{"id":"f7ba0d24-95d8-4c78-82aa-253e17c4b655","arxiv_id":"1908.05513","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"In a two-user downlink NOMA cell with reliability constraints, the user with the largest product of target error probability and average signal-to-interference ratio should be decoded last, and simple closed-form power splits achieve equal-rate or near-optimal sum-rate performance.","lead":"This paper derives simple rules for how a base station should split power and set rates for two users sharing one NOMA channel, using only average signal and interference powers plus a reliability target. If correct, it removes the need for real-time channel knowledge, which could make reliable NOMA easier to deploy in dense cellular networks.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 4's beta* rule contradicts its own Appendix G: for (phi_1^*, phi_2^*, mu)=(0.1, 0.2, 0.95), Eq. (22) selects beta*=1 while both gamma_bar and the exact sum-rate are larger at beta*=1/2.","rationale":"The reader identified the maximum-sum-rate optimality as the weak spot and attributed it to the unproven AM-GM approximation. My stress test confirms that this is the right area but finds a stronger, more concrete defect: even if one grants Proposition 1 and maximizes gamma_bar instead of (1+gamma_1)(1+gamma_2), the endpoint comparison in Appendix G is solved incorrectly, so Eq. (22) gives the wrong regime boundary. The counterexample with phi_1^*=0.1, phi_2^*=0.2, mu=0.95 is within the model's parameter range and flips the decision: the printed rule chooses beta*=1, while both the surrogate and the exact objective are larger at beta=1/2. Since Eq. (22) feeds directly into Algorithm V-A and the maximum-sum-rate simulation results, the central claim about optimal power allocation and scheduling for that problem is not established as stated. Theorem 5's OMA comparison uses the same argument and should be rechecked. This is an internal inconsistency rather than a disagreement with external consensus, so it cannot be waived as a modeling choice. The equal-rate results (Theorem 3, Corollary 1, and Corollary 2) appear self-contained and remain a useful contribution, so the paper has salvageable value. I therefore keep the reader's conditional verdict but sharpen the condition: Theorems 4 and 5 must be corrected and numerically verified over the feasible parameter grid before the maximum-sum-rate claims can be accepted. A separate abstract/body contradiction about whether the largest-phi^* user is decoded first or last should also be resolved, though it is secondary to the technical error identified here.","tokens_in":22818,"tokens_out":19369,"duration_ms":174235,"concrete_test":"Compute gamma_1 and gamma_2 from Eq. (5) for (phi_1^*, phi_2^*, mu) = (0.1, 0.2, 0.95) at beta=1/2 and beta=1, and compare gamma_bar = gamma_1 + gamma_2 and the exact objective (1+gamma_1)(1+gamma_2) with the beta* predicted by Eq. (22). If the exact objective is larger at beta=1/2, the printed rule is contradicted. To quantify the scope, grid-search beta in [1/2, 1] with the exact objective over 10^4 random feasible triples (phi_1^*, phi_2^*, mu) and report the fraction of cases where Eq. (22) selects a strictly worse candidate.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The maximum-sum-rate allocation rule is not merely unproven under the AM-GM approximation; Theorem 4 is internally inconsistent. Appendix G correctly notes that gamma_bar = gamma_1 + gamma_2 is convex in beta, so its maximum over beta in [1/2, 1] is at an endpoint, and then compares gamma_bar(1/2) = phi_1^*/(2+phi_1^*) + phi_2^*/(2+mu*phi_2^*) with gamma_bar(1) = phi_1^*. Solving gamma_bar(1/2) > gamma_bar(1) for mu gives mu < [2(phi_2^*-phi_1^*) + phi_1^*(phi_2^*-2*phi_1^*)] / [phi_1^* phi_2^* (1+phi_1^*)], not the printed threshold in Eq. (22), whose numerator contains phi_1^*(phi_2^*-2). For phi_1^*=0.1, phi_2^*=0.2, mu=0.95, the printed threshold is about 0.909, so Eq. (22) returns beta*=1; direct evaluation gives gamma_bar(1/2)=0.1389 > gamma_bar(1)=0.1, and the exact objective (1+gamma_1)(1+gamma_2) is about 1.1433 at beta=1/2 versus 1.1 at beta=1. Thus the beta* used in Algorithm V-A and in the Section VI sum-rate simulations can be strictly suboptimal. Theorem 5's OMA-comparison threshold inherits the same defect, so the maximum-sum-rate claims as printed are not valid. The equal-rate analysis and Corollary 1 are not affected by this specific error.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a distributed rate-control scheme for downlink NOMA under per-link reliability constraints, using only topological information (average desired and interfering powers) and target error probabilities rather than instantaneous CSI at the base station. It derives the SIR threshold formula in Lemma 1, a small-error-probability approximation in Theorem 1, the distribution of the allocated threshold in Poisson cellular networks in Theorem 2, and then focuses on the two-user case. For equal-rate allocation, Theorem 3 gives the power split, Corollary 1 gives the optimal decoding order, and Corollary 2 gives the condition under which NOMA beats OMA. For maximum sum-rate allocation, Proposition 1 introduces an AM-GM approximation, Theorem 4 gives a power-splitting rule, and Theorem 5 gives an OMA-comparison threshold. The claimed optimal ordering is that the user with the larger product of target error probability and average-signal-to-average-interference ratio should be decoded last.","tokens_in":23250,"tokens_out":5310,"duration_ms":51521,"significance":"The equal-rate part of the paper is a solid, analytically traceable contribution: the derivations in Lemmas 1 and 3, Theorem 3, and Corollaries 1 and 2 are careful, and the small-epsilon approximation in Theorem 1 is numerically supported. The distribution result in Theorem 2 is also useful, as it gives a density-independent characterization of the allocated SIR threshold. However, the maximum-sum-rate claims as printed are not valid: Theorem 4 contains an algebraic error in its endpoint comparison, and Proposition 1 replaces the true objective by an AM-GM upper bound without proving that the approximation preserves the argmax or the sign of the NOMA-versus-OMA comparison. Because the power-splitting rule (22) is used in Algorithm V-A and in the Section VI simulations, the maximum-sum-rate numerical results may be based on a suboptimal beta*. The equal-rate results are not affected by these errors, so the paper can likely be repaired, but the maximum-sum-rate contribution requires substantive rework.","major_comments":[{"comment":"The threshold in Theorem 4 is algebraically wrong. Solving the comparison gamma_bar(1/2) > gamma_bar(1) from Eq. (51) gives mu < [2(phi2* - phi1*) + phi1*(phi2* - 2 phi1*)] / [phi1* phi2* (1 + phi1*)], not the printed mu < [2(phi2* - phi1*) + phi1*(phi2* - 2)] / [phi1* phi2* (1 + phi1*)]. The error is the factor phi1*(phi2* - 2) in the numerator, which should be phi1*(phi2* - 2 phi1*). For example, with (phi1*, phi2*, mu) = (0.1, 0.2, 0.95), the printed threshold is approximately 0.909, so Eq. (22) returns beta* = 1; direct evaluation gives gamma_bar(1/2) = 0.1389 > gamma_bar(1) = 0.1, and the exact objective (1 + gamma1)(1 + gamma2) is approximately 1.1433 at beta = 1/2 versus 1.1 at beta = 1. Since Algorithm V-A and the Section VI maximum-sum-rate simulations use Eq. (22), the reported maximum-sum-rate results may rest on a strictly suboptimal beta*.","section":"Appendix G and Theorem 4, Eq. (22)"},{"comment":"The maximum-sum-rate claims are not robust to the AM-GM approximation (21). The paper gives the pointwise relative-error formula (48), but that formula is a function of gamma1 and gamma2, which themselves depend on beta; no proof is supplied that maximizing the approximation [1 + gamma_bar/2]^2 preserves the argmax over beta or preserves the sign of the comparison with OMA. The algebraic error in Theorem 4 is a concrete manifestation of the risk: the printed rule selects the wrong endpoint even when both the exact objective and the approximation prefer beta = 1/2. Consequently, Theorem 5's condition (24), which is derived through the same approximation, is not established as stated. The theorem statements should be revised to either prove preservation of the argmax and of the OMA comparison, or explicitly present the results as approximate scheduling rules supported by numerical checks rather than as optimality claims.","section":"Proposition 1, Eq. (21), and Appendix F; Theorem 5"}],"minor_comments":[{"comment":"The heading of Appendix F reads \"PROOF OF COROLLARY 1\", but the appendix proves Proposition 1; the heading should be corrected.","section":"Appendix F heading"},{"comment":"The notation \"zi(theta) > 1 ≦ zi(theta) <= 1\" between the two displayed equations is confusing; it should be spelled out, for example, \"for zi(theta) > 1 use (9), and for zi(theta) <= 1 use (10)\".","section":"Theorem 2, Eqs. (9)-(10)"},{"comment":"In the algorithmic summary, step 3 says beta is given by \"(3) and (22)\" for equal-rate and maximum sum-rate allocations, respectively; Eq. (3) is the interference definition, so the equal-rate reference should be Eq. (12).","section":"Section V-A, step 3"},{"comment":"Theorem 4 is stated as the optimal power allocation for maximum sum-rate, but the proof maximizes the approximate objective gamma_bar rather than the exact sum-rate objective; the statement should be qualified accordingly.","section":"Theorem 4 statement"}],"recommendation":"major_revision","confidential_remarks":"The equal-rate analysis is sound and publishable, but the maximum-sum-rate claims as printed are not valid due to the algebraic error in Eq. (22) and the unproven AM-GM argmax-preservation step. I recommend a major revision rather than rejection, since the flaws appear fixable and the equal-rate contribution stands independently. The Section VI-B simulations must be regenerated with a corrected beta* rule."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the equal-rate half of this paper is a genuine contribution, but the maximum-sum-rate claims are not ready to be cited. The printed beta* rule in Theorem 4 contradicts the paper's own derivation in Appendix G, and the derivation itself optimizes an AM-GM surrogate without proving the argmax carries over.\n\nWhat is new and good: the authors extend their CSI-free distributed rate control from OMA to NOMA in a Poisson multi-cell network. Lemma 1 gives a clean SIR-threshold formula, Theorem 2 characterizes its distribution, and the small-epsilon approximation (8) is well-supported and numerically verified. For the 2-user equal-rate problem, Theorem 3's power split (12), Corollary 1's ordering rule (largest phi* decoded last), and Corollary 2's NOMA-over-OMA condition (14) are careful and the Monte Carlo matches. That part is solid.\n\nSoft spots: the maximum sum-rate analysis has a load-bearing flaw. First, Proposition 1 replaces the true objective (1+gamma1)(1+gamma2) with the AM-GM bound [1+(gamma1+gamma2)/2]^2; no proof is given that the argmax over beta is preserved, and the relative-error formula in (48) does not establish that. Second, and more concretely, the threshold in Eq. (22) is algebraically wrong even relative to Appendix G's own comparison of gamma_bar at beta=1/2 and beta=1. The numerator should read 2(phi2*-phi1*)+phi1*(phi2*-2phi1*), not 2(phi2*-phi1*)+phi1*(phi2*-2). For (phi1*,phi2*,mu)=(0.1,0.2,0.95), Eq. (22) picks beta*=1, while the corrected threshold and direct evaluation both pick beta*=1/2, which yields a larger gamma_bar and a larger exact sum-rate. Since Algorithm V-A and the Section VI sum-rate simulations use beta from (22), those results are suspect. Theorem 5's OMA-comparison threshold inherits the same defect. There is also an abstract/body contradiction: the abstract says the largest-phi* user should be decoded first, but Corollary 1 and Section V-A correctly say the opposite (decoded last). Finally, the reliability computation treats the SIC residual fraction mu as a deterministic factor and does not include the probability that SIC itself fails; a reviewer should ask how that affects the stated reliability guarantee.\n\nWho this is for: researchers working on stochastic-geometry NOMA with reliability constraints. The equal-rate results and the SIR-threshold distribution are worth publishing; the sum-rate contribution needs either a corrected proof or a clear relabeling as approximate with numerical verification of the argmax.\n\nRecommendation: this deserves a serious referee despite the flaws. Send it to review, but make clear the sum-rate analysis needs major revision before it can be accepted. I would not cite the maximum-sum-rate claims in their current form.","headline":"Equal-rate analysis is solid, but the maximum-sum-rate optimality claims are internally inconsistent and should not be cited as-is.","tokens_in":23755,"tokens_out":5679,"would_cite":false,"duration_ms":49820,"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 two-user downlink NOMA group can meet per-user reliability targets using only average received powers, with a simple optimal decoding order and closed-form power splits.","keywords":["NOMA","distributed rate control","reliability constraints","power allocation","optimal decoding order","Poisson point process","successive interference cancellation","stochastic geometry"],"falsifier":"Numerically maximize the exact sum-rate objective $(1+\\gamma_1)(1+\\gamma_2)$ over $\\beta \\in [1/2,1]$ for a grid of $(\\phi_1^*,\\phi_2^*,\\mu)$ values, and compare the maximizing $\\beta$ and the sign of the comparison with OMA against Eqs. (22) and (24); any feasible grid point with an interior optimum or a sign mismatch would refute the universal optimality claim.","tokens_in":22557,"feed_emoji":"📶","tokens_out":13732,"duration_ms":117806,"temperature":0.7,"pith_summary":"This paper aims to show that a two-user downlink NOMA group can be run under per-user reliability constraints without any instantaneous channel state information at the base station. The only inputs are each user's target error probability $\\epsilon_i$ and its average received power from the serving and interfering base stations, which together determine a scalar 'standalone SIR threshold' $\\phi_i^* \\approx \\rho_i/\\epsilon_i$, where $\\rho_i$ is the average desired-signal-to-average-interference power ratio. From that scalar the paper derives the optimal decoding order, the closed-form power split for equal-rate and maximum-sum-rate operation, and explicit conditions under which NOMA beats an equal-partition OMA baseline. These results turn NOMA power and rate allocation into a cheap, local computation and show that NOMA's gain in dense interference-limited deployments lives or dies with the residual SIC interference.","feed_headline":"NOMA beats OMA when leftover interference stays below a limit","feed_subtitle":"Two-user NOMA gets optimal ordering and power splits from average signal powers and error budgets alone.","key_machinery":"The load-bearing object is the standalone reliability-constrained SIR threshold $\\phi_i^*$, defined as the unique positive solution of $\\prod_{j \\in \\Phi \\setminus \\{b_0\\}} (1+\\phi_i r_i^\\alpha r_{j,i}^{-\\alpha}) = (1-\\epsilon_i)^{-1}$ and approximated by $r_i^{-\\alpha}/\\sum_{j\\ne b_0} r_{j,i}^{-\\alpha}$ over $\\epsilon_i$ for small $\\epsilon_i$. This scalar converts an intractable instantaneous interference topology into one number per user: the average desired-signal-to-average-interference power ratio divided by the target error probability. For the sum-rate objective the argument is carried by the arithmetic-geometric-mean identity (Proposition 1), which replaces the product $(1+\\gamma_1)(1+\\gamma_2)$ by $\\left[1+(\\gamma_1+\\gamma_2)/2\\right]^2$; because the resulting $\\bar\\gamma=\\gamma_1+\\gamma_2$ is convex in $\\beta$, the maximum sits at an endpoint, $\\beta=1/2$ or $\\beta=1$, which yields the closed-form threshold (22) and the NOMA-versus-OMA condition (24).","core_discovery":"The central claim is that, in a Poisson-deployed multi-cell downlink with Rayleigh fading and interference-limited operation, reliability-constrained NOMA rate control reduces to ranking and power-splitting based on $\\phi_i^*$, the SIR threshold user $i$ would need if it occupied the channel alone. For target error probabilities $\\epsilon_i \\le 10^{-1}$, $\\phi_i^*$ is accurately the average desired-to-average-interference power ratio divided by $\\epsilon_i$ (Eq. (8)). The paper proves that the optimal decoding order is $\\phi_2^* \\ge \\phi_1^*$: the user with the largest $\\epsilon_i$ times the average power ratio should be decoded first. With that ordering, the power split $\\beta$ that makes the two users' transmission rates equal is Eq. (12), and the split that maximizes the approximately evaluated sum rate is Eq. (22). NOMA outperforms equal-resource OMA when the SIC residual fraction $\\mu$ is below Eq. (14) for equal rates, and for sum-rate maximization it always wins at $\\mu=0$ and generically wins below Eq. (24).","pith_inferences":["This suggests a testable scheduling heuristic: pair users with strongly contrasted values of $\\epsilon_i \\times$ (average desired-to-interference power ratio), because these are exactly the pairs for which the NOMA-over-OMA gain is largest, while near-equal values favor OMA.","The exact-sum-rate behaviour could deviate from Eq. (22) if the AM-GM approximation changes the location of the optimum; a direct numerical check over a $(\\phi_1^*,\\phi_2^*,\\mu)$ grid would tell whether the closed-form thresholds need an error bound.","The same ranking-by-$\\phi_i^*$ device may extend to $M>2$ users via ordered SIC, but the paper's proofs cover only the two-user case; extending the distribution theorem and the thresholds to larger groups is an open problem.","Because the numerical gap to the infinite-blocklength, full-CSI benchmark grows as $\\epsilon$ shrinks, a finite-blocklength version of this distributed rule would likely need to add a channel-dispersion penalty to $\\gamma_i$; the paper leaves that extension implicit."],"forward_implications":["A serving base station can meet per-user error budgets with a two-user NOMA group using only long-term average received powers and the target error probabilities, with no CSI feedback from users.","The optimal decoding order is fixed by $\\phi_i^*$: decode first the user with the largest $\\epsilon_i$ times the average desired-to-interference power ratio, i.e., the smallest $\\phi_i^*$.","With equal-rate targets, NOMA beats equal-partition OMA exactly for $\\mu$ below the closed-form bound in Eq. (14).","With sum-rate maximization, NOMA always beats that OMA baseline at $\\mu=0$ and, under the approximation, beats it almost surely for $\\mu$ below Eq. (24).","The average allocated rate under this rule is independent of the base-station density $\\lambda$, so the same rate-control calculation remains valid as the network densifies."],"supporting_citations":[{"why":"Introduces distributed rate control for high reliability in Poisson networks, the method this paper adapts from OMA to NOMA.","marker":"[24]"},{"why":"The authors' earlier OMA downlink rate-control scheme under finite blocklength that the proposed NOMA allocation extends.","marker":"[25]"},{"why":"Their joint power-control and rate-allocation work for OMA/SIMO networks, the direct predecessor of this formulation.","marker":"[26]"},{"why":"Supplies the PPP-based NOMA system model and the full-CSI decoding-order baseline that the topological rule is compared against.","marker":"[21]"},{"why":"Models SIC error propagation through the residual fraction µ used throughout the power-allocation analysis.","marker":"[28]"},{"why":"Defines the infinite-blocklength benchmark whose rate set to SIR gives the ideal upper bound for the numerical comparisons.","marker":"[33]"},{"why":"Defines the squared relative distance process used in deriving the CDF of the allocated SIR threshold (Theorem 2).","marker":"[34]"},{"why":"Provides the SIR distribution for random wireless networks used to obtain the closed form of the threshold CDF.","marker":"[35]"},{"why":"Shows the SIR meta-distribution is independent of network density, which Theorem 2's density-free CDF relies on.","marker":"[30]"},{"why":"Provides the nearest-neighbor distance distribution used in the Poisson-network derivation of the rate-threshold distribution.","marker":"[27]"}],"fun_headline_variants":["NOMA beats OMA when SIC residual stays below limit","Order NOMA users by error budget times power ratio","Reliability-driven NOMA needs only averages, not CSI","NOMA wins if co-interference is canceled efficiently","Two-user NOMA: decode highest error-power product first"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"For the sum-rate claims, the paper replaces the true objective $(1+\\gamma_1)(1+\\gamma_2)$ with the arithmetic-geometric-mean upper bound $\\left[1+(\\gamma_1+\\gamma_2)/2\\right]^2$ and does not prove that maximizing the bound gives the same power split or the same NOMA-versus-OMA comparison as maximizing the true objective.","fun_headline_variants_meta":{"raw":{"variants":["NOMA beats OMA when SIC residual stays below limit","Order NOMA users by error budget times power ratio","Reliability-driven NOMA needs only averages, not CSI","NOMA wins if co-interference is canceled efficiently","Two-user NOMA: decode highest error-power product first"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000684,"raw_usage":{"total_tokens":3134,"prompt_tokens":1003,"completion_tokens":2131,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":619,"completion_tokens_details":{"reasoning_tokens":2051}},"tokens_in":619,"tokens_out":2131,"duration_ms":16726,"temperature":1.0,"reasoning_tokens":2051,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:14:05.616457+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Numerically maximize the exact sum-rate objective $(1+\\gamma_1)(1+\\gamma_2)$ over $\\beta \\in [1/2,1]$ for a grid of $(\\phi_1^*,\\phi_2^*,\\mu)$ values, and compare the maximizing $\\beta$ and the sign of the comparison with OMA against Eqs. (22) and (24); any feasible grid point with an interior optimum or a sign mismatch would refute the universal optimality claim.","supporting_citations":[{"cited_title":"Distributed rate contro l for high reliability in poisson bipolar networks,","cited_arxiv_id":null,"evidence_quote":"Introduces distributed rate control for high reliability in Poisson networks, the method this paper adapts from OMA to NOMA."},{"cited_title":"Rate contro l under finite blocklength for downlink cellular networks with reliabili ty constraints,","cited_arxiv_id":null,"evidence_quote":"The authors' earlier OMA downlink rate-control scheme under finite blocklength that the proposed NOMA allocation extends."},{"cited_title":"Joint power control and rate allocation enabling ultra-reliability and energy effi ciency in SIMO wireless networks,","cited_arxiv_id":null,"evidence_quote":"Their joint power-control and rate-allocation work for OMA/SIMO networks, the direct predecessor of this formulation."},{"cited_title":"Stochastic geome try based performance study on 5G non-orthogonal multiple access sch eme,","cited_arxiv_id":null,"evidence_quote":"Supplies the PPP-based NOMA system model and the full-CSI decoding-order baseline that the topological rule is compared against."},{"cited_title":"Non-orthogonal multi ple access with SIC error propagation in downlink wireless MIMO networ ks,","cited_arxiv_id":null,"evidence_quote":"Models SIC error propagation through the residual fraction µ used throughout the power-allocation analysis."},{"cited_title":"Channel coding rate in the finite blocklength regime,","cited_arxiv_id":null,"evidence_quote":"Defines the infinite-blocklength benchmark whose rate set to SIR gives the ideal upper bound for the numerical comparisons."},{"cited_title":"SIR asymptotics in poisson c ellular networks without fading and with partial fading,","cited_arxiv_id":null,"evidence_quote":"Defines the squared relative distance process used in deriving the CDF of the allocated SIR threshold (Theorem 2)."},{"cited_title":"The performance of successive interfer- ence cancellation in random wireless networks,","cited_arxiv_id":null,"evidence_quote":"Provides the SIR distribution for random wireless networks used to obtain the closed form of the threshold CDF."},{"cited_title":"The meta distribution of the SIR in Poisson bipolar and cellular networks,","cited_arxiv_id":null,"evidence_quote":"Shows the SIR meta-distribution is independent of network density, which Theorem 2's density-free CDF relies on."},{"cited_title":"On distances in uniformly random networks ,","cited_arxiv_id":null,"evidence_quote":"Provides the nearest-neighbor distance distribution used in the Poisson-network derivation of the rate-threshold distribution."}],"review_version":1}