{"id":"3ca67ed7-7468-432b-b939-b88aee1ecabd","arxiv_id":"2607.09184","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.5,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Effective DD-domain channel filters of Zak-OTFS frames evolve deterministically with known phase factors, enabling ESPRIT-style inter-frame prediction that removes pilots from future frames.","lead":"Zak-OTFS channel filters can be predicted across frames tens of milliseconds and hundreds of MHz away by exploiting deterministic phase evolution of multipath components. This cuts pilot overhead and raises spectral efficiency in high-Doppler links.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified beyond the stationarity assumption already flagged by the reader.","rationale":"The reader correctly isolates the single soft spot: the physical DD spreading function must remain fixed for the phase-only extrapolation (Eq. 72) to stay accurate. All other pieces—factorization, subspace invariance, ESPRIT recovery of the unit-modulus poles, and the reported CRLB proximity—are self-contained and hold under the paper’s own assumptions. Because that stationarity caveat is already the basis of the CONDITIONAL verdict, no further adjustment is warranted. The suggested continuous-mobility test simply quantifies how far the assumption can be stretched before the headline numbers degrade.","tokens_in":25045,"tokens_out":401,"duration_ms":5016,"concrete_test":"Re-run the NMSPE heatmap of Fig. 5 after replacing the static Veh-A paths by a continuous-time trajectory in which each path’s delay and Doppler evolve linearly at rates corresponding to 30 m/s radial motion and 10 deg/s angular motion; if NMSPE at (n,m)=(60,60) remains below –10 dB the stationarity margin is larger than claimed, otherwise the horizon must be shortened.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central algebraic claim (Theorem 3 + Lemmas 1–3) is internally consistent: under the stated assumptions the effective filter factors exactly as h_{n,m}[k,l]=sum_i alpha_i^n beta_i^m A_i[k,l] and the consecutive-frame matrices share a rotationally invariant column space that ESPRIT recovers. The only load-bearing condition that can break the prediction is the one already identified—Assumption 1 (stationarity of h_phy over the horizon). No hidden inconsistency appears in the derivation, the rank arguments, or the numerical evidence under that assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper shows that inter-frame channel prediction is possible for Zak-OTFS. From the multi-frame I/O relation (Theorems 1–3), the effective DD-domain filter of the (n,m)-th frame factors exactly as h_{n,m}[k,l] = sum_i alpha_i^n beta_i^m A_i[k,l], where the unit-modulus phases alpha_i = exp(j 2 pi nu_i T') and beta_i = exp(-j 2 pi tau_i B') are determined by the physical path delays and Dopplers. Lemmas 1–3 establish that consecutive-frame filter matrices share a rotationally invariant column space. An ESPRIT-type procedure recovers the phases and the DD signature matrix A from Q training frames (past frames in time and frequency) and predicts the filter for future frames via Eq. (72). Monte-Carlo results on the six-path Vehicular-A channel report normalized prediction error of roughly -14 dB at (n,m)=(120,120) under 15 dB pilot SNR, and a 30 percent SE gain for prediction frames that omit pilots.","tokens_in":25190,"tokens_out":1047,"duration_ms":15154,"significance":"If the stationarity assumption holds over the claimed horizon, the result is a concrete, low-complexity alternative to AR or DNN CSI predictors for high-mobility Zak-OTFS. The algebraic factorization is derived from first principles rather than fitted, the ESPRIT step is deterministic, and the complexity is only O(N_t Q^3). The SE gains from pilot-free prediction frames and the potential reduction of FDD CSI feedback are practically relevant. Strengths include clean twisted-convolution derivations, explicit rank conditions, CRLB comparisons, and extensive Vehicular-A Monte-Carlo evidence under the stated assumptions.","major_comments":[{"comment":"Assumption 1 (Sections V-A and V-D) is load-bearing: the physical spreading function h_phy is required to remain essentially stationary over the entire prediction horizon (tens of ms and hundreds of MHz). Section V-D supplies only order-of-magnitude arguments (c/(v B) and angle-change estimates). No numerical experiment injects continuous path drift (linear acceleration, gradual angle change, or mild birth/death) and measures the resulting NMSPE degradation. Without such a stress test the claimed 60–120 ms / several-hundred-MHz horizon remains unquantified for realistic non-stationary channels.","section":null},{"comment":"All numerical results (Figs. 4–10) use a single synthetic six-path Vehicular-A model with fixed relative powers and i.i.d. angles. There is no evaluation on other standardized profiles (e.g., TDL, CDL), measured outdoor traces, or hardware-in-the-loop data. Consequently the reported NMSPE floors and SE gains cannot yet be taken as representative of practical deployment conditions.","section":null}],"minor_comments":[{"comment":"The abstract and introduction claim prediction “several tens of frames” away; the body (Figs. 5–6) shows usable accuracy out to n=m=120. Align the wording so that the abstract does not understate the demonstrated horizon.","section":null},{"comment":"Section V-E: the support-set threshold X = 0.01 E[|h_P|^2]/E[|h_1|^2] is stated for Vehicular-A but never varied. A short sensitivity plot of NMSPE versus X would strengthen the claim that the rule is robust.","section":null},{"comment":"Typographical consistency: “ESPIRIT” appears throughout; the conventional acronym is ESPRIT. Also “Vehicular-A” vs “Veh-A” in Table I.","section":null},{"comment":"Fig. 3 caption refers to “our work in [31]”; a self-contained description of the pilot/guard layout would improve readability for readers who do not consult the reference.","section":null},{"comment":"The complexity claim O(N_t Q^3) is stated after Step 5; a brief breakdown of the dominant SVD and Hungarian steps would help implementers.","section":null}],"recommendation":"minor_revision","confidential_remarks":"The algebraic core is solid and the self-citations are to the authors’ foundational Zak-OTFS papers that supply the I/O model; they do not create circularity. The main risk is over-claiming the prediction horizon without non-stationary or measured-channel evidence. A minor-revision decision that requires an explicit stationarity-stress experiment (or a clear limitation statement) and one additional channel model would keep the paper within scope while protecting the journal’s standards."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The new result is the exact factorization in Theorem 3: the effective DD filter of the (n,m)-th frame is simply sum_i alpha_i^n beta_i^m A_i[k,l], with the alphas and betas deterministic unit-modulus phases coming from the physical path delays and Dopplers. Lemmas 1–3 then show that consecutive-frame matrices share a rotationally invariant column space, so a standard ESPRIT step recovers the phases and the signature matrix A from a modest number of training frames. That is a genuine, previously missing piece for multi-frame Zak-OTFS.\n\nThe derivation is transparent and self-contained. The algorithm is fully specified (including the Hungarian reordering step and the support-set rule), complexity is linear in the number of taps, and the Vehicular-A Monte-Carlo campaign is thorough: MSE sits near the CRLB once Q is large enough, NMSPE stays around –14 dB out to 120 ms / 432 MHz at 15 dB pilot SNR, and the coded SE gain from dropping pilots is real (roughly 30 % at 15 dB TPNR). Self-citations are only to the authors’ earlier Zak-OTFS foundations that supply the I/O model; they are not circular.\n\nThe single load-bearing assumption is stationarity of h_phy over the prediction horizon (Section V-D). The order-of-magnitude arguments are reasonable for 50–100 ms and a few hundred MHz, but they remain untested on measured traces or hardware. Everything else—rank conditions, fractional-delay assumptions, free parameters Q and X—is either standard or minor. No hidden inconsistency appears in the algebra or the numerics under the stated model.\n\nThis is for people already working on OTFS or high-mobility CSI prediction who need a concrete, low-complexity way to cut pilot overhead and enable cross-band prediction. It deserves a serious referee; the math and the simulation evidence are solid enough that the only real debate will be about how far the stationarity claim travels in practice. I would engage with it.","headline":"Clean algebraic solution to inter-frame Zak-OTFS prediction that actually works under its stated assumptions; stationarity is the only real soft spot.","tokens_in":25805,"tokens_out":517,"would_cite":true,"duration_ms":5724,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"Zak-OTFS channel filters evolve by deterministic phases across frames, so an ESPRIT-type method can predict them tens of frames ahead from training pilots alone.","keywords":["Zak-OTFS","delay-Doppler channel","inter-frame prediction","ESPRIT","pilot overhead","high-mobility channels","spectral efficiency"],"falsifier":"Measure the normalized mean-squared prediction error on a vehicular channel whose relative velocity produces a 1 kHz Doppler shift every few tens of milliseconds; if the error rises sharply once the physical paths begin to migrate across bins, the stationarity premise is falsified.","tokens_in":25922,"feed_emoji":"📡","tokens_out":843,"duration_ms":8576,"temperature":0.7,"pith_summary":"Zak-OTFS already lets a receiver recover every carrier response inside one frame from a single pilot. This paper shows the same predictability extends across frames. The effective delay-Doppler filter of any later frame is exactly the filter of a reference frame multiplied, path by path, by two unit-modulus phases that advance linearly with frame index in time and in frequency. Because consecutive filter matrices therefore share a rotationally invariant subspace, a short ESPRIT-style procedure recovers those phases and the underlying path signatures from a few training frames that still carry pilots. The recovered parameters then synthesize the filter for any future or frequency-offset frame, so those frames need no pilots at all. Numerical results on a vehicular channel confirm that the prediction remains accurate more than 100 ms and several hundred megahertz away, raising the spectral efficiency of the predicted frames by roughly 30 percent.","feed_headline":"Zak-OTFS filters predicted tens of frames ahead from pilots alone","feed_subtitle":"Deterministic phase evolution lets ESPRIT recover the channel 120 ms and 400 MHz away, cutting pilot overhead","key_machinery":"The deterministic factorization h_{n,m}[k,l] = sum_i alpha_i^n beta_i^m A_i[k,l] together with the rotational invariance of the time- and frequency-phase Vandermonde matrices; these two facts turn inter-frame prediction into a standard ESPRIT eigen-decomposition of stacked training filters.","core_discovery":"The effective discrete delay-Doppler channel filter of the (n,m)-th Zak-OTFS frame factors exactly as a sum over paths of alpha_i^n beta_i^m A_i[k,l], where the complex scalars alpha_i and beta_i are deterministic unit-modulus phases fixed by the physical path delay and Doppler. Consequently the column space of successive filter matrices is rotationally invariant, and an ESPRIT-type algorithm recovers the phases and the signatures A_i from Q training frames, after which any later filter is obtained by simple powering.","pith_inferences":["The same phase-tracking idea may apply to any modulation whose effective channel is a twisted convolution with a slowly varying delay-Doppler kernel.","If path birth/death or large angular turns are detected by a sudden rank change in the training matrices, the algorithm can trigger a fresh training epoch automatically.","Extending the method to multi-antenna arrays would couple the spatial steering vectors into the same ESPRIT step, potentially yielding joint angle-delay-Doppler prediction."],"forward_implications":["Prediction frames need no pilot or guard carriers, so their spectral efficiency rises by the pilot overhead fraction (observed ~30 percent).","Downlink precoding can be computed from uplink training alone, eliminating CSI feedback in both TDD and FDD.","Pilot power and PAPR are reduced because only a sparse set of training frames carries pilots.","Resource allocation and beam management can be planned tens of milliseconds and hundreds of megahertz ahead of the current frame."],"fun_headline_variants":["Zak-OTFS filters predicted tens of frames ahead via ESPRIT","Deterministic phases let Zak-OTFS channels be recovered across frames","ESPRIT recovers DD channel filters from pilot frames alone","Subspace invariance enables inter-frame Zak-OTFS prediction","Predict Zak-OTFS channels 120 ms and 400 MHz away from pilots"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"The physical multipath delays and Dopplers themselves must stay fixed for the whole prediction window; if any path drifts by one delay or Doppler bin the fixed signatures A_i change and the phase-only forecast fails.","fun_headline_variants_meta":{"raw":{"variants":["Zak-OTFS filters predicted tens of frames ahead via ESPRIT","Deterministic phases let Zak-OTFS channels be recovered across frames","ESPRIT recovers DD channel filters from pilot frames alone","Subspace invariance enables inter-frame Zak-OTFS prediction","Predict Zak-OTFS channels 120 ms and 400 MHz away from pilots"]},"model":"grok-4.5","effort":"low","cost_usd":0.003802,"raw_usage":{"total_tokens":1281,"prompt_tokens":875,"num_sources_used":0,"completion_tokens":77,"cost_in_usd_ticks":38020000,"prompt_tokens_details":{"text_tokens":875,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":329,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":875,"tokens_out":77,"duration_ms":9841,"temperature":1.0,"reasoning_tokens":329,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-13T04:50:05.030542+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Measure the normalized mean-squared prediction error on a vehicular channel whose relative velocity produces a 1 kHz Doppler shift every few tens of milliseconds; if the error rises sharply once the physical paths begin to migrate across bins, the stationarity premise is falsified.","supporting_citations":[],"review_version":1}