{"id":"5f1b3abc-53de-4846-84a5-e752a770e23b","arxiv_id":"1908.00708","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Interleaved polar codes, created by putting random permutations between polar encoding stages, have an exactly computable ensemble weight distribution and give modest gains (0.1-0.3 dB) plus a parallel decoding option.","lead":"This paper inserts random shufflers between the encoding stages of polar codes, creating interleaved polar codes and deriving their average error performance. Generalists might read it because it offers a small but useful encoding tweak that also allows parallel decoding, relevant for ultra-reliable low-latency wireless links.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The simple-bound application to the ensemble-average WEF (Eq. 8) is unproven: the bound is nonlinear in A_d, and no argument shows the expectation commutes with the min and the ln(A_d)/N exponent.","rationale":"The paper's core contribution is the exact ensemble-average WEF and IOWEF recursions under the uniform interleaver assumption. Those recursions (Theorems 3-5) are mathematically clean and are independently supported by the N=32 exhaustive enumeration, where the sample average of 1000 realizations closely matches the analytical WEF(A). The polarization theorem is also plausible, since interleaving only permutes identical synthesized channels. The weakest point is the step from the average WEF to a BLER upper bound. Equation (8) is a nonlinear function of the WEF coefficients through the min operation and through r(delta) = ln[A_d]/N inside the exponent E(rho, delta). For a random code from the ensemble, the average of this bound over the ensemble is not automatically equal to the bound evaluated at the average WEF; the paper provides no convexity or concentration argument that would justify the swap. This is exactly the concern identified by the reader, and it is load-bearing for the paper's theoretical claim that the average WEF can be used to calculate an upper bound on the average BLER. The simulation demonstrations of relative performance (e.g., P=Q=2 vs P=Q=1) are empirical and would survive even if the bound were invalid, but the analytical upper-bound curves in Figures 6, 8, 10, and 11 would lose their stated status. The proposed numerical test would settle the question directly. Since the issue is a genuine gap rather than a demonstrated error, the conditional verdict remains appropriate.","tokens_in":18656,"tokens_out":10840,"duration_ms":110174,"concrete_test":"For the (32,16) i-polar ensemble of Section IV-C, generate many (e.g., 10^4) random realizations of the interleavers, compute the exact WEF of each realization, and evaluate the simple bound (8) at several SNRs (e.g., 4, 5, 6 dB) for each realization. Average these per-realization bounds and compare with the simple bound evaluated on the analytical ensemble-average WEF WEF(A) from Eq. (5). If the average-WEF bound is below the averaged per-realization bound at any tested SNR, the claimed upper bound on average BLER fails; if it is above, the bound survives this test. A supporting check is to plot the per-weight summand B_d(A) = min{ exp(-n E(rho, d/N, ln A/N)), A Q(sqrt(2 d rho)) } as a function of A: concavity in A over the relevant range would explain and justify the Jensen step, while any convex segment would show the step is not generally valid.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section IV-D applies the simple bound (8) to the ensemble-average WEF. For a fixed code, the bound is a sum over d of min{ exp(-n E(rho, delta)), A_d Q(sqrt(2 d rho)) }, with delta = d/N and r(delta) = ln[A_d]/N entering E(rho, delta), c0(delta), and f(rho, delta). The quantity actually bounded is the average BLER, whose simple bound is E_C[ sum_d min{ exp(-n E(rho, delta, ln A_d(C)/N)), A_d(C) Q(sqrt(2 d rho)) } ]. The paper instead plots sum_d min{ exp(-n E(rho, delta, ln E[A_d]/N)), E[A_d] Q(sqrt(2 d rho)) }. Because the min and the logarithmic exponent are nonlinear in A_d, these two expressions are not equal in general. A valid upper bound would need each summand to be concave in A_d so that Jensen's inequality gives E[min{...}] <= min{...} at E[A_d]; no such concavity argument is supplied. If the summand is not concave, the curves in Figures 6, 8, 10, and 11 are not guaranteed to bound either the average BLER or the BLER of any particular realization. This does not invalidate the WEF recursions (Theorems 3-5), but it directly undermines the advertised claim that the average WEF yields a BLER upper bound.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper introduces interleaved polar (i-polar) codes, obtained by inserting interleavers between consecutive stages of the polar encoder, and analyzes the ensemble average of their weight enumerating function (WEF) and input-output weight enumerating function (IOWEF) under the uniform-interleaver assumption. Theorems 3 and 4 give recursions for the average WEF/IOWEF, Theorem 5 gives WEF expressions for parallel concatenations, and the paper evaluates the Divsalar simple bound on the resulting average WEF to obtain claimed BLER upper bounds. A concatenated scheme with P outer codes and Q i-polar inner codes is also proposed; simulations show that the P=Q=2 BCH-aided construction outperforms a same-rate CRC-aided P=Q=1 construction at high SNR, and the paper argues for a parallel-decoding latency advantage.","tokens_in":18964,"tokens_out":16644,"duration_ms":176233,"significance":"If the core derivations are correct, the paper provides a parameter-free, exact ensemble-average weight spectrum for a randomized polar-code family, which is a useful analytical tool for short-blocklength code design and for concatenated-code analysis. The numerical validation in Table I (the analytical average matching a 1000-realization sample average for the (32,16) code) and the falsifiable high-SNR comparison between P=Q=2 and P=Q=1 constructions are concrete strengths, and the proposed parallel SCL decoding structure is a sensible latency-reduction idea. However, the central claim that the simple-bound evaluation on the ensemble-average WEF is an upper bound on the average BLER is not established, and one displayed formula appears truncated; these issues affect most of the analytical plots and the abstract's main claim.","major_comments":[{"comment":"The simple bound is evaluated at the ensemble-averaged coefficients A_bar_d = E[A_d(C)], and the result is labeled an upper bound on the average BLER. The simple bound is nonlinear in A_d: it contains min{...}, and the exponent E(rho, delta) depends on r(delta)=ln(A_d)/N through c0(delta) and f(rho,delta). For a random code C the quantity that must be bounded is E_C[ sum_d min{ exp(-N E(rho,d/N,ln A_d(C)/N)), A_d(C) Q(sqrt(2 d rho)) } ], whereas the paper plots sum_d min{ exp(-N E(rho,d/N,ln A_bar_d/N)), A_bar_d Q(sqrt(2 d rho)) }. These two expressions are not equal in general, and no concavity or monotonicity argument is supplied that would make the latter dominate the former. The linear union bound in Eq. (7) would commute with expectation, but the simple-bound curves in Figures 6, 8, 10, and 11 are therefore not established as upper bounds on the average BLER. This directly affects the abstract claim and the discussion in Sections IV-D and VI. At minimum, the authors should prove the needed inequality or replace the simple-bound evaluation by a bound that is linear (or otherwise proven to commute with expectation) in the WEF coefficients.","section":"Section IV-D (Eq. (8))"},{"comment":"The displayed simple bound sums over d = d_min to N-K+1. In the reference [8], the simple bound sums over all nonzero codeword weights, i.e., d = d_min, ..., N (equivalently all d with A_d > 0). A binary linear (N,K) code can contain codewords of weight greater than N-K+1; the (32,16) example in Table I has nonzero A_d for d = 28 and d = 32, both beyond N-K+1 = 17. Since all summands are nonnegative, truncating the sum at N-K+1 can only decrease the right-hand side, so without an additional argument the truncated expression is not an upper bound. Please either correct the summation range to match [8] or justify the truncation explicitly; if the actual computations used all weights, the displayed formula should be corrected.","section":"Section IV-D, Eq. (8)"}],"minor_comments":[{"comment":"The statement that the observation of only two WEF types among 1000 realizations means 'with high probability, any realizations are as good as the ensemble average' is an empirical observation, not a proven concentration result; please soften the wording or state it as a numerical observation.","section":"Section IV-C"},{"comment":"The text says that Q parallel SCL decoders each outputting L candidates produce 'LQ combined message vectors'; the number of combinations is L^Q, not LQ. Please correct the notation or define it clearly.","section":"Section V"},{"comment":"There are several typographical errors: 'inferier' should be 'inferior' in Section V, 'calso' should be 'also' in Section III, and 'Polyanskyi' should be 'Polyanskiy' in Section VI-A. Please proofread.","section":"Throughout"},{"comment":"The captions and text should state explicitly that the curves labeled 'upper bounds' are computed from the ensemble-average WEF, so that readers do not interpret them as bounds on the BLER of the single simulated realization.","section":"Figures 6, 8, 10, 11"}],"recommendation":"major_revision","confidential_remarks":"The main technical concern is the application of the nonlinear simple bound to the ensemble-average WEF; this needs either a proof or a reformulation of the claimed bound. If the authors can repair that point and correct the summation range in Eq. (8), the paper is likely publishable. I do not see a circularity or novelty problem: the WEF recursion is derived from the code structure rather than fitted to the target curves."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Hi [Name],\n\nYou asked about the Chiu i-polar paper. The genuinely new thing is the construction—interleavers placed between the internal stages of a polar encoder—and the exact ensemble-average WEF/IOWEF recursions in Theorems 3 and 4. Those are derived cleanly, and the small-case check against 1000 realizations at (32,16) is credible. The paper deserves credit for that; it is a real analytical tool for a new code family.\n\nThe soft spot is the jump from average WEF to the BLER upper bound. The simple bound in (8) is nonlinear in the WEF coefficients—it has the min{} and the ln(A_d)/N inside the exponent. Substituting E[A_d] into that bound and calling the result an upper bound on average BLER needs a Jensen/concavity argument, and the paper does not supply one. The stress-test note is on target: as it stands, Figures 6, 8, 10, 11 are not guaranteed to bound the average BLER. That does not invalidate the WEF recursions, but it does undercut the advertised theoretical claim.\n\nThere are two smaller things. First, ensemble typicality is only checked at length 32; for N=512/1024 the paper relies on the simulation matching the bound, which is evidence but not proof. Second, the performance gains over polar codes are modest (0.1–0.3 dB) and the latency argument is qualitative—fine as a motivation, not a demonstrated URLLC solution.\n\nOn balance, this is a solid paper with one load-bearing rigor gap. If I were the editor I would send it to review; a good referee can ask for either a proof of the commutation/concavity or a clear restatement of the curves as heuristic approximations. The WEF part is worth publishing even if the bound claim gets softened.\n\nMy verdict: worth engaging with, conditional on the bound issue being fixed.","headline":"Solid ensemble-WEF analysis for a new interleaved-polar-code family; the BLER upper-bound claim has a rigor gap that a referee should push on.","tokens_in":19491,"tokens_out":2595,"would_cite":true,"duration_ms":25118,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["94B05","94B35"],"pacs":[],"model":"deepseek-v4-flash","headline":"Interleaved polar codes get an exact ensemble-average weight enumerator, making high-SNR block error rates analytically predictable.","keywords":["polar codes","interleaved polar codes","weight enumerating function","uniform interleaver","block error rate bound","concatenated codes","SCL decoding","ultra-reliable low-latency communications"],"falsifier":"For a small ensemble such as (16,8) or (32,16), enumerate all interleaver realizations, compute the true mean BLER under maximum-likelihood decoding, and compare it with the simple bound evaluated on the ensemble-average WEF at several SNRs; any violation at high SNR would disprove the bound-as-average claim, as would a direct check that expectation does not commute with the min-expression in Eq. (8).","tokens_in":18454,"feed_emoji":"🔀","tokens_out":7704,"duration_ms":74690,"temperature":0.7,"pith_summary":"This paper asks whether polar codes can be made analytically tractable for maximum-likelihood performance prediction. It answers by inserting interleavers between encoding stages and averaging over all possible interleavers: the resulting interleaved-polar (i-polar) ensemble has an exact average weight enumerating function, computable recursively, which feeds an upper bound on average block error rate. A sympathetic reader would care because exact ML analysis of ordinary polar codes is absent; here the ensemble average gives a tool for predicting high-SNR behavior and designing concatenations. The paper further proposes a two-stage concatenation with P outer and Q inner codes, showing P=Q=2 beats CRC-aided i-polar P=Q=1 at the same length and rate at high SNR, while Q parallel decoders cut latency. If true, this gives a design-analytic route for ultra-reliable low-latency polar-coded systems.","feed_headline":"Interleavers give polar codes an exact average weight spectrum","feed_subtitle":"Randomizing interleavers between polar encoder stages makes high-SNR block error rates analytically predictable and enables parallel…","key_machinery":"The load-bearing object is the uniform interleaver of length n, a random permutation device that maps any weight-d input to all $\\binom{n}{d}$ permutations with equal probability. Lemma 2 and Theorem 3 define an operator $H_m$ acting on the two half-code WEFs: for codewords of weights $d_1,d_2$, the averaged weight enumerator of the joined stage is\n$$\\sum_{k=\\max(0,d_1+d_2-n)}^{\\min(d_1,d_2)} \\frac{\\binom{d_2}{k}\\binom{n-d_2}{d_1-k}}{\\binom{n}{d_1}} $Y^{{d_1+2d_2-2k}}$,$$\nwith $n=2^{m-1}$, and $F_m$ is the same operation on IOWEFs. Recursing from length-1 base codes $1+Y$ or $1$ computes the exact ensemble-average WEF and IOWEF without exhaustive interleaver enumeration. Theorem 1 shows the interleavers do not change the synthesized channels, so the same frozen-bit selection applies while the weight spectrum changes.","core_discovery":"The central claim is that the weight enumerating function and input-output weight enumerating function averaged over the (N,K,A) ensemble of i-polar codes can be computed exactly by recursive operators applied to the two half-code enumerators, using the uniform interleaver assumption. This average WEF, not an approximation to any particular code, is then used with the simple bound of [8] to upper-bound the average BLER of a random ensemble member. For concatenated schemes with P parallel outer codes and Q parallel i-polar inner codes separated by a uniform interleaver, the average WEF is obtained by raising component WEFs to powers and applying the serial-concatenation formula, so bounds can be evaluated without simulating the code. The paper reports that the resulting bounds track simulated SCL performance at high SNR, and that P=Q=2 gives a steeper BLER slope than P=Q=1 at the same length and rate, outperforming CRC-aided i-polar with P=Q=1 at high SNR while Q decoders run in parallel.","pith_inferences":["The same recursive averaging could be applied to any code family built from the x+y|y construction, such as Reed-Muller-style supercodes, giving exact ensemble WEFs wherever the base code WEFs are known.","The paper's WEF-based greedy unfrozen-bit selection could be benchmarked against Gaussian-approximation and 5G sequences by checking whether the selected set changes at finite list sizes; a testable prediction is that WEF-selected sets improve SCL performance beyond the gains shown for fixed A.","A second-moment or variance analysis of the WEF over the ensemble would quantify how representative the average is for long block lengths, since the paper only demonstrates concentration indirectly through 1000 realizations at N=32.","For Q>1, the L^Q combined-candidate step limits practical parallelism; combining the suggested local error detection with list pruning could trade a small BLER loss for much lower decoder complexity, an extension the paper sketches but does not simulate."],"forward_implications":["The exact ensemble-average WEF of the (N,K,A) i-polar ensemble is computable by recursion, giving an analytical predictor of ML BLER at high SNR that ordinary polar codes lack.","Because Theorem 1 shows polarization is unchanged, i-polar codes inherit polar bit-channel selection; the interleavers only redistribute codeword weights, which can lower the multiplicity of minimum-weight codewords.","The concatenated scheme's average WEF is derived from the outer WEF and inner IOWEF via the serial-concatenation formula, so BLER upper bounds can be evaluated for CRC-, RRA-, or BCH-style outer codes without simulation.","For the (1024,494) code with P=Q=2, the BLER curve has a steeper high-SNR slope than the P=Q=1 CRC-aided i-polar code of the same length and rate, and the paper reports it beats the P=Q=1 ML lower bound at high SNR.","With Q i-polar decoders operating in parallel, decoding latency is reduced compared to a single longer polar decoder, which the paper identifies as a practical benefit for ultra-reliable low-latency communication."],"supporting_citations":[{"why":"Establishes channel polarization and the standard graph whose stages the interleavers are inserted into.","marker":"[1]"},{"why":"Provides SCL decoding and the CRC-aided polar code baseline that the concatenated scheme is compared against.","marker":"[2]"},{"why":"Introduces the uniform interleaver assumption used to average WEFs over the ensemble.","marker":"[4]"},{"why":"Supplies the simple bound (Eq. 8) that converts the average WEF into a BLER upper bound.","marker":"[8]"},{"why":"Gives the Gaussian-approximation bit channel selection used to fix the unfrozen set A.","marker":"[14]"},{"why":"Gives the serial-concatenation WEF formula used to combine outer and inner super-codes.","marker":"[23]"},{"why":"Provides WEF computation for systematic RRA codes used as outer component codes.","marker":"[25]"},{"why":"Lists WEFs of BCH codes used for the BCH-aided i-polar outer code.","marker":"[24]"},{"why":"Defines the 5G NR polar sequence and CRC used in the practical comparison.","marker":"[29]"}],"fun_headline_variants":["Exact average weight spectrum for i-polar codes","Parallel polar decoders with provable BLER bounds","Interleaved polar codes beat CRC-aided at high SNR","I-Polar codes: analytic bounds for URLLC","Provable BLER bounds for interleaved polar codes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that evaluating the simple BLER bound on the ensemble-average weight enumerator still gives a valid upper bound on the average block error rate, since the bound is nonlinear in the enumerator coefficients and averaging is taken before, not after, that nonlinearity.","fun_headline_variants_meta":{"raw":{"variants":["Exact average weight spectrum for i-polar codes","Parallel polar decoders with provable BLER bounds","Interleaved polar codes beat CRC-aided at high SNR","I-Polar codes: analytic bounds for URLLC","Provable BLER bounds for interleaved polar codes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000304,"raw_usage":{"total_tokens":1775,"prompt_tokens":1003,"completion_tokens":772,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":619,"completion_tokens_details":{"reasoning_tokens":693}},"tokens_in":619,"tokens_out":772,"duration_ms":6882,"temperature":1.0,"reasoning_tokens":693,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:37:57.857364+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a small ensemble such as (16,8) or (32,16), enumerate all interleaver realizations, compute the true mean BLER under maximum-likelihood decoding, and compare it with the simple bound evaluated on the ensemble-average WEF at several SNRs; any violation at high SNR would disprove the bound-as-average claim, as would a direct check that expectation does not commute with the min-expression in Eq. (8).","supporting_citations":[{"cited_title":"Channel polarization: A method for constructing capacity-achieving codes for symmetric binary-input memoryless channels,","cited_arxiv_id":null,"evidence_quote":"Establishes channel polarization and the standard graph whose stages the interleavers are inserted into."},{"cited_title":"List decoding of polar codes,","cited_arxiv_id":null,"evidence_quote":"Provides SCL decoding and the CRC-aided polar code baseline that the concatenated scheme is compared against."},{"cited_title":"Unveiling turbo codes: Some results on parallel concatenated coding schemes,","cited_arxiv_id":null,"evidence_quote":"Introduces the uniform interleaver assumption used to average WEFs over the ensemble."},{"cited_title":"A simple tight bound on error probability of block codes with application to turbo codes,","cited_arxiv_id":null,"evidence_quote":"Supplies the simple bound (Eq. 8) that converts the average WEF into a BLER upper bound."},{"cited_title":"Polynomial representations of polar codes and decoding under overcomplete representations,","cited_arxiv_id":null,"evidence_quote":"Gives the Gaussian-approximation bit channel selection used to fix the unfrozen set A."},{"cited_title":"Serial concatenation of interleaved codes: Performance analysis, design, and iterative decoding,","cited_arxiv_id":null,"evidence_quote":"Gives the serial-concatenation WEF formula used to combine outer and inner super-codes."},{"cited_title":"Structured IRA codes: Performance analysis and construction,","cited_arxiv_id":null,"evidence_quote":"Provides WEF computation for systematic RRA codes used as outer component codes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Lists WEFs of BCH codes used for the BCH-aided i-polar outer code."},{"cited_title":"3GPP TS 38.212 version 15.2.0 Release 15 (available on line: https://www.etsi.org/deliver/etsi ts/138200 138299/138212/15.02.00 60/ts 138212v150200p.pdf), Jul","cited_arxiv_id":null,"evidence_quote":"Defines the 5G NR polar sequence and CRC used in the practical comparison."}],"review_version":1}