{"id":"70698d7a-44a3-43aa-8fe4-5a6b47b59256","arxiv_id":"2411.18123","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":2,"one_line_summary":"The paper derives association, coverage, and spectral efficiency for biased multi-band UAV association and proposes an adaptive closed-form bias that favors mmWave UAVs.","lead":"Stochastic geometry is used to analyze a two-tier UAV network with low-frequency and mmWave drones, and a closed-form bias rule sends users toward mmWave drones. The authors claim the rule improves coverage and per-user data rates over conventional maximum-average-power association.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Coverage analysis ignores dependence between association event and serving distance; Eq. (20) uses unconditional f_Rm, so analytical coverage/SE do not describe the proposed policy.","rationale":"The paper's central assertion is that the proposed adaptive cell range expansion with closed-form β in Eq. (14) significantly improves coverage and per-user data rates over MAP association, and that the derived association probability, coverage probability in Eq. (20), and spectral efficiency accurately describe this policy. The reader's weakest assumption is that association and SINR are treated independently in Eq. (20). I agree that this is the load-bearing step. The association event in (12) depends on r_m and r_lf through β S_m > S_lf, and the SINR in (2) depends on the same r_m; therefore the mmWave coverage probability must condition on the association event. Using the unconditional nearest-distance distribution f_Rm in PCm and f_Rlf in PClf describes the coverage of a randomly selected mmWave or low-frequency UAV, not of a user that actually associates under the policy. This is not a minor approximation: because the bias rule systematically selects users with relatively shorter mmWave distances and relatively longer low-frequency distances, the conditioned serving-distance distributions differ materially from the unconditional ones. Consequently, the analytical coverage and spectral efficiency in Section III do not support the central claim as written. The paper does provide simulation comparisons, which are real evidence that the proposed scheme improves over MAP in the simulated settings, but that does not validate the analytical expressions; the text does not describe any correction for the conditioning issue, and no code or formal verification is provided. The secondary inconsistency about β = 1 at τ = 1 (Eq. (14) gives β = ζ, not 1 unless ζ = 1) adds further doubt but is not the main point. The proposed test—recomputing coverage and SE with the conditional serving-distance distributions—would settle whether Eq. (20) is approximately valid or materially wrong. Given that the central analytical claims rest on this unsupported factorization, keeping the reader's REJECT verdict is appropriate.","tokens_in":8872,"tokens_out":5522,"duration_ms":50029,"concrete_test":"Compute the conditional serving-distance PDFs under the policy: f_{R_m|A}(r) = f_{R_m}(r) P(β P_m G_M K_m r^{-α_m} > P_lf K_lf R_lf^{-α_lf}) / A_m, and f_{R_lf|A^c}(r) = f_{R_lf}(r) P(β P_m G_M K_m R_m^{-α_m} ≤ P_lf K_lf r^{-α_lf}) / A_lf, using independent PPP distances. Re-evaluate Eqs. (24)-(28) and (33)-(36) with these conditional densities in place of f_Rm and f_Rlf, forming PC_cond = A_m PCm_cond + A_lf PClf_cond and SE_cond. Compare PC_cond(γ) and SE_cond against Eq. (20)/(29)-(36) over the simulation range (γ from -10 to 15 dB, N = 32-512). If PC_cond differs from Eq. (20) by more than about 5% probability at any operating γ, the independence assumption in Eq. (20) fails and the analytical claim is not the actual coverage under the policy.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Under the proposed association policy (12), the event 'associate with mmWave' is β S_m > S_lf, with S_m = P_m G_M K_m r_m^{-α_m} and S_lf = P_lf K_lf r_lf^{-α_lf}. This event depends on both r_m and r_lf. Yet Eq. (20) writes total coverage as PClf(γ) A_lf + PCm(γ) A_m, where A_m is correctly marginalized in Eqs. (16)-(19), but PCm(γ) in Eqs. (25)-(28) averages SINR over the unconditional nearest-mmWave distance distribution f_Rm(r) from Eq. (3). It does not condition on the bias event β S_m > S_lf. The true mmWave-user coverage is E[1{β S_m > S_lf} 1{SINR_m > γ}]/A_m, which requires the joint distribution of (r_m, r_lf). Because the bias event favors smaller r_m and larger r_lf, the conditional distribution of the serving distance is stochastically smaller than f_Rm; using f_Rm therefore overestimates PCm and SE_m. The same mis-conditioning affects PClf: users that actually associate with low-frequency tend to have smaller r_lf or larger r_m, so averaging over the unconditional f_Rlf does not describe them. A secondary internal inconsistency is the claim that τ = 1 gives β = 1: from Eq. (14), at τ = 1 the sigmoid equals β0/(1+(β0−1)) = 1, so β = ζ, which is generally not 1; the text says β = 1 without the ζ caveat. This is secondary because the main unsupported step is the independence factorization in Eq. (20).","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies downlink coverage and rate in a two-tier UAV network in which low-frequency and mmWave UAVs form two independent homogeneous Poisson point processes. It proposes a cell-range-expansion association rule that biases users toward mmWave UAVs with a closed-form bias factor β depending on an expected-spectral-efficiency ratio τ, derives association probabilities, coverage probabilities, and spectral efficiency via stochastic geometry, and introduces an analytical model for the antenna gain of interfering mmWave UAVs. Simulation results are used to claim that the proposed scheme improves coverage and per-user data rates relative to conventional maximum-average-power association.","tokens_in":9241,"tokens_out":12903,"duration_ms":124508,"significance":"If the analysis were correct, the closed-form bias factor would be practically appealing because it depends only on network statistics and system parameters, and the antenna-gain derivation would be a useful modeling contribution. The simulation comparisons also suggest a qualitatively promising offloading idea. However, the current analytical framework does not actually characterize the proposed association policy: the coverage and spectral-efficiency expressions ignore the dependence between the association event and the serving distance. Because this issue is load-bearing for every quantitative claim in the paper, the present version cannot be accepted; the core ideas are, in my view, salvageable with a re-derived conditional analysis, so I recommend major revision rather than rejection.","major_comments":[{"comment":"The factorization PC(γ) = PClf(γ)·Alf + PCm(γ)·Am is not valid for the association rule in Eq. (12). Because the event {user is served by mmWave} is {βS_m(r_m) > S_lf(r_lf)}, the set of users served by each band is selected based on r_m and r_lf. Equations (24) and (28) nevertheless integrate the SINR condition over the unconditional nearest-UAV densities f_Rlf and f_Rm from Eq. (3). For mmWave-served users the correct serving-distance density is f_Rm(r)(1 − F_Rlf(η^{1/α_lf} r^{α_m/α_lf}))/Am, with Am from Eq. (19), and an analogous conditional density is needed for low-frequency-served users. Since the conditioning event favors smaller r_m and larger r_lf, the unconditional averages overestimate PCm and SEm and distort PClf and SElf. This error propagates into the spectral-efficiency ratio τ in Eq. (13), the bias factor β in Eq. (14), and the claimed agreement between analysis and simulation in Figs. 3a–3c.","section":"Section III-D, Eq. (20)"},{"comment":"The spectral-efficiency ratio τ is defined in Eq. (13) using SElf and SEm from Eqs. (33) and (36), but those quantities are computed with the unconditional serving-distance distributions. Under the proposed policy, the users actually served by mmWave UAVs have a stochastically smaller serving distance than the unconditional nearest-mmWave distance, and users served by low-frequency UAVs have a stochastically smaller low-frequency serving distance. Therefore τ does not represent the ratio of per-user spectral efficiencies under the policy, and the beta-based adaptation in Eq. (14) is not the claimed equalizing bias. The authors need to re-derive SElf and SEm using the conditional serving-distance densities and recompute τ, β, and the resulting curves in Fig. 3 before the adaptive-CRE claim can be considered established.","section":"Section III-E and Eq. (13)-(14)"},{"comment":"The derivation of pϕ uses f_Rm, the density of the distance from a typical user to its nearest mmWave UAV, as the density of the distance r from an interfering UAV to the user it serves. These are different objects in a PPP model with a separate user point process: the former is a nearest-neighbor distance from a fixed point, while the latter is a Palm/Voronoi quantity that also depends on the user process and on the association rule. The authors should justify this replacement or re-derive the UAV-to-served-user distance distribution; this point is load-bearing because the gain distribution G_x enters the mmWave interference Laplace transform in Eq. (38) and hence all mmWave coverage and spectral-efficiency results.","section":"Section III-B, Eqs. (9)-(11)"}],"minor_comments":[{"comment":"The statement 'When τ = 1, then β = 1' is inconsistent with Eq. (14) unless ζ = 1; in general Eq. (14) gives β = ζ at τ = 1. Please clarify whether the standardization term ζ is intended to be part of the effective bias or a separate normalization.","section":"Section III-C, text after Eq. (14)"},{"comment":"The claim that β is lower-bounded by about 0.5 as τ → 0 is not supported by the simulation parameters (β0 = 5, α = 5), which give β → ζ·β0/(1+(β0−1)e^α) ≈ 0.0084ζ. Please state the parameters for which the stated lower bound holds.","section":"Section III-C, text after Eq. (14)"},{"comment":"There is a missing closing parenthesis in 'E[P(SINRlf > γ)' ; it should read E[P(SINRlf > γ)].","section":"Section III-D, Eq. (21)"},{"comment":"The word 'numer' should be 'number'.","section":"Section IV, Fig. 3c caption"}],"recommendation":"major_revision","confidential_remarks":"The main reason I do not recommend rejection is that the association-conditioning problem, although central, is a well-defined modeling error that can be corrected by re-deriving conditional distance distributions and recomputing the analytical curves. The antenna-gain approximation and the qualitative simulation comparison are worth preserving. I would ask the authors to rerun both the analysis and the simulation validation after the correction, since the current 'improved analysis' curves do not describe the policy they claim to evaluate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this paper has two real ideas—a closed-form bias factor based on a spectral-efficiency ratio, and a new approximation for interfering mmWave antenna gain—but the analytical coverage and spectral-efficiency expressions have a load-bearing flaw. They average SINR over the unconditional serving-distance distribution, while the association policy depends on that same distance. The stress-test note is correct.\n\nThe bias formula in (14) is clever and practical: it depends only on network statistics, no real-time tuning, and saturates to avoid overloading. The antenna-gain derivation in Section III-B is inventive and arguably better than the uniform beam model in [6]. The simulations look plausible and show coverage gains over MAP.\n\nThe problem: Eq. (20) writes total coverage as Alf*PClf + Am*PCm, and PCm in (28) integrates over f_Rm(r) from Eq. (3), the unconditional nearest-mmWave distance. But association under (12) uses the event beta S_m > S_lf, which depends on the same r_m. Users that actually connect to mmWave tend to have shorter serving distances and longer low-frequency distances. Averaging over the unconditional distance overestimates their SINR. The same mis-conditioning affects PClf. So the derived coverage and SE are not the quantities the proposed policy produces. This isn't a small approximation; it changes the numbers.\n\nThere's also a minor internal inconsistency: the text says tau=1 gives beta=1, but (14) gives beta=zeta at tau=1, and zeta is generally not 1. And the simulations only compare against MAP, not prior CRE methods like [10]-[12], so the practical claim of being better than existing CRE isn't tested.\n\nIf the authors fix the conditioning—properly deriving the serving distance distribution under the biased association rule—the paper could be solid. The idea is good, the execution in the analysis section is currently not. I'd send it to referee, because the novel pieces deserve scrutiny, but I'd expect major revision. As is, I would not rely on the analytical results.","headline":"The closed-form bias and antenna-gain approximation are genuinely new, but the coverage/SE analysis conditions on the wrong distance distribution, so the central analytical claims don't describe the proposed policy.","tokens_in":9754,"tokens_out":2591,"would_cite":false,"duration_ms":22905,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60D05","60G55","94A05"],"pacs":[],"model":"deepseek-v4-flash","headline":"A closed-form bias factor computed from network statistics can offload users to mmWave UAVs, improving coverage and data rates in multi-band UAV networks.","keywords":["UAV networks","millimeter-wave","cell range expansion","stochastic geometry","coverage probability","spectral efficiency","Poisson point process","biased user association"],"falsifier":"Recompute the coverage probability and spectral efficiency while explicitly conditioning the mmWave serving-distance distribution on the association event $\\beta S_m > S_{\\mathrm{lf}}$ (so the serving distance is drawn from the biased-offloaded set of users), and compare the results with the paper's Eq. (20) and spectral-efficiency expressions; a material difference would show that the unconditional nearest-UAV distance used for $P_{Cm}$ does not represent users actually served under the bias rule.","tokens_in":8616,"feed_emoji":"📡","tokens_out":9810,"duration_ms":79340,"temperature":0.7,"pith_summary":"This paper claims that user association in a two-band UAV network can be made both simple and adaptive by biasing users toward millimeter-wave UAVs with a bias factor $\\beta$ computed in closed form from network statistics. The authors model low-frequency and mmWave UAVs as independent Poisson point processes and introduce a refined distribution for the antenna gain that an interfering mmWave UAV presents toward a non-serving user. Under the proposed association rule $\\beta S_m > S_{\\mathrm{lf}}$, they derive the association probability, coverage probability, and average spectral efficiency as explicit stochastic-geometry expressions. The payoff, if correct, is that an operator can compute the bias from density, height, transmit-power, and path-loss parameters alone, and reap higher coverage and per-user rates than conventional maximum-average-power association without real-time tuning.","feed_headline":"Closed-form bias formula lifts UAV coverage from 65% to 90%","feed_subtitle":"The closed-form bias is computed from network statistics alone, so offloading to mmWave UAVs needs no real-time tuning.","key_machinery":"The central object is the closed-form bias factor $\\beta = \\frac{\\zeta\\beta_0}{1+(\\beta_0-1)\\exp(\\alpha(1-\\tau))}$, a logistic function of the expected spectral-efficiency ratio $\\tau$ between the mmWave and low-frequency tiers, standardized by $\\zeta$, the ratio of average received powers. It appears in the association rule $\\beta S_m > S_{\\mathrm{lf}}$. Alongside it, the key mechanism is a two-state model for the antenna gain of interfering mmWave UAVs: gain $G_M$ with probability $p_\\theta p_\\phi$ and $G_S$ otherwise, where the main-lobe elevation probability $p_\\phi$ is approximated from the serving-distance distribution of the nearest mmWave UAV, Eq. (11). This gain law feeds the Laplace transforms of per-tier interference, which are integrated against the serving-distance densities to produce the association probability, coverage probability, and spectral efficiency.","core_discovery":"The paper's central claim is that the usual maximum-average-power association rule is suboptimal in multi-band UAV networks, and that a rule of the form 'associate with the mmWave UAV if $\\beta S_m > S_{\\mathrm{lf}}$'--with $\\beta$ given by a closed-form sigmoid of the expected spectral-efficiency ratio between bands--captures most of the available gain. When the mmWave tier offers higher expected spectral efficiency ($\\tau > 1$), the bias favors mmWave UAVs to exploit lower interference and wider bandwidth; when the low-frequency tier is better, the bias shifts the other way, with saturation built in to avoid overloading either tier. The paper further claims that its new statistical model of mmWave antenna gain, which derives the main-lobe elevation probability from the nearest-UAV distance distribution, yields accurate interference Laplace transforms and therefore accurate coverage and spectral-efficiency predictions. Simulations are presented as evidence that the analytical curves track the simulated ones and that the proposed scheme raises coverage probability at a 0 dB SINR threshold from about 65% to nearly 90%, with larger per-user data rates as the mmWave density increases.","pith_inferences":["Editorial inference: the same construction--a logistic bias built from an expected spectral-efficiency ratio and standardized by path loss--could be transferred to terrestrial heterogeneous networks or to UAV networks with more than two bands, since the argument does not rely on anything specific to airborne nodes beyond the distance distributions.","Editorial inference: the analyzed gains likely overstate what a real deployment would achieve if the association-SINR independence in Eq. (20) is violated, because users offloaded to mmWave by the bias tend to be farther from their serving mmWave UAV than the unconditional nearest-UAV distance assumes.","Editorial inference: the antenna-gain approximation for interfering mmWave UAVs could be checked against a ray-tracing or 3GPP-style channel model; a significant discrepancy in the elevation main-lobe probability would change the interference Laplace transform and therefore the predicted benefit of the bias."],"forward_implications":["Under the proposed policy, an operator can set the cell-range-expansion parameter from network statistics alone (UAV densities, height, transmit powers, path-loss exponents, and expected spectral-efficiency ratio), with no real-time channel measurements or hand tuning.","If the simulations are representative, switching from maximum-average-power association to the proposed bias rule raises coverage probability at a 0 dB SINR threshold from roughly 65% to nearly 90%.","The bias saturates through its sigmoid form, so mmWave UAVs are favored only up to a maximum bias $\\beta_0$ and low-frequency UAVs are protected by a lower bound near $\\beta=0.5$, which should prevent either tier from being overloaded.","The derived analytical expressions for coverage and spectral efficiency make the scheme directly comparable across deployment parameters such as the mmWave-to-low-frequency density ratio and the number of mmWave antennas, without rerunning network simulations."],"supporting_citations":[{"why":"Supplies the Poisson point process machinery, the nearest-neighbor distance distribution, and the probability-generating-functional step used for the serving-distance density and interference Laplace transforms.","marker":"[4]"},{"why":"Provides the Nakagami-m fading model for the mmWave band and the simplified uniform-beam antenna-gain model that the paper's antenna-gain derivation refines.","marker":"[6]"},{"why":"Supplies the sectorized uniform planar array gain model ($G_M$, $G_S$, half-power beamwidths) used for mmWave UAV antenna patterns.","marker":"[14]"},{"why":"Supplies the integral identity with $K(z)$ used to evaluate expected spectral efficiency under Nakagami fading.","marker":"[15]"},{"why":"Establishes the rate-based cell range expansion baseline that relies on hand-tuned bias, which the paper's closed-form $\\beta$ is designed to replace.","marker":"[12]"}],"fun_headline_variants":["Closed-form bias lifts UAV coverage to 90%","Simple bias rule boosts UAV coverage to 90%","Multi-band UAV: closed-form bias raises coverage to 90%","Offload to mmWave UAVs: coverage jumps to 90%","Adaptive cell range expansion: bias formula lifts coverage to 90%"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that a user's probability of being assigned to the mmWave tier and the signal quality that user then experiences from that tier can be analyzed as independent, even though both depend on the same nearest-mmWave-UAV distance and the same bias rule; if that independence fails, the reported coverage and spectral-efficiency numbers are not the true quantities under the proposed policy.","fun_headline_variants_meta":{"raw":{"variants":["Closed-form bias lifts UAV coverage to 90%","Simple bias rule boosts UAV coverage to 90%","Multi-band UAV: closed-form bias raises coverage to 90%","Offload to mmWave UAVs: coverage jumps to 90%","Adaptive cell range expansion: bias formula lifts coverage to 90%"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000184,"raw_usage":{"total_tokens":1303,"prompt_tokens":915,"completion_tokens":388,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":531,"completion_tokens_details":{"reasoning_tokens":301}},"tokens_in":531,"tokens_out":388,"duration_ms":4085,"temperature":1.0,"reasoning_tokens":301,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:31:02.228936+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the coverage probability and spectral efficiency while explicitly conditioning the mmWave serving-distance distribution on the association event $\\beta S_m > S_{\\mathrm{lf}}$ (so the serving distance is drawn from the biased-offloaded set of users), and compare the results with the paper's Eq. (20) and spectral-efficiency expressions; a material difference would show that the unconditional nearest-UAV distance used for $P_{Cm}$ does not represent users actually served under the bias rule.","supporting_citations":[{"cited_title":"A tractable approach to coverage and rate in cellular networks,","cited_arxiv_id":null,"evidence_quote":"Supplies the Poisson point process machinery, the nearest-neighbor distance distribution, and the probability-generating-functional step used for the serving-distance density and interference Laplace transforms."},{"cited_title":"Modeling and analysis of mmWave UA V swarm networks: A stochastic geometry approach,","cited_arxiv_id":null,"evidence_quote":"Provides the Nakagami-m fading model for the mmWave band and the simplified uniform-beam antenna-gain model that the paper's antenna-gain derivation refines."},{"cited_title":"Device-to-device millimeter wave communications: Interference, coverage, rate, and finite topologies,","cited_arxiv_id":null,"evidence_quote":"Supplies the sectorized uniform planar array gain model ($G_M$, $G_S$, half-power beamwidths) used for mmWave UAV antenna patterns."},{"cited_title":"A useful technique for interference analysis in Nakagami fading,","cited_arxiv_id":null,"evidence_quote":"Supplies the integral identity with $K(z)$ used to evaluate expected spectral efficiency under Nakagami fading."},{"cited_title":"Rate-based cell range expansion for downlink massive MIMO heterogeneous networks,","cited_arxiv_id":null,"evidence_quote":"Establishes the rate-based cell range expansion baseline that relies on hand-tuned bias, which the paper's closed-form $\\beta$ is designed to replace."}],"review_version":1}