{"id":"1a981b91-a375-4ef2-9f54-f2aedcebac90","arxiv_id":"2411.16420","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A fifth-order Vandermonde-structured CP decomposition estimates multipath channel parameters for active RIS broadband links, including the direct transmitter-receiver path, with mostly linear algebra and optional refinements.","lead":"This paper presents a tensor-based method to estimate multipath wireless channels in a system where an active reconfigurable intelligent surface mounted on a UAV helps a base station communicate. The method builds a five-way data tensor, factorizes it with a structured decomposition, and recovers delays, angles, and path gains using mostly linear algebra, with optional refinement stages.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (40) is only valid when the Vandermonde generators are distinct; the paper never tests or bounds the case of equal or near-equal delays, and the similarity principle in Stage I has no separation threshold, so the central uniqueness-to-estimation chain is not robust to realistic parameter…","rationale":"The reader and I identify the same weakest point: the central claim depends on separability hypotheses that are asserted but not analyzed. The algebraic VSCPD construction itself is coherent, and for the simulation parameters (K1=15, R=6) the sufficient condition (40) is satisfied, so the observed simulation trends are not internally contradicted. However, Theorem 1 explicitly assumes distinct Vandermonde generators, and the paper does not carry that assumption explicitly into the standalone statement of (40); equal or near-equal delays therefore break the ESPRIT-based uniqueness proof. Likewise, the similarity principle in Stage I assumes that the P duplicate columns in B4/B5 remain mutually more correlated than columns from different RIS-BS subpaths, but no threshold or resolvability condition is given. A controlled delay-collision experiment would settle whether this is merely a generic-identifiability caveat or a practical failure mode. Since the paper does not currently provide such robustness analysis, the conditional verdict is appropriate; there is no basis for rejection because the reported simulations support the claimed trends in the tested regime.","tokens_in":63,"tokens_out":23131,"duration_ms":434673,"concrete_test":"Run the Section V-C multipath simulation with L=P=Q=2, but set the two direct-path delays exactly equal (tau_L^(1)=tau_L^(2)) while keeping all angles and the remaining parameters unchanged; at 20 dB over 1000 Monte Carlo trials, report the success rate of the Stage-I variance/similarity checks and the delay RMSE. If the success rate drops below 100% or the RMSE leaves the CRLB, the distinct-generator hypothesis in Theorem 1 is load-bearing rather than a harmless generic assumption.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 1's hypothesis requires A1 to have distinct generators, yet the paper's headline condition (40) is stated as if uniqueness depends only on K1 and K2. In the physical channel, two paths can share the same delay; then A1 (and its submatrices B1, B6) lose full column rank, ESPRIT cannot recover distinct generators, and the proof of Theorem 1 stops applying. The only support is the word 'generally' in Section III-C. A second, related failure point is the similarity principle (41) in Section III-D: grouping the P duplicate columns in B4 and B5 requires a correlation threshold and assumes the P copies are more similar to each other than to columns of other RIS-BS subpaths. No threshold, resolution analysis, or failure-mode study is supplied; if two RIS-BS angles are closer than the array resolution, the grouping and the L2=L3 / P4=P5 checks collapse. The sufficiency of (40) for the simulated K1=15 regime is not in question; the concern is that the central claim is presented as a general condition while silently inheriting separability assumptions that are not quantified.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper addresses channel parameter estimation for active RIS-assisted SIMO-OFDM systems with both a direct UE-BS link and a cascaded UE-RIS-BS link in full multipath. The authors construct a fifth-order CP tensor via customized pilots, RIS profiles, and a combining matrix, then apply spatial smoothing in the frequency mode and develop an algebraic Vandermonde-structured CPD (VSCPD) with the tensor having only one Vandermonde factor. A triple-stage estimator is proposed: Stage I gives path identification and coarse algebraic estimates, while Stages II and III are optional search- and ALS-based refinements. A closed-form CRLB is derived in the active-RIS case where the noise covariance depends on the multipath parameters, and simulations compare the proposed stages with ALS-CPD-based baselines and with passive RIS.","tokens_in":60,"tokens_out":10275,"duration_ms":162131,"significance":"The problem is well motivated: existing VSCPD-based channel estimation is limited to third-order tensors or to scenarios without a direct link, and replacing a 2D search with parallel 1D estimations is practically valuable. The tensor construction and the pilot/RIS/combining design are thoughtful, and the paper includes useful numerical evidence such as success rates, run times, and active-versus-passive RIS comparisons. If the uniqueness theorem and the path-identification steps are made rigorous, the proposed framework would be a solid contribution to active-RIS multipath channel estimation. However, the central advertised claim, the relaxed uniqueness condition (40), is currently stated as a general condition while relying on separability assumptions that are not quantified.","major_comments":[{"comment":"The proof of Theorem 1 is not internally consistent as written. Since B1 in (30) has K1 rows, the left-hand side J↑B123 in (35) has (K1-1)G1G2 rows, whereas the right-hand side B1⊙B2⊙B3 has K1G1G2 rows; the displayed identities cannot both hold. In addition, the text after (39) asserts that rank(B1⊙B2⊙B3)=min(K1-1,R) follows from the Khatri-Rao lower bound, but with kB2=kB3=1 that property gives only a lower bound, not the stated equality. The derivation of (40) therefore needs to be corrected, and the conclusion that K≥2R should be re-examined: under (40) the constraint is K≥2R+1.","section":"III-C, Theorem 1 and Eqs. (30)-(40)"},{"comment":"The general applicability of (40) is not established because Theorem 1 assumes that A1 has distinct generators. In the physical channel, two paths can share the same delay or have delays separated by less than the resolution of the training subcarriers; then A1 and its smoothed submatrices B1 and B6 become rank-deficient or ill-conditioned, ESPRIT in Algorithm 1 cannot recover R distinct generators, and the uniqueness claim (40) is neither necessary nor sufficient. The manuscript's only support for this assumption is the word \"generally\" before Eq. (40). The authors should state the precise separability condition, for example a minimum delay separation relative to 1/(KΔf), and provide a failure-mode analysis for equal or closely spaced delays.","section":"III-C, Theorem 1 hypothesis and Eq. (24)"},{"comment":"The similarity principle used to group the P duplicate columns in B4 and B5 has no defined threshold and no resolution analysis. The correlation criterion in (41) will fail when two RIS-BS subpaths have angular separations smaller than the array resolution or when the estimated factor columns are too noisy, and the algorithm then aborts through the checks in lines 6-9 of Algorithm 2. Since this grouping is load-bearing for the subsequent mapping of delays, angles, and gains in (43)-(46), the paper needs a quantitative separation criterion and either an error analysis for the grouping step or a remedy when grouping fails.","section":"III-D, Eq. (41) and Algorithm 2"},{"comment":"The closed-form CRLB depends on neglecting the second term in (55), the q≠q' cross term in GRGH_R. The paper justifies this by saying that the first term is larger due to aligned phases, but the neglected term contains Q(Q-1) unit-modulus phase factors and also depends on the delay and angle parameters; its magnitude relative to the first term is not established. The authors should either retain the cross term in the FIM or validate the approximation numerically, for example by comparing the approximate CRLB with the full FIM or with the empirical error of a near-ML estimator in the high-SNR asymptotic region.","section":"IV-A, Eq. (55)"}],"minor_comments":[{"comment":"The phrase \"direst UE-BS paths\" appears to be a typo for \"direct UE-BS paths.\"","section":"Eq. (25)"},{"comment":"The notation \"V andermonde\" with a space appears repeatedly; it should be written as \"Vandermonde.\"","section":"Throughout"},{"comment":"The element-space and transformed-space ESPRIT models use selection/projection matrices J↑, J↓, Q_n, and F_n, but Q_n and F_n are not defined in the text; please add their definitions or point to the exact equations in the cited references.","section":"III-B and III-D"},{"comment":"The complexity expressions in (60) would benefit from a consistency check; for example, the term O(R^2 K G^2) for delay recovery and the Stage III expression should be re-derived or commented on, since several terms appear to be missing factors of K or G.","section":"IV-B"},{"comment":"The zoomed insets in Figures 4 and 6 are small and not explained in the captions; please enlarge them and state what is being magnified.","section":"Figs. 4 and 6"}],"recommendation":"major_revision","confidential_remarks":"No ethical concerns. The main risk is that Theorem 1 and the path-identification steps are advertised as general results while hiding separability assumptions; I believe the issues are fixable with a corrected uniqueness proof, explicit assumptions, and numerical validation of the CRLB approximation."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read the paper. It is a genuine extension of the VSCPD line: prior tensor work on RIS channel estimation either blocked the direct UE-BS link or stayed in LOS, and this one builds a fifth-order CP tensor for fully multipath with the direct link present. The pilot/RIS-profile design that makes direct-path columns constant, and the variance principle that uses that to separate direct from cascaded paths, is the clever part. The algebraic pipeline - spatial smoothing, ESPRIT-based VSCPD, then two optional refinement stages - is coherent, and the simulations support the main trends: perfect VSCPD success rates where ALS-CPD fails (Tables II and III), the expected gain from the active RIS, and sensible complexity numbers. The simulation settings are described in enough detail that the results are reproducible on paper, even without released code. The soft spots are real but not fatal. The headline uniqueness condition (40) is stated as if it depends only on K1 and K2, but the proof and the ESPRIT step inherit a genericity assumption: the Vandermonde generators (delays) must be distinct, and the similarity principle (41) needs a correlation threshold that the paper never specifies. Two paths sharing a delay, or two RIS-BS angles closer than array resolution, are exactly the cases where the variance/similarity checks either fail or need a tuned threshold; the paper's response is the word 'generally' and perfect success rates in the chosen regime. I don't think this sinks the paper - it matches how the VSCPD literature normally works - but the condition should be stated with its separability assumptions made explicit, and a failure-mode study (or at least a threshold sensitivity analysis) would materially strengthen it. The CRLB derivation is a second, smaller soft spot: the noise covariance depends on the RIS-BS channel, and the paper drops the cross-terms in (55) as 'reasonable' with no bound on the approximation error. Minor, but an error-bounded or numerically verified approximation would be easy to add. No code or data is included; for this venue that is typical, but given the empirical claims, releasing the Tensorlab-based simulator would settle reproducibility questions. Net: this paper deserves a serious referee. I would ask the authors to state the distinctness and threshold assumptions explicitly, quantify the CRLB approximation, and ideally release code. As is, it is a solid conditional-accept candidate.","headline":"A solid VSCPD extension to fifth-order tensors for active RIS channel estimation with a direct link; the uniqueness claim is plausible but stated a bit too crisply about genericity, and the path-identification similarity step lacks a threshold.","tokens_in":741,"tokens_out":1908,"would_cite":true,"duration_ms":60467,"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":"This paper establishes that a fifth-order tensor with one Vandermonde factor can be uniquely decomposed after spatial smoothing even when the other four factors have repeated columns, and uses that decomposition to estimate all multipath…","keywords":["active RIS","channel estimation","tensor decomposition","Vandermonde structured CP decomposition","spatial smoothing","multipath parameter estimation","OFDM","Cramér-Rao lower bound"],"falsifier":"Simulate the multipath scenario with $L=P=Q=2$, then set two of the direct-link delays equal so $A_1$ has a repeated generator. If the VSCPD still uniquely recovers all parameters and the variance and similarity checks still pass, Theorem 1's distinct-generator assumption is not necessary; if it fails, that assumption is confirmed as load-bearing.","tokens_in":28380,"feed_emoji":"📡","tokens_out":6936,"duration_ms":65977,"temperature":0.7,"pith_summary":"This paper tries to show that channel estimation for an active-RIS-assisted broadband link in a fully multipath environment, including a direct transmitter-receiver path, can be solved by tensor decomposition without any two-dimensional search. The authors construct a fifth-order canonical polyadic tensor whose five modes match the channel's five dimensions, then prove a relaxed uniqueness condition: spatial smoothing along the single Vandermonde factor guarantees uniqueness under $\\min(K_1-1,K_2)\\geq R$ even though the other four factors have repeated columns. This yields a triple-stage algorithm: algebraic coarse estimation followed by two optional refinements, all built on one-dimensional parameter estimation. A closed-form Cramer-Rao lower bound is derived for the active-RIS case, where thermal noise at the RIS makes the noise covariance depend on multipath parameters. If correct, the result replaces exhaustive 2D search in RIS channel estimation with faster algebraic 1D recovery and extends prior LOS-only active-RIS work to general multipath.","feed_headline":"Active-RIS channel estimation cut from 2D search to parallel 1D","feed_subtitle":"Fifth-order Vandermonde tensor decomposition stays unique with repeated columns, so one algebraic pass recovers delays, angles, and gains.","key_machinery":"The central object is the spatial-smoothed fifth-order Vandermonde structured CP decomposition (VSCPD): a canonical polyadic decomposition in which one factor matrix is Vandermonde, with columns $e^{j(m-1)\\omega_r}$, and the tensor is augmented by smoothing along that mode. This machinery converts a rank-deficient tensor factorization into linear algebra: a compact SVD of the mode-3 matricization, ESPRIT applied to the shift-invariance of the Vandermonde factor to extract delays, and rank-1 SVDs to split the Khatri-Rao products into individual factor columns for the other four modes.","core_discovery":"The central claim is that a fifth-order CP tensor with only one Vandermonde factor can be uniquely decomposed after spatial smoothing, under the condition $\\min(K_1-1,K_2)\\geq R$, even though the other four factor matrices contain repeated columns and therefore violate Kruskal's classical uniqueness condition. On this basis, the paper constructs a received-signal tensor of order five matching the SIMO-OFDM channel's five dimensions, decomposes it with a VSCPD algorithm using only SVD and ESPRIT, identifies which columns correspond to direct versus cascaded paths by variance and similarity principles, and recovers delays, RIS angle-related parameters, BS arrival angles, and path gains for both links. A closed-form CRLB is derived in which the noise covariance depends on RIS-BS channel parameters, and simulations show that the triple-stage estimator achieves high accuracy with 100% decomposition success in the tested multipath scenarios.","pith_inferences":["Editorial extension: the same fifth-order VSCPD recipe is portable to other active-RIS and integrated-sensing setups where one tensor mode has Vandermonde structure, such as Doppler or another spatial dimension, so the paper's replacement of 2D search by parallel 1D estimation is not tied to the specific SIMO-OFDM model.","Editorial extension: the variance and similarity checks that the paper uses only to abort on failure could be turned into a data-driven rank and path-number selection rule, since they give a direct measure of whether the decomposition's column clustering is self-consistent.","Editorial extension: because the active-RIS noise covariance depends on RIS-BS parameters, reporting a single SNR-based NMSE may hide large per-parameter differences; the derived CRLB suggests comparing estimators parameter-by-parameter, especially at low SNR.","Editorial extension: the Stage II search refinement could likely be replaced by one Newton step from the Stage I algebraic estimate; the CRLB would then show whether the residual gap is search-resolution-limited or inherent to the ESPRIT initialization."],"forward_implications":["Channel parameters for both direct and cascaded links are recovered from one fifth-order CP decomposition with no 2D search; each parameter type is estimated by parallel 1D problems.","VSCPD uniqueness holds with $\\min(K_1-1,K_2)\\geq R$ even though four of the five factor matrices have repeated columns, so Kruskal's condition is not needed.","The algorithm runs with only linear algebra (SVD, ESPRIT, rank-1 approximations) before optional refinements, making Stage I initialization-free and faster than ALS-CPD.","Optional Stage II (correlation-based search) and Stage III (ALS initialized by Stage II) improve accuracy sequentially, and Stage I alone outperforms ALS-CPD baselines in multipath scenarios.","Active RIS amplification lifts the cascaded-link power enough to make decomposition succeed in the simulated setting, improving both success rate and tensor reconstruction error compared to the passive-RIS case."],"supporting_citations":[{"why":"Supplies the Vandermonde structured CP decomposition and its uniqueness theorem, which the paper extends from third order to fifth order.","marker":"[26]"},{"why":"Provides the Khatri-Rao rank bound used to turn the Theorem 1 rank conditions into the explicit $\\min(K_1-1,K_2)\\geq R$ generic condition.","marker":"[45]"},{"why":"Defines the prior active-RIS LOS-only joint calibration and positioning framework that this paper generalizes to fully multipath channels with a direct link.","marker":"[33]"},{"why":"Introduces the ALS-CPD plus exhaustive correlation-based search channel estimator used as a baseline and as the origin of the CBS refinement principle.","marker":"[47]"},{"why":"Earlier compressed low-rank tensor channel estimation for IRS-assisted mmWave OFDM; represents the third-order formulation whose 2D search this paper replaces.","marker":"[24]"},{"why":"Earlier Vandermonde structured tensor decomposition for RIS-aided MIMO that also relies on 2D search, used as a baseline.","marker":"[25]"},{"why":"Provides the beamspace/transformed-space ESPRIT procedure used column-wise in Stage I to recover generators from the factor estimates.","marker":"[43]"},{"why":"Supplies the standard CP tensor model, Kruskal uniqueness discussion, and least-squares factor/gain update used in the algorithm.","marker":"[42]"}],"fun_headline_variants":["Active RIS multipath channel estimation via fifth-order tensor algebra","Vandermonde CP decomposition: unique despite repeated columns","One algebraic pass for active RIS channel parameters in multipath","From 2D search to parallel 1D: active RIS estimation","Triple-stage active RIS estimation with refined accuracy"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The decomposition is unique only if the delay generators in the Vandermonde factor are distinct and the cascaded-path columns in the other factors cluster into clean, separable groups; the paper asserts this is generally true but does not analyze what happens when two paths nearly share a delay or when noise smears the clusters.","fun_headline_variants_meta":{"raw":{"variants":["Active RIS multipath channel estimation via fifth-order tensor algebra","Vandermonde CP decomposition: unique despite repeated columns","One algebraic pass for active RIS channel parameters in multipath","From 2D search to parallel 1D: active RIS estimation","Triple-stage active RIS estimation with refined accuracy"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000877,"raw_usage":{"total_tokens":3835,"prompt_tokens":1027,"completion_tokens":2808,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":643,"completion_tokens_details":{"reasoning_tokens":2726}},"tokens_in":643,"tokens_out":2808,"duration_ms":21378,"temperature":1.0,"reasoning_tokens":2726,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T13:08:39.605239+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the multipath scenario with $L=P=Q=2$, then set two of the direct-link delays equal so $A_1$ has a repeated generator. If the VSCPD still uniquely recovers all parameters and the variance and similarity checks still pass, Theorem 1's distinct-generator assumption is not necessary; if it fails, that assumption is confirmed as load-bearing.","supporting_citations":[{"cited_title":"Blind signal separati on via tensor decomposition with V andermonde factor: Canonical polyadi c decompo- sition,","cited_arxiv_id":null,"evidence_quote":"Supplies the Vandermonde structured CP decomposition and its uniqueness theorem, which the paper extends from third order to fifth order."},{"cited_title":"Almost-su re identifiability of multidimensional harmonic retrieval,","cited_arxiv_id":null,"evidence_quote":"Provides the Khatri-Rao rank bound used to turn the Theorem 1 rank conditions into the explicit $\\min(K_1-1,K_2)\\geq R$ generic condition."},{"cited_title":"JrCUP: Joint RIS calibration and user p ositioning for 6G wireless systems,","cited_arxiv_id":null,"evidence_quote":"Defines the prior active-RIS LOS-only joint calibration and positioning framework that this paper generalizes to fully multipath channels with a direct link."},{"cited_title":"L ow- rank tensor decomposition-aided channel estimation for mi llimeter wave MIMO-OFDM systems,","cited_arxiv_id":null,"evidence_quote":"Introduces the ALS-CPD plus exhaustive correlation-based search channel estimator used as a baseline and as the origin of the CBS refinement principle."},{"cited_title":"Compressed channe l estimation for IRS-assisted millimeter wave OFDM systems: A low-rank t ensor decomposition-based approach,","cited_arxiv_id":null,"evidence_quote":"Earlier compressed low-rank tensor channel estimation for IRS-assisted mmWave OFDM; represents the third-order formulation whose 2D search this paper replaces."},{"cited_title":"Slow-moving channel est imation via V andermonde structured tensor decomposition in RIS-ai ded MIMO systems,","cited_arxiv_id":null,"evidence_quote":"Earlier Vandermonde structured tensor decomposition for RIS-aided MIMO that also relies on 2D search, used as a baseline."},{"cited_title":"Tensor decomposition based beamspace ESPRIT for millimeter wave M IMO channel estimation,","cited_arxiv_id":null,"evidence_quote":"Provides the beamspace/transformed-space ESPRIT procedure used column-wise in Stage I to recover generators from the factor estimates."},{"cited_title":"Tensor decompositions and a pplications,","cited_arxiv_id":null,"evidence_quote":"Supplies the standard CP tensor model, Kruskal uniqueness discussion, and least-squares factor/gain update used in the algorithm."}],"review_version":1}