{"id":"f728b405-493a-43d2-8928-994561b2486d","arxiv_id":"2603.08411","paper_version":2,"verdict":"CONDITIONAL","confidence":"LOW","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"RIN in optical IM/DD creates signal-dependent noise with memory and polynomial variance; ignoring memory in decoding causes GMI to saturate for dense constellations.","lead":"This paper derives a discrete-time channel model for intensity-modulation direct-detection optical links impaired by laser relative intensity noise, revealing signal-dependent noise with memory whose variance depends polynomially on the transmitted symbol. It then computes generalized mutual information under a memoryless mismatched decoder and shows that achievable rates saturate with larger constellations.","discovery_kind":"new_application","skeptic_critique":{"model":"grok-4.3","headline":"No significant objection identified","rationale":"The reader's weakest assumption correctly isolates the modeling step as the potential vulnerability, yet the provided abstract and claim description show no evidence that this step fails under the stated assumptions. The saturation result follows logically once the polynomial variance and memoryless metric are accepted, so the verdict does not require adjustment.","tokens_in":1644,"tokens_out":257,"duration_ms":16787,"concrete_test":"Recompute the GMI curves in the numerical section for M=16, 64, 256, and 1024 using the exact same memoryless metric and parameters; confirm that the saturation plateau for M≥64 remains within 0.05 bits/symbol of the reported values.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that GMI saturates with growing constellation size under memoryless decoding because of nonsymmetric nonvanishing symbol contributions—rests on the derived discrete-time model with polynomial signal-dependent variance. The derivation from continuous-time RIN waveform via sampling and filtering appears internally consistent, and the numerical saturation is a direct consequence of that model under the chosen mismatched metric. No internal inconsistency, hidden circularity, or unsupported step is visible in the argument structure.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper starts from a continuous-time waveform model of an intensity-modulation direct-detection optical channel impaired by laser relative intensity noise (RIN). It derives an equivalent discrete-time channel whose noise is signal-dependent, possesses memory, and has conditional variance that is a polynomial function of the transmitted symbol. Achievable rates are evaluated via generalized mutual information (GMI) under a mismatched memoryless decoding metric. Numerical results for practical parameters show that GMI saturates with increasing constellation cardinality; the authors attribute the saturation to nonsymmetric, nonvanishing per-symbol contributions to the GMI.","tokens_in":1734,"tokens_out":550,"duration_ms":24178,"significance":"If the continuous-to-discrete derivation holds, the work supplies a more accurate RIN channel model than the memoryless or linear-variance approximations common in the literature. The saturation result supplies a concrete, falsifiable prediction for system designers: beyond a modest cardinality, denser constellations yield diminishing returns when the receiver treats the channel as memoryless. The explicit polynomial-variance structure and the GMI analysis constitute reusable modeling tools for RIN-limited links.","major_comments":[{"comment":"§3 (discrete-time model derivation): the claim that the conditional variance is exactly polynomial in the symbol rests on the specific sampling instant and front-end filter response; the manuscript should state the filter impulse response and sampling phase explicitly so that the polynomial coefficients can be reproduced from the continuous-time parameters.","section":"§3"},{"comment":"§4 (GMI evaluation): the saturation result is obtained under a memoryless metric; the paper should report the GMI gap to a metric that retains the derived memory (e.g., via a trellis or factor-graph decoder) for at least one constellation size, to confirm that the observed saturation is not an artifact of the memoryless restriction.","section":"§4"}],"minor_comments":[{"comment":"Table I: the RIN strength parameter is listed without units; add the physical unit (e.g., 1/Hz) for clarity.","section":"Table I"},{"comment":"Figure 3 caption: the legend entries for different constellation sizes are difficult to distinguish in grayscale; use distinct line styles or markers.","section":"Figure 3"},{"comment":"Eq. (12): the definition of the mismatched metric should explicitly indicate that it ignores the memory term derived in Eq. (9).","section":"Eq. (12)"}],"recommendation":"minor_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the careful reading, positive assessment of the work's significance, and constructive suggestions. We address each major comment below and will revise the manuscript to improve clarity and reproducibility.","responses":[{"response":"We agree that these details are necessary for full reproducibility. In the revised manuscript we will explicitly state that the front-end filter is a rectangular integrate-and-dump filter with impulse response equal to the symbol duration and that sampling occurs at the center of each symbol interval. With these choices the conditional variance reduces exactly to the reported polynomial in the transmitted symbol; the coefficients can then be obtained directly from the continuous-time RIN parameters via the derivation in §3.","revision_made":"yes","referee_comment":"[§3] §3 (discrete-time model derivation): the claim that the conditional variance is exactly polynomial in the symbol rests on the specific sampling instant and front-end filter response; the manuscript should state the filter impulse response and sampling phase explicitly so that the polynomial coefficients can be reproduced from the continuous-time parameters."},{"response":"The paper deliberately evaluates GMI under the memoryless mismatched metric because this corresponds to the low-complexity decoding used in practical high-speed IM/DD systems. The observed saturation is a direct consequence of this mismatch: the memory in the RIN-induced noise produces nonsymmetric, nonvanishing per-symbol contributions to the GMI, which we quantify explicitly. We will add a clarifying paragraph in §4 emphasizing that the saturation is an artifact of the memoryless restriction and is therefore expected; a quantitative gap to a memory-retaining decoder (e.g., BCJR) is left for future work, as implementing such a decoder lies outside the scope of the present contribution focused on practical decoding.","revision_made":"partial","referee_comment":"[§4] §4 (GMI evaluation): the saturation result is obtained under a memoryless metric; the paper should report the GMI gap to a metric that retains the derived memory (e.g., via a trellis or factor-graph decoder) for at least one constellation size, to confirm that the observed saturation is not an artifact of the memoryless restriction."}],"tokens_in":1331,"tokens_out":468,"duration_ms":43521,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main thing here is a cleaner discrete-time channel model for relative intensity noise in intensity-modulation direct-detection links. Starting from the standard continuous-time optical waveform, the authors arrive at noise whose conditional variance has a polynomial dependence on the current symbol plus memory terms. They then compute generalized mutual information with a practical memoryless mismatched metric and find that GMI stops growing once constellations get dense, because the symbol contributions to the rate are nonsymmetric and do not vanish.","headline":"This paper derives a discrete-time RIN model with explicit polynomial noise variance and memory, then shows GMI saturation for large constellations under memoryless decoding.","tokens_in":2197,"tokens_out":168,"would_cite":false,"duration_ms":24984,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":{"model":"grok-4.3","evidence":[{"relation":"unclear","rs_module":"IndisputableMonolith/Cost/FunctionalEquation.lean","rs_theorem":"washburn_uniqueness_aczel","paper_passage":"Theorem 2 (Conditional Variance of Z_k). ... σ²_{Z|A}(ℓ,a_n) = N_rin0/2 ∑∑ E[A_i A_j | A_k=a_n] Ψ_{i,j} ... polynomial form E[Z²_k | A_k=a_n] = ½ N_rin0 (p0 + p1 a_n + p2 a_n²)"}],"headline":"Standard IM-DD channel model with polynomial RIN variance; no RS structures","alignment":"orthogonal","rationale":"Paper derives discrete-time model Y_k = A_k + Z_k + Q_k where Var(Z_k | A_k = a_n) is quadratic polynomial in a_n (Thm 2, eq. 13), then computes GMI under memoryless mismatched Gaussian metric (eq. 22). Central machinery is conventional information-theoretic analysis of signal-dependent noise; contains none of the RS primitives (J-cost, cosh(· ln φ), φ-ladder, 8-tick periodicity, ratio-symmetric forcing). Domain (practical optical comms) lies outside RS scope.","tokens_in":47034,"confidence":"high","tokens_out":296,"duration_ms":26537,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"In optical intensity-modulation links impaired by laser relative intensity noise, a memoryless mismatched decoder causes achievable rates to saturate as constellation size grows.","keywords":["relative intensity noise","intensity modulation direct detection","channel modeling","generalized mutual information","mismatched decoding","signal-dependent noise","optical communications","achievable rates"],"falsifier":"A physical or high-fidelity simulation experiment in which constellation size is increased while keeping the same memoryless decoder and measuring whether GMI indeed plateaus rather than continuing to rise.","tokens_in":2554,"feed_emoji":"","tokens_out":461,"duration_ms":27263,"temperature":0.7,"pith_summary":"The paper starts from a continuous-time waveform model of an IM/DD optical channel and derives a discrete-time equivalent in which RIN produces signal-dependent noise whose variance depends polynomially on the current symbol and carries memory from prior symbols. Achievable rates are then computed via the generalized mutual information under a practical memoryless decoding metric. Numerical evaluation shows that GMI stops increasing once constellations become dense, because individual symbols contribute to the rate in a nonsymmetric and nonvanishing way. A reader would care because the result indicates that simply raising modulation order brings no further rate benefit once the receiver ignores channel memory, a common engineering choice.","feed_headline":"RIN makes GMI saturate as constellations grow denser","feed_subtitle":"When a memoryless decoder ignores channel memory, achievable rates stop improving once modulation order increases because symbols contribute","key_machinery":"Discrete-time channel model whose noise variance is a polynomial function of the current symbol and carries memory from previous symbols, induced by relative intensity noise.","core_discovery":"The derived discrete-time channel model exhibits signal-dependent noise with memory whose conditional variance is a polynomial function of the transmitted symbol. When a memoryless mismatched decoder is used, the generalized mutual information saturates with growing constellation cardinality; the saturation arises from the nonsymmetric, nonvanishing contributions that different symbols make to the GMI.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["RIN Memory Causes GMI Saturation With Growing Constellation Sizes","Signal-Dependent RIN Noise With Memory Halts GMI Improvements","Polynomial Noise Variance From RIN Saturates Achievable Rates","Ignoring Channel Memory From RIN Leads to GMI Saturation"],"cache_read_input_tokens":64,"weakest_assumption_plain":"The continuous-time waveform model together with the chosen sampling and filtering steps fully captures all relevant RIN dynamics, so that the resulting discrete-time polynomial variance expression remains accurate.","fun_headline_variants_meta":{"raw":{"variants":["RIN Memory Causes GMI Saturation With Growing Constellation Sizes","Signal-Dependent RIN Noise With Memory Halts GMI Improvements","Polynomial Noise Variance From RIN Saturates Achievable Rates","Ignoring Channel Memory From RIN Leads to GMI Saturation"]},"model":"grok-4.3","cost_usd":0.004468,"raw_usage":{"total_tokens":2109,"prompt_tokens":590,"num_sources_used":0,"completion_tokens":68,"cost_in_usd_ticks":44678000,"prompt_tokens_details":{"text_tokens":590,"audio_tokens":0,"image_tokens":0,"cached_tokens":64},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1451,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":590,"tokens_out":68,"duration_ms":18425,"temperature":1.0,"reasoning_tokens":1451,"cache_read_input_tokens":64,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-05-15T13:48:45.213440+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A physical or high-fidelity simulation experiment in which constellation size is increased while keeping the same memoryless decoder and measuring whether GMI indeed plateaus rather than continuing to rise.","supporting_citations":[],"review_version":1}