{"id":"93903a2f-7b07-4a3c-9093-728e20dbaaaf","arxiv_id":"2502.04827","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Uplink RSMA with finite-blocklength coding is shown to improve the successful computation probability and reduce latency in a two-user mobile edge computing system compared to NOMA.","lead":"This paper combines rate-splitting multiple access (RSMA) with short-packet transmission for mobile edge computing, where two users offload computation tasks to an edge server. The authors derive the successful computation probability and an optimization algorithm, then show in simulation that RSMA beats NOMA in reliability and latency.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"RSMA-vs-NOMA gains rest on an unverified surrogate objective: the linearized SCP and Chernoff bound are used without enforcing the small-error regime, and the convergence proof is directionally inconsistent.","rationale":"The reader's weakest assumption identifies the same core issue: the approximations leading to Eq. (18) and Eq. (21) are used without verifying their validity regimes. I agree that this is the most load-bearing point because the numerical RSMA-versus-NOMA comparison is generated by optimizing that surrogate, not the true SCP. The reader's rationale also flags the directionally inconsistent convergence proof, which reinforces the concern that the reported points may not even be optimized for the surrogate. I do not move the verdict because the qualitative claim has independent structural support: Remark 1 correctly notes that NOMA is a feasible point of RSMA, so an exact global optimization would be expected to favor RSMA. However, the paper's evidence for the magnitude of the gain and the latency reduction is quantitative and depends on the surrogate optimization. The proposed exact-SCP grid check would settle whether the reported gain is real or an artifact. This is a significant but not fatal concern, so the conditional verdict remains appropriate pending verification.","tokens_in":19072,"tokens_out":13181,"duration_ms":150393,"concrete_test":"Recompute the Fig. 2(b) operating point (M1 = 6k bits, M2 = 5.5k bits, N = 500, SNR = 15 dB) by evaluating the exact SCP, Ps = (1 − ε1)(1 − ε2) with the Q-function from Eq. (12), for both the Algorithm 1 solution and a fine grid over (λ1, λ2, β, P1,1, P1,2) with all constraints (17b)–(17e) enforced. If the grid yields a higher exact SCP than Algorithm 1, or if the exact-SCP RSMA advantage over NOMA changes sign or falls below 0.05, the reported gain is an artifact of the surrogate rather than a true SCP improvement.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is supported only by simulations that optimize the surrogate objective in Eq. (21), i.e. a Chernoff upper bound of the first-order expansion in Eq. (18) of the true SCP in Eq. (15). The surrogate is accurate only when per-stream error probabilities are small enough that products such as εaεb are negligible and when f_i ≥ 0.5 so that the Chernoff bound in Eq. (20) holds. Neither condition is checked or enforced by the algorithm. In the operating region highlighted in Fig. 2(b), SCP ≈ 0.6–0.8 implies per-stream errors on the order of 0.1–0.3, so the dropped product terms contribute roughly 0.01–0.1 to Ps; an error of 0.1 is comparable to the reported RSMA–NOMA gap at several points in Figs. 2–4. Moreover, the convergence proof in Sec. IV-D concludes that the objective sequence is non-decreasing in Eq. (45) for a minimization problem, which is the wrong direction; if the sequence is allowed to increase, the points labeled \"optimized\" need not be local minima of even the surrogate. Because the same flawed surrogate is used for both schemes, the qualitative ordering could survive, but the numerical magnitudes and the latency comparisons in Figs. 2–4 are not established.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a two-user uplink rate-splitting multiple access (RSMA) system in which users offload computation tasks to a mobile edge computing (MEC) server under finite-blocklength (FBL) constraints. It derives an expression for the successful computation probability (SCP), which is the probability that all offloaded and local tasks are completed within a delay budget, and then formulates a joint optimization of the offloading factor, transmit power allocation, and task-splitting factor. The problem is solved by alternating optimization (AO) with successive convex approximation (SCA) for the power and splitting subproblems, and a closed-form offloading factor is proposed. Numerical results compare RSMA with non-orthogonal multiple access (NOMA) and report that RSMA achieves a higher SCP and can reach the same SCP with shorter blocklengths, i.e., lower latency.","tokens_in":19359,"tokens_out":14016,"duration_ms":150631,"significance":"If the numerical claims are validated, the paper provides a useful extension of RSMA to short-packet mobile edge computing, combining a finite-blocklength reliability analysis with an iterative resource-allocation algorithm. A strength of the paper is that the system model and approximations are explicit, and the algorithm is described in sufficient detail to be reproduced. However, because NOMA is a special case of RSMA (Remark 1), the qualitative ranking RSMA-vs-NOMA is structurally expected; the contribution rests on the quantitative SCP analysis and on the reliability of the optimization. The main risks are the unvalidated surrogate objective used in the optimization and the direction of the convergence proof, both of which affect the reported numerical gains.","major_comments":[{"comment":"The optimization does not directly maximize the true SCP in Eq. (15) but a linearized surrogate 1-(2εa+2εb+εc) in Eq. (18), after dropping all product terms such as εaεb and ε1ε2. The paper does not verify that these terms are negligible in the operating region of Figs. 2-4. In the highlighted point of Fig. 2(b) (SCP≈0.6-0.8), per-stream error probabilities are of order 0.1-0.3, so the dropped product terms contribute roughly 0.01-0.1 to Ps, which is comparable to the reported RSMA-NOMA gap. Please quantify the approximation error, state explicitly whether the plotted SCP curves are computed from the true expression (15)-(16) using the optimized variables or from the surrogate (18)/(21), and if the latter, recompute the curves with the true objective.","section":"Section III-A, Eqs. (14)-(18), and Section V"},{"comment":"The transformation from Problem (19) to Problem (21) uses the bound exp(-x^2/2) ≥ Q(x) for x≥0.5, but the algorithm never enforces f(γi,Mi) ≥ 0.5 (or even f≥0). When f(γi,Mi)<0, the inequality reverses, and exp(-f^2/2) can be much smaller than Q(f); minimizing the exponential then rewards large |f| regardless of sign, potentially favoring negative f with large true error. The paper reports operating points with SCP=0, e.g., Fig. 2(a), which can correspond to f<0 for some streams. In those regimes Problem (21) is not an upper bound on Problem (19), so the optimized powers and splitting factors are not guaranteed to minimize the true error. Please enforce the condition f_i≥0.5 as a constraint, or verify after optimization that it holds at the returned solutions and restrict the operating regime accordingly.","section":"Section IV, Eqs. (19)-(21)"},{"comment":"The convergence proof has the wrong monotonicity direction. Problem (21) and its subproblems are minimizations, so solving the nth subproblem should produce an objective value no larger than the value at the previous iterate; Eq. (43) asserts the opposite inequality. Consequently Eq. (45) establishes that the sequence of the quantity being minimized is non-decreasing, which is directionally inconsistent with the goal of minimizing the error and does not prove convergence to a local minimum. Please correct the inequality chain (it should be non-increasing for a minimization algorithm) and state the resulting stationarity claim for Algorithm 1.","section":"Section IV-D, Eqs. (43)-(45)"},{"comment":"The closed-form offloading factor in Eq. (25) relies on the claim that the objective is monotonically increasing in λ. The proof in the Appendix, around Eq. (47), assumes log(1+γi) - λiMi/N ≥ 0, i.e., that the FBL rate is not above the corresponding capacity. This is the same 'good' regime required by the Chernoff step; in the low-SCP regime this condition can fail and the derivative can change sign, so the closed-form λ* may not minimize the objective. In addition, Lemma 1 enforces only the local computation constraints (23a), not the MEC computation constraint (23b), so the returned λ can violate feasibility if the server computation time is the bottleneck. Please state the regime required for Lemma 1 and check constraint (23b) explicitly, or justify that the lower-bound λ always satisfies it in the simulated parameters.","section":"Section IV-A, Lemma 1 and Appendix"}],"minor_comments":[{"comment":"Problem (38) is written as a maximization of the objective 2exp(-ta)+2exp(-tb)+exp(-tc), whereas the equivalent transformed problem (42) is written as a minimization of the same objective. Please make the sign consistent.","section":"Section IV-C, Eq. (38) vs Eq. (42)"},{"comment":"In Eq. (34a), the first term on the left-hand side appears as 't1/N'; it should be t_i/N to match the derivation in Eqs. (28)-(30).","section":"Section IV-B, Eq. (34a)"},{"comment":"The last panel is labeled '(c) N=1000' but should be labeled '(d) N=1000'; panel (c) appears twice.","section":"Figure 2 caption"},{"comment":"The complexity claim uses X=K+1 as the number of variables, but Problem (37) includes the slack variables ρ, t, and t1 in addition to the powers m; please clarify how these auxiliary variables are counted in the SOCP complexity.","section":"Section IV-D, complexity discussion"},{"comment":"The simulations are averaged over only 100 random channel realizations. Since the quoted differences in SCP between RSMA and NOMA are of order 0.1, please provide confidence intervals or error bars, or increase the number of realizations.","section":"Section V"}],"recommendation":"major_revision","confidential_remarks":"The paper is an application of known RSMA principles to a new FBL-MEC setting. The qualitative RSMA-over-NOMA result is expected from Remark 1, so the main value lies in the quantitative finite-blocklength analysis and the optimization framework. The technical issues identified above---surrogate-objective accuracy, unenforced Chernoff condition, and the convergence proof direction---are load-bearing for the numerical claims but appear fixable within the scope of the paper. I would support considering a revised version after these points are addressed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nHere's my read. The paper is the first to combine uplink rate-splitting multiple access, finite-blocklength coding, and successful computation probability in a mobile edge computing system. The literature table supports that claim: prior RSMA-MEC work used infinite blocklength, and prior FBL-MEC work used NOMA. That is a genuine new slot, and the paper works through it in a mostly careful way: the system model is clear, the closed-form offloading factor in Lemma 1 is a useful simplification, and the SCA-based alternating optimization is a standard approach for the coupled problem.\n\nThe central qualitative claim—RSMA outperforms NOMA—is structurally unsurprising because NOMA is a special case of RSMA (Remark 1). An ideal global optimizer would always do at least as well with RSMA, so the simulations showing gains are nice but not the hard part.\n\nThe soft spots are real. The convergence proof in Section IV-D shows the objective sequence is non-decreasing for a minimization problem. That is the wrong direction; as written it does not establish convergence to a local minimum, and the points labeled \"optimized\" might just be feasible points with worse objective values. Separately, the SCP approximation in (14) drops product terms like epsilon_a*epsilon_b, and the Chernoff bound in (20) requires f(gamma_i, M_i) >= 0.5. Neither condition is checked or enforced. The paper's own operating region in Fig. 2(b) has SCP around 0.6-0.8, implying per-stream errors of 0.1-0.3, where the dropped products and the Chernoff relaxation can contribute errors on the same order as the claimed RSMA-NOMA gap. So the numerical magnitudes and the latency comparisons in Figs. 2-4 are not yet established.\n\nI also note the absence of code and data, which makes the simulation claims hard to verify. That is common in this area, but it matters more when the proof has a sign error.\n\nWho is this for? Researchers working on RSMA, finite-blocklength analysis, or MEC offloading. I would send it to peer review—the application is new and the framework is likely to become a baseline—but it needs major revision before the quantitative claims are trustworthy. Specifically, fix the convergence proof, verify or enforce the small-error regime, and add reproducibility details.\n\nVerdict: conditional; major revision required.","headline":"A legitimate new combination of uplink RSMA + FBL + MEC, but the convergence proof has a sign error and the surrogates are unverified, so the reported gains are plausible yet not established.","tokens_in":19895,"tokens_out":4047,"would_cite":true,"duration_ms":41654,"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":"Uplink RSMA with one split stream achieves higher successful computation probability and lower latency than NOMA in short-packet mobile edge computing.","keywords":["rate-splitting multiple access","mobile edge computing","finite blocklength","short-packet communications","successful computation probability","NOMA comparison","task offloading","power allocation"],"falsifier":"Re-derive the SCP curves for the same channel realizations using the exact joint error probability $(1-\\varepsilon_a)(1-\\varepsilon_b)(1-\\varepsilon_c)$ instead of the first-order approximation, and check whether every optimized point satisfies the Chernoff-bound condition $x \\ge 0.5$; a visible gap or a violated condition would mean the reported RSMA gains do not correspond to the true maximum of the successful computation probability.","tokens_in":18821,"feed_emoji":"📡","tokens_out":7770,"duration_ms":67652,"temperature":0.7,"pith_summary":"This paper argues that uplink rate-splitting multiple access (RSMA) is a better multiple-access scheme than non-orthogonal multiple access (NOMA) for mobile edge computing (MEC) when task offloading uses short packets with finite blocklengths. It builds a two-user MEC model in which one user splits its task into two streams transmitted alongside the other user's unsplit stream, derives the Successful Computation Probability (SCP) under finite-blocklength constraints, and shows numerically that RSMA attains a higher SCP than NOMA and reaches the same SCP with a shorter blocklength, which translates to lower latency. The optimization jointly chooses how much of each task to offload, how to split one user's offloaded task, and how to allocate transmit power between the RSMA streams, using alternating optimization with successive convex approximation. If the claim holds, RSMA becomes a practical candidate for reliable low-latency edge offloading in future wireless systems.","feed_headline":"Splitting one user's stream boosts short-packet edge success over NOMA","feed_subtitle":"For short-packet mobile edge computing, RSMA matches NOMA's success probability with shorter blocks, lowering latency.","key_machinery":"The carrier of the argument is the two-user uplink RSMA structure in which user 1's task is split into two independently encoded streams, $s_{1,1}$ and $s_{1,2}$, while user 2 remains unsplit, decoded at the edge server in the order $s_{1,1} \\rightarrow s_2 \\rightarrow s_{1,2}$. This decoding order gives the first stream a manageable signal-to-interference-plus-noise ratio, lets the second user be decoded after partial interference cancellation, and finally decodes the interference-free remainder of user 1; the paper relies on a prior finite-blocklength analysis showing that this order yields much lower error probability than NOMA's fixed order. On top of this, the finite-blocklength achievable-rate expression $r \\approx C(\\gamma) - \\sqrt{V(\\gamma)/N}\\,Q^{-1}(\\epsilon)$ turns each stream's error probability into an explicit function of SINR, data size, and blocklength $N$, which makes the SCP an optimizable quantity. The optimization then uses the small-error approximation $1 - (2\\varepsilon_a + 2\\varepsilon_b + \\varepsilon_c)$ and the Chernoff bound $\\exp(-x^2/2) \\ge Q(x)$ to turn SCP maximization into minimizing a sum of exponentials, solved by alternating optimization with successive convex approximation using first-order Taylor surrogates for the non-convex SINR and data-size constraints.","core_discovery":"The central claim is that in an uplink RSMA-aided MEC system with finite-blocklength constraints, splitting one user's message into two streams and decoding them in the order $s_{1,1} \\rightarrow s_2 \\rightarrow s_{1,2}$ yields a higher Successful Computation Probability than conventional NOMA with decoding order $s_1 \\rightarrow s_2$, and achieves the same SCP at a shorter blocklength. The paper derives the SCP as the product of the successful offloading probability and the successful execution probability, approximates the offloading error probability using the finite-blocklength normal approximation with channel dispersion, and then maximizes a Chernoff-bound proxy of the SCP by alternately optimizing the offloading factor, the RSMA power allocation, and the task-splitting factor. Numerical results over Rayleigh fading show RSMA outperforming NOMA across blocklengths, task sizes, and signal-to-noise ratios, with the gain saturating at large blocklengths.","pith_inferences":["Because NOMA is a special case of RSMA in this setup (setting $P_{1,1}$ to $P_1$ or $0$), the reported gain can be attributed to the extra degree of freedom from splitting one user's stream; a natural extension is to quantify the gain from splitting both users or choosing which user to split adaptively per channel realization.","The Chernoff-bound proxy suggests the optimization may be conservative or mis-tuned in low-SCP regimes; optimizing the exact product form instead could change the shape of the RSMA-versus-NOMA gap in the very short-blocklength region.","The latency benefit implies a direct reliability-delay trade-off; a testable extension is to formulate a joint SCP-latency Pareto problem rather than fixing the time budget and minimizing blocklength.","The framework assumes perfect channel state information and no queueing at the edge server; extending to imperfect channel state information would likely shrink but not eliminate the RSMA advantage, since the SINR-balancing mechanism is the core of the gain."],"forward_implications":["In a two-user MEC offloading scenario with finite-blocklength links, RSMA with the $s_{1,1} \\rightarrow s_2 \\rightarrow s_{1,2}$ decoding order attains a higher SCP than NOMA at the same blocklength and signal-to-noise ratio.","RSMA reaches the same SCP as NOMA at a shorter blocklength, meaning lower offloading latency for the same reliability in short-packet edge computing.","The RSMA advantage grows with task size and saturates as blocklength becomes large, so the benefit is concentrated in the short-packet regime where MEC latency targets are tight.","The closed-form offloading factor $\\lambda_k^* = \\max\\{0,\\, 1 - (T - N T_s) f_{\\text{user}}/(M_k C_{\\text{cpu}})\\}$ determines how much each user must offload given the blocklength and time budget.","The proposed alternating optimization with successive convex approximation converges with complexity $O\\big((K{+}1)^{3.5}\\log(1/\\tau)\\big)$ and can be extended in principle to $K$ users and MIMO with revised SCP expressions."],"supporting_citations":[{"why":"Supplies the finite-blocklength achievable-rate approximation used in Eq. (11) to convert SINR, data size, and blocklength into per-stream error probabilities.","marker":"[33]"},{"why":"Provides the finite-blocklength analysis of uplink RSMA that motivates the decoding order $s_{1,1} \\rightarrow s_2 \\rightarrow s_{1,2}$ and its error-probability advantage over NOMA.","marker":"[26]"},{"why":"Introduces the Successful Computation Probability metric and the NOMA-aided MEC baseline that this paper compares against.","marker":"[19]"},{"why":"Prior RSMA-aided MEC work with SCP maximization, the direct predecessor that this paper extends to the finite-blocklength regime.","marker":"[31]"},{"why":"Frames RSMA as a 6G multiple access and motivates why NOMA's drawbacks make it less suitable for MEC networks.","marker":"[21]"},{"why":"Provides the uplink MIMO RSMA finite-blocklength design machinery referenced for the K-user MIMO extension.","marker":"[23]"}],"fun_headline_variants":["RSMA beats NOMA for short-packet edge offloading","Splitting streams improves short-packet MEC success","Rate splitting wins in finite-blocklength edge computing","Short-packet edge gains from RSMA over NOMA","Uplink RSMA speeds mobile edge task offloading"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole argument rests on the assumption that the per-stream decoding error probabilities $\\varepsilon_a,\\varepsilon_b,\\varepsilon_c$ stay small enough that products such as $\\varepsilon_a\\varepsilon_b$ can be dropped, and that the Chernoff bound's validity condition $x \\ge 0.5$ holds at every optimized point; if those fail in the low-success regimes the paper plots, the optimized decisions may not actually maximize the successful computation probability.","fun_headline_variants_meta":{"raw":{"variants":["RSMA beats NOMA for short-packet edge offloading","Splitting streams improves short-packet MEC success","Rate splitting wins in finite-blocklength edge computing","Short-packet edge gains from RSMA over NOMA","Uplink RSMA speeds mobile edge task offloading"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000744,"raw_usage":{"total_tokens":3375,"prompt_tokens":1059,"completion_tokens":2316,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":675,"completion_tokens_details":{"reasoning_tokens":2236}},"tokens_in":675,"tokens_out":2316,"duration_ms":17106,"temperature":1.0,"reasoning_tokens":2236,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T21:18:30.049534+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-derive the SCP curves for the same channel realizations using the exact joint error probability $(1-\\varepsilon_a)(1-\\varepsilon_b)(1-\\varepsilon_c)$ instead of the first-order approximation, and check whether every optimized point satisfies the Chernoff-bound condition $x \\ge 0.5$; a visible gap or a violated condition would mean the reported RSMA gains do not correspond to the true maximum of the successful computation probability.","supporting_citations":[{"cited_title":"Channel Coding Rate in the Finite Blocklength Regime,","cited_arxiv_id":null,"evidence_quote":"Supplies the finite-blocklength achievable-rate approximation used in Eq. (11) to convert SINR, data size, and blocklength into per-stream error probabilities."},{"cited_title":"Rate-Splitting Multi ple Access for Short-Packet Uplink Communications: A Finite Blocklength Analysis,","cited_arxiv_id":null,"evidence_quote":"Provides the finite-blocklength analysis of uplink RSMA that motivates the decoding order $s_{1,1} \\rightarrow s_2 \\rightarrow s_{1,2}$ and its error-probability advantage over NOMA."},{"cited_title":"Enhance Latency-Const rained Com- putation in MEC Networks Using Uplink NOMA,","cited_arxiv_id":null,"evidence_quote":"Introduces the Successful Computation Probability metric and the NOMA-aided MEC baseline that this paper compares against."},{"cited_title":"Rate-Splitting Multiple Access Aided Mobile Edge Computi ng With Randomly Deployed Users,","cited_arxiv_id":null,"evidence_quote":"Prior RSMA-aided MEC work with SCP maximization, the direct predecessor that this paper extends to the finite-blocklength regime."},{"cited_title":"Rate-Splitting Multiple Access: Fundamentals, Survey, a nd Future Research Trends,","cited_arxiv_id":null,"evidence_quote":"Frames RSMA as a 6G multiple access and motivates why NOMA's drawbacks make it less suitable for MEC networks."},{"cited_title":"Max-Min Fairness and PHY-Layer De sign of Uplink MIMO Rate-Splitting Multiple Access with Finite Blo cklength,","cited_arxiv_id":null,"evidence_quote":"Provides the uplink MIMO RSMA finite-blocklength design machinery referenced for the K-user MIMO extension."}],"review_version":1}