{"id":"77ccf5af-4855-4dbd-a316-74e54fa2dca2","arxiv_id":"1908.08866","paper_version":2,"verdict":"REJECT","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":4,"one_line_summary":"The authors propose a two-step channel and power allocation heuristic for underlay D2D multicasts and show via simulation that it outperforms several existing schemes in sum throughput.","lead":"This paper designs radio resource allocation schemes for device-to-device (D2D) multicast groups sharing cellular uplink channels, aiming to maximize total throughput while protecting cellular users' quality of service. The work is a candidate building block for 5G local content distribution, where multiple nearby users request the same data.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 5's quasi-convexity claim is not established and is likely false; the G_k=2 corner-search optimality therefore remains an uncalibrated heuristic.","rationale":"The reader's weakest assumption targets exactly the same step: Lemma 5's quasi-convexity assertion underpinning the G_k=2 corner-search optimality. I agree that this is the most load-bearing concern. The proof in Appendix G is invalid for concrete, identifiable reasons: (G.1) is not a correct characterization of quasi-convexity, and the sum of quasi-convex functions need not be quasi-convex. The final step of the proof is circular, since it derives the assumption f(p')≤f(p) rather than a consequence of it. The paper's own language in Section V.B calls the corner search an 'approximate solution', which further weakens the optimality claim. This matters because the general STIM power allocation in Section V.C only drives powers toward SINR feasibility; it does not maximize sum throughput. Thus the special-case corner optimality was the only formal basis for the abstract's claim that the algorithm maximizes throughput. The proposed global-optimization check would settle whether the corner-search result happens to be true for typical instances; absent such a check or a corrected proof, the manuscript's central claim is unsupported. The paper may still describe a useful heuristic, and the numerical comparisons can stand as empirical observations, but the verdict of reject-as-written is appropriate.","tokens_in":24338,"tokens_out":15527,"duration_ms":149967,"concrete_test":"Take the numerical instance in Appendix H (or generate 100 random feasible instances with C=1, G=2 and the paper's log-normal channel model). Enumerate the feasible polytope from inequalities (16a)-(16c) and the box power constraints using a dense grid or a global optimizer, and compare the best sum throughput against all corner points listed in Appendix H. If any feasible interior or edge point exceeds every corner, Lemma 5 is false and the corner-search claim must be downgraded to heuristic. If no counterexample appears, the test should be repeated with random instances and the result reported as an empirical property, not a proof.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The strongest formal claim for the two-MG case rests on Lemma 5 (Section V.B, Appendix G), which asserts that the sum-rate function is quasi-convex on the boundary of the feasible power region, so the maximum lies at corner points. Appendix G does not prove this. It states condition (G.1) as if it were the definition of quasi-convexity, but (G.1) is not a valid characterization for quasi-convex functions, and the subsequent argument assumes that a sum of quasi-convex functions is quasi-convex, which is false in general. The objective here is a sum of three log(1+SINR) terms, each of which is quasi-convex in the power vector, but quasi-convexity is not preserved under addition. The derivation then essentially restates the desired inequality (G.5) rather than establishing it. Consequently, there is no demonstrated reason that the maximum over the polytope defined by inequalities (16a)-(16c) and the power bounds occurs at a corner. The corner-search power allocation is therefore unproven as an optimal algorithm; if it is only a heuristic, the paper's strongest claim about the two-MG special case collapses.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies uplink underlay D2D multicast in a single cell, where multiple multicast groups may share a cellular user's channel. It formulates a sum-throughput maximization problem as a mixed-integer nonlinear program (P1), then proposes a two-stage solution: channel allocation algorithms (interference-aware IA-STIM and outage-aware OA-STIM) followed by power allocation. For the special cases of one or two multicast groups per channel, it claims optimal corner-based power allocation; for the general case, it proposes an iterative STIM power-control algorithm. The numerical section compares the proposed schemes with bipartite matching, random allocation, and greedy heuristics, reporting higher sum throughput under the chosen scenarios.","tokens_in":24569,"tokens_out":6559,"duration_ms":74446,"significance":"If the optimality claims were correct, the paper would provide a useful decomposition of a hard MINLP into a polynomial-time channel allocation step and provably optimal power allocation for small channel-sharing groups, together with a stochastic-geometry outage expression and extensive simulations. The paper's strengths include a clearly stated system model, a standard outage derivation in Lemma 1, and a broad set of numerical comparisons. However, the central load-bearing proofs for the claimed optimality of the corner-search power allocation are invalid, and the general STIM algorithm does not optimize the stated objective. These issues undermine the paper's main theoretical contribution, so the current manuscript does not meet the standard for publication.","major_comments":[{"comment":"The proof of Lemma 5 does not establish that the sum-rate function is quasi-convex on the boundary of the feasible power region. Appendix G first invokes a first-order condition (G.1) that is, in fact, a valid characterization for differentiable quasi-convex functions, but the proof then makes two unjustified steps: it assumes that a sum of quasi-convex functions is quasi-convex, which is false in general, and it analyzes a sum of SINR ratios rather than the actual objective, which is a sum of log(1+SINR) terms. The derivation from (G.4) to (G.5) essentially restates the target inequality without proving it. Consequently, the claim that the maximum of the sum rate over the polytope defined by (16a)–(16c) and the power bounds occurs at a corner point is unproven. This is load-bearing because the corner-search algorithm for Gk=2 is presented as optimal. The paper's own Section VI.B later calls corner search 'a good heuristic' and 'a lower bound to the optimal throughput,' which directly contradicts Lemma 5 and the abstract's optimality claim.","section":"Section V.B, Lemma 5 and Appendix G"},{"comment":"The proofs of Lemmas 2 and 3, which assert optimality at the corners for the Gk=1 case, are invalid as written. Appendix C scales both transmit powers by α>1 and claims C(αP_i,αP_j) > C(P_i,P_j). In an interference-limited setting (which the paper assumes in Proposition 1 and Appendix A), a common scaling of both powers leaves all SINRs invariant up to the noise term, so the displayed expressions for β1 and β2 do not imply the claimed inequality. The subsequent argument then asserts convexity of a product-like quantity T(P_i,P_j) in Appendix D without deriving it from the actual sum-rate objective. Thus the corner-optimality result for the single-MG case is not proven by the material supplied.","section":"Section V.A, Lemmas 2 and 3, Appendices C and D"},{"comment":"The STIM power-allocation algorithm for the general case does not maximize the throughput objective of Problem P1. Equation (17) merely caps each MGTX's power by an evenly split interference budget, and the update rule (18)–(19) iterates toward meeting SINR thresholds. No objective function appears in this procedure, so there is no mechanism by which it 'maximizes the system throughput,' as claimed in the abstract and in Section I.B. The numerical results show that the proposed scheme outperforms the chosen baselines under certain parameters, but outperforming heuristics does not establish throughput optimality or provide a performance guarantee relative to the optimum of P1.","section":"Section V.C, Algorithm 3, and equations (17)–(19)"},{"comment":"The manuscript is internally inconsistent about the status of the corner-search power allocation. Section V.B and the abstract present it as optimal for Gk=2, while Section VI.B states that the corner-search method 'tries to support minimum SINR thresholds' and 'may serve as a good heuristic to efficiently provide a lower bound to the optimal throughput.' A heuristic lower bound is not an optimal solution. The authors need to either provide a valid proof of optimality or explicitly reframe all such claims as heuristic, with appropriate empirical validation against exhaustive search for small instances.","section":"Section VI.B, Figure 4 and surrounding text"}],"minor_comments":[{"comment":"The pseudocode contains notation that is difficult to follow: in Algorithm 1, the set G_k of multicast groups sharing channel k is updated using objects that appear to be MGs, while in Algorithm 2 the update 'G_k = G_k ∪ k' mixes a set of MGs with a CU index k. These lines should be rewritten with explicit indices so that the set-membership operations are unambiguous.","section":"Section IV, Algorithms 1 and 2"},{"comment":"The phrase 'γth_r = γth_{r'} = γth_c = 3 watt or 5 dB' is dimensionally inconsistent: 3 watts is a linear power value and 5 dB is a logarithmic ratio; they cannot be equal. The intended units and conversion should be stated correctly.","section":"Appendix H, first paragraph"},{"comment":"The outage probability formula in (10) uses intensities λ_c and λ_g, but these are not defined in the system model in Section II. The relation of these intensities to the PPP models of CUs and MGs should be stated explicitly before Lemma 1 is used.","section":"Section II.B and Lemma 1 (Appendix B)"},{"comment":"In the Gk=2 SINR constraints, the receivers r and r' are introduced without specifying that they are the worst-case receivers of MG 1 and MG 2, respectively; this should be clarified to match the definition in (3).","section":"Section II.C, equations (12)–(14)"},{"comment":"The novelty claim 'for the first time considers the general problem of optimal resource allocation for multiple D2D multicasts' should be tempered or qualified, since references [10], [28], and [29] address closely related underlay D2D multicast resource allocation, and the authors should state explicitly what is new beyond those works and beyond their own conference papers (WCNC 2016/2017) and preprint [32].","section":"Section I.B and Section IV"}],"recommendation":"reject","confidential_remarks":"The paper is a journal submission that appears to build on two IEEE WCNC conference papers. The formal optimality claims for the special cases rest on invalid proofs, and the text itself retreats to calling the corner-search method a heuristic. The general STIM algorithm is a feasible SINR-based power control rather than a throughput optimizer. These are load-bearing issues that would require either sound new proofs or a substantial reframing of the contributions with exhaustive-search validation; under the journal's standards, I recommend rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a plausible engineering paper whose formal optimality claims don't survive contact with the appendices. The reader's report is right about the proofs, and the stress-test on Lemma 5 hits the load-bearing flaw.\n\nWhat's actually new: the formulation lets an arbitrary number of D2D multicast groups share one CU uplink channel under individual QoS constraints, and the paper proposes a two-step channel/power allocation framework (IA-STIM and OA-STIM). That is a reasonable extension of existing D2D resource allocation work, and the simulations show consistent gains over the baselines. The Hungarian matching for the single-MG case is standard but correctly applied; the corner-search idea for two MGs is a useful practical heuristic.\n\nThe soft spots are concentrated in the proofs. Appendix G is the serious one. Lemma 5 asserts the sum-rate function is quasi-convex on the feasible boundary so the maximum is at corners, but the proof uses condition (G.1) as if it characterized quasi-convexity, then assumes quasi-convexity is preserved under addition. Both steps are false: the individual SINR terms are quasi-convex, but sums of quasi-convex functions need not be. So the corner-search result is unproven as an optimality claim. The paper sometimes calls it an approximate solution, which is the honest description, but the abstract and Lemma 5 say more.\n\nAppendix C has a scaling error: multiplying both powers by alpha > 1 can leave the feasible box, and the rate comparison ignores that both signal and interference scale together. Appendix A's bound for Proposition 1 drops the noise term and then replaces a sum of logs of sums with a sum of logs of maxima; the inequality direction doesn't follow from the text. None of these kill the heuristic, but they kill the claim that the proposed power allocation maximizes system throughput.\n\nOne additional concern: the outage-aware allocation relies on a PPP-based interference formula (Lemma 1), but the simulation model is a fixed finite cell with uniformly distributed nodes. That formula is not obviously valid for the simulated topology, so the OA-STIM numerical results are shakier than the presentation suggests. The citation pattern is fine; the self-citation to the authors' earlier exclusion-zone paper is not load-bearing.\n\nBottom line: readers building practical D2D multicast resource allocation systems will find useful algorithms and benchmarks here. The paper deserves a serious referee, but it needs major revision: either the optimality claims must be properly proven, or they must be honestly downgraded to heuristics throughout, including the abstract.","headline":"A useful D2D multicast heuristic with unsupported optimality claims; the Gk=2 corner-search proof fails, but the two-step scheme is worth refereeing as a major revision.","tokens_in":25097,"tokens_out":2094,"would_cite":false,"duration_ms":20846,"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":"This paper claims that splitting the joint resource allocation problem into channel assignment and power control lets multiple D2D multicast groups share a cellular uplink while maximizing sum throughput under QoS and power limits.","keywords":["device-to-device multicast","underlay cellular networks","channel allocation","power allocation","sum throughput maximization","MINLP","QoS constraints","interference management"],"falsifier":"Pick a random channel realization with two multicast groups and one cellular user, compute the sum rate on a fine grid over the boundary faces of the feasible power region, and compare the best interior boundary point with the best of the seven corner-region candidates from Section V-B; if any interior boundary point beats the corner candidates, Lemma 5's quasi-convexity is false and the corner-search optimality claim fails.","tokens_in":24103,"feed_emoji":"📶","tokens_out":8606,"duration_ms":77446,"temperature":0.7,"pith_summary":"Underlay device-to-device (D2D) multicast lets nearby phones broadcast popular content directly to a group while reusing a cellular uplink channel, but the mutual interference between cellular users and multicast groups can erase the throughput gain. The paper studies the general setting in which any number of multicast groups can share any cellular channel, and tries to maximize total system throughput under per-user QoS constraints and per-transmitter power caps. Because the joint problem is a mixed-integer nonlinear program and therefore computationally intractable in general, the authors split it into a channel-allocation stage and a power-allocation stage. They claim that their two-step schemes—interference-aware and outage-aware channel allocation, each followed by the STIM power-allocation algorithm—outperform existing random, greedy, and bipartite-matching resource allocations in sum throughput, and that for one or two multicast groups per channel the power-allocation rules are optimal or near-optimal. If these claims hold, operators can support multicast video and file distribution in dense cells without sacrificing cellular QoS.","feed_headline":"Split channel and power allocation to maximize D2D multicast throughput","feed_subtitle":"Two-stage scheme lets multiple groups share uplinks while holding QoS and power limits, beating random and greedy baselines.","key_machinery":"The load-bearing object is the geometry of the feasible power region for a channel shared by a CU and one or two multicast groups. For one group per channel, Lemmas 2 and 3 place the optimum on the boundary of the region, at a corner where either the CU or the group transmitter transmits at maximum power, making bipartite matching plus a corner check sufficient. For two groups per channel, Lemma 5 asserts that the sum-rate function is quasi-convex on the boundary of the feasible region, so the maximum lies among the corner points generated by intersecting the three SINR planes (equations (16a)-(16c)) with the faces of the power cube; the paper lists seven candidate regions and evaluates those points. For the general many-group case, STIM replaces geometric corner search with a fixed-point iteration adapted from the GDCPC algorithm, updating each transmitter's power by the ratio of its target SINR to its current SINR and capping interference to the shared CU. This combination reduces an NP-hard MINLP to a polynomial-time channel assignment followed by a small corner search or an iterative power update.","core_discovery":"The central claim is that the intractable joint channel-and-power allocation problem for multiple D2D multicast groups sharing uplinks can be decomposed without losing much throughput. For each channel, the base station first admits the set of multicast groups that will be allowed to share that channel, using either an interference-aware rule (keep mutually interfering groups apart, admit only groups that add positive throughput) or an outage-aware rule (choose channels that minimize outage probabilities subject to an aggregate interference budget). The STIM power-allocation stage then updates each group transmitter's power through a modified GDCPC iteration so that SINR targets are met and the shared CU is protected. For the special cases of one or two groups per channel, the paper proves that the optimal power vector lies on the boundary of the feasible power region and, for two groups, that the sum-rate function is quasi-convex on that boundary, so the maximum can be located by checking only corner points. The numerical section concludes that the proposed schemes outperform existing resource allocation schemes in sum throughput.","pith_inferences":["A natural testable extension is to run the same two-step scheme in a multi-cell layout with inter-cell interference, since the paper only treats a single cell and lists multi-cell operation as future work.","Because Lemma 5's proof simplifies the SINR expression to an interference-limited form and drops the noise term at one step, one can stress-test it numerically by fine-grid search on the boundary; if a non-corner boundary point wins, the corner-search claim is only heuristic.","The simulation peak near 15 dBm suggests a system-design rule the authors do not state: even when hardware maximum power is higher, capping D2D multicast transmitters near that level can maximize sum throughput by limiting co-channel interference."],"forward_implications":["For exactly one multicast group per channel, the optimal power allocation is the corner solution used with bipartite matching, so no continuous power search is needed at all.","For exactly two groups per channel, testing the seven listed corner candidates gives the maximum sum rate, provided the quasi-convexity premise holds; this makes the two-group case computationally cheap.","The STIM iteration makes the general problem tractable and, in simulations, yields sum throughput that first increases and then decreases with the maximum D2D transmit power, with the best point near 15 dBm in the tested configuration.","As the number of multicast groups grows, sum throughput saturates once co-channel interference blocks further admission, and both IA-STIM and OA-STIM beat the random, greedy, and bipartite baselines in most simulated regimes.","Higher CU QoS thresholds reduce the room for D2D sharing, so the achievable gain shrinks as the CU rate requirement rises; the proposed schemes still outperform the baselines across the reported range."],"supporting_citations":[{"why":"Establishes that the joint problem is an NP-hard MINLP, motivating the decomposition into channel and power allocation.","marker":"[39]"},{"why":"Provides the distributed constrained power-control algorithm that the STIM power update modifies.","marker":"[42]"},{"why":"Supplies the bipartite-graph and random channel allocation schemes used as performance baselines.","marker":"[43]"},{"why":"Supplies the greedy CU-MG pairing baseline used in the numerical comparisons.","marker":"[44]"},{"why":"Provides the Hungarian algorithm used to solve the one-MG-per-channel matching problem optimally.","marker":"[41]"},{"why":"Provides the bipartite matching algorithm on which the single-MG channel allocation is built.","marker":"[40]"},{"why":"Formulates the earlier underlay D2D multicast resource allocation with QoS protection that this paper generalizes to multiple MGs per channel.","marker":"[10]"},{"why":"Justifies the minimum-rate QoS constraint on CUs used in the formulation.","marker":"[38]"}],"fun_headline_variants":["Two-step allocation maximizes D2D multicast throughput in underlay networks","Split channel-power allocation enables multiple D2D multicasts on one uplink","Interference-aware channel assignment plus STIM power improves D2D multicast throughput","Two-stage scheme: admit multicast groups, then allocate power to maximize throughput","Underlay D2D multicast: split channel and power to maximize sum rate"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The claim that the corner search is optimal for two multicast groups per channel rests on the sum-rate function being quasi-convex on the boundary of the feasible power region; if that property fails on realistic channel gains, the corner search is only an uncalibrated heuristic and the paper's strongest optimality claim collapses.","fun_headline_variants_meta":{"raw":{"variants":["Two-step allocation maximizes D2D multicast throughput in underlay networks","Split channel-power allocation enables multiple D2D multicasts on one uplink","Interference-aware channel assignment plus STIM power improves D2D multicast throughput","Two-stage scheme: admit multicast groups, then allocate power to maximize throughput","Underlay D2D multicast: split channel and power to maximize sum rate"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001306,"raw_usage":{"total_tokens":5365,"prompt_tokens":1027,"completion_tokens":4338,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":643,"completion_tokens_details":{"reasoning_tokens":4240}},"tokens_in":643,"tokens_out":4338,"duration_ms":24364,"temperature":1.0,"reasoning_tokens":4240,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:27:53.154413+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Pick a random channel realization with two multicast groups and one cellular user, compute the sum rate on a fine grid over the boundary faces of the feasible power region, and compare the best interior boundary point with the best of the seven corner-region candidates from Section V-B; if any interior boundary point beats the corner candidates, Lemma 5's quasi-convexity is false and the corner-search optimality claim fails.","supporting_citations":[{"cited_title":"On the computational complexity of integer programming problems,","cited_arxiv_id":null,"evidence_quote":"Establishes that the joint problem is an NP-hard MINLP, motivating the decomposition into channel and power allocation."},{"cited_title":"Autonomous distributed power control for cognitive radio networks,","cited_arxiv_id":null,"evidence_quote":"Provides the distributed constrained power-control algorithm that the STIM power update modifies."},{"cited_title":"Device-to-device communications underlaying cellular networks,","cited_arxiv_id":null,"evidence_quote":"Supplies the bipartite-graph and random channel allocation schemes used as performance baselines."},{"cited_title":"Efficient resource allocation for device-to-device communication underlaying LTE network,","cited_arxiv_id":null,"evidence_quote":"Supplies the greedy CU-MG pairing baseline used in the numerical comparisons."},{"cited_title":"The Hungarian method for the assignment problem,","cited_arxiv_id":null,"evidence_quote":"Provides the Hungarian algorithm used to solve the one-MG-per-channel matching problem optimally."},{"cited_title":"An nˆ5/2 algorithm for maximum matchings in bipartite graphs,","cited_arxiv_id":null,"evidence_quote":"Provides the bipartite matching algorithm on which the single-MG channel allocation is built."},{"cited_title":"A resource allocation scheme for D2D multicast with QoS protection in OFDMA-based systems,","cited_arxiv_id":null,"evidence_quote":"Formulates the earlier underlay D2D multicast resource allocation with QoS protection that this paper generalizes to multiple MGs per channel."},{"cited_title":"Resource allocation under channel uncertainties for relay-aided device-to-device communication underlaying LTE-A cellular networks,","cited_arxiv_id":null,"evidence_quote":"Justifies the minimum-rate QoS constraint on CUs used in the formulation."}],"review_version":1}