{"id":"602ee4f0-5beb-48c8-b24c-5fe30faad005","arxiv_id":"2505.04753","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A hybrid-field 6DMA THz channel model and a directional-sparsity-driven channel estimation algorithm are proposed, with simulations showing the model's capacity is close to the full near-field model.","lead":"This paper proposes a hybrid-field channel model that mixes plane waves inside each movable antenna panel with spherical waves across panels for Terahertz (THz) communications. It also introduces a channel estimator that uses directional sparsity to build a full channel map from a small number of panel position-rotation measurements.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (13) double-counts the surface-center position q_b in the phase, so the hybrid model is not the near-field approximation it claims to be.","rationale":"The reader's weakest assumption was the directional-sparsity premise and clustering robustness. That is a legitimate sensitivity concern, but a more foundational problem exists earlier: the displayed hybrid-field channel model (13), as written, does not approximate the near-field channel. The surface steering vector in (12) uses global antenna positions including q_b, while the prefactor in (13) uses e^(−j2πd_b/λ) with d_b equal to the user-to-surface-center distance. This double-counts the surface-center position in the phase. The error is not a matter of near-field versus far-field consensus; it is an internal consistency failure. It makes Remark 1 false already in the trivial N=1 case, and it biases the multi-surface refinement because the per-surface phase errors are different for different surfaces. The simulation-based claims about the model and the estimator therefore cannot be relied upon as presented. The concern is concrete and testable: a one-line comparison of (13) and (10) for N=1 settles it. Because the central mathematical object of the paper is wrong as written, I would reject this version while noting that a corrected phase reference might make the approach viable.","tokens_in":10944,"tokens_out":18759,"duration_ms":195158,"concrete_test":"Check the B=N=1 case with q_b≠0 (e.g., q_b=[0.25,0,0], user at [20,0,0], λ=3 mm): Eq. (13) gives phase −2π(d_b+f_b^T q_b)/λ while Eq. (10) gives −2π d_b/λ, so Remark 1 fails. Then compute the normalized correlation between (13) and (10) for the full simulation geometry across random user positions; if it is not exactly 1, rerun the Fig. 4 MSE experiment with a(q_b,u_b) replaced by the local-offset steering vector exp(−j2π f_b^T R(u_b)r̄_n/λ) to determine whether the claimed channel-reconstruction accuracy survives the phase correction.","verdict_should_be":"REJECT","load_bearing_attack":"The core model (13) builds each surface block as ν√g·e^(−j2πd_b/λ)·a(q_b,u_b), where a(q_b,u_b) in (12) uses the global antenna positions r_{b,n}=q_b+R(u_b)r̄_n, and d_b is defined as the user-to-surface-center distance. For a user at p, the correct first-order near-field phase at antenna n is approximately −2π/λ·(d_b + f_b^T R(u_b)r̄_n), with f_b=(q_b−p)/d_b. Equation (13) instead gives −2π/λ·(d_b + f_b^T q_b + f_b^T R(u_b)r̄_n). The extra term −2π f_b^T q_b/λ is generally not a multiple of 2π; e.g., q_b=[0.25,0,0], p=[20,0,0], λ=3 mm gives about 83.3 cycles. Thus the vector in (13) differs from the near-field model by a per-surface, q_b-dependent phase. The claimed degeneracies in Remark 1 already fail for N=1: (13) reduces to ν√g·e^(−j2π(d_1+f_1^T q_1)/λ), not to the corresponding entry of (10). Because this phase error differs across surfaces, the joint likelihood in (27a) and the reconstruction (28) are biased; only the surface-wise objective (20) is invariant to it. The numerical results therefore validate a different model than the one written, unless the phase reference in (12)–(13) is corrected.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes a hybrid-field LoS channel model for a THz base station equipped with multiple six-dimensional movable antenna (6DMA) surfaces, where each surface is a small uniform planar array with its own RF chain. The model retains spherical-wave phases across different surfaces but uses a single direction per surface for the intra-surface array response, thereby reducing the parameter count relative to a full near-field model. The estimation algorithm operates in two stages: per-surface maximum-likelihood grid search over distance and angles, followed by a distance-based clustering step that discards estimates from surfaces with weak or inconsistent reception and refines the parameters over a fine grid using only the largest cluster. The complete channel map for arbitrary position-rotation pairs is then reconstructed from the estimated distance, angles, and path gain. Numerical results are presented for channel capacity versus distance and for channel-estimation MSE versus SNR, with claims that the hybrid model closely matches the near-field model and that the estimator is accurate with low complexity.","tokens_in":11283,"tokens_out":12069,"duration_ms":119172,"significance":"The hybrid-field modeling idea is well motivated: for a large 6DMA movement region at THz frequencies, a user can be in the near field of the overall aperture while remaining in the far field of each small surface, so a model that combines per-surface plane waves with cross-surface spherical phases could provide substantial parameter savings. The two-stage estimation architecture, which processes surfaces independently and then fuses only reliable estimates, is also a sensible approach to the uneven power distribution across candidate positions. If the technical issues below are resolved, the paper would make a useful contribution to 6DMA channel modeling and estimation. However, the central model equation contains a phase double-counting error, the claimed low complexity is not supported by the stated grid sizes, and the numerical validation is incomplete because the ground-truth model used in the MSE experiment is not specified. These issues are substantive but appear fixable within the scope of a revision.","major_comments":[{"comment":"The steering vector in (12) is built from the global antenna positions r_{b,n}=q_b+R(u_b)\\bar r_n, while the block in (13) is additionally multiplied by e^{-j2π d_b/λ}, where d_b is the user-to-surface-center distance. The resulting phase is -2π/λ(d_b + f_b^T q_b + f_b^T R(u_b)\\bar r_n). The correct first-order near-field phase at antenna n, with f_b=(q_b-p)/d_b, is -2π/λ(d_b + f_b^T R(u_b)\\bar r_n). Thus (13) contains a spurious per-surface phase e^{-j2π f_b^T q_b/λ}, which is not a global phase because it differs across surfaces. This invalidates Remark 1: for N=1, (13) reduces to ν√g e^{-j2π(d_1+f_1^T q_1)/λ}, not to the corresponding entry of the near-field model (10). It also means the joint likelihood (27a) and the reconstructed channel (28) are biased relative to the near-field ground truth. The authors should correct (12) to use the local offset R(u_b)\\bar r_n, or equivalently r_{b,n}-q_b, and re-run the simulations.","section":"Section II-B3, Eqs. (12)-(13)"},{"comment":"The complexity claim in the abstract and conclusion is not supported by the stated parameters. With Δd_H=λ, D_min=20 m, D_max=800 m, and Δφ_H=Δθ_H=π/1000, the coarse grid Ξ_H contains approximately (780/0.003)×2001×1001 ≈ 5.2×10^11 points. Evaluating the correlation in (20) at each point costs O(N) operations per surface, so for M=16 and N=16 the surface-wise stage alone is on the order of 10^14 complex operations. The paper reports O(TNB|Ξ_H|_c + BN_c) but gives no runtime or flop count and no comparison with the complexity of the LS baseline. A genuinely low-complexity algorithm would need a decoupled or multi-resolution search, and the paper should either report actual complexity under the chosen parameters or substantially revise the low-complexity claim.","section":"Section III-B, Eq. (18) and Section III-C"},{"comment":"The ground-truth channel used to compute the MSE in Fig. 4 is not stated. If the data are generated from the hybrid-field model (13) and the estimator also uses (13), the experiment only measures self-consistency of a parameter estimator and does not validate the model against the physical near-field channel. If the data are generated from the near-field model (10), then, because of the phase error in (12)-(13), the estimator is mismatched and the reported MSE cannot be interpreted as the reconstruction error of the proposed model. The authors must state the generating model explicitly and, ideally, evaluate the reconstruction against the near-field model (10) as ground truth. In addition, Fig. 3 considers N=16 with λ/2 spacing at distances of at least 20 m, where the per-surface Rayleigh distance is only about 1.35 cm; this regime is unlikely to expose intra-surface near-field effects, so the close match between hybrid and near-field models is expected and is a weak test of the model.","section":"Section IV, Fig. 4"},{"comment":"The refinement step relies on the assumption that reliable per-surface position estimates form a single dense cluster and that the largest cluster corresponds to the true user position. This is the directional-sparsity premise, but the paper provides no sensitivity analysis of the clustering threshold ε, no experiment with a user visible to multiple widely separated surfaces, and no quantification of how often the largest-cluster rule fails. Because the discard rule directly determines which measurements enter the joint estimation in (27a) and hence the reconstructed channel map in (28), the estimation claims are conditional on an unvalidated clustering assumption. Please add a sensitivity study varying ε, SNR, and user location, and state the conditions under which the sparsity premise holds.","section":"Section III-C, Algorithm 1"}],"minor_comments":[{"comment":"In (17) the dependence of d_m and a(q_m,u_m) on the candidate parameters (d,φ,θ) through the geometric relationship is described in words but never written out; making this mapping explicit would improve reproducibility.","section":"Section II-B3, around Eq. (17)"},{"comment":"The fine-search ranges D, Φ, and Θ are defined symbolically but never specified in Section IV. Their values directly affect both accuracy and complexity, so they should be reported.","section":"Section III-C, Eq. (26)"},{"comment":"The paper does not clearly distinguish the number of pilot-measured position-rotation pairs M from the size of the candidate set used to select the B active surfaces; please clarify, especially since Fig. 4 uses M=16 and M=25 while Fig. 2 appears to use M×N up to 400.","section":"Section IV, Figs. 3-4"},{"comment":"The effective antenna gain pattern A(θ,φ) is only referenced as 'the 3GPP standard'; providing the exact pattern (or a reference with equation numbers) is needed for reproducibility.","section":"Section IV, simulation setup"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern about double-counting q_b is confirmed by direct inspection of (12)-(13); this is a real modeling error that must be fixed before the paper can be accepted. I would also require the authors to state the ground-truth model in the MSE experiment and to provide a complexity accounting that reflects the actual grid sizes. The paper is within the journal's scope and the issues are substantive but not beyond repair, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, the paper has a good idea wrapped in a flawed model: hybrid-field (far-field within a surface, near-field across surfaces) is exactly the right approximation for 6DMA THz. Second, the model as written in (13) has a phase error that invalidates its central claim. In (12), the steering vector uses the global antenna position r_{b,n}=q_b+R(u_b)\\bar{r}_n. Multiplying by e^{-j2π d_b/λ} in (13) gives a per-surface phase e^{-j2π f_b^T q_b/λ} on top of the intended d_b + f_b^T R(u_b)\\bar{r}_n. This extra phase is not a multiple of 2π in general and differs per surface. For N=1, the model gives ν√g e^{-j2π(d_1+f_1^T q_1)/λ}, not the near-field entry ν√g e^{-j2π d_1/λ}. Remark 1's claimed reduction to near-field is therefore false. The joint estimation in (27a) and reconstruction (28) are biased if the true channel is the near-field one.\n\nCredit where it's due: the idea is timely, the writing is clear, and the algorithm (surface-wise ML with clustering, then fine grid) is a reasonable pipeline. The literature review is fine. But the complexity claim is not supported: the coarse grid over distance with Δd_H=λ over 20–800 m gives hundreds of thousands of distance bins, and the angle grid with π/1000 steps gives ~2×10^6 angle pairs—times M surfaces and T measurements, this is not 'low complexity' unless the search is done cleverly, which isn't described. The threshold ε and fine-grid ranges are never reported, there are no error bars, and no code/data. The directional-sparsity premise is asserted; the clustering step can discard valid measurements if the user is visible to many surfaces or noise merges clusters.\n\nThe capacity result in Fig. 3 is insensitive to the phase error because per-surface common phases don't change ||h||^2, but the MSE results in Fig. 4 are only meaningful if ground truth is generated from the same (incorrect) hybrid model. As written, the paper does not validate the near-field approximation.\n\nBottom line: this is a promising but not yet reliable contribution. The phase issue is fixable, and after a careful re-derivation the estimation pipeline could be worth revisiting. I'd send it to peer review with a request for major revision, but I wouldn't cite it or rely on its conclusions until the model is corrected and the complexity/robustness questions are answered.","headline":"The hybrid-field model in Eq. (13) double-counts the surface-center phase, so the paper's central claim—that it approximates the near-field channel—fails on its own equations; the estimation results then validate a different model than the one written.","tokens_in":11863,"tokens_out":9628,"would_cite":false,"duration_ms":83690,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Hybrid-field model matches near-field THz rate at lower cost","keywords":["hybrid-field channel model","six-dimensional movable antenna","terahertz communications","near-field propagation","channel estimation","directional sparsity","6DMA","sum rate"],"falsifier":"Simulate a user placed so that every 6DMA surface is rotated toward it, then run Algorithm 1 while sweeping the clustering threshold epsilon across two orders of magnitude; if the largest-cluster refinement discards most surfaces or the reconstructed channel MSE rises sharply, directional sparsity is violated and the claim of accurate low-complexity estimation fails.","tokens_in":10754,"feed_emoji":"📡","tokens_out":3820,"duration_ms":39439,"temperature":0.7,"pith_summary":"The paper argues that a terahertz link with six-dimensionally movable antenna surfaces is best described by a hybrid-field channel model: planar (far-field) waves inside each antenna surface, spherical (near-field) waves between surfaces. It claims this model tracks the full near-field ground truth in achievable sum rate while using far fewer channel parameters, and that it contains both the pure far-field and pure near-field models as special cases. To acquire the channel, the paper proposes a two-stage estimator that exploits directional sparsity, the observation that each user is strongly served by only a few position-rotation pairs, so coarse per-surface estimates are refined only from the reliable cluster. A sympathetic reader would care because this is a path to serving many users with few RF chains and low pilot overhead in a regime where the near-field region can extend hundreds of meters.","feed_headline":"Hybrid-field model matches near-field THz rate at lower cost","feed_subtitle":"Per-surface wave fronts plus directional sparsity cut pilot overhead for movable-antenna terahertz base stations.","key_machinery":"The load-bearing object is the hybrid-field channel model in (13): for each 6DMA surface it uses a single user-to-surface distance and a single signal direction, multiplying a per-surface far-field steering vector by a surface-level spherical-wave phase. This keeps the parameter count tied to the number of surfaces rather than to the number of candidate antenna positions and rotations, while still capturing the near-field curvature across widely spaced surfaces. The estimation machinery is directional sparsity, the premise that each user has significant channel gain for only a small subset of position-rotation pairs, which justifies discarding surface-wise estimates that fall outside the largest consistent cluster before joint fine-grid refinement.","core_discovery":"The central claim is that the hybrid-field 6DMA THz channel vector in equation (13), which assigns each antenna surface its own direction-of-arrival and distance while keeping a planar-wave steering vector within each surface and a spherical-wave phase across surfaces, reproduces near-field channel capacity accurately with far fewer parameters than a full near-field model. The paper also claims that the directional-sparsity-driven estimator, which performs surface-wise maximum-likelihood estimation, clusters the resulting user-position estimates in Cartesian space, keeps only the largest cluster, and then refines the parameters on a fine grid, reconstructs the complete channel map for all candidate position-rotation pairs with low complexity. Simulation results are presented showing sum rate close to the near-field model, especially below the 500-meter Rayleigh distance, and channel-estimation MSE substantially lower than a least-squares baseline.","pith_inferences":["The paper gives no sensitivity analysis for the clustering threshold epsilon; a data-dependent or multi-threshold version would be a natural testable extension, since the largest-cluster rule can either merge scattered noise clusters or discard legitimate estimates when epsilon is poorly chosen.","If directional sparsity weakens, for example a user positioned so that many 6DMA surfaces have comparable gain toward it, the discard rule becomes the first point of failure; retaining and fusing multiple clusters instead of only the largest would be a direct robustness extension.","The parameter reduction of the hybrid-field model suggests it could be inserted into 6DMA position-rotation optimization, which the paper lists as future work, potentially making joint configuration search over the continuous movement space tractable.","A concrete stress test would be to estimate the channel for a user at the boundary between near- and far-field regions with all surfaces rotated toward the user, and to check whether the largest-cluster refinement still keeps enough surfaces to meet a target MSE."],"forward_implications":["Within the Rayleigh distance, the hybrid-field model preserves near-field capacity while remaining parameter-efficient; beyond it, the far-field, near-field, and hybrid-field models converge.","The estimator reconstructs the channel for all candidate position-rotation pairs from only M measured pairs, so pilot overhead scales with the number of measured configurations, not the full candidate set.","Because estimation is performed surface-wise and each surface has its own RF chain, the scheme is compatible with surface-based hybrid beamforming hardware in THz systems.","Directional sparsity turns unused position-rotation pairs into effectively zero-gain entries, which is what makes the largest-cluster refinement both low-complexity and accurate.","The model reduces to the conventional far-field 6DMA model when there is one surface and to a near-field model when each surface has a single antenna, so it unifies the two existing modeling regimes."],"supporting_citations":[{"why":"Supplies the conventional far-field 6DMA channel model and the 6DMA system concept that the proposed hybrid-field model reduces to when B=1.","marker":"[5]"},{"why":"Defines the six-dimensional position and rotation parameterization of 6DMA surfaces that the system model builds on.","marker":"[6]"},{"why":"Provides the Rayleigh-distance criterion and near-field parameter growth that motivate the hybrid-field model and the simulation's 500-meter near-field boundary.","marker":"[13]"},{"why":"Justifies neglecting NLoS components and modeling only the dominant LoS THz channel.","marker":"[14]"},{"why":"Supplies the maximum-likelihood localization approach used for the surface-wise coarse channel estimation stage.","marker":"[15]"},{"why":"Supplies the coarse-then-fine grid-search methodology used in both the surface-wise and multi-surface refinement stages.","marker":"[16]"},{"why":"Supplies the distance-based clustering rule used to discard unreliable position-rotation estimates and select the dominant cluster.","marker":"[17]"}],"fun_headline_variants":["Hybrid THz channel model cuts pilot cost for movable antennas","Per-surface wavefronts enable low-overhead 6DMA THz estimation","Near-field accuracy without near-field cost: hybrid THz","6DMA THz: directional sparsity shrinks channel map overhead","Hybrid-field 6DMA THz matches near-field rate at lower overhead"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The estimator assumes directional sparsity: each user has a significant channel gain for only a small subset of the 6DMA position-rotation pairs, so estimates from all other surfaces can be discarded as noise; if a user is strongly visible to many surfaces, or noisy estimates merge into one cluster, this discard rule removes valid information.","fun_headline_variants_meta":{"raw":{"variants":["Hybrid THz channel model cuts pilot cost for movable antennas","Per-surface wavefronts enable low-overhead 6DMA THz estimation","Near-field accuracy without near-field cost: hybrid THz","6DMA THz: directional sparsity shrinks channel map overhead","Hybrid-field 6DMA THz matches near-field rate at lower overhead"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000294,"raw_usage":{"total_tokens":1734,"prompt_tokens":994,"completion_tokens":740,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":610,"completion_tokens_details":{"reasoning_tokens":645}},"tokens_in":610,"tokens_out":740,"duration_ms":7647,"temperature":1.0,"reasoning_tokens":645,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T23:22:16.579188+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate a user placed so that every 6DMA surface is rotated toward it, then run Algorithm 1 while sweeping the clustering threshold epsilon across two orders of magnitude; if the largest-cluster refinement discards most surfaces or the reconstructed channel MSE rises sharply, directional sparsity is violated and the claim of accurate low-complexity estimation fails.","supporting_citations":[{"cited_title":"Challenges in LoS Terahertz MIMO,","cited_arxiv_id":null,"evidence_quote":"Justifies neglecting NLoS components and modeling only the dominant LoS THz channel."},{"cited_title":"Low-overhead localization and VR identification for subarray-based ELAA systems,","cited_arxiv_id":null,"evidence_quote":"Supplies the coarse-then-fine grid-search methodology used in both the surface-wise and multi-surface refinement stages."},{"cited_title":"Review and perspective for distance-based clustering of vehicle trajectories,","cited_arxiv_id":null,"evidence_quote":"Supplies the distance-based clustering rule used to discard unreliable position-rotation estimates and select the dominant cluster."}],"review_version":1}