{"id":"99b5425c-5cab-452a-92bf-546333246fc3","arxiv_id":"2607.09212","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"IRF-MAMP with multi-domain sparsity and residual feedback, plus spatial-frequency spreading, yields lower activity-detection error, channel NMSE and BER than prior grant-free LEO schemes under short pilots and practical SNR.","lead":"The paper introduces an iterative residual-feedback multi-measurement-vector AMP algorithm that alternates active-user detection in the spatial-frequency domain with channel estimation in the angular-delay domain, plus a joint spatial-frequency spreading modulation. The combination improves grant-free massive access performance for LEO satellites when the number of active terminals exceeds the number of onboard antennas.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.5","headline":"Claimed outperformance (Figs. 2–5) is shown only under a channel model that exactly matches the multi-domain sparsity priors IRF-MAMP exploits.","rationale":"The reader’s weakest-assumption paragraph already isolates the idealized common-AoA / residual-Doppler / constant-activity model as the soft underbelly of the performance claims. The stress-test confirms this is the single most load-bearing point: every reported gain occurs inside the exact model class for which multi-domain synergistic sparsity and residual feedback were designed, so the gains are expected by construction and do not yet demonstrate robustness. No internal mathematical inconsistency, circular derivation, or unfair baseline comparison was found; the effective-pilot-length equalization is carefully defined and the spreading-rate trade-off is acknowledged in the conclusions. Therefore the CONDITIONAL verdict (pending more realistic channel validation and public code) remains appropriate and requires no adjustment.","tokens_in":14775,"tokens_out":549,"duration_ms":27006,"concrete_test":"Re-run the ADEP/NMSE/BER curves of Figs. 2–3 with Lp=3 paths whose AoAs are drawn from a Laplacian of 1° RMS spread about the nominal AoA, and with residual Doppler of ±200 Hz left uncompensated (colored noise covariance). If at T=80, SNR=16 dB the gap between the Proposed scheme and the strongest baseline shrinks by more than 50 % in any metric, the practical-superiority claim is materially weakened.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The strongest claim—that the scheme “significantly outperforms existing GF methods” in ADEP, NMSE and BER, especially at low effective pilot length and practical SNR—rests entirely on Monte-Carlo results generated under the precise sparsity structure assumed by the algorithm. Section II states that all multipath components share a common AoA; residual Doppler after ephemeris pre-compensation is absorbed into white noise; and α_k is constant across the frame. Simulations further set Lp=1 (Sec. V), so the AD-domain channel is a single angular bin and the SF structured sparsity of Eqs. (6)–(7) holds exactly. Under these conditions residual-feedback alternating MAMP and the G-fold observation expansion operate in their most favorable regime. The paper supplies no mismatch experiments with modest angular spread, colored residual Doppler, or Lp>1. Consequently it remains unproven whether the ranking versus the six baselines survives realistic LEO channel deviations; that is the load-bearing condition for the performance claim.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper proposes a grant-free random-access framework for massive LEO satellite IoT. It introduces IRF-MAMP, which alternates AUD in the spatial-frequency domain with CE in the angular-delay domain and uses residual feedback of high-reliability users to reduce error accumulation (Alg. 1, Eqs. 10–16). To address rank-deficient multi-user detection when Ka exceeds Nr or AoAs are highly correlated, it also designs joint spatial-frequency spreading that expands the observation dimension for LMMSE data detection (Eqs. 2–5, 17–19). Monte-Carlo results (K=500, Ka=50, Nr=5\times5, G=16, Lp=1) claim clear gains in ADEP, NMSE and BER over six baselines, especially at short effective pilot length and practical SNR (Figs. 2–5).","tokens_in":15116,"tokens_out":996,"duration_ms":10656,"significance":"If the gains hold under realistic LEO channels, the work is a useful systems-level contribution: it jointly designs multi-domain sparsity exploitation for JADCE and observation-dimension expansion for overloaded DD, two bottlenecks that are acute on power- and antenna-limited satellites. The residual-feedback construction and the explicit complexity comparison with OAMP-MMV and SOMP are concrete engineering advances. Strengths include a consistent signal model (Eqs. 1–9), fully specified algorithm and free parameters, and head-to-head evaluation against recent LEO GF baselines. The main limitation is that all reported gains are obtained under a channel model that exactly matches the algorithm’s sparsity priors, so the practical significance remains conditional on robustness that is not yet demonstrated.","major_comments":[{"comment":"Section II (after Eq. 1) and Section V (Lp=1) assume that every multipath of a user shares a single common AoA, residual Doppler is absorbed into white noise, and activity is constant over the frame. Under these conditions the SF structured sparsity (Eqs. 6–7) and AD cluster sparsity hold exactly, so residual-feedback alternating MAMP and G-fold spreading operate in their most favorable regime. The central performance claim (Abstract, §V, Figs. 2–5) therefore rests on matched Monte-Carlo trials only. No mismatch experiments with modest angular spread, Lp>1, or colored residual Doppler are provided. Without such tests it is unclear whether the ranking versus the six baselines survives realistic LEO deviations; this is load-bearing for the claim of significant outperformance.","section":null},{"comment":"The residual-feedback mechanism (Alg. 1, lines 11–12) depends on free parameters ε_low=0.3, ε_high=0.9 and ζ_act that are stated without sensitivity analysis or selection rule. Because residual cancellation of the high-reliability set Π is the distinctive algorithmic ingredient, the reported ADEP/NMSE gains could be sensitive to these thresholds. A short ablation or robustness plot is needed to establish that the gains are not an artifact of a single operating point.","section":null}],"minor_comments":[{"comment":"Abstract and §V: “Effective pliot length” is misspelled; correct to “pilot”.","section":null},{"comment":"Fig. 1 is duplicated in the manuscript text; remove the redundant copy.","section":null},{"comment":"Notation for the residual channel (E^re) and the reconstructed residual (Y^p)^{i+1} is dense; a short clarifying sentence after Eq. (11) would help.","section":null},{"comment":"Section V: the definition of “effective pilot length” for TS-padded baselines versus OFDM pilot slots is carefully worded but still easy to misread; a one-sentence reminder in the figure captions would reduce ambiguity.","section":null},{"comment":"Conclusions correctly note the performance–complexity trade-off of spreading; a quantitative memory/complexity remark for G=32 versus G=16 would strengthen that caveat.","section":null}],"recommendation":"major_revision","confidential_remarks":"The technical core is sound and the multi-domain + residual-feedback idea is interesting for LEO GF access. The main risk is over-claiming generality from matched simulations. If the authors add a modest mismatch study (or clearly scope the claims to the LoS-dominant, single-AoA regime) and a short parameter-sensitivity check, the paper would be suitable for a solid systems journal. Novelty relative to the authors’ own prior multi-domain AMP work [17] should be stated more sharply in the introduction."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The useful takeaway is a clean receiver-side package for LEO grant-free massive access: residual-feedback outer iterations around multi-measurement-vector AMP, alternating AUD in the spatial-frequency domain with CE in the angular-delay domain, plus an explicit joint spatial-frequency spreading modulation that expands the observation dimension for the under-determined data stage. That combination is not in the six re-implemented baselines ([8], [10], [17] and the single-domain ablations).\n\nWhat they do well is the system construction. The signal model (Eqs. 1–9), the IRF-MAMP loop (Alg. 1, residual update after the high-reliability set Π, Eqs. 10–16), and the stacked LMMSE detector (Eqs. 17–19) are consistent. Complexity is written out. Simulations (K=500, Ka=50, Nr=5\times5, G=16, Lp=1, SNR and pilot-length sweeps) show clear gains in ADEP, NMSE and BER versus the baselines, especially at short effective pilot length. The spreading ablation in Fig. 4 is particularly useful: G>1 visibly helps when antennas are scarce. Free thresholds (ε_low=0.3, ε_high=0.9, ζ_act) are stated and do not look like post-hoc knobs that invent the ranking.\n\nThe soft spot is exactly the stress-test point, and it is real but not fatal. Section II assumes common AoA across multipaths, residual Doppler absorbed into white noise, and constant activity across the frame; simulations set Lp=1 so the AD-domain sparsity the algorithm exploits is perfect. There are no mismatch trials with modest angular spread, colored residual Doppler, or Lp>1. So the claim of “significant outperformance” is demonstrated only inside the model the priors match. That is standard for this literature, but it means the ranking versus baselines is not yet stress-tested for realistic LEO channels. No code or measurements either.\n\nThis is for people designing GF receivers for LEO IoT satellites who already live with AMP and multi-domain sparsity. It is not a theoretical breakthrough, but it is a coherent, reproducible-looking engineering step. I would send it to peer review; the math and the simulation design are solid enough that referees can demand the mismatch experiments and still have something worth improving. Worth a look if you work in this subfield; I would cite the spreading + residual-feedback construction if I am writing on overloaded satellite access.","headline":"Solid engineering integration of residual-feedback MAMP, SF/AD alternation, and SF spreading for overloaded LEO GF access; gains are real under the paper’s model but untested under mismatch.","tokens_in":15753,"tokens_out":646,"would_cite":true,"duration_ms":7705,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"Multi-domain residual-feedback detection plus spatial-frequency spreading lets LEO satellites serve more grant-free users than onboard antennas, with better activity detection, channel estimates, and bit error rates under short pilots.","keywords":["LEO satellite","grant-free random access","active user detection","channel estimation","approximate message passing","spatial-frequency spreading","angular-delay domain","massive connectivity"],"falsifier":"Re-run the same Monte-Carlo campaign with realistic per-path angular spreads of a few degrees, residual Doppler left uncompensated, or activity that can flip mid-frame; if activity-error probability, NMSE and BER then collapse to the level of the single-domain baselines, the multi-domain residual-feedback claim fails.","tokens_in":15696,"feed_emoji":"🛰️","tokens_out":755,"duration_ms":8680,"temperature":0.7,"pith_summary":"Grant-free random access is attractive for LEO satellite IoT because it avoids handshake latency, but when many terminals become active at once the problem is both sparse and underdetermined: satellites have few antennas, users can share nearly the same arrival angle, and conventional detectors fail. This paper claims that alternating active-user detection in the spatial-frequency domain with channel estimation in the angular-delay domain, while feeding back only high-confidence residual interference each iteration, recovers the active set and channels more accurately than single-domain or greedy baselines. It further claims that deliberately repeating the same data symbol across a sparse set of subcarriers expands the observation dimension enough for linear multi-user detection to remain reliable even when active users outnumber antennas or are spatially close. Simulations under realistic LEO path loss and short pilots show lower activity-error probability, lower channel NMSE, and lower BER than six published grant-free schemes, including OTFS and multi-satellite baselines. A sympathetic reader cares because the combination turns the physical limits of onboard arrays into a manageable multi-domain sparsity problem rather than an unsolvable rank-deficient one.","feed_headline":"LEO grant-free access works with fewer antennas than users","feed_subtitle":"Residual-feedback multi-domain detection plus subcarrier spreading cuts error under short pilots and practical SNR","key_machinery":"IRF-MAMP (iterative residual feedback multi-measurement vector approximate message passing): outer residual-feedback loops that re-detect on residual measurements after cancelling only a fraction of high-confidence users, nested with AMP updates that exploit multi-domain sparsity priors and an expectation-maximization hyper-parameter schedule.","core_discovery":"The authors establish that an iterative residual-feedback multi-measurement-vector approximate message passing algorithm, which alternates structured-sparsity detection in the spatial-frequency domain with clustered-sparsity estimation in the angular-delay domain and subtracts only a high-reliability subset of reconstructed interference each outer loop, jointly improves active-user detection and channel estimation for grant-free LEO access; when this is paired with joint spatial-frequency spreading of data symbols, the resulting multi-domain observation model remains well-conditioned for LMMSE data detection even under severe overload and highly correlated angles of arrival.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["Residual-feedback multi-domain AMP handles LEO grant-free overload","Joint spatial-frequency spreading fixes rank-deficient LEO data detection","Iterative sparsity alternation boosts AUD and CE with short LEO pilots","Multi-domain residual feedback cuts errors when users exceed LEO antennas","Angular-delay clustering plus residual feedback enables massive LEO GF access"],"cache_read_input_tokens":128,"weakest_assumption_plain":"Every multipath of a given user shares one common arrival angle at the satellite, residual Doppler after ephemeris compensation can be treated as ordinary noise, and each user’s activity status stays fixed for the whole pilot-plus-data frame.","fun_headline_variants_meta":{"raw":{"variants":["Residual-feedback multi-domain AMP handles LEO grant-free overload","Joint spatial-frequency spreading fixes rank-deficient LEO data detection","Iterative sparsity alternation boosts AUD and CE with short LEO pilots","Multi-domain residual feedback cuts errors when users exceed LEO antennas","Angular-delay clustering plus residual feedback enables massive LEO GF access"]},"model":"grok-4.5","effort":"low","cost_usd":0.006748,"raw_usage":{"total_tokens":1720,"prompt_tokens":797,"num_sources_used":0,"completion_tokens":96,"cost_in_usd_ticks":67480000,"prompt_tokens_details":{"text_tokens":797,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":827,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":797,"tokens_out":96,"duration_ms":7452,"temperature":1.0,"reasoning_tokens":827,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-13T04:38:41.703038+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Re-run the same Monte-Carlo campaign with realistic per-path angular spreads of a few degrees, residual Doppler left uncompensated, or activity that can flip mid-frame; if activity-error probability, NMSE and BER then collapse to the level of the single-domain baselines, the multi-domain residual-feedback claim fails.","supporting_citations":[],"review_version":1}