{"id":"54ad5667-b228-41b1-bf07-1551ecef6eeb","arxiv_id":"1908.09289","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Joint optimization of UAV hover position and uplink NOMA power control via SCA and a low-complexity heuristic improves sum rate over fixed-position NOMA and FDMA baselines.","lead":"This paper proposes algorithms to jointly choose where a drone hovers and how much power ground users transmit when they share the same frequency using non-orthogonal multiple access (NOMA). The design maximizes total data rate while guaranteeing each user a minimum rate, and numerical tests show gains over simpler baselines.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claimed <4% loss for the low-complexity algorithm is established only for one 4-user scenario; generality across deployments is untested.","rationale":"The paper's central claim has two parts: the SCA/penalty joint optimization improves sum rate, and the low-complexity Algorithm 3 stays within 4% of it. The first part is plausible: the problem reduction to a stationary deployment is sound, the closed-form power control follows from a linear program, and the numerical results match the analytical power. The second part is the load-bearing quantitative promise, but it is validated on a single 4-user configuration. Fig. 4 itself shows the true optimal position is not exactly atop a user, so Algorithm 3's restriction is an approximation whose cost is measured once. No theoretical bound or broad numerical study supports the 4% figure as a general statement. The reader's weakest assumption (free-space LoS channel) is a legitimate scope limitation but is explicitly stated and standard in UAV work; it does not reveal an internal flaw. The single-scenario validation is more directly tied to the headline claim's correctness and is actionable. I therefore recommend keeping the CONDITIONAL verdict rather than accepting the paper as established. My concern does not contradict the reader's view; it complements it, hence 'partial' agreement. The proposed Monte Carlo test would settle whether the 4% claim survives beyond the demonstrated geometry.","tokens_in":18156,"tokens_out":24594,"duration_ms":243228,"concrete_test":"Run a Monte Carlo sweep: generate at least 100 random instances with M ∈ {4,8}, users drawn uniformly or as Gaussian clusters in a 400x400 m area, H ∈ {50,100,200} m, Pmax ∈ {0.5,1,2} W, and r* swept from 0.1 to its feasibility limit. For each instance, compute the sum rate of Algorithm 3 (hover above the best user, closed-form power (55)) and compare it with the sum rate achieved by a fine-grained optimization over Q (e.g., 2 m grid search or a global search starting from Algorithm 1's solution) using the same power allocation. If the maximum or 95th-percentile gap exceeds 4% in any regime tested, the general claim 'less than 4% performance loss' must be qualified.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline quantitative result—Algorithm 3 achieves more than 96% of Algorithm 1's sum rate when r* ≤ R*—is supported only by Fig. 7, which uses one fixed geometry (M=4, Pmax=1 W, H=100 m, γ0=10^6). Algorithm 3 restricts the UAV to hover exactly above one user, yet Fig. 4 shows the continuous optimum is not exactly above any user (it is near user 3 but offset). The <4% gap therefore measures the cost of this restriction in a single configuration. There is no theorem bounding the gap as a function of M, user topology, H, or Pmax. For symmetric or clustered user placements, the optimal deployment may lie near the centroid and far from every user, so hovering above one user could incur a much larger loss. Because the abstract and contributions present the <4% figure as a general benefit, this unsupported generalization is the most load-bearing weakness. The LoS channel assumption is a stated modeling choice rather than an internal inconsistency; the unproven generality of the 4% claim is a correctness risk of the headline finding.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper investigates a UAV-enabled uplink non-orthogonal multiple access (NOMA) system in which a single UAV collects messages from multiple ground users. The authors jointly optimize the UAV deployment position and the users' transmit powers to maximize the sum rate subject to per-slot quality-of-service (QoS) constraints. The problem is formulated as a mixed-integer non-convex program (P1). The paper first proves (Proposition 1) that the UAV should remain stationary at an optimal point, reducing the trajectory problem to a deployment problem. It then develops an iterative algorithm (Algorithm 1) based on successive convex approximation (SCA) and a penalty method to handle the binary SIC ordering variables and non-convex rate constraints. For a fixed deployment, a closed-form power allocation is derived from KKT conditions (Eqs. (55)-(57)). A low-complexity algorithm (Algorithm 3) is proposed that hovers above one user and uses the closed-form power allocation; the authors report that it achieves more than 96% of the iterative algorithm's sum rate when the QoS threshold r* is not too large. Numerical results compare the proposed schemes against a fixed-position NOMA benchmark and an FDMA benchmark.","tokens_in":18395,"tokens_out":17042,"duration_ms":174537,"significance":"If the claims hold, the paper offers a practical and computationally efficient solution to a relevant problem in UAV-assisted uplink NOMA. The stationarity result is a useful simplification, and the KKT-based closed-form power allocation is elegant and is numerically verified in Fig. 6. The SCA and penalty framework is a credible approach to a difficult non-convex problem, and the low-complexity algorithm has clear practical appeal for real-time deployment. The main weakness is that the headline near-optimality claim of the low-complexity algorithm is supported by only a single numerical scenario, and the penalty exactness and convergence guarantees are imported from an external reference without a self-contained verification. These issues, rather than the central derivation, currently limit the strength of the paper's conclusions.","major_comments":[{"comment":"The claim that Algorithm 3 achieves 'less than 4% performance loss' compared with Algorithm 1 is supported only by one fixed scenario: M=4 users, Pmax=1 W, H=100 m, gamma0=10^6, and a single user geometry in a 400 x 400 m area. Fig. 4 shows that the continuous optimal deployment is not exactly above any user, so the 4% gap measures the cost of the hover-above-user restriction in this one configuration. No theorem or broader simulation establishes the gap as a function of M, user topology, H, or Pmax, and for symmetric or clustered placements the optimal deployment could be far from any single user. Since the low-complexity algorithm is a headline contribution, please either add Monte Carlo results over random user placements and parameter ranges (reporting mean and worst-case gaps) or explicitly qualify the claim as scenario-specific.","section":"Section IV.C and Fig. 7"},{"comment":"The exactness of the penalty relaxation is asserted by reference to [31], and Algorithm 1's termination and feasibility rely on phi_ij approaching zero at convergence. The cited result is for a different system model (multicast multigroup multicell transmission), and the manuscript does not verify that the required assumptions (e.g., compactness of the feasible set, constraint qualification, and behavior of the SCA inner approximations under the penalty update) hold for P4/P5. Because this exactness is the mechanism that converts the binary SIC variables into continuous variables, the paper should provide a self-contained argument adapted to this problem, or state clearly that binary feasibility is guaranteed only under the conditions of the cited theorem.","section":"Section III.B, Eqs. (17)-(19) and Algorithm 1"},{"comment":"The paper repeatedly states that non-convex constraints are 'transformed into convex constraints' via SCA, but the resulting constraints are sufficient inner approximations, not equivalent reformulations. For example, replacing a convex function by its first-order Taylor lower bound makes the constraint more restrictive, so P5 is an inner approximation of P4, not an equivalent problem. This distinction is important for the correct interpretation of Algorithm 1: its output is feasible for the original problem only if the sequence of inner approximations is maintained and the penalty term converges to zero. Please state this explicitly in Section III.B so that readers do not infer that P5 and P4 have the same feasible set.","section":"Section III.B, Eqs. (22), (25), (27), (30)-(32)"}],"minor_comments":[{"comment":"The first sentence is a sentence fragment: 'In order to overcome the inherent latency in multi-user unmanned aerial vehicle (UAV) networks with orthogonal multiple access (OMA).' It should be completed or merged with the next sentence.","section":"Abstract"},{"comment":"Equation (3) defines d_i[n] as the distance but states d_i[n] = H^2 + ||q[n]-q_i||^2, which is dimensionally a squared distance. Please denote this quantity as d_i^2[n] or use a separate notation for the squared distance.","section":"Section II.A, Eq. (3)"},{"comment":"The text refers to 'Fig. 2(a)' and 'Fig. 2(b)' for the convergence performance, but the convergence plots appear in Figs. 3(a) and 3(b); Fig. 2 is used for the algorithm-page table. Please correct the cross-references.","section":"Section V, Figs. 3 and 4"},{"comment":"There are several typographical errors: 'We proof Proposition 1' should be 'We prove Proposition 1'; 'Baesd' should be 'Based'; 'can-not' should be 'cannot'; 'poorer suers' should be 'poorer users'.","section":"Throughout"},{"comment":"The complexity statement is unclear: the text says the sort operation is O(M^2) and therefore Algorithm 3 has complexity O(M^3). Sorting M elements with a comparison-based sort is O(M log M), so the M repetitions in Algorithm 3 would be O(M^2 log M), not O(M^3), unless a non-standard sorting cost is assumed. Please clarify the complexity model.","section":"Section IV.C, paragraph on complexity"},{"comment":"The statement 'we can work out that R* ≈ 1.09 bps/Hz' is given without showing which user yields this value or how the bisection is performed. A brief explanation or a reference to Eq. (57) would improve reproducibility.","section":"Section V, first paragraph"}],"recommendation":"major_revision","confidential_remarks":"The paper is a competent and useful contribution, and the core derivations appear sound. The main risk is that the 4% near-optimality claim for the low-complexity algorithm is presented in the abstract and introduction without the scenario-specific qualification that the numerical results actually support. I recommend requiring additional simulation evidence or a careful rewording, and a stronger self-contained treatment of the penalty exactness and convergence claims."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague —\n\nThe one thing to know: this is a competent, honest optimization paper, not a breakthrough. Its genuinely useful piece is a closed-form power allocation for uplink NOMA when the UAV's position is fixed. The authors sort users by channel gain, show that for all but the strongest user the QoS constraints bind, and derive a simple expression for transmit power. The numerics match the formula, so I trust that result.\n\nThe paper also proves the right version of the stationarity proposition: there exists an optimal solution with the UAV hovering, because the per-slot problems are identical and a stationary point satisfies the mobility constraint. The proof is a bit clunky but the conclusion holds. The SCA/penalty machinery is standard but applied carefully. The uplink extension is a genuine gap in the literature, since downlink UAV-NOMA does not directly carry over due to the reversed SIC order.\n\nThe soft spots are proportionate. The headline \"less than 4% loss\" for the low-complexity algorithm comes from a single 4-user configuration. The stress-test is right: no theorem or multi-configuration simulation bounds that gap, and the continuous optimum is not exactly above any user, so the claim should be read as \"we saw <4% in one example,\" not \"the gap is always <4%.\" That is an overgeneralization in the abstract, but it is not load-bearing for the main result. The NP-hardness assertion is unproved, which is common in this literature and not central to the algorithm's value. The penalty exactness is imported from [31] without a self-contained argument; acceptable, but a referee should ask for a citation check. The LoS channel is an idealization, not an internal inconsistency. Minor typos (\"proofed\" in the conclusions) and no code release are worth cleaning up.\n\nNo circularity: the power rule follows from KKT conditions on the same objective, and the benchmarks against FDMA and fixed-position NOMA are external.\n\nWho gets value: anyone working on UAV-assisted uplink access will find the closed-form power rule handy, and Algorithm 3 is a reasonable low-complexity baseline. It is not a new research direction, but it is a solid, citable piece of engineering.\n\nRecommendation: send it to peer review. It deserves referee time, with the request to back the 4% claim with more than a single simulation.","headline":"A solid, honest optimization paper with one genuinely useful closed-form power rule; the headline 4% gap is a single-scenario observation, not a proven bound.","tokens_in":18919,"tokens_out":5975,"would_cite":true,"duration_ms":55710,"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 single hover point maximizes the sum rate in UAV-enabled uplink NOMA.","keywords":["UAV-enabled networks","uplink NOMA","successive interference cancellation","UAV deployment","power control","successive convex approximation","penalty function method","sum-rate maximization"],"falsifier":"In the paper's own four-user, 400-by-400 meter setup, add a single tall building between the predicted hover point and the strongest user so that the line-of-sight assumption fails; if the iterative algorithm then selects a non-stationary trajectory or a different hover point, or if the closed-form powers (55) leave the weakest user's rate below $r^*$, the central claim is falsified.","tokens_in":17953,"feed_emoji":"🚁","tokens_out":7882,"duration_ms":76602,"temperature":0.7,"pith_summary":"This paper studies a single UAV acting as a flying base station for ground users that transmit simultaneously in the same spectrum using uplink non-orthogonal multiple access (NOMA), where the UAV separates users by successive interference cancellation. The authors try to establish that, under a free-space line-of-sight channel, the optimal UAV policy is not a trajectory but a stationary hover at one point, and that choosing that point jointly with the users' transmit powers maximizes the total sum rate while every user meets a per-slot rate target. They show the joint problem is a mixed-integer non-convex (NP-hard) program, solve it with an iterative algorithm that combines successive convex approximation with a penalty method for the binary decoding order, and derive a closed-form power allocation: every user except the one with the strongest channel transmits just enough to hit the rate target, and the strongest user receives all remaining power. A low-complexity scheme that hovers over the best single user and applies the closed-form powers achieves more than 96% of the iterative algorithm's sum rate when the rate target is not too large, and it outperforms both fixed-base-station NOMA and FDMA. If true, this means UAV-enabled uplink NOMA can be made practical with a near-optimal deployment rule that takes seconds to compute.","feed_headline":"A UAV should hover, not fly, for uplink NOMA","feed_subtitle":"Closed-form power split gives the strongest user all spare power, with under 4% loss from full optimization.","key_machinery":"The load-bearing object is the distance-only free-space channel gain $h_i = \\sqrt{\\beta_0/(H^2+\\|q-q_i\\|^2)}$, which makes channel strength a deterministic function of horizontal distance and makes the SIC decoding order equivalent to a distance ordering. The identity that carries the argument is (11), $\\sum_i R_i = \\log_2(1+\\sum_i P_i\\tilde h_i)$, which removes the decoding order from the objective and turns the trajectory problem into a deployment problem. For the full problem, the machinery is successive convex approximation, which replaces each non-convex constraint by a convex surrogate at a current feasible point, together with a penalty term $\\lambda\\sum_{i,j}\\phi_{ij}$ that relaxes the binary SIC-order variables and drives them to $\\{0,1\\}$ at convergence. For the fast scheme, the machine is the closed-form power allocation (55), which lets the sum rate at any candidate hover point be evaluated by sorting channel gains and applying a formula without any numerical search.","core_discovery":"The central claim is that the trajectory variable can be eliminated. Because the per-slot sum rate satisfies $\\sum_i R_i[n] = \\log_2(1 + \\sum_i P_i[n]\\tilde h_i[n])$, independent of the SIC decoding order, every time slot faces the same optimization problem; the mobility constraints only force a return to the start, and the authors prove by contradiction that the optimum is therefore a stationary deployment position (Proposition 1). With the UAV fixed, ordering users by channel gain makes the QoS constraints linear in power, and Karush-Kuhn-Tucker analysis yields the closed-form powers of Eq. (55): for all but the strongest user, $P_{(i)} = (2^{r^*}-1)2^{(i-1)r^*}/\\tilde h_{(i)}$, while the strongest user gets $P_{(M)} = P_{\\max}-\\sum_{i