{"id":"ca853509-d272-4be9-a258-23e3a8f2500d","arxiv_id":"2509.25854","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"From LTE-R measurements at 371 km/h, the authors extract DD-domain channel models (TDDL-A/B/C) and find that fading coefficients remain quasi-invariant only for ~1-9 ms, much less than the 100-392 ms quasi-stationary intervals.","lead":"The paper measures the delay-Doppler (DD) domain wireless channel on a high-speed railway using LTE signals, and builds three tapped-delay-Doppler-line models for weak, moderate, and strong time-varying viaduct conditions. It introduces a 'quasi-invariant interval' metric showing that OTFS-style channel coefficients stay nearly constant only for milliseconds, far shorter than the statistical stationarity interval of 100-400 ms.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Algorithm 1's off-grid estimator is unvalidated; the ms-order quasi-invariant intervals and TDDL parameters may be artifacts of estimation noise, and the BER check cannot detect this.","rationale":"The paper's two central contributions are the TDDL-A/B/C models and the claim that the DD-domain quasi-invariant interval is on the ms order, much shorter than the quasi-stationary interval. Both are downstream of Algorithm 1: the delays, Dopplers, fading coefficients, amplitude distributions, and DD-TCC-based quasi-invariant intervals all use the estimated parameters. The paper provides no synthetic ground-truth validation of this estimator; the cited references are the authors' own related work. A systematic bias or noise-induced decorrelation would propagate into every reported value. The BER validation in Fig. 10 cannot expose this because the modeled channels are generated from the same estimates, so agreement only shows internal consistency. The reader's weakest assumption identifies the same estimator issue, and I agree. Other concerns, such as the single measurement segment and threshold dependence of T_QI, are real but secondary: they affect generalizability, whereas an unvalidated estimator threatens the validity of the measurements even for the specific scenario studied. The appropriate verdict remains CONDITIONAL: the methodology is plausible and the concern is testable, but the central claims should not be accepted as established until an independent simulation-based validation of Algorithm 1 and the DD-TCC pipeline is provided.","tokens_in":19083,"tokens_out":12536,"duration_ms":109405,"concrete_test":"Run a Monte Carlo simulation with the measurement pilot grid and SNR: (a) generate a perfectly time-invariant two-path DD channel with known off-grid delays/Dopplers, add noise, and apply Algorithm 1; (b) compute DD-TCC and T_QI from the estimated coefficients. If T_QI is finite for the truly invariant channel, or if the RMS bias in estimated l_i/k_i exceeds 0.1 resolution bin, the central quasi-invariant claim is not established. If T_QI is infinite and estimates are unbiased, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"All quantitative results depend on Algorithm 1, but its off-grid estimates are never checked against ground truth. Steps 6-7 interpolate delay/Doppler from the magnitude ratio of the peak and one neighbor, which is exact only for an isolated Dirichlet kernel; noise and nearby MPCs bias the ratio. Since the same estimated h_i enter the DD-TCC in Eq. (16), any zero-mean estimation error lowers DD-TCC even for a perfectly time-invariant channel, so the reported quasi-invariant intervals (T_min_QI = 0.93-9.33 ms in Table VIII) may reflect estimator noise rather than physical channel dynamics. The BER agreement in Fig. 10 is generated from the same biased estimates used to fit the models, so it demonstrates self-consistency, not unbiasedness, and the cited prior validation [38], [39] is the authors' own work.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a measurement-based methodology for delay-Doppler (DD) domain channel modeling in high-speed railway (HSR) viaduct scenarios, using LTE-R downlink pilots at 465 MHz and 371 km/h. Channel transfer-function measurements are converted to the DD domain, and a precise off-grid multipath estimator (Algorithm 1) extracts delays, Doppler shifts, and complex fading coefficients. The authors define quasi-stationary intervals via a collinearity metric in the DD domain, fit amplitude distributions (Rician/Rayleigh/Nakagami/Weibull) within those intervals, and define more stringent quasi-invariant intervals via a DD-domain time correlation coefficient (DD-TCC). Three tapped-delay-Doppler-line models (TDDL-A/B/C) are constructed for strong, moderate, and weak time-varying conditions, with path delays, Doppler shifts, powers, amplitude distributions, and minimum quasi-invariant intervals. The models are 'validated' by comparing OTFS bit-error-rate (BER) performance under modeled channels with BER under measured channels from the same quasi-stationary intervals, and by testing equalization with channel matrices separated by the quasi-invariant interval. The paper concludes that quasi-invariant intervals are on the order of milliseconds, much shorter than the 100-ms-order quasi-stationary intervals.","tokens_in":19353,"tokens_out":7213,"duration_ms":62902,"significance":"If the central claims hold, this would be one of the first systematic DD-domain channel models for HSR viaduct scenarios, providing quantitative guidance for OTFS/DDMC design. The strengths are the real measurement campaign, the explicit step-by-step modeling methodology, the concrete TDDL tables, and the use of BER as a physical-layer validation metric. However, the quantitative results and conclusions are only as reliable as the off-grid estimator in Algorithm 1 and the in-sample validation in Section V-B; both are currently unproven, so the model parameters and quasi-invariant intervals should be treated as provisional until those concerns are addressed.","major_comments":[{"comment":"The off-grid delay/Doppler estimator is never validated against synthetic ground truth. The interpolation formula from adjacent-bin magnitude ratios is exact only for an isolated Dirichlet kernel with no noise and no off-grid interference; for finite N (N=280) and closely spaced MPCs, the estimates are biased, and the iterative subtraction (Steps 9-10) cannot correct an error in the first detected path. Because every later quantity - Tables III-VIII, amplitude fits, and the DD-TCC in Eq. (16) - is derived from these estimates, a synthetic Monte-Carlo test (known delays/Dopplers, controlled SNR, variable path spacing) or comparison with an established estimator (e.g., SAGE/RI-SAGE) is needed to support the quantitative claims.","section":"II-D, Algorithm 1 (Steps 4-7)"},{"comment":"DD-TCC uses |hi(tb)hi*(tc)|/max(|hi(tb)|^2, |hi(tc)|^2), so phase differences between the two coefficients are invisible, despite the fading coefficient being complex-valued with phase entering the OTFS channel matrix in Eq. (17). More importantly, the same estimated coefficients from Algorithm 1 are used; additive estimation noise lowers DD-TCC even for a perfectly time-invariant coefficient, making the reported T_min_QI values (0.93-9.33 ms in Table VIII) potentially artifacts of estimator noise rather than physical channel dynamics. Please calibrate the metric on a synthetic time-invariant channel with known estimator noise, and consider a complex-correlation metric that captures phase.","section":"III-C, Eq. (16)"},{"comment":"The BER comparison is an in-sample goodness-of-fit test. The modeled channels are generated from parameters extracted from the same quasi-stationary interval from which the 'measured' benchmark is also drawn; agreement then demonstrates internal consistency of the fitted model, not predictive accuracy. The 2T_QS comparison only shows degradation when the channel becomes nonstationary, which is not an independent test of the TDDL models. To support the abstract claim that TDDL-A/B/C 'accurately capture the characteristics of realistic channels,' an out-of-sample validation (hold-out intervals, a second measurement segment, or cross-validation) is required; otherwise the models should be described as case-study fits.","section":"V-B, Fig. 10"},{"comment":"Each scenario class (weak/moderate/strong time-varying) is represented by a single measured segment, with no documented selection criteria or quantification of segment-to-segment variability. With only P=3, 4, and 5 detected MPCs in one segment each, the proposed TDDL-A/B/C are not established as scenario-class models. Provide multiple independent segments per class and report parameter variability, or clearly frame the tables as single-case examples and temper the generalization.","section":"IV, Tables III-VIII"}],"minor_comments":[{"comment":"The phrase 'simulation verifies that ... the quasi-invariant interval ... is on millisecond order' is inaccurate: the quasi-invariant intervals are computed from measurements, not from simulation. Consider rephrasing to 'measurements show' or 'analysis of measured data shows'.","section":"Abstract"},{"comment":"The text says T_max_QI rises from 317.07 ms to 388.73 ms, but Table V lists 314.07 ms at alpha=0.9; also 'nearly doubling' is inconsistent with a roughly 24% increase. Correct the discrepancy and the characterization.","section":"IV-D, Table V"},{"comment":"C^Period(.) is used without specifying that it is the product of C^Period_Delay(l_i,l) and C^Period_Doppler(k_i,k). Define unambiguously to avoid confusion.","section":"II-D, Algorithm 1 Step 8"},{"comment":"Each BER subplot lacks a legend. Use distinct line styles and a legend to identify 'measured within T_QS,' 'modeled,' and 'measured over 2T_QS'.","section":"V-B, Fig. 10"},{"comment":"The text lists 'T_min_QI = 2.8, 1.87, 1.4 and 0.93 ms' but there are five propagation paths in the strong-time-varying scenario. The value for path 4 (1.4 ms) is omitted from the list, although it appears in Table VIII.","section":"V-C"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe thing to know: this paper gives you three measured delay-Doppler TDL models (TDDL-A/B/C) for HSR viaduct at 465 MHz and 371 km/h, plus a new quasi-invariant interval metric based on DD-TCC. If you work on OTFS or DDMC link simulation, the tables in Section IV are directly usable. That is the real contribution.\n\nWhat it does well: the measurement effort is substantial. Field data from an operational LTE-R link at 371 km/h, off-grid delay/Doppler extraction, classification into weak/moderate/strong time-varying scenarios, and a validation step comparing OTFS BER under the fitted models against the measured channels. The BER agreement within the quasi-stationary intervals is a legitimate goodness-of-fit check: the model compresses the channel to a few taps and still reproduces link-level behavior. The 2*T_QS comparison is a nice touch showing the models do not extrapolate.\n\nThe soft spots are real. First, Algorithm 1's off-grid interpolation is never validated against a ground-truth or simulated channel. Steps 6–7 assume an isolated Dirichlet kernel; any spectral leakage or noise biases the delay and Doppler estimates, and all downstream parameters inherit that bias. The authors cite their own prior work for the method, but that does not replace a validation experiment. Second, the quasi-invariant interval is computed from the same fading coefficients that the estimator produces. If the estimator adds zero-mean noise, DD-TCC will drop below 1 even for a perfectly static channel, so the reported 0.93–9.33 ms intervals may partly reflect estimator noise rather than physical channel dynamics. The threshold dependence in Table V is shown, but without a noise-floor calibration the absolute numbers are shaky. Third, the sample is one measurement segment per scenario class, so the model parameters have no error bars. That is common in channel measurement papers, but it limits how much weight you can put on the specific tap values.\n\nThe circularity concern in the BER validation is less damning than it first looks: the models are summaries, not fits of the BER curve itself, and agreement is not automatic. But it does not validate estimator unbiasedness either.\n\nBottom line: this is a useful measurement paper for the OTFS channel-modeling subfield, and it should go to peer review. The authors should be asked to validate Algorithm 1 with a synthetic channel, quantify uncertainty on the TDDL parameters, and show a null test for DD-TCC on a static channel. If those are fixed, the models are citable. As it stands, treat the ms-order quasi-invariant numbers as provisional.","headline":"Useful new delay-Doppler TDL models for HSR viaduct from real measurements, but the off-grid estimator is unvalidated and the ms-scale quasi-invariant claim may be partly an artifact of estimation noise.","tokens_in":19860,"tokens_out":2909,"would_cite":true,"duration_ms":24842,"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":"The paper claims HSR delay-Doppler channels stay quasi-invariant on millisecond scales, and derives tapped delay-Doppler line models from LTE-R measurements at 371 km/h.","keywords":["delay-Doppler domain","high-speed railway","OTFS","LTE-R measurement","quasi-stationary interval","quasi-invariant interval","tapped delay-Doppler line model","multipath amplitude distribution"],"falsifier":"Take a synthetic channel with known off-grid delays and Dopplers, generate its sampled DD response through the same CRS pilot grid, run Algorithm 1, and compare the recovered parameters to ground truth; visible bias in the recovered delays or Dopplers would directly invalidate the TDDL parameters and the ms-scale quasi-invariant intervals derived from them.","tokens_in":18982,"feed_emoji":"🚄","tokens_out":4543,"duration_ms":37446,"temperature":0.7,"pith_summary":"The paper is trying to establish that the delay-Doppler (DD) channel in high-speed rail viaducts can be measured from ordinary LTE-R pilots and compressed into tapped delay-Doppler line models, and that the fading coefficient stays quasi-invariant on millisecond scales—far shorter than the 100-ms quasi-stationary window over which only second-order statistics are stable. It builds a full pipeline: extract coarse DD responses from time-frequency pilots, refine off-grid delay/Doppler estimates, partition data into quasi-stationary intervals via a collinearity metric, fit amplitude distributions per path, and define quasi-invariant intervals via a per-path time correlation coefficient. The payoff is a set of three concrete models (TDDL-A/B/C) whose OTFS bit-error-rate predictions match measurements inside their intervals, which would give future DDMC and ISAC designs a quantitative basis for choosing block lengths and equalization assumptions.","feed_headline":"Measured HSR channel fading stays quasi-invariant for milliseconds","feed_subtitle":"OTFS designers can trust the channel only inside millisecond windows; three new models show where those windows are.","key_machinery":"The central machinery is Algorithm 1, an off-grid delay/Doppler estimator that interpolates each path's true delay and Doppler from the magnitude ratio of the peak bin and its two immediate neighbours under the Dirichlet-like kernel of equation (9), plus the CDD collinearity metric for quasi-stationary intervals and the DD time-correlation coefficient (DD-TCC) for quasi-invariant intervals. The algorithm feeds a tapped delay-Doppler line (TDDL) model that assigns each multipath a delay, Doppler shift, power, and amplitude distribution.","core_discovery":"The paper's central discovery is that in HSR viaduct scenarios the DD-domain channel has two distinct time scales: a quasi-stationary interval on the order of 100 ms, over which second-order statistics hold, and a quasi-invariant interval on the order of milliseconds, over which the fading coefficient itself can be treated as constant. By measuring LTE-R signals at 371 km/h, extracting off-grid delay/Doppler parameters, and fitting amplitude distributions, the paper derives three tapped delay-Doppler line models (TDDL-A, -B, -C) for strong, moderate, and weak time-varying conditions. It then shows that OTFS bit-error-rate curves generated with these models align with curves from the measured","pith_inferences":["Editorial inference: The ms-order quasi-invariant bound suggests OTFS frame lengths in HSR should be much shorter than the quasi-stationary interval; designers should use per-path coherence times rather than global stationarity when choosing equalization windows.","Editorial inference: The same LTE-R/CRS-based extraction pipeline could be reapplied to other pilot grids or frequency bands, making DD-domain modeling available without dedicated sounders and potentially extending the TDDL approach to tunnels, urban rail, and low-altitude platforms.","Editorial inference: A natural next check would be to run Algorithm 1 on synthetic channels with known off-grid delays and Dopplers to bound estimation bias; if bias is small, the TDDL tables become reusable, and if not, the distribution fits and quasi-invariant intervals would need re-estimation with a higher-resolution method.","Editorial inference: Because only one measured segment per scenario class is used, the TDDL-A/B/C parameter tables should be treated as scenario-specific samples until repeated measurement runs confirm their representativeness."],"forward_implications":["OTFS packet durations in HSR viaduct channels should be sized by the minimum quasi-invariant interval of the multipath components, not by the 100-ms quasi-stationary window.","The three TDDL models give concrete tap delays, powers, Doppler shifts, and amplitude distributions that reproduce measured OTFS bit-error-rate behavior when the packet lies inside the corresponding quasi-stationary interval.","The quasi-invariant interval shrinks as time-varying intensity increases—from roughly 9 ms in weak scenarios down to sub-ms in strong scenarios—so channel invariance assumptions must be per-scenario.","Fading coefficients drift measurably even within a single quasi-stationary interval, so second-order stationarity does not imply coefficient invariance.","The proposed measurement method allows commercial OFDM-based LTE-R systems to act as DD-domain channel sounders without dedicated hardware."],"fun_headline_variants":["OTFS: Trust HSR fading only for milliseconds","HSR channel fading stable for just milliseconds","At 371 km/h, fading stays put for milliseconds","DD-domain fading: millisecond windows, 100 ms stats","New HSR channel models: 3 time-varying steps"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The entire parameter set—delays, Dopplers, amplitudes, and therefore the quasi-invariant intervals—rides on the peak-and-neighbor interpolation in Algorithm 1 being unbiased, which assumes each detected path is isolated and its leakage follows the ideal kernel of equation (9).","fun_headline_variants_meta":{"raw":{"variants":["OTFS: Trust HSR fading only for milliseconds","HSR channel fading stable for just milliseconds","At 371 km/h, fading stays put for milliseconds","DD-domain fading: millisecond windows, 100 ms stats","New HSR channel models: 3 time-varying steps"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000356,"raw_usage":{"total_tokens":1825,"prompt_tokens":853,"completion_tokens":972,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":597,"completion_tokens_details":{"reasoning_tokens":894}},"tokens_in":597,"tokens_out":972,"duration_ms":6270,"temperature":1.0,"reasoning_tokens":894,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T13:37:24.971301+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a synthetic channel with known off-grid delays and Dopplers, generate its sampled DD response through the same CRS pilot grid, run Algorithm 1, and compare the recovered parameters to ground truth; visible bias in the recovered delays or Dopplers would directly invalidate the TDDL parameters and the ms-scale quasi-invariant intervals derived from them.","supporting_citations":[],"review_version":1}