{"id":"9fb8dc33-9c67-4182-90f9-498634eace1c","arxiv_id":"1908.02816","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A protograph search over non-binary LDPC codes for DPSK with phase noise yields codes whose simulated error rates land within 1.2 dB of finite-length bounds down to a CER of 10^-3.","lead":"This paper designs short non-binary LDPC codes for m-ary DPSK transmission over AWGN channels with Wiener phase noise, using a joint detector-decoder factor graph. It reports coded error rates within 1.2 dB of finite-length theoretical bounds at block error probability 10^-3.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The headline '1.2 dB from finite-length bounds' rests on an unquantified Monte Carlo DT bound; absent D, variance, and an explicit information-density computation, the margin's accuracy is unverifiable.","rationale":"The reader's CONDITIONAL verdict is appropriate. The central claim is a numerical gap to a finite-length bound, and that bound is estimated by Monte Carlo with no stated D, variance, or confidence interval. The information-density computation for the non-coherent channel is delegated to references without specifying how the marginal p(r) is obtained, so a systematic bias cannot be ruled out. No other assumption is as directly tied to the headline number: the code's CER is a direct simulation result, and any inaccuracy in the protograph threshold computation would affect design optimality, not the existence of a code achieving the stated simulated performance. I therefore endorse the conditional and propose a concrete reproducibility check that would either confirm the margin or force a revised statement.","tokens_in":15186,"tokens_out":19905,"duration_ms":223771,"concrete_test":"For Example 3 (rate 3/4, 16-DPSK, N=128, σΔ=1°), rerun the DT bound of Eq. (17) with D=10^5 and D=10^6, computing p(r) via an exact forward sum-product (no DP discretization), and record the Eb/N0 at which the bound crosses 1e-3; compare this to the value implied by Figure 5. If the two D values shift the crossing by more than 0.2 dB, or if the 95% bootstrap confidence interval on the crossing exceeds ±0.1 dB, the 1.2 dB claim is not stable and needs an explicit D and confidence interval.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is a specific numerical margin: the designed codes perform within 1.2 dB of a finite-length bound. That bound is the dependency-testing bound in Eq. (16), evaluated by Monte Carlo in Eq. (17). The paper never reports D, the variance of the estimator, or a confidence interval for the bound curves in Figures 3–5. The estimator averages the bounded random variable 2^{-(i(s;r)-log2((M-1)/2))^+}; with finite D the bound curve is noisy, and near CER 1e-3 a small horizontal displacement of the bound directly changes the reported gap. Moreover, for the non-coherent channel the information density i(s;r) requires computing the marginal p(r), and the paper gives no details of that computation beyond 'as described in [32]'. If p(r) is approximated with the same phase discretization used by the detector, the DT bound would be for a mismatched channel and the reported gap would be uninterpretable. Neither source of error is quantified, so the headline margin is not fully established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper presents a design methodology for short non-binary protograph LDPC codes concatenated with m-ary DPSK over AWGN channels affected by Wiener phase noise. The receiver performs iterative detection and decoding on a joint factor graph using the discretized-phase algorithm for symbol detection and non-binary belief propagation for decoding. The authors propose a two-step protograph search based on Monte Carlo density evolution thresholds, with a refinement step to lower error floors. Performance is benchmarked against finite-length dependency-testing (DT) bounds and information-rate limits, and against a serial turbo scheme from the literature. The headline result is that the designed codes perform within 1.2 dB of the DT bounds down to a codeword error rate of 10^-3 for rate 1/2 with 8-DPSK, rate 2/3 with 8-DPSK, and rate 3/4 with 16-DPSK.","tokens_in":15383,"tokens_out":8791,"duration_ms":96382,"significance":"If the reported margins are accurate, the paper provides a useful and concrete benchmark for short-block coded modulation over phase-noise channels, a regime relevant to mMTC. The explicit protograph matrices, threshold tables, and the comparison against a serial turbo detector make the design results reproducible and practically meaningful. The main limitation is that the central claim depends on a Monte Carlo estimate of the DT bound whose statistical quality and computation details are not reported; this must be addressed before the headline margin can be fully accepted.","major_comments":[{"comment":"The DT bound curves in Figures 3-5, which are the reference for the headline \"within 1.2 dB\" claim, are obtained by averaging the summand in Eq. (17) over D simulated (s,r) tuples. The manuscript does not state D, the observed variance of the estimator, or any confidence interval for the bound. Since the horizontal displacement of the bound at CER = 10^-3 directly changes the reported gap, the central claim is not fully verifiable as written. Please report D, a variance estimate, and confidence or error bars for the bound curves, and state explicitly that the bound estimates are accurate to the claimed margin.","section":"Section III, Eq. (17)"},{"comment":"The computation of the information density i(s;r) for the non-coherent Wiener phase-noise channel is only described as done \"as described in [32]\" with a Monte Carlo approach. Reference [32] does not provide an algorithm for continuous-phase channels with memory, so it is unclear whether p(r) and p(r|s) are computed exactly (e.g., by a forward recursion on the factor graph) or with the same L-point phase discretization used by the detector. If the latter, the bound would not be a finite-length bound for the true channel and the 1.2 dB gap would be uninterpretable. Please provide the explicit recursion used to compute i(s;r), state the distribution over s, and clarify that the channel is not discretized for the bound.","section":"Section III, Eq. (17)"}],"minor_comments":[{"comment":"There is a typographical error in Eq. (7): \"p(ψi+1|,ψ i, ai+1)\" should read \"p(ψi+1|ψi, ai+1)\".","section":"Section II-C, Eq. (7)"},{"comment":"In the abstract, \"1:2 dB\" should be \"1.2 dB\".","section":"Abstract"},{"comment":"The notation \"K log2 m − 1\" in Eq. (17) is an approximation of log2((M−1)/2); the text should state this explicitly, since the expression is not obviously equal for finite M.","section":"Section III, Eq. (17)"},{"comment":"The comparison with the serial turbo scheme does not report the interleaver size, number of turbo iterations, or other implementation parameters; please provide these details to make the benchmark reproducible.","section":"Section V, Examples 1-3"},{"comment":"The Monte Carlo density evolution used for the thresholds in Tables III-V does not report the number of decoding attempts or any measure of the accuracy of the threshold estimates; adding these would strengthen the design selection procedure.","section":"Section IV-B"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the scope of IEEE Transactions on Communications and the design results are interesting. The main concern is purely about the verification of the Monte Carlo DT bound, which is fixable in revision. No issues with novelty or citation behavior."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is a competent engineering contribution. It designs non-binary protograph LDPC codes for m-DPSK over a Wiener phase-noise channel, using iterative detection/decoding on a joint factor graph. The genuinely new parts are the direct threshold optimization of the concatenated scheme for the non-coherent channel (the earlier ICC version used an AWGN surrogate), an error-floor refinement step that adds a single edge to the protograph, and finite-length DT benchmarks for this channel. The method is sensible: Monte Carlo density evolution for thresholds, exhaustive search over small base matrices, then a refinement stage. The simulations are self-consistent and show gaps to the DT bound of roughly 1–1.2 dB down to CER 1e-3, beating the serial turbo benchmark by about 0.7 dB. I believe the central claim as stated, within the usual caveats of simulation-based results.\n\nThe soft spots are mostly reporting gaps rather than methodological errors. The headline margin is measured against a DT bound computed by Monte Carlo, but the paper does not report the number D of (s,r) tuples, the variance of the estimator, or any confidence interval on the bound curves. Near CER 1e-3, a small horizontal shift in the bound directly changes the reported gap. The information density i(s;r) for the non-coherent channel also gets only a reference to [32], with no detail on how the marginal p(r) is computed. If the same phase discretization as the detector is used, the bound could be for a mismatched channel, though the authors' coherent-case DT bounds matching known results suggests this is not a serious problem. The detector parameter P_Delta is tuned by simulation and fixed at 0.1 without a sensitivity sweep; that is minor since L=8m is already justified by prior work. No code or data is shipped, so independent verification requires reimplementation.\n\nThis is a paper for people designing short LDPC codes for phase-noise channels or mMTC-style links. It deserves a serious referee: the methodology is clearly presented, the comparisons are fair, and the designs are reusable. I would accept it for peer review and ask the authors to report the DT bound Monte Carlo parameters and information-density computation details. Once those are added, the 1.2 dB margin becomes properly supported.","headline":"A solid, reusable design methodology for short non-binary LDPC codes on phase-noise channels; the 1.2 dB claim is plausible, but the DT bound supporting it is under-quantified.","tokens_in":15929,"tokens_out":1487,"would_cite":true,"duration_ms":18710,"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":"Short non-binary LDPC codes with DPSK reach within 1.2 dB of finite-length bounds on phase-noise channels down to a CER of 1e-3.","keywords":["non-binary LDPC codes","non-coherent detection","phase noise","DP algorithm","protograph codes","finite-length bounds","DPSK","turbo detection"],"falsifier":"Recompute the dependency-testing bound for the same channel parameters with a much larger Monte Carlo sample or deterministic numerical integration, and check whether the reported gap of about 1.2 dB at a CER of $10^{-3}$ persists; also test the same codes with the unquantized phase-increment density instead of the P_Delta=0.1 approximation.","tokens_in":14949,"feed_emoji":"📡","tokens_out":8507,"duration_ms":76041,"temperature":0.7,"pith_summary":"This paper shows that short non-binary protograph LDPC codes, concatenated with m-ary DPSK and decoded by a joint factor-graph receiver, can operate within 1.2 dB of a finite-length bound on channels with Wiener phase noise, at codeword error rates down to $10^{-3}$. The supporting designs cover rate 1/2 with 8-DPSK (160 symbols), rate 2/3 with 8-DPSK (120 symbols), and rate 3/4 with 16-DPSK (128 symbols). The design recipe is explicit: exhaustively search small protograph base matrices using iterative-threshold estimates from Monte Carlo density evolution, then refine the best ensemble by adding a single edge to remove error floors. If the claim holds, it gives a concrete, pilot-free benchmark for short-packet systems with low-cost oscillators and a target against which future code designs for phase-noise channels can be measured.","feed_headline":"Phase-noise codes land within 1.2 dB of finite-length bound","feed_subtitle":"Short non-binary LDPC-DPSK designs reach 1e-3 CER without pilots, beating turbo baselines by 0.7 dB.","key_machinery":"The engine is the protograph base matrix -- a small integer matrix whose entries count edges between check-node and variable-node types. The design searches exhaustively over matrices with entries in {0,1,2,3}, expurgates those with zero-weight columns or too many degree-1 columns, reduces to a minimal set modulo row/column permutations, and selects the ensemble with the best iterative threshold computed by Monte Carlo density evolution; a second step adds a single edge to the winning base matrix to lower the error floor. At the receiver, the discretized-phase (DP) algorithm runs forward-backward recursions on a phase-quantized trellis with L=8m levels and a simplified three-point increment distribution, producing extrinsic symbol probabilities that feed a non-binary belief-propagation decoder over GF(m). Detector and decoder exchange extrinsic messages iteratively as parts of one factor graph.","core_discovery":"The central claim is that a serially concatenated scheme consisting of a non-binary protograph LDPC code over GF(m) and an m-ary DPSK modulator, with iterative detection and decoding on a joint factor graph, reaches codeword error rates within 1.2 dB of the dependency-testing bound at $10^{-3}$ over a Wiener phase-noise AWGN channel. This is demonstrated for an 8-ary (160,80) rate-1/2 code, an 8-ary (120,80) rate-2/3 code, and a 16-ary (128,96) rate-3/4 code, with phase-noise standard deviations of 2° and 1° respectively. The authors further show that the protographs selected for the non-coherent channel also have the best thresholds among the searched candidates on the coherent AWGN channel, so the phase-noise-tolerant design is not achieved at the cost of coherent performance. They also report gains of about 0.7 dB over a reference serial turbo scheme with binary convolutional codes.","pith_inferences":["If the 1.2 dB gap holds under more careful bound estimation, it suggests the short-block performance penalty on this channel is mostly captured by the dependency-testing bound, and the phase discretization (L=8m) and simplified increment model are not limiting.","The same search procedure could be applied to other channels with memory -- for example, channels with frequency offset or time-varying fading -- by replacing the channel model in the density evolution and the DP detector.","Because the best non-coherent protographs coincide with the best coherent ones, a practical system could use a single code and switch between coherent and non-coherent receiver processing without redesign.","A natural test is to scale the design to longer blocks (N=500 or 1000) and see whether the 1.2 dB gap stays constant or grows, revealing whether the method is specific to the ultra-short regime."],"forward_implications":["Rate-1/2, 2/3 and 3/4 short non-binary LDPC codes with m-DPSK operate within 1.2 dB of the dependency-testing bound down to 10^-3 codeword error rate on Wiener phase-noise channels, without pilots.","Protographs that are threshold-optimal on the non-coherent channel also win on the coherent AWGN channel among the searched candidates, so a single code can serve both receiver modes.","The two-step design (threshold search plus single-edge refinement) removes visible error floors above 10^-3 at a small waterfall penalty.","The designed codes outperform a reference serial turbo scheme with binary convolutional codes by about 0.7 dB in both coherent and non-coherent operation."],"supporting_citations":[{"why":"Provides the Wiener phase-noise channel model and the discretized-phase detection framework used throughout.","marker":"[6]"},{"why":"Supplies the simplified three-point approximation of the phase-increment density used by the detector.","marker":"[8]"},{"why":"Defines the soft-output DP detection algorithm that computes extrinsic symbol probabilities on the joint factor graph.","marker":"[12]"},{"why":"Introduces non-binary belief-propagation decoding over GF(q), the decoder used in the iterative receiver.","marker":"[22]"},{"why":"Introduces non-binary protograph LDPC codes, the code family whose ensembles are searched in the design.","marker":"[24]"},{"why":"Derives the dependency-testing bound used as the finite-length performance benchmark.","marker":"[32]"},{"why":"Defines protograph-based LDPC code ensembles and the lifting procedure used to obtain finite-length codes.","marker":"[33]"},{"why":"Provides the Monte Carlo density evolution method used to compute iterative decoding thresholds of the protograph ensembles.","marker":"[38]"}],"fun_headline_variants":["Non-binary LDPC-DPSK: 1.2 dB from finite-length bound","Short non-binary LDPC codes within 1.2 dB of phase-noise bound","Phase-noise LDPC: 1.2 dB to bound, 0.7 dB over turbo","LDPC for phase noise: 1.2 dB from finite-length limit","Non-binary LDPC codes: 1.2 dB gap to phase-noise bound"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The benchmark against which the 1.2 dB gap is measured is a Monte Carlo estimate of the dependency-testing bound whose sample count and statistical accuracy are not reported, so a biased or noisy benchmark would change the headline gap.","fun_headline_variants_meta":{"raw":{"variants":["Non-binary LDPC-DPSK: 1.2 dB from finite-length bound","Short non-binary LDPC codes within 1.2 dB of phase-noise bound","Phase-noise LDPC: 1.2 dB to bound, 0.7 dB over turbo","LDPC for phase noise: 1.2 dB from finite-length limit","Non-binary LDPC codes: 1.2 dB gap to phase-noise bound"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000488,"raw_usage":{"total_tokens":2393,"prompt_tokens":927,"completion_tokens":1466,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":543,"completion_tokens_details":{"reasoning_tokens":1348}},"tokens_in":543,"tokens_out":1466,"duration_ms":12073,"temperature":1.0,"reasoning_tokens":1348,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:32:49.852001+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the dependency-testing bound for the same channel parameters with a much larger Monte Carlo sample or deterministic numerical integration, and check whether the reported gap of about 1.2 dB at a CER of $10^{-3}$ persists; also test the same codes with the unquantized phase-increment density instead of the P_Delta=0.1 approximation.","supporting_citations":[{"cited_title":"Communications over phase-noise channels: A tuto- rial review,","cited_arxiv_id":null,"evidence_quote":"Provides the Wiener phase-noise channel model and the discretized-phase detection framework used throughout."},{"cited_title":"Iterative decoding for coded noncoherent MPSK communications over phase-noisy AWGN channel,","cited_arxiv_id":null,"evidence_quote":"Supplies the simplified three-point approximation of the phase-increment density used by the detector."},{"cited_title":"Soft-output decoding of rotationally invariant codes over channels with phase noise,","cited_arxiv_id":null,"evidence_quote":"Defines the soft-output DP detection algorithm that computes extrinsic symbol probabilities on the joint factor graph."},{"cited_title":"Low density parity check codes over GF(q),","cited_arxiv_id":null,"evidence_quote":"Introduces non-binary belief-propagation decoding over GF(q), the decoder used in the iterative receiver."},{"cited_title":"Non- binary protograph LDPC codes for space communications,","cited_arxiv_id":null,"evidence_quote":"Introduces non-binary protograph LDPC codes, the code family whose ensembles are searched in the design."},{"cited_title":"Channel coding rate in the finite blocklength regime,","cited_arxiv_id":null,"evidence_quote":"Derives the dependency-testing bound used as the finite-length performance benchmark."},{"cited_title":"Low-density parity-check (LDPC) codes constructed from protographs,","cited_arxiv_id":null,"evidence_quote":"Defines protograph-based LDPC code ensembles and the lifting procedure used to obtain finite-length codes."},{"cited_title":"MacKay, Information Theory, Inference & Learning Algorithms","cited_arxiv_id":null,"evidence_quote":"Provides the Monte Carlo density evolution method used to compute iterative decoding thresholds of the protograph ensembles."}],"review_version":1}