{"id":"c55a6fbc-3e9b-4453-9026-a19be4985286","arxiv_id":"2509.01201","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"New Markov-chain models for Wi-Fi 7 multi-link NSTR channel access predict per-device throughput in coexistence with legacy Wi-Fi, showing multi-link devices are heavily disadvantaged under saturation.","lead":"This paper builds mathematical models for how Wi-Fi 7 devices that use two radio links at once share channels with older single-link Wi-Fi devices. It finds that under heavy traffic, the multi-link devices are starved while legacy devices dominate.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Standards-compliance of the modeled NSTR channel-access rules is the deciding premise; the paper's own 'most conservative interpretation' caveat is untested against the actual 802.11be spec.","rationale":"The reader's weakest_assumption exactly matches my identified load-bearing concern: the modeled channel-access rules may not be actual standard-mandated behavior. The manuscript text even contains an explicit limitation—'most conservative interpretation of the standard'—but does not substantiate it with specific clauses from the 802.11be specification. This is the most serious threat because the central claim is not merely that the Markov chains are internally consistent, but that they describe real Wi-Fi 7 behavior. If the conservative interpretation is not the standard's actual requirement, then the X_AP, X_MLD, and Y probabilities, which govern every transition and every closed-form throughput, are irrelevant to real devices. The internal issues (undefined b3, tau_I_1 using tau_MLD_AP instead of tau_SLD_AP) and the large AP-MLD throughput discrepancies are important but secondary: they could be corrected or explained without changing the fundamental modeling approach. However, if the channel-access rules are wrong, no amount of algebraic repair can make the framework standards-compliant. Therefore, the test should settle this premise before accepting the framework. Since the reader already conditioned acceptance on addressing exactly this issue, my assessment does not change the verdict.","tokens_in":19697,"tokens_out":5509,"duration_ms":59060,"concrete_test":"Obtain the IEEE 802.11be specification (final or latest draft) and locate the clauses governing NSTR MLD backoff and start-time alignment (e.g., draft 5.0 §35.3.4 and related). For the exact scenario of Section II—one link's counter reaches 0 while the other link is busy with a third-party transmission—determine whether the standard mandates restart, permits holding the counter at 0, or allows transmission. If any alternative is permitted, re-derive or simulate the model with that permitted behavior and compare the per-device throughput. If the throughput results change materially, the paper's 'standards-compliant' claim is invalidated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that the proposed Markov chains are a 'standards-compliant analytical framework' that accurately models MLO NSTR channel access. The load-bearing premise is Section II's channel-access rules, especially the rule that a link whose backoff counter reaches zero while the other link is busy must restart its backoff. The paper itself labels these rules 'a most conservative interpretation of the standard' (Section II, near end), and provides no clause-level citations to the IEEE 802.11be specification. If the actual standard permits a different action—e.g., transmitting if NSTR allows simultaneous TX, freezing the counter, or deferring differently—then the MC transitions involving X_AP, X_MLD, and Y no longer describe 802.11be behavior. Since these probabilities appear in every stationary probability and throughput expression, the entire analytical framework and the qualitative conclusion that MLO devices are nearly starved would be artifacts of the interpretation, not of the standard. This concern is independent of internal typographical errors (undefined b3, tau_I_1 mix-up), which are correctable; if the modeled rules are not mandated, the central claim fails even with a correct derivation.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a Markov-chain analytical model for a WLAN coexistence scenario with one STR-capable AP MLD, NMLD NSTR non-AP MLDs, and NSLD legacy single-link devices on two links. It introduces per-STA Markov chains with new states for backoff restart when the other link is busy, models start-time and end-time alignment, and derives closed-form per-device per-link throughput expressions. The analysis is evaluated against a custom ns-3 implementation of NSTR. The reported validation shows good agreement for legacy SLD UL throughput (about 5–11% error), but the MLD-specific throughputs are either effectively zero in both analysis and simulation or show large discrepancies (up to a factor of about 6.7 for AP MLD DL to legacy devices).","tokens_in":20025,"tokens_out":8296,"duration_ms":94344,"significance":"If the model and its interpretation of IEEE 802.11be were correct, the paper would be a useful step: it avoids the combined-chain simplification of [14], provides explicit per-link Markov chains for AP MLD and non-AP MLD, and predicts severe starvation of NSTR MLDs under saturation, a non-obvious coexistence insight. The closed-form expressions and the ns-3 NSTR implementation are concrete contributions. The current manuscript, however, does not yet establish the central claim of ‘accurate per-device throughput’: the only quantity validated with meaningful nonzero values is legacy SLD UL throughput, and the standards-compliance of the modeled access rules is asserted rather than demonstrated.","major_comments":[{"comment":"The paper’s novelty rests on being a ‘standards-compliant analytical framework,’ but the channel-access rules in Section II are introduced as ‘a most conservative interpretation of the standard’ without clause-level citations to IEEE 802.11be. In particular, the rule that a link whose backoff counter reaches zero while the other link is busy must restart its backoff (Section III, Fig. 4) is load-bearing: it enters X_AP, X_MLD, and Y, and therefore every throughput expression. If the actual standard permits freezing, deferring, or transmitting in that situation, the Markov chains in Figs. 6 and 9 describe a non-standard scheme and the qualitative NSTR-starvation conclusion becomes an artifact of the assumption. Please provide specific 802.11be clauses or validate the rule against a spec-compliant implementation, or narrow the claim from ‘standards-compliant’ to ‘one conservative interpret","section":"Section II (system model, p. 3)"},{"comment":"The expression for the probability that the AP MLD alone transmits a packet destined to a legacy STA in Case 1 is given as tau_I_1 = tau_MLD_AP * (1 - p_SLD_AP). This contradicts the definitions in Section III: Case 1 should use tau_SLD_AP, not tau_MLD_AP. As printed, this typo feeds Eq. (70) and accounts for the large underprediction of AP MLD DL throughput to legacy devices in Fig. 10 (analysis 2.2 Mbps vs simulation 14.7 Mbps at N=2). Correct this to tau_SLD_AP and re-run the numerical evaluation.","section":"Section V-B (definitions of tau_I_1)"},{"comment":"The abstract and Section I claim ‘accurate estimates on the per-device throughput performance.’ However, the validation only supports legacy SLD UL throughput (5–11% error). The non-AP MLD UL throughput is effectively zero in both analysis and simulation (0.000007–0.00247 Mbps vs 0 Mbps), so it provides no discriminating validation. The AP MLD DL throughput to non-AP MLDs shows large relative errors (e.g., N=2: analysis 0 vs simulation 1.1 Mbps; N=3: 0.6 vs 1.7 Mbps). The paper itself notes in Section VI-A that these comparisons are ‘less meaningful,’ yet retains the broad accuracy claim. Either validate the MLD-specific quantities with a meaningful nonzero regime or restrict the accuracy claim to SLD UL throughput.","section":"Section VI-A and VI-B (validation)"}],"minor_comments":[{"comment":"The formula for p_II_c,2 contains an undefined symbol ‘b3’: p_II_c,2 = 1 - (p_II_idle + b3) - (tau_II_1a + tau_II_1b + tau_II_2 + tau_II_3) - p_II_c,1. Since this quantity enters the average slot duration in Eq. (59), the Case-2 throughput equations are not computable as printed. Please define or remove the erroneous term.","section":"Section V-B (p_II_c,2)"},{"comment":"There are typographical inconsistencies in the exponents for tau_II_2 and tau_II_3: ‘tau_II_MD’ and ‘NM LD’ appear where ‘tau_II_MLD’ and ‘NMLD’ are intended. These should be fixed for reproducibility.","section":"Section V-B (tau_II_2 and tau_II_3)"},{"comment":"The function Nth(·) is borrowed from [14] but is not defined in the manuscript. Since it directly affects the DL throughput to non-AP MLDs, a brief definition or equation reference to [14] is needed.","section":"Section V-E (Eq. (71))"},{"comment":"The closed-form expressions are dense and the superscript/subscript notation (e.g., tau_SLD_AP vs tau_MLD_AP) is easy to confuse. A summary table of all tau/p symbols and their Case-1/Case-2 meanings would improve readability and reduce the risk of the type of typo noted above.","section":"General presentation"}],"recommendation":"major_revision","confidential_remarks":"The central modeling idea is reasonable and the SLD UL validation is credible, but the manuscript currently overclaims validation for the novel MLD quantities. The tau_I_1 typo is likely the cause of the factor-6.7 discrepancy and should be straightforward to fix. The deeper concern is the unsupported ‘most conservative interpretation’ of 802.11be; I have not independently checked the standard clauses, but this should be verified by someone with specification expertise before the paper can be accepted as a standards-compliant model. The authors' own admission that the MLD comparisons are ‘less meaningful’ should be addressed head-on."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing you should know: this is a real modeling contribution—per-STA Markov chains for NSTR MLO with restart and start-time alignment states that genuinely fix gaps in Song & Kim and Korolev et al.—but the paper's central claim that it models standard-mandated behavior rests on an untested interpretation, and the validation of the new quantities is not as clean as the abstract implies.\n\nThe good parts are real. The authors build separate chains for AP MLD and non-AP MLD STAs instead of consolidating them into one chain, add transitions for backoff restart when the peer link is busy, and introduce states for start-time alignment with a Y probability that the waiting link survives until the other counter hits zero. That is a sensible framework and a step beyond prior analytical models. The closed-form throughput expressions for AP DL, non-AP UL, and legacy UL are derived in full, and the SLD UL throughput matches ns-3 within 5–11%, which is respectable for a Bianchi-style model.\n\nNow the soft spots, in order of load-bearingness. First, the standards-compliance premise. Section II describes the channel access rules as 'a most conservative interpretation of the standard' and gives no clause-level citations. The critical rule—if the other link is busy when your backoff counter hits zero, you restart the backoff—is what generates the XAP/XMLD transitions, and everything downstream depends on it. If the spec actually permits freezing, transmitting anyway, or a different deferral, the model describes a made-up protocol, not Wi-Fi 7. The paper has to demonstrate that this is not the case before the 'standards-compliant' label is defensible.\n\nSecond, validation. The SLD UL match is fine, but the two quantities the model is claimed to predict accurately—AP MLD DL throughput and non-AP MLD UL throughput—are not validated. AP DL to SLD is off by a factor of up to 6.7 at N=2 (analysis 2.2 vs sim 14.7 Mbps). The paper rationalizes the gap but doesn't explain it. Non-AP MLD UL is essentially zero in both analysis and simulation, so it gives no evidence that the model correctly predicts non-zero UL behavior. No simulation variance is reported, and no code or data is released, which is fixable and should be required.\n\nThird, typos/notation: 'b3' appears undefined in pII_c_2, and tau_I_1 uses tau_MLD_AP where tau_SLD_AP seems intended. Both are easy to correct, but they suggest the manuscript wasn't carefully proofread.\n\nThe qualitative finding—NSTR MLDs are nearly starved under saturation—is plausible, but it is a direct consequence of the conservative interpretation. If the standard allows more permissive behavior, the starvation result may be an artifact.\n\nWho is this for? Researchers modeling MLO coexistence, especially those extending Bianchi-style analysis to multi-link. It deserves a serious referee and major revision. I would not reject it outright; the framework is novel and the derivation is a substantial piece of work. But it needs the spec verification, the validation fixes, and the code/data before it can be trusted.","headline":"A genuine analytical contribution to NSTR MLO modeling, but its standards-compliance claim rests on an unverified interpretation and the validation of the new quantities is weak.","tokens_in":20495,"tokens_out":2637,"would_cite":true,"duration_ms":28968,"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":"Wi-Fi 7's own NSTR rules nearly starve non-AP MLDs under saturation, first standards-compliant Markov-chain analysis shows.","keywords":["Wi-Fi 7","IEEE 802.11be","Multi-Link Operation","NSTR","Markov chain","coexistence","throughput analysis","ns-3"],"falsifier":"Run an IEEE 802.11be testbed or an independently implemented NSTR simulator with two links, one AP MLD, several non-AP MLDs, and several legacy devices, all saturated; measure non-AP MLD uplink throughput. If that throughput is substantially above zero—say more than 5% of legacy per-device throughput—while still using start-time alignment, then either the conservative backoff-restart rule is not what devices do, or the model's fixed-point equations miss a mechanism that lets NSTR MLDs transmit.","tokens_in":19603,"feed_emoji":"📶","tokens_out":1551,"duration_ms":20106,"temperature":0.7,"pith_summary":"The paper proposes the first analytical framework, built on Markov chains, that models IEEE 802.11be Multi-Link Operation with the per-STA backoff and the mandatory start- and end-time alignments that earlier models omitted. It derives closed-form per-device throughput for AP MLD, non-AP MLDs, and legacy single-link devices coexisting in one WLAN. The central result is that under saturated traffic, the standard's alignment rules force non-AP MLD uplinks to almost zero throughput while legacy devices take most transmission opportunities; the paper verifies this with an ns-3 simulator it extended with NSTR behavior. A sympathetic reader should care because this quantifies a structural fairness cost of NSTR that was missing from prior aggregated models, and gives a reusable analytic core for future MLO coexistence design.","feed_headline":"Backoff rules nearly starve Wi-Fi 7 NSTR devices","feed_subtitle":"A standards-compliant Markov-chain model shows legacy Wi-Fi devices dominating the channel under saturation.","key_machinery":"The central objects are two new Markov chains: one for the AP MLD's per-link backoff, split into an MLO part (downlink to non-AP MLDs) and an SLO part (downlink to legacy devices), and one for a non-AP MLD's per-link backoff with additional states (i′,0) and (i′′,0) capturing backoff restart and start-time waiting. The AP chain introduces a transition that restarts backoff when the counter hits zero while the other link is busy with a transmission not from the AP; the non-AP chain freezes one link at zero until the other reaches zero, restarting if either channel becomes busy. These chains feed transmit probabilities τ and collision probabilities p, which are solved as a fixed point along wi","core_discovery":"The paper claims that standard-compliant NSTR channel access, when modeled with separate per-STA Markov chains and with the rule that a link whose backoff reaches zero while the other link is busy restarts its backoff, yields closed-form transmit and collision probabilities and per-device throughput for AP MLD, non-AP MLDs, and legacy SLDs. It further claims that under saturated conditions, a non-AP MLD's start-time alignment requirement—both links' counters must reach zero simultaneously—makes its uplink throughput effectively zero, and that the AP MLD's downlink to non-AP MLDs is also severely throttled because transmitting on one link while the other is busy is forbidden to avoid in-devic","pith_inferences":["The model's most conservative interpretation—that a link whose counter hits zero while the other link is busy must restart backoff—predicts near-total starvation of NSTR MLDs in saturation; if real 802.11be devices instead defer without restarting or transmit when feasible, the quantitative starvation result could soften while the qualitative tension between alignment and throughput would persist.","The same Markov-chain structure could be adapted to unsaturated or finite-buffer traffic by adding queue-state dimensions, which would test whether the starvation effect is an artifact of full saturation or a general coexistence property.","A natural testable extension is to measure whether dynamic per-link contention-window adjustment—for instance, smaller windows on the link that waits for alignment—can restore non-AP MLD throughput without violating the standard.","The framework implicitly suggests that NSTR MLDs may benefit from selecting one primary link and parking the other, reducing the joint-probability penalty of simultaneous counter arrival; such an extension is not in the paper and would need its own validation."],"forward_implications":["If the model is correct, saturated Wi-Fi 7 NSTR coexistence is inherently unfair: legacy single-link devices dominate the channel, and non-AP MLD uplink throughput is effectively zero.","The closed-form throughput expressions allow fast parameter sweeps over contention windows, numbers of devices, and MCS choices without running simulations.","The framework extends to asymmetric link configurations because each link is modeled by its own chain, unlike the earlier combined-chain model.","The AP MLD's downlink to non-AP MLDs is shown to be severely constrained by the need to avoid IDC interference, suggesting that NSTR downlink scheduling must be jointly designed with the backoff rules.","The proposed chains are positioned as a building block for future MLO channel-access mechanisms aimed at throughput and latency fairness among coexisting device types."],"supporting_citations":[{"why":"Supplies the Bianchi model from which legacy STA transmit probabilities, collision probabilities, and slot-duration formulas are adapted.","marker":"[17]"},{"why":"Provides the earlier combined-Markov-chain NSTR model that this paper identifies as misrepresenting per-STA backoff and alignment, and supplies the Nth function used in the AP MLD downlink throughput expression.","marker":"[14]"},{"why":"Provides the prior analytical NSTR model that this paper extends by adding start-time alignment.","marker":"[15]"},{"why":"The preliminary conference version that presented only the AP MLD chain without stationary-probability derivations or simulation validation.","marker":"[16]"},{"why":"The ns-3 simulator base that the authors extended with NSTR functionalities for validation.","marker":"[19]"},{"why":"Earlier ns-3 coexistence evaluation that the paper contrasts with and validates against in spirit, providing the scenario of AP MLD, NSTR MLD, and legacy devices.","marker":"[10]"},{"why":"Supplies OFDM symbol duration and slot-timing formulas used in deriving event durations.","marker":"[18]"}],"fun_headline_variants":["Wi-Fi 7 NSTR MLO backoff rules cause starvation","Legacy Wi-Fi beats NSTR MLO in saturation modeling","Markov-chain model reveals Wi-Fi 7 NSTR MLO weakness","NSTR MLO throughput collapses under saturated channel","Why NSTR Wi-Fi 7 uplink hits zero under load"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The model's accuracy hinges on the assumption that real Wi-Fi 7 devices follow the paper's most conservative reading of the standard—that a link whose backoff counter reaches zero while the other link is busy must restart its backoff rather than transmit, defer, or align differently.","fun_headline_variants_meta":{"raw":{"variants":["Wi-Fi 7 NSTR MLO backoff rules cause starvation","Legacy Wi-Fi beats NSTR MLO in saturation modeling","Markov-chain model reveals Wi-Fi 7 NSTR MLO weakness","NSTR MLO throughput collapses under saturated channel","Why NSTR Wi-Fi 7 uplink hits zero under load"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000299,"raw_usage":{"total_tokens":1604,"prompt_tokens":821,"completion_tokens":783,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":565,"completion_tokens_details":{"reasoning_tokens":706}},"tokens_in":565,"tokens_out":783,"duration_ms":8354,"temperature":1.0,"reasoning_tokens":706,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T12:45:57.886636+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run an IEEE 802.11be testbed or an independently implemented NSTR simulator with two links, one AP MLD, several non-AP MLDs, and several legacy devices, all saturated; measure non-AP MLD uplink throughput. If that throughput is substantially above zero—say more than 5% of legacy per-device throughput—while still using start-time alignment, then either the conservative backoff-restart rule is not what devices do, or the model's fixed-point equations miss a mechanism that lets NSTR MLDs transmit.","supporting_citations":[{"cited_title":"Performance Analysis of the IEEE 802.11 Distributed Coordination Function,","cited_arxiv_id":null,"evidence_quote":"Supplies the Bianchi model from which legacy STA transmit probabilities, collision probabilities, and slot-duration formulas are adapted."},{"cited_title":"Performance Analysis of Synchronous Multi- radio Multi-link MAC Protocols in IEEE 802.11be Extremely High Throughput WLANs,","cited_arxiv_id":null,"evidence_quote":"Provides the earlier combined-Markov-chain NSTR model that this paper identifies as misrepresenting per-STA backoff and alignment, and supplies the Nth function used in the AP MLD downlink throughput expression."},{"cited_title":"Analytical Model of Multi- link Operation in Saturated Heterogeneous Wi-Fi 7 Networks,","cited_arxiv_id":null,"evidence_quote":"Provides the prior analytical NSTR model that this paper extends by adding start-time alignment."},{"cited_title":"Modeling the Coexistence Performance between Wi-Fi 7 and Legacy Wi-Fi,","cited_arxiv_id":null,"evidence_quote":"The preliminary conference version that presented only the AP MLD chain without stationary-probability derivations or simulation validation."},{"cited_title":"Network Simulations with the ns-3 Simulator,","cited_arxiv_id":null,"evidence_quote":"The ns-3 simulator base that the authors extended with NSTR functionalities for validation."},{"cited_title":"Analyses of NSTR Multi-link Operation in the Presence of Legacy Devices in an IEEE 802.11be Network,","cited_arxiv_id":null,"evidence_quote":"Earlier ns-3 coexistence evaluation that the paper contrasts with and validates against in spirit, providing the scenario of AP MLD, NSTR MLD, and legacy devices."},{"cited_title":"Performance Evaluation of 802.11ax OFDMA through Theoretical Analysis and Simulations,","cited_arxiv_id":null,"evidence_quote":"Supplies OFDM symbol duration and slot-timing formulas used in deriving event durations."}],"review_version":1}