{"id":"2a998951-766b-4aac-b202-fc52694a334c","arxiv_id":"1908.01463","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"For the square-law distortion-noise profile, the minimum robust transmission energy lies between 0.9057√α and 2.3203√α, improving prior bounds of 0.4507√α and 3.1846√α.","lead":"This paper tightens the known lower and upper bounds on the minimum energy needed to transmit a Gaussian source robustly over a Gaussian channel when the noise level is unknown, using a square-law distortion profile. The improved bounds narrow the uncertainty about the true energy cost by more than half, helping system designers understand how much energy robust transmission really requires.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 4's upper bound omits an explicit check that the inclined staircase stays above the convex profile; this is a genuine proof gap, though a short convexity argument appears to fill it.","rationale":"The central claim is the improved bounds 0.9057√α ≤ E_min ≤ 2.3203√α. The lower bound in Theorem 3 is a valid specialization of Lemma 1 with K=2; the numerical optimization values check out (rough evaluation of (8) at q1=1.5496, q2=5.6679, τ=0.1285 gives ≈0.9054, close to the reported 0.9057). The upper bound in Theorem 4 is the delicate part. The proof asserts, without demonstration, that the piecewise-linear fidelity function stays above the convex profile. The reader flagged this as the weakest assumption. We agree that the assertion is not proved in the text: Lemma 2 proves only that (19) makes β_k increasing, and the left-endpoint condition F(Q_k^+)≥1+αQ_k^2 is never stated or checked. However, the gap is easily closed: because the profile is convex, the difference between the line and the parabola is concave, so it suffices to check the two endpoints of each segment. The right endpoint is equality by (17), and the left endpoint difference reduces to (Q_{k+1}-Q_k)[α(Q_{k+1}+Q_k)-A_{k+1,total}]. For the proposed parameters, this is nonnegative exactly when (1-d^{k+1})/(1-d) ≤ 2k+1, which holds for all k≥1 and 0<d<1 (the left side is ≤ k+1 ≤ 2k+1). Thus the staircase property is true, but the manuscript should include this check before the proof can be considered complete. We also considered the infinite-layer limit; a finite-K truncation can satisfy the distortion constraint within ε for all Q because the distortion gap decays as 1/K^2, so this is not a serious obstacle. No other load-bearing errors were found. The reader's conditional verdict is appropriate.","tokens_in":8519,"tokens_out":42119,"duration_ms":381636,"concrete_test":"For the parameters of Theorem 4, compute the left-endpoint gap G_k = β_k+Q_k A_{k+1,total} - (1+αQ_k^2). Substitute (17) to get G_k = (Q_{k+1}-Q_k)[α(Q_{k+1}+Q_k)-A_{k+1,total}]. With Q_k=kΔ, A_{k+1,total}=αΔ(1-d^{k+1})/(1-d), verify G_k ≥ 0 for all k≥1 by proving (1-d^{k+1})/(1-d) ≤ 2k+1. If this inequality holds, the missing endpoint check is satisfied and the staircase-above-profile assertion follows by concavity of the difference; the proof of Theorem 4 is then complete.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In the proof of Theorem 4, the authors assert that the piecewise-linear fidelity function F(Q)=β_k+Q A_{k+1,total} on [Q_k,Q_{k+1}) is always above the convex profile 1+αQ^2 and coincides with it at the jump points. The proof of Lemma 2 verifies only the parameter constraint (19), which ensures β_k is increasing; it does not explicitly verify the left-endpoint inequality F(Q_k^+)≥1+αQ_k^2. Since the difference between the linear segment and a convex parabola is concave, checking the two endpoints suffices: the right endpoint is equality by (17), and the left-endpoint difference reduces to (Q_{k+1}-Q_k)[α(Q_{k+1}+Q_k)-A_{k+1,total}]. For the chosen parameters this requires (1-d^{k+1})/(1-d) ≤ 2k+1, which holds for 0<d<1. Thus the claimed staircase property is true but not demonstrated in the paper; a reader cannot verify the upper bound without reconstructing this argument.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper considers the minimum energy per source symbol required for robust transmission of a unit-variance Gaussian source over an AWGN channel whose noise variance may be any positive value, subject to a distortion-noise profile D(N). Working with the fidelity-quality profile F(Q)=1/D(1/Q), the authors study the square-law profile F(Q)=1+alpha Q^2. They prove an improved lower bound E_min(F) >= 0.9057 sqrt(alpha) (Theorem 3) by instantiating the K=2 case of a multi-receiver broadcast outer bound from prior work, and they propose a layered joint source-channel coding scheme with multiple uncoded transmissions and Wyner-Ziv binning that yields an improved upper bound E_min(F) <= 2.3203 sqrt(alpha) (Theorem 4). The main results are supported by algebraic derivations and by numerical optimization of the free parameters.","tokens_in":1358,"tokens_out":1387,"duration_ms":106776,"significance":"If the bounds are correct, the paper meaningfully tightens the previously known constants for a nontrivial robustness profile, and the coding scheme is an elegant extension of the hybrid digital-analog approach: sending each quantization error uncoded adds energy but enriches the side information for the subsequent digital layers. The lower-bound technique demonstrates a clean adaptation of a broadcast outer bound to the energy-limited setting, and the paper provides explicit algebraic expressions for the achievable fidelity function. The main concerns are a gap in the proof of the upper bound's dominance property and the lack of a reproducible numerical verification of the claimed constants; both appear addressable in revision and do not undermine the underlying approach.","major_comments":[{"comment":"The claim that the piecewise-linear function F(Q)=beta_k + Q A_{k+1,total} is always above the profile F(Q)=1+alpha Q^2 is not established by the argument given. Equality at the jump points, expressed in (17), fixes the right endpoint of each segment, and condition (19) only ensures that beta_k is increasing. A linear segment with these parameters can dip below a convex parabola inside the interval unless the left-endpoint value is also checked. The proof must explicitly verify, for each k, that beta_k + Q_k A_{k+1,total} >= 1+alpha Q_k^2, equivalently A_{k+1,total} <= alpha(Q_{k+1}+Q_k). With the choices Q_k=k*Delta, A_0=alpha*Delta, and A_k=d^k*alpha*Delta for k>=1, this reduces to (1-d^{k+1})/(1-d) <= 2k+1. This inequality in fact follows from the condition proved in Lemma 2 after rearrangement, but the paper does not show that connection; the reader cannot reconstruct the above-profile claim without doing additional work. Because the upper bound's achievability rests on this domination property, this is a load-bearing gap in the proof of Theorem 4.","section":"IV-B (Theorem 4 proof, Eqs. (16)-(19))"},{"comment":"The numerical constants 0.9057 and 2.3203 are presented as results of numerical optimization, but the manuscript does not provide a verifiable calculation. For Theorem 3, the paper should state explicitly that evaluating the expression in (8) at the reported q1*, q2*, tau* yields a value equal to 0.9057 sqrt(alpha) (or at least >= 0.9057 sqrt(alpha)). For Theorem 4, the infinite series in (27) at c*=0.00137, d*=0.999 needs to be evaluated with a rigorous bound on the truncation error, or the authors should provide the partial-sum data or code. Without this, the claimed constants are not formally supported by the text, even though they are likely correct and easily checkable from the explicit formulas.","section":"Theorems 3 and 4"}],"minor_comments":[{"comment":"The sentence containing 'Furthermore,it is represented that exponential profiles are not achievable' should read 'Furthermore, it is reported that exponential profiles are not achievable'.","section":"Introduction"},{"comment":"The phrase 'we can point military situation, indoor fires and emergency conditions' should be reworded, for example 'we can point to military situations, indoor fires, and emergency conditions'.","section":"Introduction"},{"comment":"The phrase 'In other words, (17)' after stating that the fidelity-quality tradeoff is 'always above the profile' is misleading: equation (17) only imposes equality at the jump points, not dominance over each interval. The dominance condition should be stated and proved separately.","section":"IV-B (proof of Theorem 4)"},{"comment":"The expression '(d^k-1)/(d-1)' is written without parentheses in the displayed equation; it should be typeset as (d^k-1)/(d-1) or, since d<1, preferably as (1-d^k)/(1-d) to avoid sign ambiguity.","section":"Lemma 2 proof, Eq. (20)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript fits the scope of the journal and, conditional on fixing the proof gap in Theorem 4 and making the numerical constants reproducible, would be a solid contribution. The proof gap is real but seems fillable with a short convexity or algebra argument; the numerical issues are a matter of rigor and reproducibility rather than of correctness. I would encourage the authors to submit a revised version addressing these points."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a real advance on a narrow problem. The multi-layer uncoded transmission scheme is new and it genuinely tightens the gap: upper bound down from 3.1846√α to 2.3203√α, lower bound up from 0.4507√α to 0.9057√α via a K=2 outer bound. Both constants are concrete and the comparison table is honest.\n\nThe derivations are mostly solid. The lower bound follows from the known energy-distortion tradeoff and the broadcast outer bound; the algebra in the upper bound (the MMSE estimate, the rate expressions via the matrix determinant lemma, the Sherman-Morrison-Woodbury step) checks out. The authors also correctly identify that their scheme reduces to [10] when d=0 and A0=E0, which is a good sanity check.\n\nThe soft spots are real but not fatal. The proof of Theorem 4 asserts that the inclined staircase F(Q)=β_k+Q A_{k+1,total} stays above the convex profile 1+αQ^2 and coincides with it at the jump points. Eq. (17) enforces the coincidence at the jump points and constraint (19) makes β_k increasing, but neither statement alone implies the line segments are above the parabola. The stress-test note is right: a short convexity argument (check the left endpoint) fills the gap, but the paper does not supply it, and a reader cannot verify the upper bound without reconstructing that argument. That is a genuine omission in a proof of the main result.\n\nTwo smaller items: the K→∞ limit is taken informally; since all terms are nonnegative and the sums converge numerically, it is probably justifiable with a monotone convergence argument, but it should be said. And the numerical optimizations for the lower bound (gradient ascent) and upper bound (optimal c,d) are not independently verified; the values are plausible and the code is not shipped, so this is a reproducibility nit, not a red flag.\n\nVerdict: the paper deserves peer review. The core scheme is new, the bounds are meaningfully improved, and the proof gap is fixable without changing the result. For anyone working on energy-limited robust JSCC or distortion-noise profiles, this is worth citing and worth reading carefully. I would accept it for review and ask the authors to add the omitted convexity check and justify the limit.","headline":"Genuine improvement on both bounds for the square-law profile, with a real but fixable gap in the upper-bound proof.","tokens_in":9275,"tokens_out":1914,"would_cite":true,"duration_ms":18255,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["94A15","94A29","94A40"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that the minimum energy for the square-law distortion-noise profile lies between $0.9057\\sqrt{\\alpha}$ and $2.3203\\sqrt{\\alpha}$, significantly tightening previously known bounds.","keywords":["robust joint source-channel coding","distortion-noise profile","fidelity-quality profile","minimum energy","Gaussian source","Gaussian channel","square-law profile","energy-distortion tradeoff"],"falsifier":"Evaluate the inclined staircase $F(Q)=\\beta_k+Q A_{k+1,\\mathrm{total}}$ with the paper's optimized parameters $d^*=0.999$ and $c^*=0.00137$ (so $Q_k=k c^*\\sqrt{\\alpha}$) on each interval $[Q_k,Q_{k+1})$; if for any $k$ and any $Q$ in that interval $F(Q)<1+\\alpha Q^2$, then the proof of the upper bound as written is incomplete, and if no such dip exists, the missing monotonicity check is satisfied.","tokens_in":8354,"feed_emoji":"📡","tokens_out":7971,"duration_ms":74922,"temperature":0.7,"pith_summary":"The paper studies the minimum energy needed to send a Gaussian source over a Gaussian channel when the transmitter does not know the noise level but the receiver does, and the system must honor a prescribed distortion-noise profile. For the square-law fidelity-quality profile $F(Q)=1+\\alpha Q^2$, where $Q$ is channel quality and $\\alpha>0$, it claims the minimum energy $E_{\\min}(F)$ satisfies $0.9057\\sqrt{\\alpha} \\le E_{\\min}(F) \\le 2.3203\\sqrt{\\alpha}$. These bounds improve the previously known $0.4507\\sqrt{\\alpha}$ and $3.1846\\sqrt{\\alpha}$, shrinking the gap from a factor of about 7 to about 2.6. The point of tightening the bounds is to give system designers a reliable energy budget when the noise level is completely unknown.","feed_headline":"Minimum energy for robust Gaussian coding: 0.906–2.320 sqrt(alpha)","feed_subtitle":"Tighter bounds for the square-law distortion-noise profile shrink the unknown-energy gap from 7x to 2.6x.","key_machinery":"The load-bearing object for the upper bound is the inclined staircase fidelity function. On each noise-quality interval $Q_k\\le Q<Q_{k+1}$, the scheme's achieved fidelity is piecewise linear, $F(Q)=\\beta_k+Q A_{k+1,\\mathrm{total}}$, where $\\beta_k$ is the inverse quantization-error variance and $A_{k+1,\\mathrm{total}}$ is the cumulative uncoded energy sent in the first $k+1$ layers. If that staircase remains at or above the convex profile $1+\\alpha Q^2$ and touches it at the jump points $Q_k$, then summing the uncoded energies $A_k/\\beta_k$ and digital energies $B_k$ gives the upper bound. The lower bound rests on the $K=2$ case of a general $K$-receiver outer bound for Gaussian source broadcast, converted into a two-variable numerical optimization.","core_discovery":"The paper's central claim is that the minimum energy for the square-law profile is proportional to $\\sqrt{\\alpha}$ with a constant between $0.9057$ and $2.3203$. The lower bound comes from a two-receiver version of a broadcast-channel converse bound: optimizing the auxiliary parameters yields $0.9057\\sqrt{\\alpha}$. The upper bound comes from a $K$-layer coding scheme in which the source is successively quantized, each quantization error is also sent uncoded, and the digital indices are sent with binning against receiver side information; as $K\\to\\infty$ and after numerical optimization of the layer spacings and energies, the total energy is no more than $2.3203\\sqrt{\\alpha}$. The paper positions this as a significant tightening of the earlier bounds rather than an exact characterization.","pith_inferences":["The same inclined-staircase construction should extend to any convex fidelity-quality profile: if one can prove the piecewise-linear majorization condition, the energy sum will give an upper bound; the square-law profile is the first test case.","Running the same numerical lower-bound optimization with $K=3$ or more receivers is a natural next step and could push the lower constant further toward the true value.","The unproved above-curve condition in Theorem 4 is checkable by direct computation for the optimized parameters; if it fails, the numerical constant $2.3203$ would need to be re-derived, though the general scheme may still work with adjusted parameters.","The model assumes the receiver knows the noise variance exactly; relaxing that to estimated noise would likely require a different robust-coding formulation, so the energy bounds here are a baseline for unknown-channel design rather than a full solution."],"forward_implications":["For any square-law distortion-noise profile, no transmission scheme can use less than $0.9057\\sqrt{\\alpha}$ energy per source symbol, and the proposed layered scheme achieves $2.3203\\sqrt{\\alpha}$.","The gap between the best known lower and upper bounds shrinks from about a factor of 7.1 to about 2.56, so the true minimum-energy constant is pinned to a much narrower interval.","Because both bounds scale linearly with $\\sqrt{\\alpha}$, doubling the profile parameter doubles the energy budget's scale: the required energy scales as the square root of the fidelity-curve curvature.","The upper-bound construction generalizes the earlier uncoded-first-layer scheme: setting all but the first uncoded energy to zero recovers the previous scheme, so the improvement is a strict generalization rather than a separate method."],"supporting_citations":[{"why":"Supplies the baseline energy-distortion relation $D=\\exp(-E/N)$ and the energy-distortion problem formulation used in the lower bound.","marker":"[6]"},{"why":"Gives the previous lower and upper bounds for the square-law profile and the uncoded-plus-digital scheme that this paper generalizes.","marker":"[10]"},{"why":"Provides the $K$-receiver outer bound for the Gaussian source broadcast distortion region, which yields the family of lower bounds in Lemma 1.","marker":"[15]"},{"why":"Gives the two-receiver outer bound that the $K$-receiver bound generalizes, supporting the $K=2$ lower-bound special case used for the improved constant.","marker":"[16]"},{"why":"Supplies the matrix-determinant and rank-one update identities used to simplify the rate and distortion expressions in the upper-bound scheme.","marker":"[17]"}],"fun_headline_variants":["Energy gap for robust Gaussian coding drops from 7x to 2.6x","Square-law profile: minimum energy constant now known between 0.906 and 2.320","Robust Gaussian coding: tighter bounds close the energy gap to 2.6x","Minimum energy for square-law profile: constant lies in [0.906, 2.320]","Improved energy bounds for robust Gaussian coding: gap cut to 2.6x"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The upper bound stands on the assumption that the inclined staircase fidelity function stays above the convex square-law profile for all channel qualities, meeting it only at the chosen jump points; the proof shows only that the staircase parameters are increasing, not that the staircase never dips below the profile.","fun_headline_variants_meta":{"raw":{"variants":["Energy gap for robust Gaussian coding drops from 7x to 2.6x","Square-law profile: minimum energy constant now known between 0.906 and 2.320","Robust Gaussian coding: tighter bounds close the energy gap to 2.6x","Minimum energy for square-law profile: constant lies in [0.906, 2.320]","Improved energy bounds for robust Gaussian coding: gap cut to 2.6x"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001492,"raw_usage":{"total_tokens":5901,"prompt_tokens":771,"completion_tokens":5130,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":387,"completion_tokens_details":{"reasoning_tokens":5015}},"tokens_in":387,"tokens_out":5130,"duration_ms":36093,"temperature":1.0,"reasoning_tokens":5015,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:13:02.966711+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Evaluate the inclined staircase $F(Q)=\\beta_k+Q A_{k+1,\\mathrm{total}}$ with the paper's optimized parameters $d^*=0.999$ and $c^*=0.00137$ (so $Q_k=k c^*\\sqrt{\\alpha}$) on each interval $[Q_k,Q_{k+1})$; if for any $k$ and any $Q$ in that interval $F(Q)<1+\\alpha Q^2$, then the proof of the upper bound as written is incomplete, and if no such dip exists, the missing monotonicity check is satisfied.","supporting_citations":[{"cited_title":"Energy- distortion tradeoffs in Gaussian joint source-channel coding problems,","cited_arxiv_id":null,"evidence_quote":"Supplies the baseline energy-distortion relation $D=\\exp(-E/N)$ and the energy-distortion problem formulation used in the lower bound."},{"cited_title":"On minimum energy for robust Gaussian joint source-channel coding with a distortion-noise profile,","cited_arxiv_id":null,"evidence_quote":"Gives the previous lower and upper bounds for the square-law profile and the uncoded-plus-digital scheme that this paper generalizes."},{"cited_title":"Approximate characterization for the Gaussian source broadcast distortion region,","cited_arxiv_id":null,"evidence_quote":"Provides the $K$-receiver outer bound for the Gaussian source broadcast distortion region, which yields the family of lower bounds in Lemma 1."},{"cited_title":"Distortion bounds for broadcast- ing with bandwidth expansion,","cited_arxiv_id":null,"evidence_quote":"Gives the two-receiver outer bound that the $K$-receiver bound generalizes, supporting the $K=2$ lower-bound special case used for the improved constant."},{"cited_title":"Eigenvalues of rank-one updated matrices with some applications,","cited_arxiv_id":null,"evidence_quote":"Supplies the matrix-determinant and rank-one update identities used to simplify the rate and distortion expressions in the upper-bound scheme."}],"review_version":1}