{"id":"1499b472-e87e-4e3d-9fc6-b0ae86392247","arxiv_id":"2501.04328","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A genie-aided decoder that searches over scaling factors α can achieve lower word error rate than the MMSE one-shot decoder, with a proven lower bound via the covering sphere.","lead":"A decoder that can retry with many scale factors, guided by an oracle that says whether a trial worked, can beat the standard one-shot lattice decoder by about half a decibel on some lattices. The paper proves a lower bound on its error rate and gives a practical CRC-based retry scheme for polar lattice codes.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claimed lower bound for power-constrained lattice codes is unsupported: the modulo operation enlarges the decodable region beyond V(x), so D(x)⊂D_c does not imply Pe,Dec > Pe_cover for such codes.","rationale":"The reader's weakest assumption identifies exactly the load-bearing gap: the lower-bound proof treats the decoder as a quantizer on the infinite lattice and uses the single Voronoi region V(x), while the actual decoder for power-constrained lattice codes applies a modulo operation with respect to the shaping lattice. I re-derived the decision region for the nested-lattice decoder: for a transmitted codeword x, Dec(αy) = x iff αy lands in the union of Voronoi regions of all translates x+Λ_s. This union is strictly larger than V(x), so the inclusion D(x) ⊂ D_c used in (5) does not imply that the true decodable region is contained in D_c. Consequently Pe,Dec can be smaller than Pe_cover, invalidating the claimed lower bound for power-constrained codes. The paper's Section III-B explicitly asserts validity for power-constrained lattice codes without proving the needed inclusion. This is a correctness risk for the abstract's broadest claim, though the infinite-lattice Theorem 1 and its proof appear internally coherent, and the asymptotic exchange in Section III-D is explicitly flagged by the authors as requiring formal justification. The concrete test with a small nested lattice code would settle whether the code-level bound actually fails; the reader's conditional verdict already captures this concern, so no verdict change is needed.","tokens_in":8026,"tokens_out":11091,"duration_ms":120071,"concrete_test":"Simulate the genie decoder of Section III-A for the 2-D nested lattice code C = Z²/mZ² with hypercube shaping, transmit x = (1,0), and search α over a dense grid (e.g., step 10^-3) on R. Measure WER over AWGN at SNRs where Eq. (6) gives Pe_cover = 10^-2 and 10^-3, for m = 2, 4, 8. If the measured WER is at or below Pe_cover for any setting, Theorem 1's claim for power-constrained codes is refuted; if WER remains above Pe_cover, the paper should still supply a proof using the enlarged region D_s(x) rather than D(x).","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section II-B states 'the decoding succeeds if αy falls into the Voronoi region of x', while Fig. 2 includes a mod Λ_s operation after Q_{Λ_c}. For a nested lattice code C = Λ_c/Λ_s, correctness requires αy ∈ ∪_{λ∈Λ_s} V_{Λ_c}(x+λ), not merely αy ∈ V(x). Section III-B defines D(x) from V(x) and uses V(x) ⊂ S_c(x) to conclude D(x) ⊂ D_c and hence Pe,Dec > Pe_cover. This chain is valid for an infinite lattice Λ, but for a power-constrained lattice code the true decodable region D_s(x) contains the union of cones over all translates x+Λ_s, which is much larger than D(x). Therefore the code WER can be smaller than the infinite-lattice error probability and can fall below Pe_cover. The assertion in Section III-B that 'the bound is valid for both power-constrained lattice codes C and lattices Λ' is stated without proof; with the modulo operation it is false for generic nested lattice codes. This affects the abstract's central claim and the interpretation of Fig. 6 as a lower bound. The infinite-lattice version of Theorem 1 is not challenged here.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a genie-aided exhaustive-search decoder for finite-dimensional lattice decoding. Instead of using only the MMSE scaling factor alpha_MMSE, the decoder tries many scaling factors alpha and uses a genie (later implemented by CRC) to detect whether the decoded word is correct. The main theoretical result, Theorem 1, is a lower bound on the word error rate obtained by replacing the Voronoi region of the transmitted lattice point by its covering sphere and integrating the Gaussian noise over the resulting cone-like decodable region. The paper also gives an effective-sphere estimate of the WER, an asymptotic closed form for the lower bound as the per-dimension power tends to infinity, and numerical evaluations for E8, BW16, and CRC-aided polar code lattices, reporting gains of about 0.4-0.5 dB at WER 10^-4 relative to one-shot decoding with alpha_MMSE.","tokens_in":8309,"tokens_out":13021,"duration_ms":131614,"significance":"If the main claim were valid, the paper would provide a parameter-free geometric lower bound for finite-dimensional lattice decoding and an interesting practical decoding strategy. A clear strength is that the infinite-lattice version of the derivation is elegant and uses only published lattice constants such as the covering radius; no quantity is fitted to the simulated WER curves. The reported gains are plausible because the exhaustive alpha-search contains alpha_MMSE as a candidate. However, the paper's central claim that the bound is valid for power-constrained lattice codes with modulo shaping is unsupported and, for generic nested lattice codes, false. This undermines the interpretation of the numerical lower-bound curves for E8 and BW16 lattice codes and requires either a substantial new proof or a restriction of the paper's scope.","major_comments":[{"comment":"The lower bound is proved for a decoder whose correct-decision region is D(x) = {y : there exists alpha in R with alpha y in V(x)}. For the power-constrained scheme of Fig. 2 the decoder applies Q_Λc followed by mod Λs, so for a codeword x in C the correct decision region is D_s(x) = union over λ in Λs of {y : there exists alpha with alpha y in V_Λc(x+λ)}. Since V_Λc(x) is a subset of this union, D_s(x) is generally larger than D(x), and the inclusion D(x) subset of D_c does not imply Pe,Dec > Pe,cover for the code decoder. The assertion in Section III-B that the proposed decoder and its lower bound are valid for both power-constrained lattice codes C and lattices Λ is therefore stated without proof and, for generic nested lattice codes, is false. The theorem should be restricted to infinite lattices, or a separate proof must be supplied for the modulo-shaped decoder.","section":"Section III-B, Eq. (5)"},{"comment":"The closed-form asymptotic expression is obtained by exchanging the limit P→∞ and the integral over z1. The paper itself states that formal justification is required, but as presented the step is unproved. The numerical evaluation in Fig. 5 for finite Px up to 500 does not establish the limit, and the asymptotic claim in the abstract is therefore not supported. The authors need to provide a dominated-convergence or explicit error-bound argument, or withdraw the asymptotic claim.","section":"Section III-D, Eq. (11)"}],"minor_comments":[{"comment":"The sentence 'if existing x′ in Λ having Px′ ≤ r_c^2/n, such x′ may not exist for all lattices, the D_c doesn't form a cone-like region and Pe_cover=0' is confusing and should be rewritten; the intended point is that when ||x|| ≤ r_c, the covering sphere contains the origin, making D_c the whole space and the bound trivial.","section":"Section III-B"},{"comment":"The phrase 'the even and odd are opposite of (1) and (2)' is unclear; because Ps(rz) is an (n-1)-dimensional Gaussian sphere probability, the parity in the CDF formula is swapped relative to the n-dimensional formulas (1) and (2), and this should be stated explicitly.","section":"Equation (10)"},{"comment":"For n=2, the sum in h(z) for even n is empty; state explicitly that empty sums are taken to be zero.","section":"Theorem 1"},{"comment":"The caption says that r is 5.4512 and 6.5552 for n=8 and 16, respectively, but the horizontal asymptotes are for Px→∞; clarify how r is chosen and why Px can be varied independently of a fixed lattice code.","section":"Figure 5"},{"comment":"The genie-aided exhaustive search is implemented as a grid search over alpha in (0.5,1.5) with step 0.01, so the simulated WER is for a restricted search and is an upper bound on the ideal search over all real alpha; this distinction should be stated explicitly.","section":"Section IV"},{"comment":"Theorem 1 is stated for a fixed transmitted point with power Px, but Fig. 6 uses average message power; the paper should specify whether the bound is averaged over the codebook or applied with a representative Px.","section":"Section IV"}],"recommendation":"major_revision","confidential_remarks":"The infinite-lattice version of Theorem 1 appears sound and could form the basis of a publishable paper if the claims about power-constrained lattice codes are either proved or removed. The current overclaim in Section III-B is not a mere presentation issue: it affects the abstract, the theorem's stated scope, and the interpretation of Fig. 6. I would encourage the authors to either develop a correct lower bound for the modulo-shaped decision region or rewrite the paper to restrict the bound to infinite lattices and present the lattice-code results as heuristic estimates."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The new idea is the cone-shaped decodable region: for a fixed lattice point, the received vectors that can be correctly decoded by some scaling factor form a cone over the Voronoi region. Replacing the Voronoi region by its covering sphere gives a clean lower bound, and the Gaussian-slice integral is standard and appears correct. The exhaustive-search genie decoder is a natural extension of one-shot MMSE lattice decoding, and the simulations show modest but real gains: about 0.5 dB for E8 and 0.4 dB for BW16 at 10^-4. The CRC retry implementation is a sensible practical step, even if the gain shrinks to 0.1 dB.\n\nThe soft spot is the claim that the bound applies to power-constrained lattice codes. The proof treats the decoder as a pure lattice quantizer on the infinite lattice, with one Voronoi region V(x). But the actual code decoder in Fig. 2 includes mod Λ_s; correct decoding occurs whenever αy falls into any Voronoi region x+Λ_s, not just V(x). That union is much larger, so D(x) ⊂ D_c does not imply the code decoder's error probability is above the covering-sphere bound. Section III-B states the code claim without proof, and as written it is not justified. The infinite-lattice version of Theorem 1 may be fine; the gap is specifically the extension to nested lattice codes. The asymptotic exchange of limit and integral in Section III-D is also waved at rather than proven, though that is a minor issue compared to the code claim.\n\nThe numerics would be more convincing with details on the search grid, channel model, and confidence intervals, but the reported gains are small enough not to be suspicious. The citation pattern is fine; the paper builds on Erez-Zamir and Tarokh-Vardy-Zeger.\n\nThis is a worthwhile paper for researchers in lattice coding and HARQ-style retry. It deserves a serious referee, but the authors should be asked to either prove the bound for the modulo decoder or restrict the claim to lattices. As it stands, the abstract overclaims.\n\nRecommendation: send to peer review, conditional on fixing the code-lattice issue.","headline":"A clean cone-based lower bound for infinite lattice decoding, but the claimed extension to power-constrained lattice codes rests on an unproven and likely false modulo-region step.","tokens_in":8823,"tokens_out":2213,"would_cite":false,"duration_ms":20928,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["94B35","94A15"],"pacs":[],"model":"deepseek-v4-flash","headline":"A genie-aided decoder that exhaustively searches all real scaling factors before quantizing is shown to lower word error rate below one-shot MMSE lattice decoding, with a finite-dimensional lower bound derived from the covering sphere of…","keywords":["genie-aided decoding","lattice codes","word error rate","MMSE scaling","covering sphere","Voronoi region","AWGN channel","retry decoding"],"falsifier":"Run the genie-aided exhaustive-search decoder on a hypercube-shaped E8 lattice code in the SNR range used in Fig. 6, stepping $\\alpha$ from 0.5 to 1.5, and compare the measured WER with the right-hand side of (6). If any measured WER for a power-constrained code falls below the bound, then the paper's assertion that the bound is valid for power-constrained lattice codes fails.","tokens_in":7813,"feed_emoji":"📡","tokens_out":6760,"duration_ms":56175,"temperature":0.7,"pith_summary":"A genie-aided lattice decoder that exhaustively tries every real scaling factor $\\alpha$ before quantizing can decode with lower word error rate than the standard one-shot MMSE decoder. The paper's main result is a lower bound on this decoder's WER: because the decodable region of a lattice point is a cone through the Voronoi region, replacing that region by the larger covering sphere gives a cone whose Gaussian volume can be computed exactly, yielding Theorem 1. The paper shows numerically that using the genie at the decoder gives a 0.5 dB gain for the E8 lattice code and a 0.4 dB gain for the BW16 lattice code at WER $10^{-4}$, with a CRC-based retry decoder giving about 0.1 dB gain on an $n=128$ polar lattice. The bound matters because it gives a finite-dimensional, non-asymptotic guarantee for retry decoding, whose benefit is not captured by the asymptotic optimality of $\\alpha_{MMSE}$.","feed_headline":"Genie-aided decoding beats MMSE lattice decoding by 0.5 dB","feed_subtitle":"Searching over all scaling factors lowers word error; a covering-sphere bound predicts the gain.","key_machinery":"The load-bearing object is the covering-sphere decodable region $D_c = \\{y \\mid \\exists \\alpha \\in \\mathbb{R},\\ \\alpha y \\in S_c\\}$, where $S_c$ is the sphere of radius $r_c$ that covers the Voronoi region $V(x)$. This cone replaces the true decodable region $D(x)=\\{y \\mid \\exists \\alpha \\in \\mathbb{R},\\ \\alpha y \\in V(x)\\}$; since $V \\subset S_c$, we have $D(x) \\subset D_c$ and the Gaussian volume of $D_c$ is an upper bound on the correct-decoding probability. The volume is computed by rotating coordinates so the cone's axis is the $z_1$ axis, slicing the cone into $(n-1)$-dimensional spheres, and using the closed form (10) for the probability that a Gaussian vector falls inside such a spherical slice. The equal-volume effective sphere, radius $r_e$ with $V(S_e)=V_n$, yields an estimate rather than a bound, because neither $V \\subset S_e$ nor $S_e \\subset V$ holds; the paper uses it to predict WER accurately for E8 and BW16, whose Voronoi regions are sphere-like.","core_discovery":"The central claim is that the word error rate of lattice decoding is not minimized by the MMSE scaling factor in finite dimensions, and that a decoder which searches over all $\\alpha \\in \\mathbb{R}$ can provably do better. The paper proves Theorem 1: for a nonzero lattice point $x$ in an $n \\geq 2$ lattice with covering radius $r_c$ and per-dimension power $P_x = \\|x\\|^2/n$, with $r_c^2 < nP_x$, the WER of the genie-aided decoder on the AWGN channel satisfies $$P_{e,\\mathrm{Dec}} > 1 - \\int_{-\\infty}^{\\infty} \\frac{1}{\\sqrt{2\\pi\\$sigma^{2}$}} $e^{{-z^2/(2\\sigma^2)}}$ (1-h(z)) \\, dz,$$ where $h(z)$ is a closed-form expression in the complementary error function and a finite exponential sum. The strict inequality comes from $V \\subset S_c$, which makes the covering-sphere decodable region a strict superset of the true decodable region; the bound tightens as the Voronoi region becomes sphere-like. The paper further reports that replacing the covering sphere by the equal-volume effective sphere gives an accurate WER estimate for E8 and BW16, and that the empirical gains relative to one-shot $\\alpha_{MMSE}$ decoding are 0.5 dB and 0.4 dB at WER $10^{-4}$.","pith_inferences":["A natural extension the paper leaves implicit: the benefit of searching $\\alpha$ is essentially a directional-noise effect, so lattices whose Voronoi regions are least sphere-like should show the largest retry gains; the covering-sphere bound cannot capture this variation.","The unproved step for power-constrained codes suggests a direct test: simulate the shaped-coset decoder and check whether the bound (6) remains a lower bound once the correct region is a union of Voronoi cells.","CRC-based retry decoding is a generic ingredient: any code with an error-detecting outer check can serve as the genie, so the scheme could be combined with other decoders (e.g., list decoders) where candidate checks are already available."],"forward_implications":["MMSE scaling is not optimal for finite-dimensional lattice decoding; retrying a few alternative scaling factors is a legitimate route to lower WER.","Any lattice with a known covering radius gets a computable lower bound on the WER of retry decoding, without simulating the decoder.","At WER $10^{-4}$, the E8 and BW16 lattice codes can be improved by 0.5 dB and 0.4 dB respectively by allowing the decoder to search $\\alpha$, at the cost of a genie (or CRC) to verify candidates.","In the high-power/high-rate limit $P \\to \\infty$, the bound approaches a closed-form cylinder expression, giving a simple asymptotic benchmark.","A CRC-embedded polar lattice of dimension 128 already shows about 0.1 dB gain over one-shot MMSE even after the SNR penalty of the CRC, with only three retry candidates."],"supporting_citations":[{"why":"Defines the MMSE scaling factor $\\alpha_{MMSE}$ and its asymptotic optimality; the one-shot MMSE decoder is the baseline the proposed decoder is compared against.","marker":"[2]"},{"why":"Supplies the universal bound and the integral method used to evaluate the probability that Gaussian noise falls into a spherical slice, giving closed forms (10).","marker":"[7]"},{"why":"Introduces the covering sphere and effective sphere and their relation to the Voronoi region, which the lower bound and estimate replace the Voronoi region by.","marker":"[6]"},{"why":"Provides the known covering radii for the E8 and BW16 lattices used in the numerical evaluation.","marker":"[8]"},{"why":"Gives the finite-dimension polar lattice design used in the CRC-embedded $n=128$ practical implementation.","marker":"[9]"}],"fun_headline_variants":["Genie-aided lattice decoding: 0.5 dB gain over MMSE","A lower bound for genie-aided lattice decoder error rate","Search over all scaling factors cuts lattice decoding errors","Covering-sphere bound predicts genie decoder gains","E8 and BW16: genie decoding beats MMSE by up to 0.5 dB"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof of the lower bound treats the decoder as a pure lattice quantizer whose correct region is a single Voronoi region; for power-constrained lattice codes that use a shaping lattice and modulo operation, the correct region is a union of Voronoi regions, and the paper asserts without proof that the covering-sphere cone still lower-bounds the error probability in that larger setting.","fun_headline_variants_meta":{"raw":{"variants":["Genie-aided lattice decoding: 0.5 dB gain over MMSE","A lower bound for genie-aided lattice decoder error rate","Search over all scaling factors cuts lattice decoding errors","Covering-sphere bound predicts genie decoder gains","E8 and BW16: genie decoding beats MMSE by up to 0.5 dB"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000763,"raw_usage":{"total_tokens":3453,"prompt_tokens":1078,"completion_tokens":2375,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":694,"completion_tokens_details":{"reasoning_tokens":2296}},"tokens_in":694,"tokens_out":2375,"duration_ms":15689,"temperature":1.0,"reasoning_tokens":2296,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T21:37:08.458253+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the genie-aided exhaustive-search decoder on a hypercube-shaped E8 lattice code in the SNR range used in Fig. 6, stepping $\\alpha$ from 0.5 to 1.5, and compare the measured WER with the right-hand side of (6). If any measured WER for a power-constrained code falls below the bound, then the paper's assertion that the bound is valid for power-constrained lattice codes fails.","supporting_citations":[{"cited_title":"Erez and R","cited_arxiv_id":null,"evidence_quote":"Defines the MMSE scaling factor $\\alpha_{MMSE}$ and its asymptotic optimality; the one-shot MMSE decoder is the baseline the proposed decoder is compared against."},{"cited_title":"Tarokh, A","cited_arxiv_id":null,"evidence_quote":"Supplies the universal bound and the integral method used to evaluate the probability that Gaussian noise falls into a spherical slice, giving closed forms (10)."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the covering sphere and effective sphere and their relation to the Voronoi region, which the lower bound and estimate replace the Voronoi region by."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the known covering radii for the E8 and BW16 lattices used in the numerical evaluation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the finite-dimension polar lattice design used in the CRC-embedded $n=128$ practical implementation."}],"review_version":1}