{"id":"89a8ad7a-9c31-4571-b921-7278d43bf6b1","arxiv_id":"1908.05889","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Polar code error analysis and construction are expressed through the weight distribution of polar subcodes, yielding explicit bounds and linear-complexity construction metrics.","lead":"This paper introduces a weight distribution tool called the polar spectrum for analyzing and constructing polar codes, replacing iterative channel reliability calculations with explicit formulas. The authors use it to derive error bounds and two construction metrics that they report match or beat conventional designs in simulations.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"UBW/SUBW construction claims depend on per-configuration tuning of design SNR α; no evidence a single fixed α preserves the reported gains.","rationale":"The reader's weakest assumption identifies the same load-bearing point: the construction metrics depend on α and are tuned per configuration. I examined other potential concerns. Theorem 1's derivation in Eq. (10) is not a set equality, but the bound can be repaired using the coset structure D(i)_N = C(i+1)_N ⊕ g_i, which gives Sum_right = L(g_i)·Sum_left and hence E_i ⊆ ⋃_{c(1)} { W_N(0) ≤ W_N(c(1)) }; so that proof gap is not fatal. The enumeration algorithm's apparent ordering issue in Algorithm 1 is resolved by the separate loops for solving S(l) and computing A(l). The alpha-tuning issue, by contrast, directly bears on the paper's central practical claim and on its internal consistency: the text calls the construction channel-independent by fixing α, while the simulations vary α across configurations. Since the reader already recommends CONDITIONAL, the verdict should remain unchanged, with the condition being a demonstration of robustness to α and release of code/data.","tokens_in":23103,"tokens_out":19060,"duration_ms":186142,"concrete_test":"Re-run the UBW/SUBW constructions for every (N,R) in Section VI-B with one fixed α=4 dB and simulate SC and SCL BLER; if any configuration loses more than roughly 0.2 dB compared with the tuned-α curves in Figs. 4–7, the claim of a simple, channel-independent construction with the reported gains is not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central construction claim rests on treating UBW and SUBW as simple, explicit, channel-independent metrics. In Eqs. (54) and (58), both metrics depend on a design SNR α: UBW(i)_N = max_d { L_N^{(i)}(d) − d α } and SUBW(i)_N = L_N^{(i)}(d_min) − d_min α. The text states that fixing α to a constant yields a channel-independent construction, but Section VI-B tunes α separately for every (N,R): N=128, R=1/2 uses 4 dB; N=1024 uses 1.5/4/4.5 dB (UBW) and 1/3.5/4 dB (SUBW) for R=1/3,1/2,2/3; N=4096, R=2/3 uses 4.5/3.5 dB. This is eight different α values spanning 1–4.5 dB. Because the term −d α reweights all distances, the argmax in UBW and the relative ordering of channels can change as α varies; no stability analysis or universal α is provided. Without such evidence, the reported SC similarity and SCL superiority in Figs. 4–7 could be artifacts of per-configuration tuning rather than intrinsic properties of the polar-spectrum construction. The log-sum-exp approximation in Eq. (50) and the min-weight truncation in Eq. (55) add further looseness, so the link from Theorem 3 to UBW/SUBW is heuristic. No code or data are released to reproduce the curves, which makes the tuning dependence impossible to audit.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces the polar spectrum, the weight distribution of a certain polar subcode associated with each polarized channel, and uses it as the basis for a framework to analyze and construct polar codes. The main theoretical results are a union bound (Theorem 3) and a union-Bhattacharyya (UB) bound (Theorem 6) on the block error probability under successive cancellation decoding, upper and lower bounds on the symmetric capacity of polarized channels (Theorem 7 and Eqs. (45)-(46)), and an iterative enumeration algorithm for the polar spectrum based on the MacWilliams identities (Algorithm 1). The paper then proposes two construction metrics, UBW and SUBW, derived from the UB bound, and reports simulations showing that these constructions achieve similar SC performance and superior SCL performance compared with Tal-Vardy, Gaussian approximation, and polarized-weight constructions for selected code lengths and rates.","tokens_in":23497,"tokens_out":2809,"duration_ms":28376,"significance":"If the polar-spectrum framework is accepted, it offers a useful analytic counterpart to iterative channel-reliability computations and a new way to interpret polarization through distance spectra. The paper has clear strengths: Theorem 3 is a genuine first-principles union bound with explicit PEP formulas for BEC, BSC, and AWGN (Theorems 4-5); Algorithm 1 is a concrete, polynomial-time enumeration procedure with an explicitly stated complexity; and the relation between polar subcodes and dual subcodes (Theorem 8, Proposition 6) is a clean algebraic observation. The contribution is conditional, however: the UBW/SUBW construction claims depend heavily on a design SNR that is tuned separately for each (N,R) configuration, and the passage from the UB bound to the construction metrics involves uncontrolled approximations. These issues do not invalidate the core bound framework but do undermine the paper's headline claim that UBW/SUBW are simple, channel-independent, and superior in SCL decoding.","major_comments":[{"comment":"The claimed channel-independence of the UBW and SUBW constructions is not supported by the reported experiments. In Section VI-B the design SNR α is set to 4 dB for N=128; to 1.5/4/4.5 dB for UBW and 1/3.5/4 dB for SUBW at N=1024 for R=1/3, 1/2, 2/3; and to 4.5/3.5 dB at N=4096, R=2/3. Because the term −d·α in (54) reweights the spectrum, changing α can change the argmax in UBW and the relative ordering in SUBW. Without a stability analysis over α, a single fixed α, or a demonstration that the reported gains are insensitive to α in a neighborhood, the SCL gains in Figs. 4-7 may be artifacts of per-configuration tuning rather than properties of the polar-spectrum metric. This is load-bearing for the paper's central construction claim and needs to be addressed directly.","section":"VI-B and Eqs. (54), (58)"},{"comment":"The approximation Z(W_N^{(i)}) ≳ (Z(W))^{d_min^{(i)}} is stated without proof and is used as a lower bound to derive the symmetric-capacity upper bounds in Eqs. (44) and (46). As written, Eq. (41) is not a rigorous inequality: it is only asserted that the minimum-weight term 'can be intended to serve as' a lower bound. The subsequent capacity upper bounds are therefore not justified. The authors should either prove a valid lower bound on Z(W_N^{(i)}) in terms of the polar spectrum, or clearly mark the capacity upper bounds as heuristic rather than proven.","section":"III-D, Eq. (41)"},{"comment":"The definition of UBW rests on the approximation ln(Σ_d exp(L_N^{(i)}(d) + d ln Z(W))) ≈ max_d {L_N^{(i)}(d) + d ln Z(W)}. This log-sum-exp approximation has no stated error bound, and SUBW further truncates to the single minimum-weight term in Eq. (55). Consequently, the link from the exact UB bound of Theorem 6 to the ordering used for code construction is heuristic. The paper should quantify the approximation error or provide simulations showing that the UBW/SUBW ordering coincides with the UB-bound ordering in the parameter regimes of interest.","section":"V-A, Eq. (50)"}],"minor_comments":[{"comment":"The word 'Poltkin' appears twice in the induction argument; it should be 'Plotkin'.","section":"IV-A, Proof of Theorem 8"},{"comment":"In the statement of MacWilliams identities for length 2N, the notation mixes indices (l, 2N+2−l) in a way that is easy to misread; a short derivation or a note that S_{2N}^{(2N+2−l)} is the dual of S_{2N}^{(l)} would improve readability.","section":"Algorithm 1, Eq. (49)"},{"comment":"The table would benefit from a column header that separates the index i, the weight d, and the multiplicity A_N^{(i)}(d) explicitly; currently the repeated vertical-pair formatting (e.g., '1 1(31) 32 32 32 1') is difficult to parse.","section":"Table I"},{"comment":"No code or data files are mentioned, so the numerical comparisons, especially the tuned α values, are not independently auditable. A brief reproducibility statement would strengthen the paper.","section":"VI-B"}],"recommendation":"major_revision","confidential_remarks":"The core theoretical framework (polar spectrum, union bound, MacWilliams-based enumeration) is a solid contribution and the algebraic duality result is clean. The main risk to the manuscript is the construction section: the apparent per-configuration tuning of the design SNR α, together with the unquantified log-sum-exp approximation, currently makes the headline performance claims not reproducible from the text alone. I recommend major revision rather than rejection because these concerns are addressable: the authors could fix α across all configurations, provide a sensitivity analysis, or clearly re-scope the claims as channel-dependent with a tuned parameter. The paper would also benefit from explicitly labeling Eq. (41) and the capacity upper bounds as heuristic unless a proof is supplied."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. The polar-spectrum framework is real: the mapping from a polarized channel to its polar subcode, and the union-bound Theorem 3, give a legitimate way to think about SC error events that goes beyond Bhattacharyya-parameter counting. The MacWilliams-based Algorithm 1 is also a substantive advance: it computes these spectra recursively at O(N^3), which is far better than brute-force weight enumeration, and the duality results (Theorem 8, Propositions 6–8) look correct to me. The construction metrics UBW/SUBW, however, are not as clean as advertised, because the design SNR alpha is tuned separately for nearly every (N,R) in the simulations and no evidence is given that a single fixed alpha preserves the ordering.\n\nThe strength of the paper is the analysis section. Theorem 3 follows from standard union-bound logic once you accept the subcode/PEP setup; the BEC, BSC, and AWGN PEP formulas are standard; and the numerical comparisons among the union, UB, and simplified UB bounds are honest about which bounds are loose. The enumeration algorithm and its complexity discussion are the most citable part of the work.\n\nNow the soft spots, in proportion. Equation (41) is the clearest gap: it asserts a lower bound on the Bhattacharyya parameter using only the minimum-weight term, marked with an unproved \"approximately less than or equal to,\" and this feeds the capacity upper bound in (44)/(46). Since the capacity bounds are explicitly coarse, this is a moderate issue, not fatal, but it needs either a proof or a clear \"heuristic\" label. The bigger practical concern is alpha. Section VI-B uses 4 dB for N=128,R=1/2; 1.5/4/4.5 and 1/3.5/4 for N=1024; and 4.5/3.5 for N=4096. Because the -alpha*d term reweights every distance, the UBW argmax and SUBW ordering can change as alpha varies. The paper says a fixed constant makes the construction channel-independent, but it never shows that one constant works across configurations or that the reported SCL gains survive an alpha sweep. Without that stability check, or released code and data, I read the SCL gains as promising but not established. This is parameter fitting, not circularity: the bound derivations do not assume the construction, and the enumeration algorithm stands on its own.\n\nMinor: the log-sum-exp approximation in (50) is standard, and the min-weight truncation in SUBW is explicitly a simplification, so those are fine.\n\nWho should read this: anyone working on polar-code construction or distance-spectrum analysis of polar codes. The framework and enumeration algorithm deserve a serious referee. I would send it out and ask for: a proof or qualification of (41), an alpha-stability study or a universal-alpha result, and ideally the enumeration code and simulation scripts. With those, the construction claims could become solid; without them, the paper's lasting value is the framework, which is already worth publishing.","headline":"A genuinely new polar-spectrum framework with a sound enumeration algorithm, but the UBW/SUBW construction story leans on per-configuration SNR tuning that is not yet shown to be robust.","tokens_in":23939,"tokens_out":2013,"would_cite":true,"duration_ms":21870,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["94B05","94B35","94A24"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that the block error probability of polar codes under successive cancellation decoding can be bounded by a weight spectrum of the code, and that this bound leads to explicit construction metrics.","keywords":["polar codes","polar spectrum","polar subcode","weight distribution","MacWilliams identities","union bound","successive cancellation decoding","polar code construction"],"falsifier":"Find one code length and rate where every fixed $\\alpha$ used in UBW/SUBW produces an information set whose simulated SC or SCL block error rate is worse than a standard iterative construction at the operating SNR, or show that the $\\alpha$ reproducing the best information set changes with the channel, so that no channel-independent sequence exists.","tokens_in":22916,"feed_emoji":"📡","tokens_out":8404,"duration_ms":76936,"temperature":0.7,"pith_summary":"This paper sets out to give polar codes the same kind of distance-spectrum analysis that classical linear codes enjoy. It introduces the polar spectrum, the weight distribution of the subcode attached to each polarized channel, and proves a union bound: under successive cancellation decoding the block error probability is at most the sum, over the chosen information channels, of the spectrum-weighted pairwise error probabilities. From that bound it derives two explicit construction metrics, UBW and SUBW, which order polarized channels using the polar spectrum plus a single design signal-to-noise ratio. The reported simulations show codes made this way perform about as well as standard iterative constructions under SC decoding and better under list decoding, while avoiding the error floor seen with an empirical weight-based construction. If the bound and metrics hold up, polar-code design no longer needs per-channel Monte Carlo or density-evolution calculations.","feed_headline":"One weight spectrum bounds polar-code errors and designs codes","feed_subtitle":"UBW/SUBW metrics match iterative construction under SC decoding and win under list decoding.","key_machinery":"The load-bearing object is the polar subcode $D_N^{(i)}$, the set of codewords generated by input vectors whose first $i-1$ bits are $0$ and whose $i$-th bit is $1$; its weight distribution $A_N^{(i)}(d)$ is the polar spectrum. Each polarized channel corresponds to one such subcode, so the SC error bound becomes a spectrum calculation: the channel's contribution to BLER is a weighted count of codewords by Hamming weight. The computation uses two structural facts: the Plotkin decomposition of polar codes, and the duality $C_N^{(N+2-i)} = (C_N^{(i)})^\\perp$ for $i \\geq N/2+1$, which lets the algorithm obtain one subcode's weight distribution from its dual through the MacWilliams identities. This machinery converts a problem normally solved by iterative channel-tracking into an exact combinatorial count.","core_discovery":"The central claim is Theorem 3: for a fixed length $N$, information set $A$, the block error probability under SC decoding satisfies $P_e(N,K,A) \\leq \\sum_{i\\in A}\\sum_d A_N^{(i)}(d)P_N^{(i)}(d)$, where $A_N^{(i)}(d)$ is the number of weight-$d$ codewords in the polar subcode $D_N^{(i)}$ and $P_N^{(i)}(d)$ is the pairwise error probability between the zero codeword and a weight-$d$ codeword. The paper derives $P_N^{(i)}(d)$ explicitly for the binary erasure, binary symmetric, and AWGN channels, and replaces it by $(Z(W))^d$ to obtain the looser union-Bhattacharyya bound. It then shows that the spectra can be enumerated exactly off-line: pairs of subcodes are dual, so their weight distributions are tied by the MacWilliams identities, and a recursive Plotkin-based algorithm computes all spectra for a fixed $N$ in $O(N^3)$ time. Taking logarithms of the UB bound gives the UBW metric, and retaining only the minimum-weight term gives SUBW; both are linear-time once the spectrum is known.","pith_inferences":["The mechanism suggests a sharper, testable conjecture: information sets from UBW/SUBW are close to maximizing the minimum polar-subcode distance among early indices, and that property, not SNR tuning, drives the list-decoding gains.","The same spectrum-bound framework should extend to polar codes with kernels other than the $2\\times2$ matrix, since it needs only a Plotkin-type decomposition and subcode duality; a MacWilliams enumeration for generalized kernels would give explicit constructions there.","If fixed-$\\alpha$ ordering fails, a natural fallback is an $\\alpha$-free partial order derived by comparing UBW curves as functions of SNR rather than at one design point."],"forward_implications":["Polar-code construction becomes a one-time combinatorial computation: enumerate spectra off-line, then select information bits by sorting UBW/SUBW values in linear time.","The union-Bhattacharyya bound gives an analytical BLER estimate for SC decoding, so code performance can be predicted from the weight distribution without running density evolution or Monte Carlo simulation.","Because the spectra for all information sets of a fixed length come from one enumeration, the same precomputation serves every code rate at that length.","The reported SCL gains and the disappearance of the empirical construction's error floor imply that spectrum-based construction chooses information sets better matched to list decoding, not only SC decoding."],"supporting_citations":[{"why":"Supplies the polar transform, SC decoding model, and the sum-of-channel-parameters upper bound that the new spectrum bound refines and replaces.","marker":"[1]"},{"why":"Provides density evolution, the high-accuracy iterative construction baseline whose accuracy the spectrum bound is compared against.","marker":"[4]"},{"why":"Supplies the high-precision quantization construction used as a performance baseline in the simulations.","marker":"[5]"},{"why":"Provides the Gaussian-approximation construction baseline at medium complexity.","marker":"[6]"},{"why":"Gives the empirical polarized-weight construction whose performance and error floor the paper compares with UBW/SUBW.","marker":"[9]"},{"why":"Introduces the list decoding that the paper uses to demonstrate the spectrum-based constructions' advantage.","marker":"[13]"},{"why":"States the MacWilliams identities that link dual subcodes and enable the exact enumeration algorithm.","marker":"[20]"}],"fun_headline_variants":["Polar spectrum: one weight distribution for bounds and code design","Weight spectrum unlocks polar-code bounds and construction","Polar spectrum yields simple construction metrics for polar codes","From weight distribution to polar-code design: UBW and SUBW","Polar spectrum: exact error bounds and fast construction"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The construction's claimed channel-independence rests on fixing one design signal-to-noise ratio $\\alpha$ in the UBW/SUBW formulas; the simulations choose $\\alpha$ separately for each code length and rate, so if no single $\\alpha$ orders channels correctly across configurations the method's practical advantage over iterative construction is weakened.","fun_headline_variants_meta":{"raw":{"variants":["Polar spectrum: one weight distribution for bounds and code design","Weight spectrum unlocks polar-code bounds and construction","Polar spectrum yields simple construction metrics for polar codes","From weight distribution to polar-code design: UBW and SUBW","Polar spectrum: exact error bounds and fast construction"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000245,"raw_usage":{"total_tokens":1582,"prompt_tokens":1037,"completion_tokens":545,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":653,"completion_tokens_details":{"reasoning_tokens":466}},"tokens_in":653,"tokens_out":545,"duration_ms":4754,"temperature":1.0,"reasoning_tokens":466,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:02:20.108532+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find one code length and rate where every fixed $\\alpha$ used in UBW/SUBW produces an information set whose simulated SC or SCL block error rate is worse than a standard iterative construction at the operating SNR, or show that the $\\alpha$ reproducing the best information set changes with the channel, so that no channel-independent sequence exists.","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":"Supplies the polar transform, SC decoding model, and the sum-of-channel-parameters upper bound that the new spectrum bound refines and replaces."},{"cited_title":"Performance of polar codes with the construc- tion using density evolution,","cited_arxiv_id":null,"evidence_quote":"Provides density evolution, the high-accuracy iterative construction baseline whose accuracy the spectrum bound is compared against."},{"cited_title":"How to construct polar codes,","cited_arxiv_id":null,"evidence_quote":"Supplies the high-precision quantization construction used as a performance baseline in the simulations."},{"cited_title":"Efficient design and decoding of polar codes,","cited_arxiv_id":null,"evidence_quote":"Provides the Gaussian-approximation construction baseline at medium complexity."},{"cited_title":"β-expansion: A Theoretical Framework for Fast and Recursive Construction of Polar Codes,","cited_arxiv_id":null,"evidence_quote":"Gives the empirical polarized-weight construction whose performance and error floor the paper compares with UBW/SUBW."},{"cited_title":"List decoding of polar codes,","cited_arxiv_id":null,"evidence_quote":"Introduces the list decoding that the paper uses to demonstrate the spectrum-based constructions' advantage."},{"cited_title":"A theorem on the distributionof weights in a systematic code,","cited_arxiv_id":null,"evidence_quote":"States the MacWilliams identities that link dual subcodes and enable the exact enumeration algorithm."}],"review_version":1}