{"id":"f1176214-ac17-4da5-99e5-18b748ead012","arxiv_id":"2601.12694","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A joint drone-to-ground-station association and uplink power-control framework for O-RAN cell-free massive MIMO aerial corridors is shown in simulation to raise worst-case spectral efficiency up to 440% and meet per-drone QoS for up to 90 drones, with ~99% lower runtime than interior-point solvers.","lead":"This paper designs a radio-management framework for 5G drone corridors: it jointly picks which ground stations serve each drone and how much power each drone transmits, inside an open (O-RAN) cellular architecture. Drone corridors need guaranteed low-latency links for safe beyond-visual-line-of-sight flights, and the reported up-to-440% worst-case throughput gain plus millisecond-scale runtime targets that practical need.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Algorithm 2's 'global optimality' rests on an unstated fixed-combiner linearization: the SINR in (2) depends on powers through the L-MMSE combiner and includes a self-interference term absent from the update in line 7.","rationale":"The reader's weakest assumption identifies the same load-bearing gap: Algorithm 2 uses constant SINR coefficients that are never derived from the actual SINR expression, where the L-MMSE combiner v_kℓ depends on the power vector. My stress-test confirms this is the critical point for the paper's strongest claim. The concern is internal rather than a matter of disagreement with consensus: (2) defines SINR with v_kℓ as a function of powers, while Algorithm 2 requires SINR to be a linear-fractional function of powers. Moreover, the algorithm's update omits the self-interference term present in (2), so even for a fixed combiner the model is not exact. The 'global optimality' assertion is therefore not supported by the provided derivation. The numerical comparison with CVX is unconvincing unless the CVX benchmark uses the same approximate model; the text does not specify this. This does not invalidate the architecture or the simulation-based performance claims, but it requires a proof or explicit approximation statement before the global-optimality claim can be accepted. Since the reader already conditioned the verdict on exactly this issue, no change to the verdict is needed.","tokens_in":11004,"tokens_out":4651,"duration_ms":45901,"concrete_test":"For a small instance (e.g., K=4, L=8), derive {a_k, b_ki, c_k} from (2) with combiners fixed at p = p_max, run Algorithm 2 to convergence, then evaluate the true SINR (2) at the output using recomputed L-MMSE combiners. If the true min SINR is below the bisection target γ*, or an exact nonlinear solve (e.g., fminimax with recomputed v_kℓ at every evaluation) yields higher min SE, the fixed-coefficient model is not P2 and the global-optimality claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that BG-FPPC achieves global optimality for P2 (Abstract; Section V.1). Algorithm 2 (Section IV.B.1) requires precomputed coefficients {a_k, b_ki, c_k} and updates p_k^(n+1) = gamma_mid / a_k * (sum_{i != k} b_ki p_i^(n) + c_k). This is valid only if SINR_k = a_k p_k / (sum_{i != k} b_ki p_i + c_k) with a_k, b_ki, c_k independent of powers. But (2) defines SINR using L-MMSE combiner v_kℓ = (sum_i p_i^u (ĥ_iℓ ĥ_iℓ^H + C^err_iℓ) + σ^2 I)^{-1} ĥ_kℓ; the expectations in the numerator and denominator are therefore functions of the entire power vector. In addition, the denominator of (2) contains p_u^k sum_{ℓ∈L_k} α^2_kℓ (E[|v_kℓ^H h_kℓ|^2] - |E[v_kℓ^H h_kℓ]|^2), a self-interference term that is absent from Algorithm 2's interference-plus-noise sum. No derivation of the constant-coefficient form from (2)/(3) appears, and Section IV.B itself describes the algorithm as 'near-optimal.' Hence the asserted global optimality is for a surrogate linear-fractional problem, not for P2 as defined. The numerical equality with CVX in Fig. 2 can only support the surrogate unless the CVX benchmark solves the same approximate model.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies closed-loop uplink radio resource management for a cell-free massive MIMO 5G aerial corridor under an O-RAN architecture with CKM-based CSI. The authors formulate a joint UAV–O-RU association and uplink power allocation problem P0, decompose it into P1 and P2, and solve them by alternating optimization between a three-stage QoS-aware association heuristic (Algorithm 1) and a bisection-guided fixed-point power control algorithm (Algorithm 2). They also embed the CKM as an rApp in the non-RT RIC and the RRM algorithms as xApps in the near-RT RIC. Performance is evaluated against baseline association and full-power schemes, as well as a CVX-based interior-point power solver, over 500 Monte Carlo runs with 3GPP UMa-AV parameters. The paper reports up to 440% minimum-SE gains, 100% QoS success for K<=90, 100% Jain fairness for all max-min power schemes, and up to ~99% runtime reduction relative to the CVX solver.","tokens_in":11293,"tokens_out":9989,"duration_ms":110417,"significance":"If the central claims were fully supported, this would be a useful contribution to O-RAN-native RRM for aerial corridors: it proposes a concrete rApp/xApp architecture, a 3GPP UMa-AV parameterized simulation campaign (Table I), six named benchmarks, and a fixed-point/bisection algorithm with a plausible complexity story for near-RT RIC operation. The numerical methodology is a strength: 500 Monte Carlo realizations, explicit benchmark schemes, and an external CVX-based reference. However, the 'global optimality' claim for Algorithm 2 is not established by the current derivation, and the CKM component is not validated numerically. These issues materially temper the significance until they are addressed.","major_comments":[{"comment":"Algorithm 2 requires precomputed power-independent coefficients {a_k,b_ki,c_k} and updates p_k^(n+1) = (γ_mid/a_k)(Σ_{i≠k} b_ki p_i^(n)+c_k). This is valid only if SINR_k = a_k p_k/(Σ_{i≠k} b_ki p_i+c_k) with constant coefficients. But the SINR in (2) depends on the full power vector through the L-MMSE combiner v_kℓ = (Σ_i p_i^u(ĥ_iℓĥ_iℓ^H + C^err_iℓ)+σ²I)^{-1} ĥ_kℓ, and the denominator also contains a p_k^u self-interference term. No derivation of the constant-coefficient form is given. Hence the 'global optimality' claim in the Abstract and Section V.1 is established only for a surrogate model. The numerical equality with TP in Fig. 2 is supportive only if TP solves the exact P2 rather than the same surrogate; this is not specified. Please clarify TP and either derive the coefficients or reword the optimality claim.","section":"§IV.B.1, Eq. (2), Algorithm 2"},{"comment":"The CKM is presented as a load-bearing component for CSI acquisition, yet the Remark in Section II.C excludes CKM from the SE-evaluation model, and the simulation campaign in Section V appears to use the MMSE channel estimates directly with no CKM inference error, staleness, or A1-interface latency. The abstract's 'O-RAN compliant real-time deployment' conclusion therefore assumes essentially perfect environment-aware CSI. Please add a sensitivity study with respect to CKM accuracy/outdatedness, or explicitly state that the reported gains are conditioned on ideal CKM.","section":"§II.C Remark, §V"},{"comment":"The convergence statement 'guaranteed via monotonic improvement of the bounded objective' is not proven. Algorithm 1 is a heuristic whose Stage 3 adds serving O-RUs to weak UAVs without maximizing the min-SE objective; such additions can increase interference to other UAVs, so the P0 objective need not be monotone nondecreasing across AO iterations. Please prove monotonicity or provide an objective-versus-iteration plot for the settings of Fig. 2.","section":"§IV.C"}],"minor_comments":[{"comment":"Abstract reports 'up to 99.7%' runtime reduction, but Fig. 5/text reports up to 99.1% for PA+PP and up to 99.9% for BA+PP; harmonize the numbers.","section":"Abstract, §V.4"},{"comment":"The pseudocode assigns a_{k,C(x)}=1 without checking the per-O-RU capacity constraint (4c), although the text says 'provided capacity is available.' Add the capacity check to the pseudocode.","section":"Algorithm 1, Stage 3"},{"comment":"Algorithm 2 is described as 'near-optimal' in one sentence and as solving the max-min problem 'globally' in the next; make the terminology consistent once the surrogate issue is resolved.","section":"§IV.B.1"},{"comment":"The Remark is ambiguous: it should state explicitly whether CKM-derived CSI is used in the performance evaluation or only in the architecture description.","section":"§II.C"}],"recommendation":"major_revision","confidential_remarks":"The main risk is the fixed-combiner linearization. If the authors can either provide a rigorous derivation of the constant-coefficient SINR representation or explicitly downgrade the optimality claim to 'near-optimal within a fixed-combiner surrogate,' and then validate against an exact-SINR solver, the paper would be a solid systems contribution. The CKM sensitivity and AO monotonicity issues also need attention. No ethical or citation-pattern concerns."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things about this paper. First, the engineering package is real: QoS-driven multi-connectivity association, a low-complexity bisection/fixed-point power allocator, and an O-RAN xApp/CKM framing for a genuine BVLOS drone-corridor problem. The simulations are well specified—3GPP UMa-AV, 500 Monte Carlo realizations, six named benchmarks, full parameter table—and the headline directions (large min-SE gains, ~99% runtime reduction, 100% fairness for power-optimized schemes) are internally consistent. Second, the central optimality claim does not survive close reading. Algorithm 2 needs precomputed SINR coefficients {a_k, b_ki, c_k} that are power-independent, but the SINR in (2) uses an L-MMSE combiner that depends on the whole power vector, and it includes a self-interference term absent from the update in line 7. No derivation reconciles these. So the algorithm solves a linear-fractional surrogate, not P2 as defined. The abstract and conclusion say “global optimality,” but Section IV.B itself calls it “near-optimal.” The numerical equality with the CVX benchmark is suggestive, but if the CVX benchmark uses the same implicit approximate model, it cannot rescue the claim. This is the main fix: state the fixed-combiner approximation explicitly, prove or cite standard interference-function theory for the surrogate, and scope the optimality claim accordingly.\n\nThe other soft spots are real but smaller. The CKM is excluded from the SE evaluation by the Remark in Section II.C, so the real-time deployment claim rests on an unvalidated CSI-inference component; you should either simulate CKM error or scope the claim. The abstract's “100% QoS satisfaction” contradicts Fig. 3's ~80% at K=100; qualify it as K≤90. The AO convergence assertion is hand-wavy—Algorithm 1 is a heuristic, so monotonic improvement is not automatic—though this is minor compared to the optimality issue.\n\nThere is also a novelty caveat: the “adaptive target SINR” mechanism is a standard feasibility-search/bisection technique, and the underlying fixed-point iteration is classical. The novelty is in the specific combination and the O-RAN architectural integration, not in the power-control theory.\n\nWho should read this? People designing O-RAN xApps for UAV corridors or benchmarking practical cell-free resource allocation. It is not a theory paper and should not be cited for the global optimality claim in its current form. It deserves a serious referee—yes, send it to review—but the referee should push hard on the derivation of the SINR coefficients and on the scoping of the CKM contributions. I would bring it to a reading group to discuss where simulation-based engineering claims cross into unsupported optimality assertions.","headline":"A practically useful O-RAN cell-free UAV-corridor resource-allocation package with a solid simulation campaign, but the 'global optimality' claim for the power-control algorithm rests on an unstated fixed-combiner approximation and should be reframed before the paper is used as a design reference.","tokens_in":11907,"tokens_out":3076,"would_cite":false,"duration_ms":35203,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that jointly optimizing UAV–O-RU association and uplink power in a cell-free O-RAN aerial corridor raises the minimum spectral efficiency by up to 440% while cutting compute time by ~100×, enabling real-time closed-loop con","keywords":["cell-free massive MIMO","O-RAN","aerial corridor","uplink power control","UAV association","spectral efficiency","fixed-point algorithm","channel knowledge map"],"falsifier":"Take a small network (e.g., 2 UAVs, 2 O-RUs), run BG-FPPC to a target SINR, then recompute the L-MMSE combiners using the resulting powers and evaluate the true SINR; if the true minimum SINR falls materially below the target (beyond bisection tolerance), the linearization is invalid.","tokens_in":10724,"feed_emoji":"📡","tokens_out":7516,"duration_ms":71373,"temperature":0.7,"pith_summary":"To make 5G aerial corridors—dedicated three-dimensional routes for drone traffic—reliably serve every UAV, the paper proposes a closed-loop uplink resource manager built on O-RAN's near-real-time RIC. The central claim is that a joint UAV-to-radio-unit association and power-allocation scheme can maximize the worst-case (minimum) spectral efficiency while guaranteeing per-UAV quality-of-service, and that this can be done fast enough for real-time control. Because the joint problem is NP-hard, the authors decompose it and contribute a QoS-driven multi-connectivity association algorithm plus a bisection-guided fixed-point power control algorithm they assert reaches global optimality. Simulation results show up to 440% minimum-SE gain, 100% QoS satisfaction and fairness for up to 90 UAVs, and up to 99.7% runtime reduction relative to an interior-point solver, making O-RAN-compliant real-time deployment the intended payoff.","feed_headline":"O-RAN aerial-corridor scheme lifts weakest UAV uplink SE 440%","feed_subtitle":"Bisection fixed-point power control matches CVX's optimum with 99.7% less runtime, enabling near-RT RIC deployment.","key_machinery":"The load-bearing mechanism is the bisection-guided fixed-point power control (BG-FPPC), which combines an outer bisection over the target SINR with an inner Foschini–Miljanić-style fixed-point update that computes the minimal power meeting each target. It assumes the SINR can be captured by precomputed coefficients {a_k, b_ki, c_k} that make interference and noise linear in powers. The complementary mechanism is the association algorithm: a QoS-driven, multi-connectivity assignment that adds serving O-RUs to weak UAVs, which is what makes the QoS constraints satisfiable.","core_discovery":"On its own terms, the paper establishes a practical decomposition for the NP-hard max-min spectral-efficiency problem in an O-RAN-enabled cell-free massive MIMO aerial corridor. It splits the joint UAV–O-RU association and uplink power allocation into two subproblems solved in alternation: a three-stage QoS-driven association algorithm (strongest-link initialization, O-RU-centric load balancing, and refinement for QoS-violating UAVs) and a bisection-guided fixed-point power control (BG-FPPC) that searches the maximum feasible target SINR with an inner fixed-point update. The paper claims BG-FPPC reaches the same minimum SE as an interior-point solver at O(K^2) complexity, and that the full f","pith_inferences":["If the linearization holds, the BG-FPPC result extends naturally to other max-min resource allocation problems where interference is approximately linear in powers, but that condition is not derived in the paper.","The 440% gain is relative to a full-power baseline with a static association; against an already-tuned power-controlled network, the gain would be far smaller, so the headline number depends on the baseline choice.","CKM accuracy is excluded from the SE evaluation (per the Remark in Section II.C), so the real-time-deployment claim implicitly assumes the CKM-inferred CSI is perfect; a sensitivity analysis to CKM error would be a natural extension.","For K=100 the success rate drops to ~80%, indicating that resource scarcity, not algorithmic quality, becomes the bottleneck; joint trajectory optimization or dynamic pilot assignment could push the QoS frontier further."],"forward_implications":["If correct, the framework moves cell-free mMIMO uplink resource management from offline optimization to near-real-time operation, with decision times in the tens-of-milliseconds range.","The proposed power controller reproduces the interior-point optimum in simulation, suggesting that max-min fair power allocation can be obtained without a general-purpose solver.","The QoS-driven association step, not power control alone, is the main enabler of per-UAV QoS satisfaction; power control is the main driver of fairness.","The CKM-based CSI path removes the need for high-rate E2 CSI exchange, aligning the solution with O-RAN interface constraints."],"fun_headline_variants":["O-RAN aerial scheme: weakest UAV SE up 440%","UAV uplink SE soars 440% via O-RAN CF-MIMO","Bisection power control: 99.7% faster, same optimality","Aerial 5G: O-RAN xApp maxes min SE, 440% gain","CF-O-RAN: 440% min-SE boost, 99.7% runtime cut"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The power-control algorithm treats interference as a straight-line function of transmitter powers that does not change as powers change; if the way receivers combine signals shifts with power, the algorithm may be solving an easier problem than the real one.","fun_headline_variants_meta":{"raw":{"variants":["O-RAN aerial scheme: weakest UAV SE up 440%","UAV uplink SE soars 440% via O-RAN CF-MIMO","Bisection power control: 99.7% faster, same optimality","Aerial 5G: O-RAN xApp maxes min SE, 440% gain","CF-O-RAN: 440% min-SE boost, 99.7% runtime cut"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000657,"raw_usage":{"total_tokens":2898,"prompt_tokens":850,"completion_tokens":2048,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":594,"completion_tokens_details":{"reasoning_tokens":1936}},"tokens_in":594,"tokens_out":2048,"duration_ms":14093,"temperature":1.0,"reasoning_tokens":1936,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T09:46:32.598198+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a small network (e.g., 2 UAVs, 2 O-RUs), run BG-FPPC to a target SINR, then recompute the L-MMSE combiners using the resulting powers and evaluate the true SINR; if the true minimum SINR falls materially below the target (beyond bisection tolerance), the linearization is invalid.","supporting_citations":[],"review_version":1}