Pith. sign in

REVIEW 4 major objections 5 minor 26 references

CSI Reconstruction in Fluid Antenna Systems Without Spatial Covariance Priors

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A fluid antenna's full-port channel can be reconstructed without any covariance prior as soon as the number of observed ports reaches the modal dimension (about 2W+1), and is impossible below it regardless of SNR.

desk verdict The low-rank modal model is sensible and the oracle bounds are mostly right, but the abstract overclaims: the sharp threshold is proven only inside the reduced-rank model, not for the true Clarke channel. read the letter →

arxiv 2608.09499 v1 pith:O73UPWTI submitted 2026-08-10 cs.IT math.IT

Zhentian Zhang , Kaitao Meng , Tuo Wu , Kai-Kit Wong , Hao Xu , Liang Liu , Pei Xiao , Chao Wang
show 1 more author
Kin-Fai Tong
This is my paper · ORCID
classification cs.ITmath.IT MSC 94A1294A15
keywords fluidantennasystemschannelstateinformationreconstructionprior-freeestimationmodal-domainrepresentationClarkeisotropicscatteringphasetransitionspatialcovariancesparseportobservations
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper asks whether a fluid antenna can reconstruct the channel at all $N$ candidate ports from observations at only $M$ active ports when no spatial covariance prior is available. The answer is yes under the Clarke isotropic scattering model, because the channel is effectively confined to a low-dimensional modal subspace whose dimension $r \approx 2W+1$ is set by the physical aperture in wavelengths, not by the port count $N$. The central result is a sharp phase transition: for $M < r$ reliable reconstruction is impossible regardless of SNR, while for $M \ge r$ recovery becomes feasible and accuracy improves with more observed ports, higher SNR, and more training snapshots. This matters because it provides hardware designers with a concrete rule for sizing the number of active ports and shows that expensive covariance learning is not a prerequisite for full-port channel reconstruction.

What carries the argument

The central object is the modal-domain basis $B(\vartheta)=F_{N,\mathcal{K}(\vartheta)}$, the $N\times r$ matrix of DFT columns whose frequencies fall inside the band $|\omega|\le 2\pi d_\vartheta$. Szegő's theorem makes the Toeplitz Clarke covariance asymptotically diagonal in the DFT basis, so only $r\approx 2W+1$ modes carry significant energy; this is what reduces the reconstruction to an $r$-dimensional linear system $y_t = \Psi_t S_t B(\vartheta) z_t + n_t$. The proof machinery is the three-term MSE decomposition $\epsilon_{\mathrm{sub}}+\epsilon_{\mathrm{est}}+\epsilon_{\mathrm{learn}}$, the eigenvalue analysis of the modal sensing Gram matrix $G^{\mathsf{H}}G$, and a Fisher-information calculation for learning the unknown modal powers and noise variance.

What would settle it

Take a Clarke-model channel with $W=2$ (so $r=5$), observe $M=4$ uniformly spaced ports, and measure the NMSE of the best reduced-rank estimator at very high SNR; the paper predicts an irreducible floor of at least $(r-M)/r=1/5$ under the flat spectrum, so observing the floor fall substantially below $1/5$ as SNR grows would falsify the sharp threshold. Conversely, use a non-isotropic angular power spectrum with significant energy outside $|\omega|\le 2\pi d_\vartheta$ and check whether the truncation error at $M=r$ remains large; if it is small despite out-of-band energy, the modeling assumption is conservative, and if it is large, the negligible-truncation premise fails.

Watch

Extended reading notes

Core claim

The paper establishes that under the Clarke isotropic scattering model with a rectangular angular power spectrum, the $N$-port fluid-antenna covariance is asymptotically diagonalized by the DFT and is effectively low-rank, with rank $r = 2\lfloor W\rfloor + 1$ where $W$ is the aperture measured in wavelengths. Modeling the channel as $g_t = B(\vartheta) z_t$, with $B$ collecting the $r$ active DFT columns and $z_t$ Gaussian with an unknown diagonal covariance, turns sparse-port observation and interpolation into an $r$-dimensional linear inverse problem. The oracle reduced-rank MMSE decomposes into a modal truncation error, a sensing/estimation error, and a learning error; the truncation term vanishes as $N\to\infty$ when the modeled band contains the true spectrum. The estimation term obeys a sharp threshold: when $M<r$ every sensing matrix has a null space of dimension at least $r-M$, producing an SNR-independent NMSE floor, whereas when $M\ge r$ a generic port selection gives full column rank and the NMSE decays with SNR. The paper concludes that prior-free full-port recovery is information-theoretically possible exactly in the regime $M\ge r$, and it quantifies the learning cost through a snapshot requirement $T\ge\lceil(r+1)/M\rceil$.

Load-bearing premise

The load-bearing premise is that the true channel lies almost entirely in the $r$-dimensional DFT modal subspace with $r=2\lfloor W\rfloor+1$, meaning the scattering is isotropic with a rectangular angular power spectrum whose band is covered by the modeled frequencies and whose out-of-band energy $\epsilon_{\mathrm{sub}}$ is negligible; if real propagation concentrates significant energy outside that band, the $M\ge r$ rule no longer guarantees accurate recovery.

Editorial extensions

If this is right

  • Hardware designers can size an FAS by aperture: prior-free reconstruction needs only $M \ge 2\lfloor W\rfloor+1$ active ports, so the active-port count no longer needs to grow with the candidate-port count $N$.
  • Below the threshold, the floor is a structural limit: with $M<r$, the oracle reduced-rank NMSE is at least $(r-M)p_{\min}/(r\bar{p})$, which is $(r-M)/r$ under a flat Clarke spectrum and cannot be lowered by increasing SNR.
  • Above the threshold, uniform port selection is near-optimal in the large-$N$ regime: it gives $G^{\mathsf{H}}G\approx(M/N)I_r$ and an NMSE of about $1/(1+(M/N)\mathrm{SNR}_m)$, making performance predictable from the observation ratio and modal SNR.
  • The pilot/training cost has an explicit bound: learning the $r+1$ hyperparameters from $T$ snapshots requires $T\ge\lceil(r+1)/M\rceil$ for identifiability, and one snapshot suffices when $M\ge r+1$.
  • System design becomes a three-way tradeoff among aperture (which sets truncation), the number of observed ports and RF chains and SNR (which set estimation), and training data (which sets the learning error).

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper's claims: if the threshold persists under measured non-isotropic scattering, then the active-port count can be chosen from the aperture width alone during deployment, before any channel statistics are collected, which would simplify FAS hardware planning considerably.
  • The same modal-subspace reasoning suggests a natural test for richer scattering: an angular power spectrum with $K$ separated rectangular supports would plausibly raise the modal dimension to roughly $2W$ times the occupied bandwidth plus $K$, so comparing the measured modal dimension with $2W+1$ would locate exactly where the Clarke idealization breaks down.
  • Because the impossibility statement is proved for the oracle that knows the true modal powers, it is a limit of the reduced-rank model itself rather than of empirical Bayes; extending the impossibility beyond that model would require a separate minimax argument.
  • A concrete experiment: switch a $2W$-wavelength aperture so that $M=r-1$ and $M=r$ ports are sampled uniformly and compare NMSE at high SNR; the predicted cliff at $M=r$ would confirm the modal dimension as the controlling system parameter.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper studies reconstruction of the full N-port fluid-antenna channel from M active-port observations without a pre-acquired covariance prior. It models the channel as lying in an r-dimensional DFT modal subspace determined by the normalized aperture W (Eq. (15)), derives a three-term MSE decomposition (Proposition 1), oracle reduced-rank MMSE bounds (Propositions 3\u20135), and a phase transition at M = r (Proposition 4), and reports numerical verification for N = 64, W = 2. The central claim is that reliable prior-free recovery becomes possible once M \u2265 r and is impossible when M < r.

Significance. If the claims held unconditionally, the paper would provide useful engineering guidance for sizing FAS hardware according to the aperture rather than the total port count. The conceptual decomposition into truncation, estimation, and learning errors is helpful, and the exact reduced-rank model is analyzed with explicit bounds. However, as discussed below, the main threshold as stated is conditional on an exact-subspace model, and one of the key oracle trace formulas is not valid for noncommuting factors; these issues must be resolved before the threshold can be regarded as established for the Clarke channel.

major comments (4)
  1. [Section III-C, Eq. (28)] The oracle reduced-rank MMSE formula is not valid for general diagonal P. The posterior covariance is (P^{-1} + \sigma^{-2} G^H G)^{-1}, and its trace is not equal to \sum_k p_k \sigma^2/(\sigma^2 + p_k \mu_k) unless P and G^H G commute, which does not hold for arbitrary port selections because G = S_O B is a generic sub-Gram matrix of DFT columns. This affects Corollaries 1\u20133 and the achievability bound in Proposition 4(ii); the authors should either restrict the exact expression to scalar P or provide a correct trace formula or inequality for the general diagonal case.
  2. [Section III-B/Remark 1 and Section IV] The central M \u2265 r threshold is stated for the Clarke model, but it is proved for the assumed model g_t = B(\vartheta) z_t in Eq. (15). Proposition 2 gives only an asymptotic O(1/N) truncation bound, and Remark 1 concedes that \epsilon_sub is nonzero under basis mismatch. The numerical section does not display the total NMSE including \epsilon_sub; the reduced-rank oracle curves exclude it by construction. Thus the finite-N achievability claim for the true Clarke covariance is currently unsupported, and the threshold should either be stated conditionally on \epsilon_sub \approx 0 or verified with \epsilon_sub included in the plots.
  3. [Section III-D, Prop. 4(i)] The impossibility statement is established only for estimators confined to the r-dimensional subspace col(B). The bound NMSE \geq (r-M) p_min/(r \bar{p}) follows from the rank deficiency of G = S_O B inside that subspace and does not rule out arbitrary estimators that use a full-rank covariance or a different basis. The contribution text's phrase "information-theoretically impossible ... regardless of SNR or prior knowledge" therefore overstates the proof's scope; please add a qualifier or prove a lower bound for the full Clarke model with \epsilon_sub > 0.
  4. [Section III-B, Prop. 2] The proof of the truncation bound invokes Szeg\u0151's theorem for eigenvalue samples, but \epsilon_sub is a subspace-projection error relative to a fixed DFT column space, not an eigenvalue truncation. The O(1/N) rate in Eq. (27) needs a direct argument relating the projection error to the DFT-based spectral approximation, and it should be checked numerically, because Eq. (27) is the bridge connecting the exact-subspace model to the Clarke model.
minor comments (5)
  1. [Section II-B, Eq. (13)] Eq. (13) uses r \approx 2\vartheta + 1 while Proposition 4 uses r = 2\lfloor W \rfloor + 1; please define the cardinality of K(\vartheta) exactly for non-integer apertures and state the floor assumption consistently.
  2. [Section IV, Fig. 2] The reduced-rank oracle curve is described as possibly lying below the full-rank oracle because it excludes \epsilon_sub; this should be marked clearly in Fig. 2 so readers do not interpret it as a violation of optimality.
  3. [Section III-D, Prop. 5] Proposition 5's statement that the cross terms vanish "without aliasing" is an approximation when N/M is not an integer; please state the asymptotic condition explicitly.
  4. [Section III-E, Eq. (38)] Eq. (38) is asserted without derivation; a sketch of the Fisher-information computation would help readers verify the prefactor and the claimed T^{-1} scaling.
  5. [Section II-C, Eq. (18)] The notation L_t in Eq. (18) is later specialized to M without comment; please make the specialization explicit.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the M>=r threshold is a rank argument under the explicitly stated reduced-rank Clarke model, and the truncation approximation is disclosed.

full rationale

The central feasibility threshold is derived from the stated observation model y_t = G z_t + n_t with G = S_O B(ϑ) ∈ C^{M×r}. Proposition 4's impossibility regime is the elementary rank bound dim null(G) ≥ r−M (Corollary 3, Eq. 32), and the achievability regime is full column rank of a generic DFT submatrix plus Corollary 2's high-SNR inversion. These conclusions are mathematical consequences of the model, not restatements of the data. The modal subspace itself is justified from the Clarke spectrum: K(ϑ) in Eq. (12) is the set of DFT frequencies inside the band-limited support [−2πdϑ, 2πdϑ], yielding r ≈ 2ϑ+1 by counting grid points. Eq. (15) intentionally models the channel as exactly in that subspace, and the paper is explicit in Remark 1 that finite-N leakage gives nonzero ε_sub and O(1/N) is an idealized asymptotic rate. The numerical evaluation is similarly transparent: the reduced-rank oracle 'reports only ε_est/tr(Σ_true)' and 'excludes the subspace truncation term ε_sub,' so the phase-transition curves are not presented as an empirical test of the approximation-free Clarke channel. The separate full-rank oracle and the [18] ideal bound are external comparisons, not load-bearing self-citations; [18] is used only for the standard rank-M eigenvalue truncation benchmark. No fitted parameter is renamed as a prediction; the modal powers and noise variance are learned (Prop. 6), but the M≥r threshold is independent of their values. The possible issue that Prop. 3's trace formula (Eq. 28) requires G^H G and P to commute in the general non-diagonal eigenbasis is a mathematical correctness concern, not a circularity, and does not affect the rank-deficiency lower bound. Overall, the derivation chain is self-contained under its stated assumptions.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central results rest on the assumed exact low-rank modal structure of the Clarke channel. The phase transition M=r is a direct consequence of this assumption. The main free parameters are the learned aperture theta, modal powers P, and noise variance sigma^2, along with the chosen modal order r. No new physical entities are introduced.

free parameters (4)
  • theta (effective normalized aperture) = learned via EM, not reported in the paper
    Controls the active modal set K(theta) and the modal dimension r. The entire feasibility threshold depends on it. In the numerics, theta is set to the true W=2.
  • Modal powers p_k = learned via EM, not reported in the paper
    Diagonal entries of the modal covariance P. They appear in the oracle MSE formulas (Prop. 3, 4) and in the learning-error expressions (Prop. 6).
  • Noise variance sigma^2 = SNR = 1/sigma^2 in the numerics
    Affects estimation and learning errors; it is assumed known or learned.
  • Modal dimension r = r=5 for W=2 in the numerics
    Model order; chosen from the -20 dB eigenvalue threshold in Fig. 1(c) and the approximation r=2W+1. The sharp threshold and all results depend on this choice.
assumptions (5)
  • domain assumption Clarke isotropic scattering with rectangular spatial power spectrum f_theta(omega) = 1/(2d_theta) for |omega| <= 2*pi*d_theta
    Used in Eq. (8)-(10) and Prop. 2 to establish the bandlimited spectrum and low-rank structure. If scattering is not isotropic, the modal structure deviates, as acknowledged in Future Work.
  • standard math Szego's theorem for Toeplitz matrices diagonalized asymptotically by the DFT
    Invoked after Eq. (10) and in Prop. 2 to justify the DFT-basis representation and the O(1/N) truncation rate.
  • ad hoc to paper The channel is exactly modeled as g_t = B(theta) z_t with z_t ~ CN(0,P) and P diagonal (Eq. 15)
    Core low-rank model. All subsequent bounds (Prop. 3-5) assume this model, and the numerical experiments generate data from it. The paper notes in Remark 1 that if the true covariance is not in col(B), then the truncation error is nonzero.
  • ad hoc to paper Truncation error negligible (epsilon_sub approximately 0) when deriving oracle MMSE and phase transition
    Prop. 3 and Prop. 4 assume epsilon_sub is approximately zero, so the impossibility result applies to the reduced-rank oracle, not to the full Clarke channel with leakage.
  • standard math Asymptotic regime N large; eigenvalues of G^H G under uniform selection converge to M/N
    Used in Prop. 5 and Prop. 2. At finite N there is leakage and edge effects.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CSI Reconstruction in Fluid Antenna Systems Without Spatial Covariance Priors." pith.science (2026). https://pith.science/paper/O73UPWTI

@misc{pith2026260809499,
  author       = {Pith},
  title        = {Pith review of: CSI Reconstruction in Fluid Antenna Systems Without Spatial Covariance Priors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/O73UPWTI}},
  note         = {Machine review of arXiv:2608.09499}
}
abstract

Fluid antenna systems (FASs) exploit many candidate ports for spatial diversity, but hardware constraints allow channel observations at only a few active ports. Whether full-port CSI can be recovered without pre-acquired channel statistics remains open. Under the Clarke isotropic scattering model, we show that the channel lies in a low-dimensional spatial modal subspace determined by the scattering environment rather than the total port count. Consequently, recovery becomes feasible when the number of observed ports reaches the modal dimension (i.e., $M\geq r$), even when $M\ll N$. We further establish a sharp feasibility threshold: reliable recovery is impossible below this dimension regardless of SNR, whereas accuracy improves with additional observations above it. By decomposing the recovery error into modal truncation, estimation, and learning components, we derive explicit tradeoffs among RF chains, pilot overhead, transmit power, and training data. These results enable scalable prior-free full-port CSI recovery with few active ports.

Figures

Figures reproduced from arXiv: 2608.09499 by the authors.

Figure 1
Figure 1. Eigenvalue spectra and effective ranks of the Clarke [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. NMSE versus SNR with M = 10 uniformly selected ports. System parameters, N = 64, W = 2, and r = 5. E. Learning Error and Snapshot Requirements The learning term quantifies the cost of estimating, rather than knowing, the hyperparameters. Proposition 6 (Snapshot requirements). Consider EM estima￾tion of the r modal powers {pk} and noise variance σ 2 from T independent snapshots, each with M noisy observations. (i) Id… view at source ↗
Figure 3
Figure 3. NMSE versus the number of observed ports [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 26 canonical work pages

  1. [1]

    Fluid antenna systems,

    K. K. Wong et al. , “Fluid antenna systems,” IEEE Transactions on Wireless Communications, vol. 20, no. 3, pp. 1950–1962, Mar. 2021

    work page 1950

  2. [2]

    Closed-form expressions for spatial correlation pa- rameters for performance analysis of fluid antenna systems,

    K. K. Wong et al. , “Closed-form expressions for spatial correlation pa- rameters for performance analysis of fluid antenna systems, ” Electronics Letters, vol. 58, no. 11, pp. 454–457, Apr. 2022

    work page 2022

  3. [3]

    W. C. Jakes and D. C. Cox, Microwave Mobile Communications . Hoboken, NJ, USA: Wiley, 1974

    work page 1974

  4. [4]

    When is clarke’s approximat ion valid?

    T. W. C. Brown and U. M. Ekpe, “When is clarke’s approximat ion valid?” IEEE Antennas and Propagation Magazine , vol. 52, no. 3, pp. 171–181, Jun. 2010

    work page 2010

  5. [5]

    Jointly correlated dual-side fluid antenna system,

    Z. Zhang et al., “Jointly correlated dual-side fluid antenna system,” IEEE Wireless Communications Letters , 2026, early Access

    work page 2026

  6. [6]

    Cram´ er–rao bounds for activity detection in con- ventional and fluid antenna systems,

    Z. Zhang et al. , “Cram´ er–rao bounds for activity detection in con- ventional and fluid antenna systems,” IEEE Wireless Communications Letters, vol. 15, pp. 3059–3063, 2026

    work page 2026

  7. [7]

    On fundamental limits of slow-fluid antenna multiple access for unsourced random access,

    Z. Zhang et al. , “On fundamental limits of slow-fluid antenna multiple access for unsourced random access,” IEEE Wireless Communications Letters, vol. 14, no. 11, pp. 3455–3459, Nov. 2025

    work page 2025

  8. [8]

    On fundamental limits for fluid antenna-assisted integrated sensing and communications for unsourced rando m access,

    Z. Zhang et al. , “On fundamental limits for fluid antenna-assisted integrated sensing and communications for unsourced rando m access,” IEEE Journal on Selected Areas in Communications , vol. 44, pp. 136– 149, 2026

    work page 2026

Show all 26 references
  1. [9]

    Fluid antenna systems enabling 6g: Principles, applica- tions, and research directions,

    T. Wu et al. , “Fluid antenna systems enabling 6g: Principles, applica- tions, and research directions,” IEEE Wireless Communications , 2025, early Access

    2025

  2. [10]

    A contemporary survey on fluid antenna systems: Fundamentals and networking perspectives,

    H. Hong et al. , “A contemporary survey on fluid antenna systems: Fundamentals and networking perspectives,” IEEE Transactions on Network Science and Engineering , vol. 13, pp. 2305–2328, 2026

    2026

  3. [11]

    Scalable fluid antenna systems: A new paradigm for array signal processing,

    T. Wu et al. , “Scalable fluid antenna systems: A new paradigm for array signal processing,” IEEE Journal of Selected Topics in Signal Processing, vol. 20, no. 3, pp. 389–406, Apr. 2026

    2026

  4. [12]

    Finite-blocklength fluid antenna systems with spatial block-correlation channel model,

    Z. Zhang et al. , “Finite-blocklength fluid antenna systems with spatial block-correlation channel model,” IEEE Wireless Communications Let- ters, vol. 15, pp. 1911–1915, 2026

    1911

  5. [13]

    Finite-aperture fluid antenna array design: Analysis and algorithm,

    Z. Zhang et al. , “Finite-aperture fluid antenna array design: Analysis and algorithm,” IEEE Wireless Communications Letters , vol. 15, pp. 3199–3203, 2026

    2026

  6. [14]

    Finite-aperture planar fluid antenna array,

    Z. Zhang et al. , “Finite-aperture planar fluid antenna array,” arXiv preprint, 2026. [Online]. Available: https://arxiv.org/abs/2605 .22040

    2026

  7. [15]

    Fluid antenna-enhanced flexible beamforming,

    J. Xu et al., “Fluid antenna-enhanced flexible beamforming,” in Proceed- ings of the IEEE Wireless Communications and Networking Con ference W orkshops (WCNCW), Kuala Lumpur, Malaysia, 2026, pp. 1–6

    2026

  8. [16]

    Peak sidelobe suppression in planar fluid antenna array,

    H. Liang et al. , “Peak sidelobe suppression in planar fluid antenna array,” arXiv preprint , 2026. [Online]. Available: https://arxiv.org/abs/2606.31149

    2026 arXiv

  9. [17]

    REMAA: Reconfigurable pixel antenna-based electronic movable-antenna arrays for multiuser communications,

    K. Chen et al., “REMAA: Reconfigurable pixel antenna-based electronic movable-antenna arrays for multiuser communications,” IEEE Transac- tions on Communications, vol. 73, no. 11, pp. 12 913–12 928, Nov. 2025

    2025

  10. [18]

    Beyond covariance: Generative spatial correlation modeling and channel interpolation for fluid antenna system s,

    Z. Zhang et al. , “Beyond covariance: Generative spatial correlation modeling and channel interpolation for fluid antenna system s,” arXiv preprint, 2026. [Online]. Available: https://arxiv.org/abs/2604 .16639

    2026

  11. [19]

    Neural networks-enabled channel reconstruction for fluid antenna systems: A data-driven approach,

    H. Liang et al. , “Neural networks-enabled channel reconstruction for fluid antenna systems: A data-driven approach,” in Proceedings of the IEEE Wireless Communications and Networking Conference W o rkshops (WCNCW), Kuala Lumpur, Malaysia, 2026, pp. 1–6

    2026

  12. [20]

    Geometry-structured channel reconstruction for conventional and fluid antenna systems: Bayesian inference and fundamental limits,

    Z. Zhang et al. , “Geometry-structured channel reconstruction for conventional and fluid antenna systems: Bayesian inference and fundamental limits,” arXiv preprint , 2026. [Online]. Available: https://arxiv.org/abs/2606.04001

    2026 arXiv

  13. [21]

    Sparse bayesian learning-based channel estimation for fluid antenna systems,

    B. Xu et al. , “Sparse bayesian learning-based channel estimation for fluid antenna systems,” IEEE Wireless Communications Letters , vol. 14, no. 2, pp. 325–329, Feb. 2025

    2025

  14. [22]

    Learned-approximate message passing under karhunen– lo` eve modeling for fluid antenna systems,

    Y . Wu et al. , “Learned-approximate message passing under karhunen– lo` eve modeling for fluid antenna systems,” IEEE Wireless Communica- tions Letters , vol. 15, pp. 2719–2723, 2026

    2026

  15. [23]

    Channel estimation for fas-assisted multiuser mmwave systems,

    H. Xu et al. , “Channel estimation for fas-assisted multiuser mmwave systems,” IEEE Communications Letters , vol. 28, no. 3, pp. 632–636, Mar. 2024

    2024

  16. [24]

    Joint activity detection and channel estimation for fluid antenna system exploiting geographical and angular in formation,

    Z. Zhang et al. , “Joint activity detection and channel estimation for fluid antenna system exploiting geographical and angular in formation,” IEEE Journal of Selected Topics in Signal Processing , vol. 20, no. 3, pp. 354–370, Apr. 2026

    2026

  17. [25]

    A new spatial block-correlation model for fluid antenna systems,

    P . Ram´ ırez-Espinosa et al. , “A new spatial block-correlation model for fluid antenna systems,” IEEE Transactions on Wireless Communications, vol. 23, no. 11, pp. 15 829–15 843, Nov. 2024

    2024

  18. [26]

    Toeplitz and circulant matrices: A review,

    R. M. Gray, “Toeplitz and circulant matrices: A review, ” F oundations and Trends in Communications and Information Theory , vol. 2, no. 3, pp. 155–239, 2006

    2006

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.