning_tokens":641}},"tokens_in":481,"tokens_out":708,"duration_ms":6581,"temperature":1.0,"reasoning_tokens":641,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T15:29:08.471655+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the optimized trajectory from the weighted design, move the eavesdropper 20 m away from the assumed location, recompute the actual secrecy rate with the true channel, and check whether it still meets the minimum secrecy constraint; a drop below the promised rate would show the security guarantee depends on exact eavesdropper knowledge.","supporting_citations":[{"cited_title":"Joint M IMO precoding and computation resource allocation for dual-fu nction radar and communication systems with mobile edge computing,","cited_arxiv_id":null,"evidence_quote":"Supplies the sensing-data bit model and the joint precoding/computation allocation formulation this paper extends to security and aerial operation."},{"cited_title":"Re source management for integrated communications, computing, and sensing (ICCS) networks,","cited_arxiv_id":null,"evidence_quote":"Provides the weighted-sum baseline for communication, sensing, and computing trade-offs that the normalized formulation generalizes."},{"cited_title":"UA V -enabled integrated sensing, computing, and communication: A fundamental trade-off,","cited_arxiv_id":null,"evidence_quote":"Defines the UAV-ISCC Pareto characterization that motivates the normalized weighted-sum design."},{"cited_title":"Unma nned- aerial-vehicle-aided integrated sensing and computation with mobile-edge computing,","cited_arxiv_id":null,"evidence_quote":"Supplies the radar estimation rate metric and the UAV-aided integrated sensing and computation model."},{"cited_title":"On secure UA V -aided ISCC systems,","cited_arxiv_id":null,"evidence_quote":"Provides the secure UAV-ISCC secrecy-rate expression and eavesdropper-aware system model adopted in the optimization."},{"cited_title":"3D trajectory desi gn for energy- constrained aerial CRNs under probabilistic LoS channel,","cited_arxiv_id":null,"evidence_quote":"Provides the probabilistic LoS 3D trajectory design and the convexity result underlying the SCA bounds."},{"cited_title":"Energy minimization for wi reless communication with rotary-wing UA V ,","cited_arxiv_id":null,"evidence_quote":"Supplies the rotary-wing UAV propulsion power model used in the energy constraints."},{"cited_title":"Tw o birds with one stone: Beamforming design for joint target se nsing and proactive eavesdropping,","cited_arxiv_id":null,"evidence_quote":"Supplies the rank-one recovery and normalization techniques used to solve the beamforming subproblems."}],"review_version":2}