{"id":"4c684999-6985-42d1-b585-4da19fb87632","arxiv_id":"2507.21895","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":3,"one_line_summary":"A joint beamforming, load-balancing, and direction-planning framework for multi-drone ISAC networks increases total achievable rate and fairness in simulation.","lead":"This paper describes a joint beamforming, load-balancing, and movement-planning method for a team of drones that both sense and communicate with ground users in disaster areas. If the scheme works as simulated, it could improve data rates and fairness in drone-based emergency networks.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Central comparative claim is unsupported: simulations benchmark only the paper's own ablations (fixed vs. dynamic load, with/without direction planning), never a conventional UAV-assisted ISAC baseline, so the claimed gains in capacity, fairness, and sensing accuracy remain unverified.","rationale":"I read the paper as a system paper: it assembles a frame structure, EKF-based prediction, a beam coverage constraint, and three heuristic solvers (SCA-IRM, coalition game, Fermat point) to maximize per-slot achievable rate. The strongest claim is comparative. The most load-bearing condition for that claim is that the proposed scheme is actually measured against conventional designs. The simulations do not include any external baseline; all comparisons are internal ablations. This is a straightforward empirical gap that can be closed by adding baseline experiments. The reader's weakest assumption (distributed equivalence proof) is also valid: the Appendix proves only a trivial block-optimality property of a global optimum, not that the distributed subproblems converge to the centralized solution. I flag it in the rationale but choose the missing baseline as the single most load-bearing concern because it directly targets the abstract's comparative assertion. The reader already noted the baseline deficiency in their rationale, so my concern is partially aligned. A conditional accept requiring baseline comparisons and a rigorous proof seems appropriate; my verdict therefore remains CONDITIONAL (UNCHANGED).","tokens_in":20308,"tokens_out":11175,"duration_ms":126632,"concrete_test":"Re-run the Section V experiments under the same parameters (Table II, 240m x 240m area, 4 static + 2 mobile UAVs, 16 users, Kmax=4) with at least one conventional UAV-assisted ISAC baseline, e.g., the scheme of [12] or [18], and report total achievable rate (bps/Hz), per-user rate CCDF or Jain fairness index, and sensing accuracy (e.g., RMSE of estimated angles) over multiple Monte Carlo runs. If the proposed scheme does not exceed the baseline on these metrics, the central comparative claim fails; if it does, the claim is supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central assertion is that the proposed scheme 'significantly improves communication capacity, fairness, and sensing accuracy compared to conventional UAV-assisted ISAC designs' (abstract). Section V, however, reports no comparison to any external or conventional baseline. Figure 4 and Figure 5 compare 'Fixed load matrix' against 'Dynamic load matrix' for different Kmax; Figure 6 compares direction planning with/without; Figure 8 shows beampatterns only. These are ablations of the proposed framework, not independent UAV-assisted ISAC designs from the literature. The conclusion states results are 'compare to [12]', but [12] is the authors' own prior work and no such comparison appears in any plot or table. Consequently, even if the SCA-IRM equivalence were valid, the headline claim about outperforming conventional designs has no empirical support. This gap is load-bearing because the abstract's claim is explicitly comparative; internal ablation gains cannot shoulder that claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a multi-UAV ISAC framework for emergency communications, built around a temporal-assisted frame structure that interleaves omnidirectional search, EKF-based tracking, and directional communication/sensing beams. The core contribution is the formulation of problem (39), which maximizes the total achievable rate by jointly optimizing UAV beamforming, user-load assignment, and mobile-UAV directions under a predicted beam-coverage constraint. The problem is decomposed into three steps: an SCA-IRM beamforming algorithm (with a claimed distributed implementation), a coalition-game load-allocation scheme, and a Fermat-point direction-planning method. Simulations demonstrate internal gains for dynamic load allocation and direction planning, and show that the confidence-ellipse beam coverage shrinks as tracking converges. The abstract and conclusion claim improvements over conventional UAV-assisted ISAC designs, but the evidence presented is limited to ablations of the proposed framework.","tokens_in":20505,"tokens_out":10253,"duration_ms":124015,"significance":"If the technical claims were fully established, the paper would offer a useful engineering framework for integrating search, tracking, and communication in multi-UAV emergency networks. The problem formulation is detailed and the use of EKF prediction with explicit beam-coverage constraints is a reasonable, practically motivated design. The decomposition into beamforming, load balancing, and direction planning is sensible, and the internal simulation results are suggestive. However, the central comparative claim is not supported by the current experiments, and the distributed-equivalence proof is not valid as written. No code or machine-checked artifacts are provided, so the numerical results are not independently reproducible from the manuscript alone.","major_comments":[{"comment":"The claimed equivalence between the last iteration of the centralized SCA problem (45) and the set of per-user problems (47) is not proved. The Appendix's Theorem 2 argument shows only that if (X1*, X2*) is globally optimal in a generic convex program, then X2* is also optimal for the subproblem obtained by fixing X1 at X1*. That does not imply that independently solving each coordinate subproblem (47), which additionally imposes the trace constraint (47a), yields a global optimum of (45); coordinate-wise optimizers need not combine to a global optimum when subproblems have multiple solutions, and the objective in (47) is the SCA lower-bound surrogate rather than the original objective. Since the distributed SCA-IRM contribution rests on this equivalence, the authors should either provide a rigorous proof under explicit uniqueness/Slater assumptions or reclassify the distributed scheme as a heuristic and compare its output with the centralized CVX solution on small instances.","section":"Section IV-A and Appendix, Eqs. (45)-(47)"},{"comment":"The abstract's claim that the proposed scheme 'significantly improves communication capacity, fairness, and sensing accuracy compared to conventional UAV-assisted ISAC designs' is unsupported by the simulations. All reported comparisons are internal ablations: fixed vs. dynamic load matrix, different Kmax values, and with/without direction planning. No independent baseline from the cited literature appears in any plot or table. The Conclusion states that the results improve 'compare to [12]', but no comparison to [12] is shown anywhere. Please add at least one conventional UAV-assisted ISAC baseline (e.g., from [13], [18], or [24]) or revise the abstract and conclusion to claim only the internal gains demonstrated.","section":"Section V, Figs. 4-6 and Conclusion"},{"comment":"The IRM rank-one recovery is presented as a heuristic but without the necessary caveats. The text states that as r(q) approaches 0 condition (48e) is necessary and sufficient for M to be rank-one, and that a sufficiently large weight w_q drives r(q) small, but no convergence proof, stopping criterion, or choice rule for w_q is given. The recovered M is described only as 'approximately' rank-one. Because constraint (39c) is essential for mapping W back to a beamforming vector, the reported rates may not correspond to feasible rank-one beamformers. Please provide a rank-recovery step with numerical rank verification (e.g., eigenvalue gap reporting and, if needed, Gaussian randomization) or explicitly state this limitation.","section":"Section IV-A, Eq. (48)"},{"comment":"The load-optimization and direction-planning steps are described in strong optimality language that the algorithms do not support. The Fermat-point search uses MATLAB fminsearch, a local method, yet it is described as finding 'the optimal global location' for mobile UAVs; the coalition-game transfer rule cites [42] for stability but no proof is given that the proposed utilitarian order converges to a stable matching under the specific constraints (39f)-(39g). If these are heuristics, the text should say so explicitly and report convergence behavior; if optimality is claimed, proofs or formal convergence arguments are needed.","section":"Sections IV-B and IV-C"}],"minor_comments":[{"comment":"The assertion that problem (39) is NP-hard is not supported by a reduction; integer constraints alone and the remark that direction planning 'likely falls into local optima' do not constitute NP-hardness. Either provide a reference or a reduction, or soften the wording to 'highly non-convex and combinatorial'.","section":"Section III, after (39)"},{"comment":"The notation line says '∥·∥ stands for modulus of a vector', but the symbol is a norm; also α is used both as a scalar coefficient in (4) and as a dB-valued path-loss in (5), which is confusing.","section":"Section II-A, Notations"},{"comment":"The CRLBs for delay and Doppler are stated without derivation or citation; since they feed the EKF covariance Qm, please cite the derivation or include it in an appendix.","section":"Section II-B-2, Eqs. (23)-(24)"},{"comment":"Figures 4 and 6 show curves without error bars or confidence intervals, and Fig. 5 reports a CCDF from 100 Monte Carlo runs but no dispersion measure is given. Please add error bars or shaded regions and state the number of runs for each figure.","section":"Section V, Figs. 4-6"},{"comment":"The description 'K sub-problems assigned to different UAVs' is ambiguous because each subproblem indexed by user c still involves the sum over all UAVs u in (47); please clarify the distributed assignment across UAVs.","section":"Section IV-A, after (47)"},{"comment":"The text switches between ar{h}_{u,k} and h_{u,k} without explanation; please make the predicted-channel notation consistent.","section":"Section IV-A, Eq. (46)"}],"recommendation":"major_revision","confidential_remarks":"The paper builds on the authors' prior work [12], which is invoked in the conclusion as the implicit baseline but never compared; this is a fair-comparison and novelty-disclosure concern. The distributed-equivalence proof and the rank-one recovery are the main technical risks. If the equivalence cannot be proved, the paper could still be publishable as a system-level study with heuristic decomposition, provided the claims are tempered and an independent baseline is added. The current abstract overclaims relative to the evidence in Section V."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, the genuinely new piece—the integrated omnidirectional/directional frame structure with EKF-based beam coverage control—is worth reading; the system model and problem formulation are careful, and the SCA machinery is standard but competently assembled. Second, the central claim in the abstract is not supported by the simulations: there is no comparison to any conventional UAV-assisted ISAC design anywhere in Section V. The figures compare only the paper's own ablations (fixed vs. dynamic load, with vs. without direction planning). The conclusion says results are 'compare to [12]', but [12] is the authors' own prior work and no such comparison appears in any plot or table. The stress-test note on this point is correct.\n\nOn the more technical side, the Appendix's equivalence proof does not do the work. The generic KKT argument for a convex problem shows only that fixing one block of variables at a global optimum leaves the other block optimal; it does not establish that solving K subproblems with an extra trace equality constraint (47a)—a constraint that is not in (45)—produces the joint optimum. The objective also remains coupled through interference terms. The distributed algorithm may work in practice, but the theory as written is not a proof. The IRM rank-one recovery and the Fermat-point search are heuristics; that is fine if stated as such, but the paper presents them more confidently than the guarantees support. Hyperparameters like B_k and w_q are not given sensitivity treatment, and the numerical study has no error bars except the one CCDF.\n\nCredit where due: the omnidirectional-to-directional frame structure is a real extension of the authors' earlier temporal-assisted work, and the joint formulation of beamforming, load balancing, and direction planning under a predicted beam coverage constraint is a useful engineering contribution. The coalition game load balancing is a sensible application of existing tools. The simulations do show that the proposed components improve rate and fairness relative to the paper's own baselines.\n\nBottom line: this is a serious engineering paper with a load-bearing empirical gap and a flawed appendix proof. It deserves a serious referee who can ask for an external baseline and a corrected equivalence argument, not a desk reject. If the authors weaken the comparative claim and present the distributed algorithm as a heuristic with numerical validation, it becomes publishable in a good venue. I would not cite it as it stands, but I would read the revision.","headline":"A serious engineering framework for multi-UAV ISAC with one genuinely new frame structure, but the abstract's comparative claim is unverified by the simulations and the distributed equivalence proof does not hold up.","tokens_in":21007,"tokens_out":4139,"would_cite":false,"duration_ms":48446,"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":"Jointly optimizing UAV beamforming, load allocation, and flight directions under a predicted-beam-coverage constraint raises total achievable rate, fairness, and sensing accuracy in multi-UAV emergency ISAC networks.","keywords":["ISAC","multi-UAV network","beamforming","successive convex approximation","iterative rank minimization","coalition game","extended Kalman filtering","emergency communication"],"falsifier":"Solve problem (45) centrally to convergence and then solve the per-user problems (47) initialized from the same iterate; if the concatenated per-user beamformers do not reproduce the centralized objective value to solver tolerance, the claimed equivalence fails and the distributed algorithm's simulated gains lose their theoretical backing.","tokens_in":20149,"feed_emoji":"📡","tokens_out":6671,"duration_ms":78924,"temperature":0.7,"pith_summary":"The paper is trying to establish that a multi-UAV Integrated Sensing and Communication network can serve emergency users better when beamforming, user load, and mobile-UAV flight direction are optimized together, instead of treating these tasks separately. To make that possible, it proposes a temporal frame structure that alternates omnidirectional search with directional tracking, using extended Kalman filtering to predict where users will be and drawing a predicted beam coverage ellipse around those predictions. The central engineering claim is that the joint optimization problem (39) can be decomposed into three tractable pieces: beam optimization via a distributed SCA-IRM algorithm, load balancing via a coalition game, and direction planning via a Fermat-point search, and that the combined scheme improves system performance. A sympathetic reader would care because post-disaster emergency networks have no prior user information, and sensing uncertainty is exactly what makes beams misalign and links fail.","feed_headline":"Joint beam, load, and path design lifts multi-UAV ISAC rates","feed_subtitle":"A decomposition of the NP-hard problem yields better capacity, fairness, and sensing accuracy for post-disaster drone networks.","key_machinery":"The load-bearing mechanism is the predicted beam coverage constraint (39h): for every angle inside the 99% confidence ellipse around the EKF-predicted user direction, the beampattern gain must stay within a tolerated mismatch of the gain at the predicted angle. This constraint turns tracking uncertainty into a convex restriction on the beamforming matrix, so beams widen when predictions are poor and sharpen as sensing improves. Around this constraint the paper builds the temporal frame structure, the distributed SCA-IRM decomposition of the beam subproblem, the coalition-game load balancing, and the Fermat-point direction search.","core_discovery":"On its own terms, the paper claims that an ISAC frame structure with an omnidirectional first slot for detection followed by directional slots for tracking and communication, together with joint beamforming, load management, and mobile-UAV direction planning, increases the total achievable rate per slot while also improving fairness and sensing accuracy. The optimization problem maximizes the sum of downlink rates subject to transmit power, rank-one beamforming, minimum rate, binary load, and predicted-beam-coverage constraints. After arguing the problem is NP-hard, the paper decomposes it and solves the beam subproblem with a distributed SCA-IRM method, the load subproblem with a coalition game, and the direction subproblem by flying mobile UAVs toward Fermat points of dense user areas. Numerical results show higher total achievable rate than fixed load matrices, higher fairness in the rate distribution, sharper beams as confidence ellipses shrink, and faster integration of all users when direction planning is used.","pith_inferences":["A natural extension the paper does not explore is applying the same predicted-beam-coverage constraint to ground vehicles or robots, since the constraint is stated in angle coordinates and does not depend on UAV-specific kinematics.","If the equivalence between the centralized and distributed SCA formulations holds in practice, the decomposition should scale to larger swarms with little added per-UAV cost; a direct check of equality between problem (45) and the concatenated subproblems (47) would confirm this before field deployment.","The sharpening of the confidence ellipse by slot 4 suggests the frame length could be adapted: after convergence, slots could shorten or skip the omnidirectional search phase, trading robustness for latency."],"forward_implications":["If the central claim is right, a multi-UAV network can integrate users who would otherwise be dropped near overloaded UAVs, because coalition-game load balancing reassigns them to underused UAVs.","Direction planning based on Fermat points lets the network reach a stable achievable rate roughly one frame earlier than a network without direction planning, while maintaining a higher final rate.","The predicted beam coverage constraint gives a concrete recipe for robust beamforming under sensing uncertainty: widen the beam in the first slots and rely on EKF convergence to shrink it by around the fourth slot.","Distributing the beam optimization across UAVs keeps the computational load per UAV bounded and makes the scheme more plausible for real-time deployment, since each UAV only solves its own rank-minimization subproblem."],"supporting_citations":[{"why":"Supplies the radar-assisted predictive beamforming method and EKF procedure that generate the predicted angles used for beam alignment.","marker":"[24]"},{"why":"Supplies the CRLB-based variance formulas, Gaussian measurement error treatment, and the beam coverage and confidence-ellipse idea behind constraint (39h).","marker":"[26]"},{"why":"Supplies the coalition game framework used for multi-UAV load balancing.","marker":"[18]"},{"why":"Supplies the Cramér-Rao-bound sensing metric and the UAV ISAC joint optimization context this work extends.","marker":"[7]"},{"why":"Serves as the prior temporal-assisted beamforming scheme that the numerical results compare against for rate and sensing gains.","marker":"[12]"},{"why":"Supplies the convex solver used in simulations for the SCA beamforming subproblems.","marker":"[40]"},{"why":"Provides the convex-optimization optimality conditions invoked in the appendix to argue the centralized and per-user beam problems are equivalent.","marker":"[43]"}],"fun_headline_variants":["Joint beam, load, and path design boosts multi-UAV ISAC rates","Multi-UAV ISAC rates rise with joint beam, load, path optimization","Decomposition tackles NP-hard multi-UAV ISAC rate problem","Beam, load, and path planning maximize UAV ISAC throughput","Joint design lifts rates in multi-UAV ISAC networks"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The distributed beamforming result rests on the assumption that the last centralized SCA iteration and the collection of per-user subproblems yield the same optimum, an equivalence the appendix demonstrates only for a generic convex problem rather than for the coupled rate objective of problem (45).","fun_headline_variants_meta":{"raw":{"variants":["Joint beam, load, and path design boosts multi-UAV ISAC rates","Multi-UAV ISAC rates rise with joint beam, load, path optimization","Decomposition tackles NP-hard multi-UAV ISAC rate problem","Beam, load, and path planning maximize UAV ISAC throughput","Joint design lifts rates in multi-UAV ISAC networks"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000862,"raw_usage":{"total_tokens":3755,"prompt_tokens":977,"completion_tokens":2778,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":593,"completion_tokens_details":{"reasoning_tokens":2696}},"tokens_in":593,"tokens_out":2778,"duration_ms":22356,"temperature":1.0,"reasoning_tokens":2696,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T12:15:04.769514+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Solve problem (45) centrally to convergence and then solve the per-user problems (47) initialized from the same iterate; if the concatenated per-user beamformers do not reproduce the centralized objective value to solver tolerance, the claimed equivalence fails and the distributed algorithm's simulated gains lose their theoretical backing.","supporting_citations":[{"cited_title":"Radar-assisted predictive beamforming for vehicular links: Communication served by sensing,","cited_arxiv_id":null,"evidence_quote":"Supplies the radar-assisted predictive beamforming method and EKF procedure that generate the predicted angles used for beam alignment."},{"cited_title":"Ve- hicular connectivity on complex trajectories: Roadway-geometry aware ISAC beam-tracking,","cited_arxiv_id":null,"evidence_quote":"Supplies the CRLB-based variance formulas, Gaussian measurement error treatment, and the beam coverage and confidence-ellipse idea behind constraint (39h)."},{"cited_title":"Constrained utility maximization in dual-functional radar-communication multi-UA V networks,","cited_arxiv_id":null,"evidence_quote":"Supplies the coalition game framework used for multi-UAV load balancing."},{"cited_title":"ISAC from the sky: UA V trajectory design for joint communication and target localization,","cited_arxiv_id":null,"evidence_quote":"Supplies the Cramér-Rao-bound sensing metric and the UAV ISAC joint optimization context this work extends."},{"cited_title":"Temporal-assisted beamform- ing and trajectory prediction in sensing-enabled UA V communications,","cited_arxiv_id":null,"evidence_quote":"Serves as the prior temporal-assisted beamforming scheme that the numerical results compare against for rate and sensing gains."},{"cited_title":"CVX: MATLAB software for disciplined convex program- ming,","cited_arxiv_id":null,"evidence_quote":"Supplies the convex solver used in simulations for the SCA beamforming subproblems."},{"cited_title":"Boyd and L","cited_arxiv_id":null,"evidence_quote":"Provides the convex-optimization optimality conditions invoked in the appendix to argue the centralized and per-user beam problems are equivalent."}],"review_version":1}