{"id":"80adade7-264d-4f89-80ad-56d79a313b05","arxiv_id":"2512.04418","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Modified special-node SC decoding for polar IR-HARQ achieves the same node-traversal count as non-HARQ decoding, cutting traversals by 72% with no FER penalty.","lead":"This paper adapts fast 'special-node' decoders for polar error-correcting codes so they can work with Incremental Redundancy HARQ retransmission schemes. The result is a 72% reduction in decoder node traversals for a length-2048 code without loss of frame-error performance, targeting high-throughput 6G links.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"PC-frozen bit timing assumption unproven: special-node decoding requires each PC-frozen bit's mapped info bit to be decoded earlier, which [6]'s construction may not guarantee.","rationale":"The reader's weakest assumption identifies exactly the same load-bearing condition: the decoder knows pc at the time each special node is decoded. This is the crux because all new special-node algorithms are derived algebraically from pc and would collapse if pc were not yet known. The paper gives no proof or explicit statement that the IR-HARQ construction guarantees this causality; Section III simply introduces the formulas. This is an unverified external dependency inherited from [6], not an internal inconsistency in the derivations. The stress-test concludes that the reader's CONDITIONAL verdict is appropriate: the central claims are sound if the timing assumption holds, but they are not self-contained. We see no reason to escalate to REJECT without checking the mapping, and no reason to downgrade to UNVERDICTED because the paper does present a coherent method and simulation. The proposed concrete test—a simple index-ordering check on the [6] mapping—directly settles the uncertainty. We agree with the reader's assessment and therefore recommend no change to the verdict.","tokens_in":7700,"tokens_out":7064,"duration_ms":64843,"concrete_test":"Reproduce the exact IR-HARQ bit-position mapping of [6] for the length-2048 configuration used in Fig. 1. For every PC-frozen bit (at any stage of the code tree), compute the SC decoding index of its mapped new information bit. Then, for each special node detected in configuration A, check that the mapped info index for every PC-frozen bit inside the node is strictly less than the smallest leaf index of that node. If a counterexample exists, the proposed special-node rules cannot be evaluated in the assumed decoding order. If no counterexample exists, the timing assumption is satisfied and the traversal/FER claims remain plausible.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proposed fast SC decoding assumes that when a special node is processed, the encoded PC-frozen vector pc for that node is already known. This is necessary for the modified REP rule (Eq. 4), SPC rule (Eq. 16), RPC rule (Eq. 23), and PCR modification (Section III.G). The paper never states or proves that every PC-frozen bit in a detected special node has its mapped new information bit located strictly to the left (lower index) in the SC decoding order. Under the matrix-extension IR-HARQ of [6], original information bits are reclassified as PC-frozen and mapped one-to-one to new information bits, but the paper does not show that every new info bit precedes its PC-frozen partner. If any PC-frozen bit's source lies to the right—possibly inside the same special node or later in the decoding order—the PC-frozen value is not yet available, and the proposed special-node decoders cannot be executed at the claimed point in the SC tree. This would invalidate both the 72% node-traversal reduction and the FER-equivalence claim. The manuscript contains no statement of this timing condition in Section III or elsewhere, making the central claim conditional on an unverified property of the underlying HARQ construction.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes modifications to SC special-node decoding (Rate-0/1, REP, REP-2, SPC, SPC-2, RPC, PCR) so that polar IR-HARQ with matrix extension can be decoded with fast SC decoding, despite the presence of PC-frozen bits with arbitrary values. It derives modified path-metric and parity-check rules, reports a 72% reduction in node traversals for a length-2048 IR-HARQ code compared to the unmodified special-node decoder, and presents FER curves claiming no performance degradation.","tokens_in":7974,"tokens_out":12742,"duration_ms":116067,"significance":"If the central assumption holds, this is a useful and timely contribution: it extends the latency benefits of special-node SC decoding to polar IR-HARQ, which is relevant for 6G data-channel decoding. The derivations are transparent and the REP proof is a clean maximum-likelihood argument. The paper does not resort to fitting: the traversal reduction is a direct count from Table I, and the FER simulation is a direct comparison. The main risk is that the modified decoders assume the encoded PC-frozen vector is available at node-processing time, a property that is neither proven nor explicitly verified.","major_comments":[{"comment":"All modified special-node decoders require the encoded PC-frozen vector pc of the node to be known when the node is processed. Section II.D states only that 'the new information bit is first estimated, after which the corresponding PCFrozen bit is fixed.' That statement is insufficient for special-node decoding: SC treats a node as an atomic block, so if a PC-frozen bit and its mapped new information bit lie in the same special node, pc is not available at node level. For example, in the REP rule Eq. (4), a term with pc_j equal to the repeated bit being decoded would make the decision circular. The manuscript must prove from the matrix-extension construction of [6], or state and empirically verify, that every PC-frozen bit in every detected special node has its source information bit strictly left of that node. Without this, the 72% traversal claim and the FER-equivalence claim are condi","section":"Section III, Eqs. (4), (16), (23), Section III.G"},{"comment":"The FER simulation is not reproducible: no simulation parameters are reported (reliability ordering/construction metric, CRC polynomial, number of simulated blocks, Eb/N0 grid, channel model, bit-reversal usage). The claim that the proposed modifications cause 'no degradation' is central, but without these details it cannot be assessed. Please report the parameters and check bit-level equivalence between the modified node decoders and full SC decoding for random PC-frozen assignments. Also clarify how Configuration B is decoded with unmodified nodes when PC-frozen bits are nonzero: those nodes assume frozen bits are zero, so the comparison may be unfair unless the PC-frozen bits are all zero in that configuration.","section":"Section IV, Fig. 1"}],"minor_comments":[{"comment":"Typo: 'a72%' should be 'a 72%'.","section":"Abstract"},{"comment":"The term 'PCFrozen' is used without defining the acronym at first use; define it explicitly and consider using consistent spacing (PC-Frozen or PCFrozen).","section":"Section II.D"},{"comment":"Use 'Hence' instead of 'Henceforth' in the REP-2 paragraph.","section":"Section III.C"},{"comment":"The PCR modification is described in prose; presenting the changed line of Algorithm 2 in pseudocode would improve clarity.","section":"Section III.G"},{"comment":"The 72% reduction is for one code length and one construction. The node-type distribution, and hence the reduction ratio, depends on rate and reliability ordering. State the scope of this result.","section":"Table I"},{"comment":"In the RPC definition, the parity equations in (2a) are redundant; consider simplifying to the minimal set, as is later used in Section III.F.","section":"Section II.C.7"}],"recommendation":"major_revision","confidential_remarks":"The core derivations appear sound under the PC-frozen timing assumption, but that assumption is load-bearing and unproven. The paper is also thin on simulation details. I would recommend major revision, not rejection, because the missing guarantee may be provable from [6] or verifiable empirically; if it fails, the approach needs a preprocessing step to reorder or split nodes."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a competent, incremental extension of fast SC decoding to polar IR-HARQ. The genuinely new piece is the REP path-metric derivation (Eqs. 4–11), which is correct and easy to verify; the SPC, SPC-2, RPC, and PCR modifications are the same change-of-variables trick in different clothing. The 72% traversal reduction is just a count from Table I and it matches the baseline comparison. The paper deserves a referee.\n\nWhat is new: previous special-node work assumed zero-valued frozen bits. Here the authors adapt the full special-node family to the PC-frozen setting from [6] and correctly distinguish their SC approach from the SCL candidate-generation method in their own [8]. That distinction is not hand-waving; the complexity arguments are sensible.\n\nThe weak spots, in proportion. First, the PC-frozen timing assumption. Section II.D says the new info bit is \"first estimated\" and then the PC-frozen bit is fixed, which implies the source precedes the PC-frozen bit in SC order. But the paper never proves, or even states, that this placement holds for every special node—that a PC-frozen bit's mapped info bit never lies inside the same special node or to its right. Without that, the modified REP/SPC formulas use inputs that may not be available when the node is processed. I suspect the construction in [6] guarantees this by design, and it may be obvious to people who work on this family, but the authors should say so explicitly and cite or give a one-line proof. This is a missing paragraph, not a missing argument.\n\nSecond, there are no simulation parameters: no code construction tables, no trials, no details on the AWGN setup, no LLR quantization. For a conference paper this is common, but it makes the \"no FER degradation\" claim harder to verify. Third, the 72% figure is for one length-2048 configuration; it is a scheduling result, not a general bound.\n\nWho this is for: hardware-oriented polar code researchers, especially anyone building SC decoders with IR-HARQ for 6G. General coding theorists will not be surprised, but it is a solid contribution to the fast-decoding subfield.\n\nBottom line: the math is sound, the claim is modest, and the gap is fixable in a short revision. Send it to peer review.","headline":"A clean, useful extension of fast SC special-node decoding to IR-HARQ with PC-frozen bits; the derivations check out, and the main gap is an unstated placement assumption that should be fixed in revision.","tokens_in":8492,"tokens_out":5434,"would_cite":true,"duration_ms":47198,"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":"IR-HARQ polar codes can reuse fast SC special-node decoding without performance loss.","keywords":["polar codes","SC decoding","IR-HARQ","special nodes","PC-frozen bits","matrix extension","node traversals","6G"],"falsifier":"Construct a matrix-extension IR-HARQ polar code of length 2048 in which a PC-frozen bit's source information bit is scheduled later in SC order than the special node containing that PC-frozen bit; run the modified decoder and show it either needs pc before it is known or produces a different node decision than full SC. Alternatively, simulate FER for several rates/lengths beyond 2048 and observe a gap between the modified-node decoder and unmodified leaf-wise IR-HARQ decoding.","tokens_in":7592,"feed_emoji":"📡","tokens_out":4081,"duration_ms":33258,"temperature":0.7,"pith_summary":"This paper seeks to close the gap between two otherwise attractive features of polar codes: incremental-redundancy HARQ and fast successive-cancellation decoding. Incremental redundancy introduces PC-frozen bits whose values are not fixed to zero, which breaks the assumptions of the special-node shortcuts used to speed up SC decoding. The paper proposes small modifications to each special-node decoder—repetition, single-parity-check, and their variants—that account for the known encoded PC-frozen pattern. With these changes, an IR-HARQ SC decoder at length 2048 visits 72% fewer nodes and shows no frame-error-rate degradation against baselines. The implication is that high-throughput polar decoders can offer HARQ retransmission capability without sacrificing the complexity savings of special nodes.","feed_headline":"IR-HARQ polar decoding cuts node traversals by 72%","feed_subtitle":"Modified special nodes let SC decoders handle IR-HARQ with no FER loss at length 2048.","key_machinery":"The load-bearing object is the encoded PC-frozen vector pc: the binary vector produced by passing the parity-check frozen bits through the polar transform. Because the polar codeword splits as β = i ⊕ pc between the encoded information vector and the encoded PC-frozen vector, the pc pattern tells the decoder which 'frozen' positions carry known nonzero values. Each modified node decoder performs a change of variables on its parity or repetition accumulator so that the known pc contribution is absorbed before applying the standard fast-decode rule. That single adjustment is what lets the whole library of scalar, repetition, and parity-check special nodes operate under IR-HARQ.","core_discovery":"The central claim is that the encoded PC-frozen vector pc, whose components give the binary values of parity-check frozen bits after polar encoding, is known at decode time, so every special node can be decoded against a known nonzero frozen-bit pattern. Under the identity β = i ⊕ pc, the usual repetition and parity-check decisions are replaced by modified ones—for example, a repeated bit is decided by the sign of Σ l_j(1−2pc_j), and a single-parity-check node flips its least reliable bit when (XOR of hard decisions) ⊕ pc_0 ≠ 0. Applying these rules to the RPC and PCR nodes likewise reduces to initializing the parity accumulators with the pc pattern. The paper reports that, with a length-204","pith_inferences":["The change-of-variable trick is generic: any new special-node family added to the fast-decoder library can likely absorb pc in the same way, by XOR-ing the PC-frozen pattern into its parity constants.","The method's practical reach depends on the IR-HARQ bit ordering; if a future polar code design places a PC-frozen bit before its source information bit in SC order, the modified node cannot be used at that position and the decoder would need a fallback.","A similar pc-aware adjustment could be applied to other low-complexity polar decoders (e.g., simplified SC with early termination) to support IR-HARQ without full subtree traversal."],"forward_implications":["An IR-HARQ SC decoder can match the node-traversal count—and thus the latency—of a plain fast SC decoder at the same total block length.","Retransmission-based reliability can be offered to very high-throughput polar decoders (e.g., 6G data channels) without giving up special-node acceleration.","The 72% traversal reduction reported at length 2048 scales the per-transmission decoding effort down to roughly one third of the unmodified IR-HARQ scheme.","The modified node rules preserve FER, so the coding gain promised by matrix-extension IR-HARQ is not traded away for speed."],"fun_headline_variants":["Polar SC decoding gets 72% faster with IR-HARQ","Modified SC decoders enable IR-HARQ, slash traversals by 72%","Special nodes accelerate polar SC decoding for IR-HARQ","No FER loss: polar SC IR-HARQ with 72% fewer node traversals","IR-HARQ polar decoding: 72% less traversal, same FER"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The decoder must already know the value of every PC-frozen bit inside a special node when that node is processed, which requires that the corresponding information bits appear earlier in the SC decoding order.","fun_headline_variants_meta":{"raw":{"variants":["Polar SC decoding gets 72% faster with IR-HARQ","Modified SC decoders enable IR-HARQ, slash traversals by 72%","Special nodes accelerate polar SC decoding for IR-HARQ","No FER loss: polar SC IR-HARQ with 72% fewer node traversals","IR-HARQ polar decoding: 72% less traversal, same FER"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00024,"raw_usage":{"total_tokens":1320,"prompt_tokens":677,"completion_tokens":643,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":421,"completion_tokens_details":{"reasoning_tokens":540}},"tokens_in":421,"tokens_out":643,"duration_ms":5239,"temperature":1.0,"reasoning_tokens":540,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T18:36:02.588916+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Construct a matrix-extension IR-HARQ polar code of length 2048 in which a PC-frozen bit's source information bit is scheduled later in SC order than the special node containing that PC-frozen bit; run the modified decoder and show it either needs pc before it is known or produces a different node decision than full SC. Alternatively, simulate FER for several rates/lengths beyond 2048 and observe a gap between the modified-node decoder and unmodified leaf-wise IR-HARQ decoding.","supporting_citations":[],"review_version":1}