{"id":"aa23bbd6-e9ee-4b0a-bcfa-9af63e0f0ba2","arxiv_id":"2502.04003","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"The paper derives the MSE of BEM-based channel estimation and a BER lower bound for OTFS under transceiver hardware impairments, validated by simulation.","lead":"This paper analyzes how imperfect transceiver hardware affects channel estimation and error rates in OTFS, a modulation for high-speed mobile links. It derives closed-form formulas for the channel-estimation error and a bit-error-rate lower bound, showing that even small hardware flaws noticeably hurt performance.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 2's SINR derivation conflates true and estimated channel; Eq. (29) does not follow from Eq. (15), so the BER lower bound (33) is not established.","rationale":"The reason I pick this over the reader's Gaussian-impairment objection is that the Gaussian model is a standard, cited assumption (refs [10],[11]) and the paper's own simulations enforce it; a mismatch with unknown real-world impairments would be a scope limitation, not a defect in the derivation. The H_t/\\Hat H_t conflation, in contrast, is an internal inconsistency in the proof of Theorem 2 that directly undermines the claimed BER lower bound, which is one of the paper's two headline analytic results. If the concern lands, the paper can likely be repaired by redefining T with \\Hat H_t and clearly stating whether the BER is averaged conditionally on the channel estimate; but as currently written the theorem is not proved. I still recommend CONDITIONAL rather than REJECT because the MSE analysis (Theorem 1) appears sound, the simulations are reproducible in structure, and the BER formula may be the intended standard result with a single mis-specified matrix. The reader's verdict of CONDITIONAL is therefore preserved, but for a different and more specific reason.","tokens_in":9865,"tokens_out":20877,"duration_ms":185262,"concrete_test":"Compute, for one simulated realization with M=64, N=16, ξ_i=ξ_o=0.95, SNR=15 dB, the Monte Carlo estimates of E|\\Hat x_d[i]|^2 for all i and compare them (a) to sqrt(ξ_o ξ_i)σ_d^2 T[i,i] with T from Theorem 2 (using H_t), and (b) to sqrt(ξ_o ξ_i)σ_d^2 \\Hat T[i,i] with \\Hat T = sqrt(ξ_o ξ_i) F G_t \\Hat H_t F^H. If (a) deviates beyond Monte Carlo error while (b) matches, Theorem 2's stated equality (29) is false and the BER bound (33) cannot be justified as written.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing defect is in the BER analysis. Eq. (15) defines the input to the MMSE detector as \\r_hat_o = sqrt(ξ_o ξ_i) \\Hat H_t s_d + n, with n = sqrt(ξ_o ξ_i) \\tilde H_t s + ..., so the effective signal channel after pilot cancellation is \\Hat H_t, not H_t. The detector G_t in (16) is also built from \\Hat H_t. Therefore the correct data estimate is \\Hat x_d = sqrt(ξ_o ξ_i) F G_t \\Hat H_t F^H x_d + F G_t n, and the MMSE identity that yields E|\\Hat x_d[i]|^2 = sqrt(ξ_o ξ_i) σ_d^2 T[i,i] holds only for T built from \\Hat H_t. However Theorem 2 and Eqs. (28)-(29) define T = sqrt(ξ_o ξ_i) F G_t H_t F^H and use H_t in the power calculation. Because n already contains the \\tilde H_t s term, using H_t in the signal part double-counts the channel estimation error. With G_t based on \\Hat H_t but H_t in the covariance, the identity (29) acquires missing cross terms sqrt(ξ_o ξ_i) G_t(\\Hat H_t R_sd \\tilde H_t^H + \\tilde H_t R_sd \\Hat H_t^H) G_t^H; these are not generally zero. Hence SINR[i] = T[i,i]/(1-T[i,i]) in (31) is not a consequence of the model, and the Jensen lower bound (33) is unproven. The paper's simulations use 4-QAM and accurate channel estimates, so the numerical mismatch may be small, but the theoretical claim as written is invalid. This is distinct from the Gaussian-impairment modeling question: it is an internal derivation error.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This letter analyzes BEM-based channel estimation and MMSE detection for OTFS systems under transceiver hardware impairments, modeled through quality factors xi_i and xi_o with Gaussian distortion noises. The paper derives the MMSE channel estimator and a closed-form MSE expression (Eq. 13), then proposes a time-domain MMSE detector and derives a lower bound on the average BER (Eq. 33). Numerical simulations are presented to validate the MSE and BER analyses.","tokens_in":10231,"tokens_out":11913,"duration_ms":101765,"significance":"If the derivations were sound, the paper would provide a useful extension of BEM-OTFS channel estimation to practical hardware-impaired transceivers; the MSE analysis is structurally standard and the Fig. 3 match between simulation and theory supports that portion. The BER analysis, however, is not supported by the derivation as written due to inconsistent use of the estimated versus true channel and algebraic errors in Appendix B. The qualitative conclusion that hardware impairments degrade BER may still be correct, but the theoretical backing needs substantial repair.","major_comments":[{"comment":"The first displayed expression for \\hat x_d[i] in Eq. (28) does not follow from Eq. (15). Since (15) gives \\hat r_o = sqrt(xi_o xi_i) \\hat H_t s_d + n, with n containing sqrt(xi_o xi_i) \\tilde H_t s, the DD-domain estimate should be \\hat x_d = sqrt(xi_o xi_i) F G_t \\hat H_t F^H x_d + F G_t n. Replacing \\hat H_t by H_t in (28) and in T = sqrt(xi_o xi_i) F G_t H_t F^H double-counts the channel estimation error, because n already includes the \\tilde H_t s term. The missing cross terms sqrt(xi_o xi_i) G_t( \\hat H_t R_sd \\tilde H_t^H + \\tilde H_t R_sd \\hat H_t^H ) G_t^H do not vanish in general. Consequently, the SINR expression (31) and the lower bound (33) are not derived from the stated model.","section":"Appendix B, Eqs. (28)-(29)"},{"comment":"The algebra in Eq. (29) is incorrect: the noise term from the first line is f_i G_t R_n G_t^H f_i^H, so the second line should be f_i G_t( xi_o xi_i H_t R_sd H_t^H + R_n ) G_t^H f_i^H, not xi_o xi_i f_i G_t( H_t R_sd H_t^H + R_n ) G_t^H f_i^H. As a result, the equality to xi_o xi_i sigma_d^2 T[i,i] does not hold dimensionally or algebraically; with T defined as in the paper, the factor of xi_o xi_i is inconsistent.","section":"Appendix B, Eq. (29)"},{"comment":"The denominator of the averaged T[i,i] term is written as N, but N is the Doppler dimension of the OTFS grid; the average over the Nnum data symbols should be divided by Nnum. As written, the lower bound is not well-defined, and the numerical value of the bound would depend on this normalization. This should be corrected and the derivation should state which quantity is actually averaged.","section":"Appendix B, Eqs. (17) and (33)"},{"comment":"The Jensen step requires convexity of eta(x) = erfc( sqrt(b_M x/(1-x)) ) on (0,1), which is asserted by citation [14] but not proven or stated as a lemma in this paper; since this is load-bearing for the lower bound, the authors should provide a proof or a precise statement of the convexity condition for the b_M used. In addition, the paper never gives an explicit expression for R_n = E[n n^H] used in G_t and T; without it, the theoretical BER curve cannot be reproduced from the equations, and the paper should state whether R_n is computed analytically from (14) or estimated from simulation.","section":"Appendix B, Eq. (33) and R_n definition"}],"minor_comments":[{"comment":"The first sentence contains a duplicated word: \"with with zero mean\" should be \"with zero mean\".","section":"Appendix A"},{"comment":"\"exhibits a increase in volatility\" should read \"exhibits an increase in volatility\".","section":"Section IV"},{"comment":"\"Pilot Patten\" is a typo for \"Pilot Pattern\".","section":"Figure 2 caption"},{"comment":"The sentence \"Rhh denotes the covariance matrix of the BEM coefficient vectors and the CIR vectors\" is ambiguous; it should clarify that Rhh is the covariance of the CIR vector h and Rc is the covariance of the BEM coefficient vector c.","section":"Section III-B, after Eq. (10)"},{"comment":"The summation index starts at i=0 in Eq. (32) but at i=1 in Eq. (33); the indexing should be made consistent, presumably summing over i=1,...,Nnum.","section":"Theorem 2, Eq. (32)"}],"recommendation":"major_revision","confidential_remarks":"The MSE analysis appears sound and is a useful contribution, but the BER lower bound is not established as written. The errors are localized to Appendix B and Eqs. (28)-(33) and seem repairable by replacing H_t with \\hat H_t in the signal term, correcting Eq. (29), and specifying R_n. I recommend major revision rather than reject because the central approach remains valid and the fixes are within the scope of the manuscript."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The short version: the MSE part is worth reading, and the BER bound is not. The stress-test note is right. Equation (15) puts \\hat H_t in the signal path, but (28)–(29) switch to H_t. Since n already includes \\tilde H_t s, that double-counts the estimation error, and the MMSE identity that gives E|\\hat x_d[i]|^2 = \\sigma_d^2 T[i,i] only holds if T is built from \\hat H_t. As written, Theorem 2's SINR and the Jensen bound (33) do not follow. This is a genuine internal inconsistency, not a typo, though it may be repairable by redefining T with \\hat H_t and re-deriving the noise covariance.\n\nWhat's new here is the BEM-OTFS estimation under residual hardware impairments. The derivation of the MSE in (13) follows the standard MMSE line, and the simulation curves in Fig. 3 back it up. That part is useful and honestly presented. The extension is incremental—it reduces to the ideal-hardware results when \\xi_i = \\xi_o = 1—but it quantifies something designers care about.\n\nThe soft spots beyond the BER issue: the convexity of \\eta(x) is imported from [14] without a proof; the notation in (17) and (33) has a confusing N/Nnum mix; and the impairment model is the usual Gaussian distortion model, untested against measurements. Those are minor by comparison. The novelty claim is a bit strong, since the components are known, but the combination is a reasonable data point.\n\nThis is a paper for OTFS receiver designers. I'd send it to review because the MSE result is solid and the BER error is fixable, but a referee needs to check Theorem 2 carefully. I wouldn't cite the BER bound as it stands.","headline":"MSE part is solid and useful; the BER lower bound has an internal channel-model inconsistency that should be fixed before it is cited.","tokens_in":10801,"tokens_out":8665,"would_cite":false,"duration_ms":74572,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["94A12","94A40"],"pacs":[],"model":"deepseek-v4-flash","headline":"Hardware impairments sharply degrade OTFS channel estimation and BER, with closed-form MSE and lower-bound expressions.","keywords":["OTFS","channel estimation","basis expansion model","hardware impairments","MMSE detector","bit error rate","mean square error","delay-Doppler domain"],"falsifier":"Measure a real OTFS transceiver's MSE and BER against equations (13) and (33) across SNR values while independently measuring the hardware quality factors; if the high-SNR error floor departs from the predicted floor, or moves with modulation format in a way (33) cannot capture, the Gaussian distortion model is falsified.","tokens_in":9663,"feed_emoji":"📡","tokens_out":7329,"duration_ms":62938,"temperature":0.7,"pith_summary":"This letter tries to establish that OTFS systems are sensitive to residual hardware impairments and that this sensitivity is exactly quantifiable. Using a basis expansion model for the time-varying channel, the authors derive a closed-form mean square error expression for MMSE channel estimation that includes transmitter and receiver quality factors, distortion noise, and basis expansion modeling error. They then derive a lower bound on the average bit error rate of an MMSE detector from per-symbol SINR expressions. The formulas show that even small departures from ideal hardware noticeably degrade error performance, especially at high SNR where thermal noise no longer dominates. A sympathetic reader should care because the results set a concrete hardware-quality target for OTFS to deliver its Doppler-resilience advantage.","feed_headline":"Hardware flaws degrade OTFS error rates, analysis shows","feed_subtitle":"New closed-form MSE and BER formulas let designers predict exactly when transceiver quality limits OTFS.","key_machinery":"The central machinery is the generalized complex exponential basis expansion model (GCE-BEM), which represents each time-varying channel tap as a small sum of exponential basis functions plus a modeling error term, reducing the channel to a coefficient vector of size $(Q_L+1)(L+1)$. Hardware impairments enter through two scalar quality factors $\\xi_i$ and $\\xi_o$ that scale the useful signal and add Gaussian distortion noise at the transmitter and receiver. An MMSE estimator for the BEM coefficients and an MMSE detector in the time domain carry the analysis; their error covariance matrices and per-symbol SINR expressions produce the MSE formula (13) and the BER lower bound (33).","core_discovery":"The paper claims that for an OTFS link whose transmitter and receiver suffer residual hardware impairments, modeled as independent zero-mean Gaussian distortion noises with covariances $(1-\\xi_i)E[|s|^2]I$ and $(1-\\xi_o)E[|H_t s_i|^2]I$, the MSE of the BEM-based MMSE channel estimator is given by (13) and the average BER of the MMSE detector is lower-bounded by (33). The MSE expression splits the estimation error into BEM coefficient error plus modeling error, while the BER bound follows by applying Jensen's inequality to the per-symbol SINR. The paper validates both formulas by simulation and concludes that even minor hardware impairments cause significant performance degradation, most visibly in the high-SNR regime where AWGN is no longer the dominant error source.","pith_inferences":["If the Gaussian distortion model were replaced by measured impairment statistics, such as IQ imbalance or amplifier nonlinearity, the same derivation structure could produce adjusted MSE and BER expressions, but the paper does not test such distributions.","Equation (33) suggests a practical hardware calibration target: transceivers need quality factors above some threshold before OTFS Doppler resilience is realized, and extracting that threshold from the bound is a natural follow-up.","The simulations' comparison with OFDM implies the framework could be extended to quantify hardware-impairment penalties across modulation waveforms, though the paper only gestures at that comparison."],"forward_implications":["Equation (13) lets system designers predict channel estimation accuracy from transceiver quality factors alone, without running Monte Carlo simulations.","The BER lower bound (33) exposes an SNR ceiling set by hardware quality: beyond it, increasing transmit power stops improving the error rate.","The analysis identifies the high-SNR region as the regime where hardware impairments dominate, so OTFS performance there is limited by hardware quality rather than by thermal noise.","The BEM-based estimator keeps pilot overhead low (7.5% in the simulations) while making the hardware-impaired MSE analytically tractable."],"supporting_citations":[{"why":"Supplies the Gaussian distortion-noise model for residual transmitter hardware impairments used in equations (1) and (2).","marker":"[10]"},{"why":"Supplies the companion receiver distortion-noise modeling and motivates the covariance structure in (1)-(2).","marker":"[11]"},{"why":"Provides the GCE-BEM channel representation, the BEM coefficient estimation approach, and the modeling-error covariance expressions used in the MSE derivation.","marker":"[5]"},{"why":"Provides the embedded pilot-aided OTFS channel estimation baseline that the paper's pilot pattern extends.","marker":"[4]"},{"why":"Supplies the per-symbol SINR to BER conversion that Theorem 2 adapts and simplifies.","marker":"[12]"},{"why":"Provides the modulation-specific constants $a_M$ and $b_M$ used in the BER bound.","marker":"[13]"},{"why":"Supplies the convexity of the complementary error function used with Jensen's inequality to obtain the BER lower bound (33).","marker":"[14]"}],"fun_headline_variants":["Minor hardware flaws spike OTFS error rates","OTFS degrades sharply with slight hardware impairments","Closed-form MSE, BER expose OTFS hardware sensitivity","Even small hardware flaws hurt OTFS performance","OTFS error rates climb as hardware quality drops"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central assumption is that residual hardware impairments are independent zero-mean Gaussian distortion noises whose powers scale with the signal via the quality factors; if real impairments are nonlinear, non-Gaussian, or correlated with the signal, the derived MSE and BER formulas break down.","fun_headline_variants_meta":{"raw":{"variants":["Minor hardware flaws spike OTFS error rates","OTFS degrades sharply with slight hardware impairments","Closed-form MSE, BER expose OTFS hardware sensitivity","Even small hardware flaws hurt OTFS performance","OTFS error rates climb as hardware quality drops"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000959,"raw_usage":{"total_tokens":4031,"prompt_tokens":836,"completion_tokens":3195,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":452,"completion_tokens_details":{"reasoning_tokens":3125}},"tokens_in":452,"tokens_out":3195,"duration_ms":26973,"temperature":1.0,"reasoning_tokens":3125,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T23:54:08.485710+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure a real OTFS transceiver's MSE and BER against equations (13) and (33) across SNR values while independently measuring the hardware quality factors; if the high-SNR error floor departs from the predicted floor, or moves with modulation format in a way (33) cannot capture, the Gaussian distortion model is falsified.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the Gaussian distortion-noise model for residual transmitter hardware impairments used in equations (1) and (2)."},{"cited_title":"Tubail, B","cited_arxiv_id":null,"evidence_quote":"Supplies the companion receiver distortion-noise modeling and motivates the covariance structure in (1)-(2)."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the GCE-BEM channel representation, the BEM coefficient estimation approach, and the modeling-error covariance expressions used in the MSE derivation."},{"cited_title":"Raviteja, K","cited_arxiv_id":null,"evidence_quote":"Provides the embedded pilot-aided OTFS channel estimation baseline that the paper's pilot pattern extends."},{"cited_title":"Signal Processing, vol","cited_arxiv_id":null,"evidence_quote":"Supplies the per-symbol SINR to BER conversion that Theorem 2 adapts and simplifies."},{"cited_title":"Goldsmith, Wireless communications","cited_arxiv_id":null,"evidence_quote":"Provides the modulation-specific constants $a_M$ and $b_M$ used in the BER bound."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the convexity of the complementary error function used with Jensen's inequality to obtain the BER lower bound (33)."}],"review_version":1}