Pith. sign in

REVIEW 3 major objections 6 minor 30 references

EM-Guided Graph Learning for Fluid Antenna Beamforming under Current-Domain Constraints

T0 review · 3 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read A graph network predicts fluid-antenna port layouts in one forward pass, matching a 200-step EM-aware optimizer closely in rate and current loading while cutting configuration latency from about 90 ms to 3 ms.

desk verdict A coherent simulation-level method paper with a genuinely new architecture and honest evaluation; the hybrid EM model is the main caveat, but it deserves a serious referee. read the letter →

arxiv 2608.13254 v1 pith:EYQ2AHS2 submitted 2026-08-13 cs.IT math.IT

Yuanhui Wu , Hao Jiang , Zaichen Zhang This is my paper · ORCID
classification cs.ITmath.IT
keywords fluidantennaarraymutualcouplinggraphneuralnetworkcurrent-domainbeamformingelectromagneticfeasibilityportplacementmulti-userdownlinkconfigurationlatency
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

EMG-CC, a graph neural network with mutual-impedance-aware edge features, learns to place the movable ports of a fluid antenna array in a single forward pass rather than per-channel iterative search. The paper's claim is that, under a common post-projection current-domain evaluation, this one-pass generator lands close to a 200-step projected optimizer (mean 24.203 vs 24.871 bit/s/Hz) while cutting median configuration latency from about 90 ms to about 3 ms, and that it improves current-loading uniformity over a rate-only learner (mean concentration 2.708 vs 3.088) at a small rate cost. The training objective couples communication rate with mutual coupling, radiation conditioning, accepted-power, current, voltage, and current-concentration penalties, so the network is explicitly trained for a prescribed rate-current-loading operating point. The reason to care is that real-time electromagnetic-aware reconfiguration of compact arrays is otherwise expensive, and this provides a fast approximation whose feasibility metrics are computed by the same circuit model for every method.

What carries the argument

The central object is the mutual-impedance-aware graph coordinate generator: a convolutional encoder maps channel samples on an observation grid to a latent field, anchor-based initialization produces bounded starting coordinates, and S message-passing layers refine them using edge features that include relative position and the complex mutual impedance Z_mn(P). The companion evaluator is EM-RZF, a regularized zero-forcing rule with a scalar κ that scales raw currents to satisfy accepted-power, total-current, and source-voltage budgets, making every layout feasible by construction. The loss couples rate with soft penalties on spacing, radiation conditioning, coupling energy, raw loads, and current concentration, so the graph is trained for a chosen rate-current-loading operating point while the hard feasibility rule is applied identically at test time.

What would settle it

Build or simulate a 4λ×2λ fluid antenna array with 32 ports whose mutual impedances are obtained by full-wave simulation or measurement, and compare predicted versus measured accepted power, radiated power, port currents, and source voltages under the same EM-RZF scaling rule; if the measured radiation efficiency or current distribution deviates from the induced-EMF prediction by more than the margin implied by the reported feasibility rates, the claimed tradeoff does not transfer to that hardware.

Watch

Extended reading notes

Core claim

The central claim is that EMG-CC, trained end-to-end with the current-domain EM-RZF evaluator, predicts continuous fluid antenna array port coordinates that, after the same spacing projection and feasibility scaling applied to all methods, provide a controllable tradeoff among downlink sum rate, current loading, and configuration latency. On 2000 paired test channels with a 4λ×2λ aperture, 32 ports, 8 users, and the induced-EMF half-wave-dipole model, EMG-CC achieves mean rate 24.203 bit/s/Hz and mean current concentration 2.708 with median latency 2.79 ms, versus 24.871, 2.613, and 89.86 ms for the per-realization projected optimizer EM-PO; it also reduces mean current concentration relative to Rate-only (3.088) at a 0.08 bit/s/Hz mean-rate cost. The authors present EMG-CC as a low-latency approximation to iterative EM placement, not a replacement for its final optimization quality, and they report that accepted power, current, and voltage each set the final scaling in 50.7%, 13.9%, and 35.4% of test cases.

Load-bearing premise

The tradeoff rests on the assumption that the induced-EMF model of identical parallel center-fed half-wave dipoles in free space captures the mutual coupling, radiation conditioning, and source-voltage behavior of a real compact fluid antenna; the paper itself flags that device-level validation with calibrated element responses, port quantization, and actuator dynamics is still needed.

Editorial extensions

If this is right

  • With the free-space dipole model as test bed, one-pass placement yields mean rate within about 0.67 bit/s/Hz of the 200-step iterative reference while reducing median layout latency from roughly 90 ms to 3 ms.
  • The current-concentration penalty moves the operating point: mean current concentration drops from 3.088 (Rate-only) to 2.708 (EMG-CC) and the 95th percentile from 4.914 to 4.083, at a mean-rate loss of 0.08 bit/s/Hz, demonstrating that the rate-current-loading tradeoff is controllable.
  • Circuit information during training matters: No-EM-training, which omits mutual-impedance and circuit-aware evaluation, trails EMG-CC by about 1.15 bit/s/Hz mean rate under the same evaluator.
  • All compared methods, including fixed and heuristic baselines, are judged by the same post-projection current-domain evaluator, so the reported ordering reflects placement quality rather than differing power conventions.
  • EMG-CC satisfies the spacing constraint before projection, and the three operating budgets all bind in different fractions of test cases, so the trained network produces layouts that are not trivially dominated by one constraint.

Reading between the lines

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

  • If the free-space induced-EMF model transfers to hardware, the same graph architecture can be retrained on measured or full-wave impedance matrices without structural change, because mutual impedance already enters as an edge feature.
  • The current-concentration metric CI could serve as a cheap proxy for per-port power-amplifier or thermal stress; a hardware study could test whether lowering CI indeed equalizes PA temperatures or lifetimes, which the paper does not claim.
  • The two-port radiation-resistance screening rule for d_min could be reused as a fast pre-check in other fluid antenna array placement algorithms, including non-learned ones, to avoid infeasible dense configurations early.
  • A practical extension would be a two-stage procedure: pretrain with the cheap dipole model, then fine-tune with calibrated element responses and actuator dynamics; the paper stops at identifying these as required for device-level validation.
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

3 major / 6 minor

Summary. The paper proposes EMG-CC, a graph-neural-network layout generator that maps sampled channel fields to continuous port coordinates for a fluid antenna array, with training and evaluation under a current-domain multiport model. The model combines a geometric multi-user channel model (Eq. (6)) with induced-EMF mutual impedances (Eqs. (8)-(12)), uses a circuit-weighted RZF precoder with hard scaling (Eqs. (23)-(26)), and adds penalties for coupling, radiation conditioning, loading, and current concentration. On 2000 paired channels, EMG-CC is compared against an iterative EM-aware optimizer (EM-PO), a rate-only learner, a no-EM learner, and several heuristics under a common post-projection evaluator. The main results are that EMG-CC approaches EM-PO in sum rate (24.20 vs 24.87 bit/s/Hz) and current concentration (2.71 vs 2.61) at a median configuration latency of 2.79 ms instead of 89.86 ms, and that it improves current-loading uniformity over Rate-only with a small mean-rate loss. The paper explicitly limits its scope to a normalized far-field isotropic-element model and states that device-level validation requires calibrated element responses and actuator dynamics.

Significance. If the results hold, the paper demonstrates a useful amortization: a one-pass layout generator can approximate an iterative EM-aware placement at a fraction of the latency while preserving most of the rate and current-loading benefits. The evaluation design is a strength: all nine methods are subjected to the same spacing projection and the same current-domain EM-RZF evaluator, so the reported differences cannot be attributed to inconsistent power conventions. The ablations (Rate-only and No-EM-training) cleanly separate the value of circuit information and of current-concentration control. The paper also states its limitations openly, including the normalized isotropic-element channel model and the absence of full-wave validation. No code or data are provided, and the central physical assumption is that the channel can be modeled with g_m=1 while feasibility is judged through coupled-dipole impedances; this assumption is acknowledged but not tested. As a result, the practical significance of the claimed rate/current-loading/latency tradeoff depends on an additional validation that is not currently present.

major comments (3)
  1. [II-B and II-C (Eqs. (5)-(12))] The channel model in Eq. (6) assumes g_m=1, so H(P) depends on the layout only through the phase e^{j2π p_m^T u}, whereas Section II-C uses the induced-EMF mutual-impedance matrix Z(P) of parallel half-wave dipoles. With d_min=0.125λ, mutual coupling substantially alters embedded element patterns, so the channel that generates the reported rates cannot be radiated by the same dipole array whose impedance matrix supplies the circuit constraints. The conclusion acknowledges this ('normalized far-field isotropic-element model'), but the central comparative claims (EMG-CC approaches EM-PO, improves current loading over Rate-only) are established only inside this hybrid model. I request either an additional simulation in which the element response g_m is derived from the same coupled-dipole model (e.g., active element patterns), or a clear re-framing of the claims as a proof-of-concept under a decoupled channel/circuit model. Without one of these, the transfer of the rate/current-loading tradeoff even to the simulated dipole array remains unsupported.
  2. [IV (validation and Table I)] The abstract and conclusion state that EMG-CC provides a 'controllable tradeoff' among rate, current loading, and latency, and Section III-D says η_cc and Γ_cc define the operating point. However, the reported results contain only a single trained EMG-CC operating point; the validation in Section IV-a is limited to three candidate pairs and the selected one is used in all figures. No curve or table shows how rate and current concentration change as (Γ_cc, η_cc) or other loss weights vary. A single point compared with Rate-only and EM-PO demonstrates a tradeoff across methods, but it does not demonstrate controllability by the stated parameters. Please add a sweep over the operating-point parameters, or explicitly state that the tradeoff is only across methods rather than controlled by the network's hyperparameters.
  3. [III-E and Section IV-a] The common evaluator is described as applying a 'deterministic collision-resolution projection' to every layout, but the projection algorithm is never specified, and EM-PO is described only as 200 Adam steps from uniform anchors. Since the paper's central claim is a comparison under a common post-projection feasibility rule, the projection details are needed for reproducibility and to assess whether the projection itself could change the relative ordering of methods. Please specify the projection algorithm (or provide code) and state whether EM-PO uses any convergence check or multi-restart.
minor comments (6)
  1. [Section I, second paragraph] The phrase 'a broad diversity diversity including random fading' contains a duplicated word and should read 'a broad diversity including random fading'.
  2. [Section IV, setup paragraph] The text refers to 'activation frequencies reported in Section IV-0a'; the binding frequencies actually appear in Section IV-e (Discussion).
  3. [Section III-D, Eqs. (31)-(32)] The quantity eCI is used in Eq. (31) but defined only after Eq. (32); please move the definition before its first use.
  4. [Section III-F] The evaluation-metrics paragraph promises to report accepted-power, total-current, and source-voltage feasibility rates together with minimum spacing and minimum eigenvalue of R_rad, but the results section reports only the binding activation frequencies. Please add the promised metrics or remove the sentence.
  5. [Section IV-a (EM-PO description)] The label 'optimization-quality reference' should be qualified as 'a fixed-budget iterative reference', since no convergence criterion or multi-restart is specified.
  6. [Figure 4 caption] The selection criterion 'joint median distance over rate and current concentration across methods' is unclear; please define the distance used.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: all methods share the same EM model, feasibility projection, and current-domain evaluator, and the learned generator's held-out performance is measured against independent iterative and heuristic baselines.

full rationale

The paper's derivation chain is not circular. The induced-EMF multiport model from [15] supplies the common mutual-impedance matrices Z(P), the radiation/spacing screening rule, and the budgets used by the EM-RZF evaluator; this model is a shared input applied identically to every method, not a quantity fitted to the proposed method's outputs. EMG-CC is trained with the same current-domain objective that EM-PO optimizes per channel, so the comparison 'EMG-CC approaches EM-PO' is an amortized-optimization evaluation on 2000 held-out realizations, not an identity: the reported rate, current-concentration, and latency statistics are measured after a common post-projection evaluator and are not equal to the training loss by construction. The current-concentration penalty does explicitly push EMG-CC toward lower CI, but the paper transparently labels this as an implementation-oriented regularizer defining an operating point and reports the paired rate tradeoff rather than presenting the CI reduction as an emergent prediction. The acknowledged limitations - the normalized far-field isotropic-element channel model and the absence of device-level calibration - are correctness risks about transfer to hardware, not circularity in the derivation. No equation reduces to another by definition, and no fitted parameter is renamed as a prediction. Self-citations appear frequently, but the load-bearing model [15] is parameter-free with stated assumptions and does not include the target comparison as an input, so it does not raise the circularity score.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

No new physical particles, forces, conserved quantities, or dimensions are postulated. The novel objects (EMG-CC, CI, U) are software and mathematical constructs, not physical entities; they do not require independent falsifiability outside the paper.

free parameters (6)
  • Current-domain budgets (P_acc, I_max, V_max^2) = P_acc=1, I_max=0.03621, V_max^2=443.4
    Hand-chosen normalized operating limits; they define the feasibility envelope and the scaling in Eq. (26).
  • Spacing screening margin epsilon_rad = 9 ohm
    Chosen margin in Eq. (12); it sets d_min=0.125273 lambda and therefore the minimum spacing constraint active in training and projection.
  • EM-RZF constants (alpha, epsilon, zeta) = (0.1, 0.05, 0.02)
    Hand-picked diagonal loading, voltage weighting, and RZF regularization; the authors state EM-RZF is not a claimed precoding optimum.
  • Loss weights and smoothing constants = (eta_em, eta_s, eta_e, eta_cc, eta_p, eta_i, eta_v, tau_s, tau_e, tau_cc…
    Manual and validation-selected weights that define the rate-current-loading operating point of Eq. (29).
  • Current-concentration target Gamma_cc (with eta_cc) = Gamma_cc=3, eta_cc=0.1
    Selected by validation among (4,0.1), (3,0.1), and (3,0.3); this is the explicit knob for the claimed controllable tradeoff.
  • Anchor offset scale rho and anchor-cell size c = not reported
    Eq. (20) uses these values to bound the initial coordinate offsets, but no numerical values are given in Section IV, making initialization not fully reproducible.
assumptions (5)
  • domain assumption The induced-EMF model for identical parallel center-fed thin half-wave dipoles in free space produces the mutual-impedance matrix Z(P) and radiation matrices used in Eqs. (8)-(12).
    This is the physical model for all EM feasibility metrics; if inaccurate, the tradeoff claim does not transfer to hardware. The authors flag this in the conclusion.
  • domain assumption The geometric multipath channel with isotropic element factor g_m=1 and fixed number of paths and users is representative for evaluating FAA placement.
    Section II-B and Section IV setup; no full-wave element-pattern variation, substrate, or hardware impairments are modeled.
  • domain assumption A single forward pass of the trained graph network generalizes across the channel distribution well enough to approximate EM-PO.
    This is the amortized-inference premise tested by Table I; it has no analytical guarantee and is only demonstrated on 2000 test realizations.
  • domain assumption The EM-RZF rule of Eq. (23) with the scale factor of Eq. (26) is an adequate common precoder for comparing layout generators.
    All methods use the same rule, so layout quality is isolated, but the rule itself is a heuristic and affects every reported number.
  • standard math Standard multiport circuit identities for accepted power, radiated power, and source voltage hold for the FAA current-domain model.
    Eqs. (14)-(16) rely on i^H R_acc i, i^H R_rad i, and i^H Q_v i from multiport array theory and [15].

how reviews work

0 comments
Cite this review

Pith. "Pith review of EM-Guided Graph Learning for Fluid Antenna Beamforming under Current-Domain Constraints." pith.science (2026). https://pith.science/paper/EYQ2AHS2

@misc{pith2026260813254,
  author       = {Pith},
  title        = {Pith review of: EM-Guided Graph Learning for Fluid Antenna Beamforming under Current-Domain Constraints},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EYQ2AHS2}},
  note         = {Machine review of arXiv:2608.13254}
}
read the original abstract

Fluid antenna arrays (FAAs) reconfigure a finite set of radiating ports within a prescribed aperture. In compact apertures, however, channel-driven placement may cluster ports, strengthen mutual coupling, degrade radiation conditioning, increase source-voltage demand, and produce uneven current loading. This paper studies downlink multi-user beamforming with jointly optimized port placement and current-domain transmission. An electromagnetic-guided graph network predicts port layouts from channel observations and refines them using geometric and mutual-impedance information. The training objective jointly considers communication performance and electromagnetic feasibility, while a common evaluation procedure is applied to all methods. The results show that, under a common feasibility standard, the proposed method provides a controllable tradeoff among communication rate, current loading, and configuration latency.

Figures

Figures reproduced from arXiv: 2608.13254 by the authors.

Figure 1
Figure 1. Distribution of the EM-aware downlink sum rate at SNR [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 3
Figure 3. Mean EM-aware downlink sum rate versus SNR over [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Representative active-port layouts for one common test-channel realiza [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 22 canonical work pages

  1. [15]

    Electromagnetic-aware fluid antenna array,

    Z. Zhang,et al., “Electromagnetic-aware fluid antenna array,”arXiv preprint, 2026. [Online]. Available: https://arxiv.org/abs/2607.21375

    arXiv 2026

  2. [23]

    Graph Neural Network Enabled Fluid Antenna Systems: A Two-Stage Approach

    C. He,et al., “Graph neural network enabled fluid antenna systems: A two-stage approach,”arXiv preprint, 2025. [Online]. Available: https: //arxiv.org/abs/2502.03922

    work page Pith review arXiv 2025

  3. [1]

    Fluid antenna system for 6G: When Bruce Lee inspires wireless communications,

    K.-K. Wong,et al., “Fluid antenna system for 6G: When Bruce Lee inspires wireless communications,”Electron. Lett., vol. 56, no. 24, pp. 1288–1290, Nov. 2020, doi: 10.1049/el.2020.2788

    arXiv 2020

  4. [2]

    Fluid antenna systems,

    K.-K. Wong,et al., “Fluid antenna systems,”IEEE Trans. Wire- less Commun., vol. 20, no. 3, pp. 1950–1962, Mar. 2021, doi: 10.1109/TWC.2020.3037595

    arXiv 1950

  5. [3]

    Finite-blocklength fluid antenna systems,

    Z. Zhang,et al., “Finite-blocklength fluid antenna systems,” IEEE Trans. Wireless Commun., early access, 2026, doi: 10.1109/TWC.2026.3723456

    work page arXiv 2026

  6. [4]

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

    Z. Zhang, K.-K. Wong, D. Morales-Jimenez, H. Jiang, P. Ramirez- Espinosa, C.-B. Chae, and C. Masouros, “Finite-blocklength fluid an- tenna systems with spatial block-correlation channel model,”IEEE Wireless Commun. Lett., vol. 15, pp. 1911–1915, 2026

    work page 1911

  7. [5]

    Cram´er–Rao bounds for activity detection in conventional and fluid antenna systems,

    Z. Zhang, K.-K. Wong, H. Jiang, C. Masouros, and C.-B. Chae, “Cram´er–Rao bounds for activity detection in conventional and fluid antenna systems,”IEEE Wireless Commun. Lett., vol. 15, pp. 3059– 3063, 2026

    work page 2026

  8. [6]

    Performance limits of fluid antenna systems,

    K.-K. Wong,et al., “Performance limits of fluid antenna systems,” IEEE Commun. Lett., vol. 24, no. 11, pp. 2469–2472, Nov. 2020, doi: 10.1109/LCOMM.2020.3006554

    arXiv 2020

Show all 30 references
  1. [7]

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

    Z. Zhang,et al., “On fundamental limits for fluid antenna-assisted integrated sensing and communications for unsourced random access,” IEEE J. Sel. Areas Commun., vol. 44, pp. 136–149, 2026

    2026

  2. [8]

    Bruce Lee-inspired fluid antenna system: Six research topics and the potentials for 6G,

    K.-K. Wong,et al., “Bruce Lee-inspired fluid antenna system: Six research topics and the potentials for 6G,”Front. Commun. Netw., vol. 3, Mar. 2022

    2022

  3. [9]

    A tutorial on fluid antenna system for 6G net- works: Encompassing communication theory, optimization methods and hardware designs,

    W. K. New,et al., “A tutorial on fluid antenna system for 6G net- works: Encompassing communication theory, optimization methods and hardware designs,”IEEE Commun. Surveys Tuts., vol. 27, no. 4, pp. 2325–2377, 2025

    2025

  4. [10]

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

    Z. Zhang,et al., “Joint activity detection and channel estimation for fluid antenna system exploiting geographical and angular information,” IEEE J. Sel. Topics Signal Process., vol. 20, no. 3, pp. 354–370, 2026

    2026

  5. [11]

    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 Commun. Lett., vol. 14, no. 11, pp. 3455–3459, 2025

    2025

  6. [12]

    Jointly correlated dual-side fluid antenna system,

    Z. Zhang, Y . Wu, K.-K. Wong, H. Jiang, and A. Li, “Jointly correlated dual-side fluid antenna system,”IEEE Wireless Commun. Lett., early access, 2026, doi: 10.1109/LWC.2026.3710907

    2026

  7. [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 Commun. Lett., vol. 15, pp. 3199–3203, 2026

    2026

  8. [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 arXiv

  9. [16]

    Antenna positioning and beamforming design for fluid- antenna enabled multi-user downlink communications,

    H. Qin,et al., “Antenna positioning and beamforming design for fluid- antenna enabled multi-user downlink communications,”arXiv preprint,

  10. [17]

    Sum-rate maximization for fluid antenna enabled multiuser communications,

    Z. Cheng,et al., “Sum-rate maximization for fluid antenna enabled multiuser communications,”IEEE Commun. Lett., vol. 28, no. 5, pp. 1206–1210, May 2024

    2024

  11. [18]

    Near-field beamforming and port selection for fluid antenna systems,

    S. Chen,et al., “Near-field beamforming and port selection for fluid antenna systems,”IEEE Commun. Lett., vol. 30, pp. 2248–2252, 2026

    2026

  12. [19]

    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 J. Sel. Topics Signal Process., vol. 20, no. 3, pp. 389–406, Apr. 2026

    2026

  13. [20]

    Fluid antenna-enhanced flexible beamforming,

    J. Xu,et al., “Fluid antenna-enhanced flexible beamforming,” inProc. IEEE Wireless Commun. Netw. Conf. Workshops (WCNCW), 2026, pp. 1–6

    2026

  14. [21]

    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

  15. [22]

    Learned blockwise port activation for real time beamform- ing in fluid antenna arrays,

    Y . Wu,et al., “Learned blockwise port activation for real time beamform- ing in fluid antenna arrays,”arXiv preprint, 2026. [Online]. Available: https://arxiv.org/abs/2607.25365

    2026 arXiv

  16. [24]

    Toward practical fluid antenna systems: Co-optimizing hardware and software for port selection and beamforming,

    S. Xu,et al., “Toward practical fluid antenna systems: Co-optimizing hardware and software for port selection and beamforming,”arXiv preprint, 2025. [Online]. Available: https://arxiv.org/abs/2507.14035

    2025 arXiv

  17. [25]

    LLM-based port selection and beamforming for mul- tiuser MISO with fluid antenna systems,

    W. Guo,et al., “LLM-based port selection and beamforming for mul- tiuser MISO with fluid antenna systems,”IEEE Trans. Netw. Sci. Eng., vol. 13, pp. 4026–4042, 2026

    2026

  18. [26]

    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 Commun. Lett., vol. 15, pp. 2719–2723, 2026

    2026

  19. [27]

    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,” inProc. IEEE Wireless Commun. Netw. Conf. Workshops (WCNCW), 2026, pp. 1–6

    2026

  20. [28]

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

    Z. Zhang,et al., “Beyond covariance: Generative spatial correlation modeling and channel interpolation for fluid antenna systems,”arXiv preprint, 2026. [Online]. Available: https://arxiv.org/abs/2604.16639

    2026 arXiv

  21. [29]

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

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

    2026 arXiv

  22. [2023]

    Available: https://arxiv.org/abs/2311.03046

    [Online]. Available: https://arxiv.org/abs/2311.03046

    arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.