{"id":"a55ec1b3-454a-4252-bdbe-929646b4f915","arxiv_id":"1908.01607","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Tailored protograph LDPC codes with symmetric protection of packet edges increase supported traffic in asynchronous random access by about 11 to 17 percent over 5G codes in simulations.","lead":"Researchers designed special error-correcting codes for a wireless random access scheme where packet collisions hit only the start or end of a message. The codes let more terminals share the channel without synchronization, with gains of 11 to 17 percent over standard 5G codes in simulations.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The abstract's headline gain of 17% at PLR 10^-2 is contradicted by the body's 11% at that operating point, so the quantitative central claim as stated is not supported by the paper's own Figure 8.","rationale":"I read the paper in good faith. The central claim is that protograph LDPC codes tailored to edge interference improve asynchronous random access throughput relative to 5G codes. The design story is coherent: a surrogate constant-interference model, a symmetry constraint on the base matrix, threshold-based optimization, and then validation via both abstracted and finite-length physical-layer simulations. The reader's weakest assumption, that edge-only protection is sufficient for multi-user collisions, is a reasonable concern, but it is substantially tested by the full physical-layer simulations, which include realistic multi-user collisions and still show a consistent relative gain. The most load-bearing weakness is instead the mismatch between the paper's headline numbers and its own plotted results. The abstract and conclusions claim about 17% at PLR 10^-2, whereas Section IV-A3 reports 11% at PLR 10^-2 and 17% at PLR 10^-3. This is a direct factual inconsistency in the central claim's quantitative statement, not a stylistic issue. It does not necessarily invalidate the design, but it does mean the paper as written overstates the headline result. The reader already noted this discrepancy in the strongest-claim summary and rationale, so my finding partially agrees with the reader, even though the reader's formal weakest assumption was about collision geometry. The appropriate verdict remains conditional: the paper should be accepted only after the reported gains in the abstract, conclusions, and body are reconciled, and ideally after the underlying simulation data or code are made available for independent verification. I therefore recommend no change to the reader's conditional verdict.","tokens_in":19431,"tokens_out":18598,"duration_ms":180392,"concrete_test":"Using the simulation data behind Figure 8, read off the supported channel load for the 5G/5G-perm and ad-hoc curves at PLR 10^-2 and PLR 10^-3, and compute (G_ad-hoc - G_5G)/G_5G at each target. If the values are 1.0 versus 0.9 at PLR 10^-2, the gain is 11%, not 17%; if the values are 0.7 versus 0.6 at PLR 10^-3, the gain is 17%. Then correct the abstract and conclusion to report 11% at PLR 10^-2 and 17% at PLR 10^-3. Additionally, to verify that the finite-length physical-layer results support a similar statement, run at least 10^5 independent packet transmissions per load point and report the supported-load difference at PLR 10^-2 with confidence intervals.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim, as stated in the abstract and conclusions, is that the optimized protograph LDPC codes provide gains of around 17% in supported channel traffic at a packet loss rate of 10^-2 relative to off-the-shelf 5G codes. Section IV-A3 and Figure 8 report instead that, at PLR 10^-2, the supported load increases from 0.9 to 1.0 b/s/Hz, which is an 11% gain, while the 17% gain is observed at PLR 10^-3 (0.6 to 0.7 b/s/Hz). The conclusion repeats the 17% figure but attaches it to PLR 10^-2, and the finite-length physical-layer results in Section IV-B do not state a percentage gain at 10^-2. The design methodology may still be sound, but the quantitative headline of the paper is not supported by the reported data. Because the contribution is framed as a specific throughput gain, this discrepancy is load-bearing: a reader relying on the abstract or conclusion would overstate the achieved improvement by roughly six percentage points at the stated target. The correction is straightforward, but it must be made for the central claim to be accurately represented.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies physical-layer code design for an uncoordinated asynchronous random-access protocol with two replicas per packet and successive interference cancellation (SIC). It observes that asynchronous collisions typically affect only the beginning or end of a codeword, proposes a surrogate block-interference channel with Gaussian interference over a fraction alpha of the codeword, and uses EXIT/density-evolution-based optimization with a symmetry constraint on the protograph base matrix so that both packet edges are equally protected. It reports an ad-hoc protograph ensemble, compares its thresholds with a 5G raptor-like protograph and a column-permuted version, and evaluates all three in two settings: an abstracted physical layer using decoding regions, and a full finite-length physical-layer simulation of (960, 480) LDPC codes. The paper claims gains of about 17% in supported traffic at PLR 10^-2 relative to off-the-shelf 5G codes and notes that the abstracted model overestimates performance by about 10%.","tokens_in":19665,"tokens_out":6169,"duration_ms":63386,"significance":"The paper has clear strengths: the base matrices are given explicitly, the design principle is simple and transferable, and the inclusion of both a threshold-based abstraction and a full finite-length physical-layer simulation is valuable. The full-PHY simulation is an important independent check that avoids circularity in the threshold-based design loop. If the quantitative claims are corrected and the PHY results are reported with error bars or run counts, the paper makes a useful contribution to practical asynchronous random-access design. As it stands, the headline gain figure is not consistently supported by the reported data.","major_comments":[{"comment":"The abstract and the conclusions state gains of \"around 17% at a packet loss rate of 10^-2\", but Section IV-A3 reports exactly the opposite mapping: at PLR 10^-2 the supported load rises from 0.9 to 1.0 b/s/Hz, which is an 11% gain, while the 17% gain corresponds to PLR 10^-3 (0.6 to 0.7 b/s/Hz). This is not a stylistic point: the central quantitative claim of the paper is misreported by about six percentage points at the stated operating point. Please correct the abstract and conclusions to match Figure 8, or explicitly re-define what is meant by the reported gain.","section":"Abstract, Section V, and Section IV-A3/Figure 8"},{"comment":"The finite-length physical-layer results are the main independent validation of the design, but the paper reports no run counts, confidence intervals, or error bars for the PLR curves, and it does not state the percentage gain at PLR 10^-2 for the PHY curves. Since the abstract's 17% claim is traced to abstracted curves that the paper itself says overestimate performance by about 10% in supported load, the actual PHY gain at 10^-2 may be materially different from 17% and may be within simulation noise. Please report the number of simulated packets/users, error bars or a standard-error analysis, and the explicit load values and gains from Figure 9 at both PLR 10^-2 and 10^-3.","section":"Section IV-B/Figure 9"}],"minor_comments":[{"comment":"The decision to ignore packets with interference at both ends is justified by a qualitative SIC argument; a quantitative breakdown of collision types from the simulated traffic (for example, the fraction of replicas with beginning-only, end-only, and both-end interference as a function of G) would make the design rationale easier to check.","section":"Section III-D1/Equation (6)"},{"comment":"Please state explicitly that the threshold curves for the ad-hoc base matrix are the result of the same optimization objective used to select that matrix, so that Figure 6 is a design/tuning result rather than an out-of-sample prediction; the independent confirmation is provided by Figure 9.","section":"Figure 6 and Section III-D"},{"comment":"There are numerous typos and small language errors, including \"correseponds\" in the captions of Figures 6 and 8, \"Gaussain\" and \"intererfer\" in Section III-E, \"asnychronous\" in the acknowledgments, and \"photograph\" for \"protograph\" in Section III-A; these should be corrected in a final pass.","section":"Throughout"},{"comment":"The sentence \"For a fixed channel load operating point, the gain is even more remarkable\" is not backed by a specific number; either remove it or give the operating point and the resulting gain.","section":"Section IV-A3"},{"comment":"The notation for noise and interference powers is not consistently per-dimension: Equation (3) uses 1/(2 sigma_n^2) inside the capacity expression, while Equation (9) writes per-dimension powers as sigma_n^2 + j/2. Please make the per-dimension convention explicit throughout to avoid confusion.","section":"Equations (3), (7), and (9)"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for a communications journal and the design methodology is credible. The main risk is not methodological but reporting: the headline percentage must be reconciled with Section IV-A3, and the finite-length PHY results should be accompanied by confidence information before acceptance. The conference precursor is acknowledged, so I do not see a novelty concern."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this paper is worth a serious look, but the abstract and conclusions state a headline gain that the body's own numbers contradict. Read it for the finite-length physical-layer validation; don't quote the 17% at 10^-2 until they fix it.\n\nThe genuinely new content is the design objective — symmetric protection of both codeword edges in the base matrix — plus the permuted 5G baseline, and most importantly the full PHY simulation of the asynchronous RA scheme with SIC using the designed (960,480) codes. That last piece is what makes the paper worth a referee's time: the abstraction (decoding regions) tells a consistent story, and the full simulation confirms the ranking, which is not a given for threshold-based designs. The comparison against off-the-shelf 5G codes is fair and useful. The Gaussian-interference assumption is also stress-tested via quantized density evolution for a single interferer, which is more than most papers at this level do.\n\nThe soft spots are real but not disqualifying. First, the gain inconsistency: the abstract and Section V say 17% at PLR 10^-2, but Section IV-A3 reports 11% at 10^-2 and 17% at 10^-3 (0.6→0.7 vs 0.9→1.0 b/s/Hz in Figure 8). This is exactly the kind of error that propagates into citations and system budgets. It needs a correction, not a rewrite. Second, there are no error bars, run counts, or code/data release, so the PHY curves are hard to reproduce exactly; for a paper whose punchline is a throughput number, that's a real gap. Third, the design premise explicitly sets aside packets with interference at both ends (Section III-D1) and argues SIC favors other packets first; the full PHY simulation includes those cases and the code still wins, so the qualitative case holds, but the magnitude of the advantage across collision geometries is less certain than the text suggests. The threshold curves in Figure 6 are also the output of the same optimization, so they're not an independent prediction — though again, the PHY simulation is independent and that's what matters.\n\nVerdict: yes, send it to peer review. The methodology is coherent, the validation is appropriate, and the central claim — that edge-protecting LDPC codes improve asynchronous RA throughput — is supported by the data. A serious referee should ask for the gain figures to be reconciled and for some indication of simulation uncertainty before acceptance. I'd bring it to a reading group focused on random access or code design, and I'd likely cite the finite-length validation approach after the numbers are fixed.","headline":"Solid engineering paper with a genuine finite-length validation, but the headline 17% gain at PLR 10^-2 is contradicted by the body's 11% — fix that before quoting.","tokens_in":20218,"tokens_out":2335,"would_cite":true,"duration_ms":22796,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Designing LDPC codes to protect the edges of a codeword boosts asynchronous random-access traffic by about 17%.","keywords":["protograph LDPC codes","asynchronous random access","successive interference cancellation","surrogate channel model","block interference channel","unequal error protection","5G eMBB codes","packet loss rate"],"falsifier":"Run the same asynchronous random-access simulation but with a traffic scenario engineered so that a large fraction of replicas (say half) are sandwiched between two interferers, so interference corrupts both the beginning and the end. If the ad-hoc code's supported-load advantage over the 5G code at PLR $10^{-2}$ falls to zero or reverses, the edge-only interference premise is the active ingredient.","tokens_in":19232,"feed_emoji":"📶","tokens_out":7468,"duration_ms":70837,"temperature":0.7,"pith_summary":"This paper argues that the channel code inside an asynchronous random-access receiver should be designed for the interference pattern the protocol actually produces, rather than for a clean AWGN channel. In the contention-resolution scheme studied here, collisions are mostly two-user events, so a successfully decodable packet is typically corrupted at only its beginning or only its end. The authors define a surrogate channel model in which a constant-power interferer hits a fraction $\\alpha$ of the codeword, and they optimize protograph LDPC base matrices whose protection is symmetric between the two edges. They report that the resulting codes support roughly 17% more channel traffic than off-the-shelf 5G LDPC codes at a packet loss rate of $10^{-2}$, with the detailed results in Section IV-A3 showing 11% at $10^{-2}$ and 17% at $10^{-3}$. This matters because uncoordinated asynchronous access is a low-overhead way to serve many intermittently active IoT or satellite terminals, and the physical-layer code is currently a generic off-the-shelf component.","feed_headline":"Edge-protecting LDPC codes lift random-access traffic by ~17%","feed_subtitle":"Asynchronous collisions hit only packet edges; codes built for that geometry carry more traffic than 5G codes.","key_machinery":"The central object is the protograph base matrix $\\mathbf{B}_A$, a small integer matrix whose entries count connections between variable-node and check-node types; lifting it produces a finite LDPC code. The design imposes a reversal symmetry, $b_{(m_b-i-1),(n_b-p_b-j-1)} = b_{i,j}$, so the code protects the beginning and end of a codeword equally. The search uses a multi-target gain function $g = \\prod_\\ell \\sigma^2_{\\iota,th}(\\alpha^{(\\ell)},b) / \\sigma^2_{\\iota,o}(\\alpha^{(\\ell)}) \\cdot \\sigma^2_{\\iota,th}(\\alpha^{(\\ell)},e) / \\sigma^2_{\\iota,o}(\\alpha^{(\\ell)})$, evaluated with EXIT (extrinsic information transfer) analysis under the surrogate Gaussian block-interference model at two overlap fractions $\\alpha^{(0)}=6/10$ and $\\alpha^{(1)}=9/10$, with differential evolution exploring candidate base matrices. The same gain function is used to column-permute the 5G baseline matrix for fair comparison. The paper then replaces the surrogate channel with a decoding-region abstraction and finally with full finite-length physical-layer simulation.","core_discovery":"The central discovery is that an LDPC code whose base matrix is symmetric under reversal, so that the beginning and end of a codeword receive equal and strong protection, can decode replicas that a 5G eMBB LDPC code misses in asynchronous random access. The paper models a collision as a block of constant Gaussian interference covering a fraction $\\alpha$ of the codeword, and optimizes the protograph through differential evolution with a gain function built from EXIT thresholds at two values of $\\alpha$. The resulting ad-hoc base matrix $\\mathbf{B}_A$ tracks the outage-capacity Shannon limit over the whole range of $\\alpha$, whereas the 5G base matrix is asymmetric and performs much worse when the interferer hits the unprotected edge; a column-permuted 5G matrix improves but does not match it. Finite-length $(960,480)$ physical-layer simulations of the full asynchronous random-access protocol with successive interference cancellation confirm the ranking and show that the abstracted model overestimates the supported channel load by about 10%.","pith_inferences":["Beyond the paper's claims, the same edge-protection principle should transfer to other asynchronous protocols with two replicas or partial overlaps: a testable prediction is that optimizing for a heavier mix of three-packet collisions would move the optimal protection profile inward from the two edges.","The binary protected/unprotected surrogate channel is a coarse proxy; one could design codes against the actual distribution of overlap fractions $\\alpha$ seen in simulation, which might capture part of the remaining gap to the capacity-achieving bound.","If successive interference cancellation is non-ideal and leaves residual power, the edge-protecting code is plausibly more robust than a generic code, because residual interference appears exactly where the code is strongest; this could be tested by simulating imperfect cancellation.","The discrepancy between the abstract's 17% at PLR $10^{-2}$ and the body's 11% at that operating point suggests the advantage is operating-point dependent; a full waterfall comparison would clarify where the design should be deployed."],"forward_implications":["At a target packet loss rate, the physical layer can accept a higher channel load: about 0.1 b/s/Hz more than a 5G eMBB code, corresponding to roughly 11% at PLR $10^{-2}$ and 17% at PLR $10^{-3}$ in the detailed results.","The ranking of codes established with the abstracted physical layer survives finite-length simulation: the ad-hoc design beats both the 5G and the permuted 5G designs, while the abstraction overestimates absolute supported load by about 10%.","A 5G eMBB LDPC base matrix can be improved for asynchronous random access by permuting its columns to place stronger variable nodes at the edges, but this does not close the gap to the ad-hoc design.","Because a small reduction in error-correction capability causes a large PLR degradation in the random-access scheme (a 95%-of-capacity code performs like the permuted 5G code), code design deserves the same attention as the access protocol itself."],"supporting_citations":[{"why":"Supplies the asynchronous random-access protocol with two replicas and SIC that the paper builds on.","marker":"[25]"},{"why":"Establishes the asynchronous CRDSA setting and the collision geometry that motivates edge protection.","marker":"[26]"},{"why":"Provides the threshold-based physical-layer abstraction that the paper refines with decoding regions and finite-length simulation.","marker":"[29]"},{"why":"Is the authors' earlier conference paper that introduced the multidimensional decoding-region design for this channel.","marker":"[30]"},{"why":"Defines protograph construction and lifting that turns base matrices into finite LDPC codes.","marker":"[38]"},{"why":"Supplies the differential-evolution search procedure used to optimize the base matrix.","marker":"[39]"},{"why":"Provides the protograph EXIT analysis used to compute the iterative decoding thresholds in the gain function.","marker":"[40]"},{"why":"Is the block-interference channel model that justifies the surrogate constant-power interference channel.","marker":"[41]"},{"why":"Is the 5G eMBB LDPC base matrix used as the off-the-shelf baseline.","marker":"[43]"},{"why":"Introduces decoding regions, the abstraction method used for the first set of random-access simulations.","marker":"[44]"}],"fun_headline_variants":["Edge-focused LDPC design lifts async random-access load 17%","Reversal-symmetric LDPC codes outdo 5G codes in async access","Protect both ends: LDPC code gains 17% traffic in async collisions","Asynchronous collisions hit edges, so LDPC codes reinforce them: +17% load","Symmetric LDPC base matrix wins 17% more users in async random access"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The design assumes collisions are dominated by two-packet overlaps that corrupt only one edge of a codeword, so protecting the two edges symmetrically is enough.","fun_headline_variants_meta":{"raw":{"variants":["Edge-focused LDPC design lifts async random-access load 17%","Reversal-symmetric LDPC codes outdo 5G codes in async access","Protect both ends: LDPC code gains 17% traffic in async collisions","Asynchronous collisions hit edges, so LDPC codes reinforce them: +17% load","Symmetric LDPC base matrix wins 17% more users in async random access"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000775,"raw_usage":{"total_tokens":3409,"prompt_tokens":905,"completion_tokens":2504,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":521,"completion_tokens_details":{"reasoning_tokens":2395}},"tokens_in":521,"tokens_out":2504,"duration_ms":17020,"temperature":1.0,"reasoning_tokens":2395,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:08:36.375567+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same asynchronous random-access simulation but with a traffic scenario engineered so that a large fraction of replicas (say half) are sandwiched between two interferers, so interference corrupts both the beginning and the end. If the ad-hoc code's supported-load advantage over the 5G code at PLR $10^{-2}$ falls to zero or reverses, the edge-only interference premise is the active ingredient.","supporting_citations":[{"cited_title":"Performance Enhancements for Asynchronous Random Access Protocols over Satellite,","cited_arxiv_id":null,"evidence_quote":"Supplies the asynchronous random-access protocol with two replicas and SIC that the paper builds on."},{"cited_title":"Asynchronous Contention Resolution Diversity ALOHA: Making CRDSA Truly Asynchronous,","cited_arxiv_id":null,"evidence_quote":"Establishes the asynchronous CRDSA setting and the collision geometry that motivates edge protection."},{"cited_title":"LDPC code performance and optimum code rate for contention resolution diversity ALOHA,","cited_arxiv_id":null,"evidence_quote":"Provides the threshold-based physical-layer abstraction that the paper refines with decoding regions and finite-length simulation."},{"cited_title":"LDPC code design for asynchronous random access,","cited_arxiv_id":null,"evidence_quote":"Is the authors' earlier conference paper that introduced the multidimensional decoding-region design for this channel."},{"cited_title":"Low-density parity-check (LDPC) codes constructed from protographs,","cited_arxiv_id":null,"evidence_quote":"Defines protograph construction and lifting that turns base matrices into finite LDPC codes."},{"cited_title":"Design of Non-Precoded Protograph-Based LDPC Codes,","cited_arxiv_id":null,"evidence_quote":"Supplies the differential-evolution search procedure used to optimize the base matrix."},{"cited_title":"Protograph LDPC Codes Design Based on EXIT Analysis,","cited_arxiv_id":null,"evidence_quote":"Provides the protograph EXIT analysis used to compute the iterative decoding thresholds in the gain function."},{"cited_title":"Channels with Block Interference,","cited_arxiv_id":null,"evidence_quote":"Is the block-interference channel model that justifies the surrogate constant-power interference channel."},{"cited_title":"R1-1706970 LDPC design for eMBB data,","cited_arxiv_id":null,"evidence_quote":"Is the 5G eMBB LDPC base matrix used as the off-the-shelf baseline."},{"cited_title":"Unequal Diversity LDPC Codes for Relay Channels,","cited_arxiv_id":null,"evidence_quote":"Introduces decoding regions, the abstraction method used for the first set of random-access simulations."}],"review_version":1}