Pith. sign in

REVIEW 4 major objections 5 minor 46 references

Recovery of UAV Swarm-enabled Collaborative Beamforming in Low-altitude Wireless Networks under Wind Field Disturbances

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read The paper claims that a memory-augmented reinforcement learning policy can restore the directivity and sidelobe performance of a wind-displaced UAV virtual antenna array by adjusting only the per-UAV excitation current weights in real time.

desk verdict Magnitude-only weights cannot undo wind-induced phase errors, so the paper's recovery claim is physically unsupported despite a clean DRL benchmark. read the letter →

arxiv 2507.08507 v1 pith:YQTPT7WS submitted 2025-07-11 cs.NI

classification cs.NI
keywords UAVswarmcollaborativebeamformingvirtualantennaarraywindfielddisturbancesdeepreinforcementlearningproximalpolicyoptimizationLSTMlow-altitudewirelessnetwork
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

The paper sets out to fix a practical fragility of UAV collaborative beamforming: wind pushes the UAVs off their ideal array positions, deforming the beam pattern and cutting communication range and reliability. Its proposal is to leave the UAVs in the displaced positions and, at each timeslot, have a deep reinforcement learning agent reset each UAV's excitation current weight using an LSTM-augmented PPO policy with an adaptive learning rate. The authors model constant, shear, and turbulent wind fields, cast recovery as a long-term non-convex optimization of directivity versus maximum sidelobe level, and report that their policy restores the degraded beam and outperforms DDPG, DQN, SAC, and standard PPO in simulation. The payoff, if the mechanism holds, is that low-altitude wireless links can stay reliable under realistic weather without fast UAV repositioning or per-wind-profile retraining.

What carries the argument

The load-bearing object is the array factor, $AF_t(\theta_d,\phi_d)=\sum_{i=1}^N \omega^U_{i,t}\exp(jk(x^U_{i,t}\sin\theta_d\cos\phi_d+y^U_{i,t}\sin\theta_d\sin\phi_d+z^U_{i,t}\cos\theta_d))$, which turns UAV positions and excitation current weights into the beam pattern from which directivity $D$ and maximum sidelobe level $M$ are computed. The control mechanism is PPO-LA: an actor-critic policy whose LSTM layers carry temporal context about the wind and whose adaptive learning rate adjusts step sizes to the local loss curvature, with the clipped policy objective of standard PPO keeping updates stable. Its action at each timeslot is the full weight vector, and its reward $r_t=D/M-p_0$ folds the two beam objectives together with a penalty for constraint violations.

What would settle it

Take one fixed wind-distorted UAV geometry, compute the array factor $AF_t(\theta_d,\phi_d)$, and search exhaustively over all weight vectors $\omega^U_{i,t}\in[0,1]$ for the one that best maximizes $D/M$; if the best achievable pattern is indistinguishable from the equal-weight pattern, or if the reported recovery vanishes when the displaced positions are inserted into the array factor, the claimed recovery mechanism is not doing the work.

Watch

Extended reading notes

Core claim

The central claim is that a wind-perturbed UAV virtual antenna array can recover most of its lost beam quality by changing only the per-UAV excitation current weights, in real time, while the UAV positions remain displaced. In the paper's terms, the recovery is achieved by the PPO-LA policy: a proximal policy optimization agent whose LSTM layers capture temporal wind patterns and whose adaptive learning rate stabilizes updates, mapping the observed state to a new weight vector $\omega^U_{i,t}\in[0,1]$ at each timeslot. The reported simulations show stabilized directivity $D\approx 8.25\pm0.15$ dB with 8 UAVs and $D\approx 15$ dB with 16 UAVs (with maximum sidelobe level $M$ near $-3.5$ dB), which the paper reads as evidence that the recovery mechanism works across constant, shear, and turbulent wind fields and scales with swarm size.

Load-bearing premise

The claim rests on the assumption that turning each UAV's transmission power up and down can compensate for its antenna being pushed out of place, even though being displaced corrupts phase relationships that power-only adjustments cannot directly repair.

Editorial extensions

If this is right

  • Deployed UAV swarms could maintain communication through changing wind without fast mechanical repositioning, because only the per-UAV excitation weights need to be updated each timeslot.
  • A single PPO-LA policy is claimed to generalize across constant, shear, and turbulent wind, so operators would not need to retrain a separate controller for each weather condition.
  • Because the reward couples directivity gain with sidelobe suppression, recovered beams would also radiate less power toward unintended directions, which aids interference management and communication security in dense low-altitude networks.
  • In the reported 16-UAV setup the stabilized directivity rises to about 15 dB with the sidelobe level near $-3.5$ dB, suggesting the recovery benefit scales with swarm size, although the observed fluctuation range also widens.

Reading between the lines

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

  • Beyond the paper, the recovery ceiling is set by the physics of amplitude-only weights: since the array factor phase is locked to the displaced positions, no choice of $\omega^U_{i,t}\in[0,1]$ can cancel a displacement of a substantial fraction of the wavelength, so comparing against a phase-compensating baseline would expose how much of the gain comes from sidelobe trade-offs rather than true pha
  • Extension: augment the state with each UAV's displacement relative to its intended array position, since the beam pattern depends on relative phases and this may improve generalization across swarm translations or rotations.
  • Extension: re-run the comparison on recorded turbulence spectra; the paper's turbulent wind is a deterministic sine fluctuation, so the LSTM's temporal advantage may not survive broadband gusts.
Share X Bluesky LinkedIn Reddit HN

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 collaborative beamforming by a UAV swarm acting as a virtual antenna array in low-altitude wireless networks, under constant, shear, and turbulent wind disturbances. Wind-induced positional offsets perturb the array factor, and the authors formulate a long-term optimization problem whose control variables are the real-valued excitation current weights of the UAVs, bounded in [0,1]. They propose PPO-LA, a PPO variant with LSTM layers and an adaptive learning rate, and report simulations for 8- and 16-UAV swarms claiming that PPO-LA recovers degraded directivity and sidelobe performance and outperforms DDPG, DQN, SAC, and standard PPO.

Significance. If the central mechanism were sound, an online DRL-based compensator for wind-perturbed virtual antenna arrays would be practically valuable, and the paper has some commendable design choices: it compares against four DRL baselines, considers two swarm sizes, and attempts to model three wind regimes. However, the load-bearing physical assumption is inconsistent with the stated control variable: real-valued weights in [0,1] cannot apply the phase corrections needed to undo wind-induced positional phase errors. In addition, the evaluation does not include an uncompensated perturbed baseline, and the reported stable directivity is below the initial value, so the claimed recovery is not quantitatively demonstrated. The contribution is therefore not established by the present manuscript.

major comments (4)
  1. [§III-B, §III-E, §IV-B] In Eq. (1), the phase of each array element is determined exclusively by the UAV coordinates through exp(jk(x_U_i,t sinθ cosφ + y_U_i,t sinθ sinφ + z_U_i,t cosθ)), so a wind-induced displacement ΔP_i,t enters only through this phase term. Constraint (13d) restricts the excitation weight ω_U_i,t to real values in [0,1]. A real magnitude weight can attenuate or switch off an element, but it cannot rotate the per-element phase back to the unperturbed value, i.e., it cannot apply the compensating phase exp(−jk ΔP_i,t · u). Thus the claimed recovery mechanism—"adaptively adjusting excitation current weights" (Abstract; §IV-B, Eq. (15))—cannot, under the stated model, undo wind-induced phase errors. This is a load-bearing inconsistency: either the control must be changed to complex/phasor excitation weights, or the claim must be restricted to amplitude-only partial compensation and evaluated as such.
  2. [§V-B, Figs. 4–5] No baseline is reported against either the undisturbed array or the uncompensated wind-perturbed array. The text states that the average initial D is 8.38 dB and that PPO-LA stabilizes at D = 8.25±0.15 dB. This stable value is below the initial value, so the reported numbers do not by themselves demonstrate "recovery" to the original beam. Without the uncompensated perturbed case, the improvement attributable to the proposed compensation cannot be separated from the effect of amplitude tapering or element selection, and the central performance claim is not quantitatively grounded.
  3. [§IV-B, Eq. (16)] The reward r_t = D/M − p0 mixes the linear directivity ratio D defined in Eq. (2) with the logarithmic maximum sidelobe level M defined in Eq. (3) as 10 log10 of a power ratio. Since D is not expressed in dB in its definition, the ratio D/M is dimensionally inconsistent, and the learning signal does not cleanly represent the stated objective of maximizing directivity while minimizing sidelobe level. The paper should specify a consistent transformation (e.g., converting both quantities to dB) and justify the chosen combination. Additionally, the penalty term p0 is never defined.
  4. [§III-D, Eqs. (9)–(10)] The "Dryden turbulence model" is not implemented as such. Eq. (9) gives V_tur(z_tur) = V_ref ln(z_tur/z0), whereas the standard logarithmic wind profile is V_ref / ln(z_tur/z0); and Eq. (10) models fluctuations as a single sinusoid with random phase, not as a Dryden spectral process. Since the turbulent-wind scenario is one of the three core cases in the paper, this mislabeling weakens the claim that the framework captures realistic wind field disturbances.
minor comments (5)
  1. [§V-B] The text refers to a "dual-phase reward mechanism (Eq. (13))" that dynamically prioritizes between directivity and sidelobe suppression, but Eq. (13) is the optimization problem and no dual-phase reward is defined in Section IV; please add the missing definition or correct the reference.
  2. [§IV-D, Eq. (18)] The LSTM equations use x_t as the input vector, but the state in Eq. (14) is a heterogeneous collection of coordinates, weights, and wind parameters; the paper should specify how this state is encoded and normalized before being fed to the LSTM.
  3. [§III-D, Eqs. (6), (8), (11)] The positional deviations are computed as time integrals of wind velocities, but the paper does not state how these deviations translate into the coordinates (x_U_i,t, y_U_i,t, z_U_i,t) used in Eq. (1), nor whether the UAVs actively counteract the wind; please clarify the relationship between ΔP_i,t and the array-factor coordinates.
  4. [§III-C, Eq. (3)] The expression "max Pt(θd, ϕd) θd,ϕd /∈ψ" is not a valid mathematical formulation; it should be written as the maximum of P_t(θ_d, φ_d) over the angular region outside the mainlobe ψ.
  5. [§V-B] The phrase "compared to benchmark and original algorithms" is unclear; the subsequent text compares PPO-LA with DDPG, DQN, SAC, and standard PPO, so the caption and text should name these baselines consistently.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the recovery claim rests on simulation and independent baseline comparisons, not on a fitted parameter or self-citation chain.

full rationale

The paper's central derivation is a DRL policy that adjusts excitation current weights to maximize a reward built from directivity and sidelobe level. The reward in Eq. (16) is the same objective being optimized, but that is not circular: the array factor in Eq. (1), directivity in Eq. (2), and sidelobe level in Eq. (3) define a nontrivial mapping from the action variables to the reported metrics, and the paper evaluates the trained policy against independent DRL baselines (DDPG, DQN, SAC, and standard PPO) on the same metrics. No parameter is fitted to a subset of the evaluation data, and no reported 'prediction' is a renamed fit or a statistic forced by construction. The self-citations that appear (such as references [8], [15], [18], and [42]) are contextual, motivational, or related-work citations and are not used to justify the central recovery mechanism; the wind model uses standard formulas including the Dryden turbulence model, and the PPO and baseline algorithms are cited to external standard works. The concern that real-valued weights constrained to [0,1] in Eq. (13d) cannot correct per-element phase errors caused by position displacements is a physical correctness limitation, not a circularity, because the paper's equations do not reduce the claimed recovery to its own inputs by definition. The comparison against standard benchmarks provides independent grounding for the outperformance claim. Therefore no circular steps are identified.

Assumptions & free parameters 10 free parameters · 4 assumptions · 0 invented entities

The central result rests on hand-picked wind parameters and idealized assumptions about far-field radiation, perfect state sensing, and a simplified kinematic displacement model. No external data or benchmarks are used; the simulation environment is entirely constructed by the authors.

free parameters (10)
  • wind speed magnitude V_c (constant wind) = not specified
    Eq. (5) defines constant wind velocity; the value is hand-picked for simulation, not measured.
  • wind direction angle θ_c = not specified
    Eq. (5) angle parameter chosen by hand, affecting displacement direction in Eq. (6).
  • ground-level wind speed V_0 = not specified
    Eq. (7) wind shear parameter chosen by hand.
  • wind speed gradient k_w = not specified
    Eq. (7) shear gradient chosen by hand.
  • reference wind speed V_ref = not specified
    Eq. (9) turbulence model parameter chosen by hand.
  • surface roughness length z_0 = not specified
    Eq. (9) terrain parameter chosen by hand.
  • turbulence intensity σ_V = not specified
    Eq. (10) amplitude of sinusoidal fluctuation chosen by hand.
  • turbulence frequency ω_turb = not specified
    Eq. (10) frequency chosen by hand.
  • random phase φ_turb = not specified
    Eq. (10) random phase; simulation outcomes depend on its draw.
  • DRL hyperparameters (learning rate, clip range, batch size, LSTM hidden size, discount) = not reported
    Algorithm 1 lists no explicit hyperparameter values; Section V-A refers to reference [43] for communication parameters but no DRL training table is provided.
assumptions (4)
  • domain assumption Far-field, isotropic element array factor model
    Eq. (1) assumes idealized far-field superposition with isotropic elements and no mutual coupling; this is not discussed.
  • domain assumption Perfect real-time state observability
    The MDP state in Eq. (14) includes exact 3D positions and wind parameters; the paper does not model sensing noise or estimation error.
  • domain assumption Kinematic wind-to-position mapping
    Eqs. (6), (8), and (11) compute positional deviations as time integrals of wind speed, ignoring UAV attitude dynamics, control loops, and drag; no aerodynamic model is used.
  • ad hoc to paper Sinusoidal approximation of turbulence
    Eq. (10) replaces the Dryden spectral model with a deterministic sine fluctuation, discarding stochastic frequency content; the paper labels this as the Dryden model, which it is not.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Recovery of UAV Swarm-enabled Collaborative Beamforming in Low-altitude Wireless Networks under Wind Field Disturbances." pith.science (2026). https://pith.science/paper/YQTPT7WS

@misc{pith2026250708507,
  author       = {Pith},
  title        = {Pith review of: Recovery of UAV Swarm-enabled Collaborative Beamforming in Low-altitude Wireless Networks under Wind Field Disturbances},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YQTPT7WS}},
  note         = {Machine review of arXiv:2507.08507}
}
read the original abstract

Unmanned aerial vehicle (UAV) swarms utilizing collaborative beamforming (CB) in low-altitude wireless networks (LAWN) demonstrate significant potential for enhanced communication range, energy efficiency, and signal directivity through the formation of virtual antenna arrays (VAA). However, environmental disturbances, particularly wind fields, significantly degrade CB performance by introducing positional errors that disrupt beam patterns, thereby compromising transmission reliability. This paper investigates the critical challenge of maintaining CB performance in UAV-based VAAs operating in LAWN under wind field disturbances. We propose a comprehensive framework that models the impact of three distinct wind conditions (constant, shear, and turbulent) on UAV array performance, and formulate a long-term real-time optimization problem to maximize directivity while minimizing maximum sidelobe levels through adaptive excitation current weight adjustments. To address the inherent complexity of this problem, we propose a novel proximal policy optimization algorithm with long short-term memory (LSTM) structure and adaptive learning rate (PPO-LA), which effectively captures temporal patterns in wind field disturbances and enables real-time adaptation without requiring extensive prior training for specific wind conditions. Our simulation results demonstrate that the proposed PPO-LA algorithm successfully recovers degraded CB performance across various wind scenarios, and thus significantly outperforming benchmark algorithms.

Figures

Figures reproduced from arXiv: 2507.08507 by the authors.

Figure 1
Figure 1. Communication between the UAV swarm and a BS of LAWN in a [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Framework of Recurrent PPO-LA. computational complexity for this phase is expressed as O(2|θπ| + |θv|). • Action Sampling: This phase entails generating actions according to the current policy πθ. The complexity of generating actions for each timeslot is O(|θπ|). Since this is done for T timeslots per episode and M episodes, the total complexity for action sampling is O(MaNaT|θπ|). • Replay Buffer Collection: The co… view at source ↗
Figure 3
Figure 3. Simulation results with 8 UAVs. (a) Cumulative rewards training curve. (b) Dual-objective CB optimization analysis. (c) Quantitative impact of wind [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (1 more)
Figure 5
Figure 5. Figure 5: Simulation results with 16 UAVs. (a) Directivity Training curve. (b) [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

46 extracted references · 37 canonical work pages

  1. [1]

    High-mobility satellite-UA V com- munications: Challenges, solutions, and future research trends,

    J. Wang, C. Jiang, and L. Kuang, “High-mobility satellite-UA V com- munications: Challenges, solutions, and future research trends,” IEEE Commun. Mag. , vol. 60, no. 5, pp. 38–43, 2022

  2. [2]

    Multi-UA V disaster environment coverage planning with limited-endurance,

    H. Song, J. Yu, J. Qiu, Z. Sun, K. Lang, Q. Luo, Y . Shen, and Y . Wang, “Multi-UA V disaster environment coverage planning with limited-endurance,” in Proc. IEEE ICRA , 2022, pp. 10 760–10 766

  3. [3]

    Simulation of intelligent unmanned aerial vehicle (UA V) for military surveillance,

    M. A. Masum, M. K. Arrofi, G. Jati, F. Arifin, M. nanda Kurniawan, P. Mursanto, and W. Jatmiko, “Simulation of intelligent unmanned aerial vehicle (UA V) for military surveillance,” in Proc. IEEE ICACSIS , 2013, pp. 161–166

  4. [4]

    Toward realization of low-altitude economy networks: Core architecture, integrated tech- nologies, and future directions,

    Y . Wang, G. Sun, Z. Sun, J. Wang, J. Li, C. Zhao, J. Wu, S. Liang, M. Yin, P. Wang, D. Niyato, S. Sun, and D. I. Kim, “Toward realization of low-altitude economy networks: Core architecture, integrated tech- nologies, and future directions,” arXiv preprint arXiv:2504.21583 , vol. abs/2504.21583, 2025

  5. [5]

    From ground to sky: Architectures, applications, and challenges shaping low-altitude wireless networks,

    W. Yuan, Y . Cui, J. Wang, F. Liu, G. Sun, T. Xiang, J. Xu, S. Jin, D. Niyato, S. Coleri, S. Sun, S. Mao, A. Jamalipour, D. I. Kim, M.-S. Alouini, and X. Shen, “From ground to sky: Architectures, applications, and challenges shaping low-altitude wireless networks,” arXiv preprint arXiv:2506.12308, 2025

  6. [6]

    Securing the sky: Integrated satellite-UA V physical layer security for low-altitude wireless networks,

    J. Li, G. Sun, X. Sun, F. Mei, J. Wang, X. Hou, D. Tian, and V . C. M. Leung, “Securing the sky: Integrated satellite-UA V physical layer security for low-altitude wireless networks,” arXiv preprint arXiv:2506.23493 , 2025

  7. [7]

    Joint trajectory and communication design for multi-UA V enabled wireless networks,

    Q. Wu, Y . Zeng, and R. Zhang, “Joint trajectory and communication design for multi-UA V enabled wireless networks,” IEEE Trans. Wirel. Commun., vol. 17, no. 3, pp. 2109–2121, 2018

  8. [8]

    UA V swarm-enabled collaborative post-disaster communications in low altitude economy via a two-stage optimization approach,

    X. Zheng, G. Sun, J. Li, J. Wang, Q. Wu, D. Niyato, and A. Jamalipour, “UA V swarm-enabled collaborative post-disaster communications in low altitude economy via a two-stage optimization approach,” IEEE Trans. Mobile Comput. , vol. abs/2501.05742, 2025

Show all 46 references
  1. [9]

    Distributed and collaborative beamforming in wireless sensor networks: Classifications, trends, and research directions,

    S. Jayaprakasam, S. K. A. Rahim, and C. Y . Leow, “Distributed and collaborative beamforming in wireless sensor networks: Classifications, trends, and research directions,” IEEE Commun. Surv. Tutorials , vol. 19, no. 4, pp. 2092–2116, 2017

  2. [10]

    Satellite-assisted low- altitude economy networking: Concepts, applications, and opportunities,

    S. He, J. Wang, Y . Liang, G. Sun, and D. Niyato, “Satellite-assisted low- altitude economy networking: Concepts, applications, and opportunities,” arXiv preprint arXiv:2505.04098 , vol. abs/2505.04098, 2025

  3. [11]

    Positional perturbations analysis for micro-UA V array with relative position-based formation,

    F. Y . P. Feng, M. Rihan, and L. Huang, “Positional perturbations analysis for micro-UA V array with relative position-based formation,” IEEE Commun. Lett. , vol. 25, no. 9, pp. 2918–2922, 2021

  4. [12]

    Location error analysis for collaborative beamforming in UA Vs random array,

    J. Du, W. Guo, M. Yan, H. Zhao, and Y . Tang, “Location error analysis for collaborative beamforming in UA Vs random array,” IEEE Wirel. Commun. Lett. , vol. 13, no. 3, pp. 904–907, 2024

  5. [13]

    Reinforcement learning algorithm for non-stationary environments,

    S. Padakandla, P. K. J., and S. Bhatnagar, “Reinforcement learning algorithm for non-stationary environments,” Appl. Intell. , vol. 50, no. 11, pp. 3590–3606, 2020

  6. [14]

    An accurate UA V 3-D path planning method for disaster emergency response based on an improved multiobjective swarm intelligence algorithm,

    Y . Wan, Y . Zhong, A. Ma, and L. Zhang, “An accurate UA V 3-D path planning method for disaster emergency response based on an improved multiobjective swarm intelligence algorithm,” IEEE Trans. Cybern., vol. 53, no. 4, pp. 2658–2671, 2023

  7. [15]

    QoE maximization for multiple-UA V-assisted multi-access edge computing: An online joint optimization approach,

    L. He, G. Sun, Z. Sun, Q. Wu, J. Kang, D. Niyato, Z. Han, and V . C. M. Leung, “QoE maximization for multiple-UA V-assisted multi-access edge computing: An online joint optimization approach,” IEEE/ACM Trans. Netw., vol. abs/2406.11918, 2024

  8. [16]

    Experimental evaluation of multi-antenna receivers for UA V communication in live LTE networks,

    T. Izydorczyk, M. Bucur, F. M. L. Tavares, G. Berardinelli, and P. Mogensen, “Experimental evaluation of multi-antenna receivers for UA V communication in live LTE networks,” in Proc. IEEE Globecom , 2018, pp. 1–6

  9. [17]

    Security energy efficiency analysis of analog collaborative beamforming with stochastic virtual antenna array of UA V swarm,

    H. Jung, I. Lee, and J. Joung, “Security energy efficiency analysis of analog collaborative beamforming with stochastic virtual antenna array of UA V swarm,”IEEE Trans. V eh. Technol., vol. 71, no. 8, pp. 8381–8397, 2022

  10. [18]

    Time and energy minimization communications based on collaborative beamforming for UA V networks: A multi-objective optimization method,

    G. Sun, J. Li, Y . Liu, S. Liang, and H. Kang, “Time and energy minimization communications based on collaborative beamforming for UA V networks: A multi-objective optimization method,” IEEE J. Sel. Areas Commun., vol. 39, no. 11, pp. 3555–3572, 2021

  11. [19]

    Toward a connected sky: Performance of beamforming with down-tilted antennas for ground and UA V user co-existence,

    R. Amer, W. Saad, and N. Marchetti, “Toward a connected sky: Performance of beamforming with down-tilted antennas for ground and UA V user co-existence,”IEEE Commun. Lett. , vol. 23, no. 10, pp. 1840– 1844, 2019

  12. [20]

    Cooperative downlink interference transmission and cancellation for cellular-connected UA V: A divide-and-conquer approach,

    W. Mei and R. Zhang, “Cooperative downlink interference transmission and cancellation for cellular-connected UA V: A divide-and-conquer approach,” IEEE Trans. Commun. , vol. 68, no. 2, pp. 1297–1311, 2020

  13. [21]

    3-D sidelobe suppression for collaborative beamforming with UA Vs,

    J. Han, J. Kim, I. Lee, and H. Jung, “3-D sidelobe suppression for collaborative beamforming with UA Vs,” IEEE Wirel. Commun. Lett. , vol. 14, no. 3, pp. 796–800, 2025

  14. [22]

    Decentralized frequency alignment for collaborative beamforming in distributed phased arrays,

    H. Ouassal, M. Yan, and J. A. Nanzer, “Decentralized frequency alignment for collaborative beamforming in distributed phased arrays,” IEEE Trans. Wirel. Commun. , vol. 20, no. 10, pp. 6269–6281, 2021

  15. [23]

    Performance of cellular-connected UA V in cell-free radio access network with network-assisted full-duplex,

    Z. Wan, J. Li, P. Zhu, D. Wang, H. Zhang, F. Liu, and X. You, “Performance of cellular-connected UA V in cell-free radio access network with network-assisted full-duplex,” IEEE Trans. Wirel. Commun. , vol. 23, no. 10, pp. 14 848–14 863, 2024

  16. [24]

    Distributionally robust collaborative beamforming in D2D relay networks with interference constraints,

    S. Gong, S. X. Wu, A. M. So, and X. Huang, “Distributionally robust collaborative beamforming in D2D relay networks with interference constraints,” IEEE Trans. Wirel. Commun. , vol. 16, no. 8, pp. 5048–5060, 2017

  17. [25]

    Secrecy rate analysis of open-loop analog collaborative beamforming under position estimation error of virtual antenna array,

    H. Jung and I. Lee, “Secrecy rate analysis of open-loop analog collaborative beamforming under position estimation error of virtual antenna array,” IEEE Wirel. Commun. Lett. , vol. 8, no. 5, pp. 1337–1340, 2019

  18. [26]

    A novel control approach for flight-stability of fixed-wing UA V formation with wind field,

    J. Zhang and J. Yan, “A novel control approach for flight-stability of fixed-wing UA V formation with wind field,” IEEE Syst. J. , vol. 15, no. 2, pp. 2098–2108, 2021

  19. [27]

    Chase or wait: Dynamic UA V deployment to learn and catch time-varying user activities,

    Z. Wang and L. Duan, “Chase or wait: Dynamic UA V deployment to learn and catch time-varying user activities,” IEEE Trans. Mob. Comput. , 2021, early access, doi: 10.1109/TMC.2021.3107027

  20. [28]

    Modeling quadrotor dynamics in a wind field,

    J. Heegyun, S. Jeyoung, L. Hanbit, and E. Yongsoon, “Modeling quadrotor dynamics in a wind field,” IEEE/ASME Trans. Mechatronics , vol. 26, no. 3, pp. 1401–1411, 2021

  21. [29]

    Wind turbine power modelling and optimization using artificial neural network with wind field experimental data,

    H. Sun, C. Qiu, L. Lu, X. Gao, J. Chen, and H. Yang, “Wind turbine power modelling and optimization using artificial neural network with wind field experimental data,” Applied Energy, vol. 280, p. 115880, 2020

  22. [30]

    Geometric adaptive control with neural networks for a quadrotor in wind fields,

    M. Bisheban and T. Lee, “Geometric adaptive control with neural networks for a quadrotor in wind fields,” IEEE Trans. Control. Syst. Technol., vol. 29, no. 4, pp. 1533–1548, 2021

  23. [31]

    Multibody model of a UA V in presence of wind fields,

    M. C. D. Simone, S. Russo, Z. B. Rivera, and D. Guida, “Multibody model of a UA V in presence of wind fields,” in Proc. IEEE ICCAIRO , 2017, pp. 83–88

  24. [32]

    Joint resource management for energy-efficient UA V- assisted SWIPT-MEC: A deep reinforcement learning approach,

    Y . Chen, H. Kang, J. Li, G. Sun, B. Wang, J. Wang, C. Liang, S. Liang, and D. Niyato, “Joint resource management for energy-efficient UA V- assisted SWIPT-MEC: A deep reinforcement learning approach,” IEEE Internet Things J. , vol. abs/2505.03230, 2025

  25. [33]

    Uabeam: UA V-based beamforming system analysis with in-field air-to-ground channels,

    Y . Shi, R. Enami, J. Wensowitch, and J. Camp, “Uabeam: UA V-based beamforming system analysis with in-field air-to-ground channels,” in Proc. IEEE SECON , 2018, pp. 127–135

  26. [34]

    Collision recovery control of a foldable quadrotor,

    K. Patnaik, S. Mishra, Z. Chase, and W. Zhang, “Collision recovery control of a foldable quadrotor,” in Proc. IEEE AIM , 2021, pp. 418–423

  27. [35]

    ADMOST: UA V flight data anomaly detection and mitigation via online subspace tracking,

    Y . He, Y . Peng, S. Wang, and D. Liu, “ADMOST: UA V flight data anomaly detection and mitigation via online subspace tracking,” IEEE Trans. Instrum. Meas. , vol. 68, no. 4, pp. 1035–1044, 2019

  28. [36]

    Vision- based autonomous landing using an mpc-controlled micro UA V on a moving platform,

    A. Mohammadi, Y . Feng, C. Zhang, S. Rawashdeh, and S. Baek, “Vision- based autonomous landing using an mpc-controlled micro UA V on a moving platform,” Proc. IEEE ICUAS , pp. 771–780, 2020

  29. [37]

    Covert prompt transmission for secure large language model services,

    R. Zhang, Y . Liu, S. Tang, J. Wang, D. Niyato, G. Sun, Y . Li, and S. Sun, “Covert prompt transmission for secure large language model services,” arXiv preprint arXiv:2504.21311 , vol. abs/2504.21311, 2025

  30. [38]

    Online collaborative resource allocation and task offloading for multi-access edge computing,

    G. Sun, M. Yuan, Z. Sun, J. Wang, H. Du, D. Niyato, Z. Han, and D. I. Kim, “Online collaborative resource allocation and task offloading for multi-access edge computing,” IEEE Trans. Mobile Comput. , vol. abs/2501.02952, 2025

  31. [39]

    LAMeTA: Intent-aware agentic network optimization via a large AI model-empowered two-stage approach,

    Y . Liu, G. Liu, J. Wang, R. Zhang, D. Niyato, G. Sun, Z. Xiong, and Z. Han, “LAMeTA: Intent-aware agentic network optimization via a large AI model-empowered two-stage approach,” arXiv preprint arXiv:2505.12247, vol. abs/2505.12247, 2025

  32. [40]

    Proximal policy optimization via enhanced exploration efficiency,

    J. Zhang, Z. Zhang, S. Han, and S. L ¨u, “Proximal policy optimization via enhanced exploration efficiency,” Inf. Sci. , vol. 609, pp. 750–765, 2022

  33. [42]

    Multi- objective aerial collaborative secure communication optimization via generative diffusion model-enabled deep reinforcement learning,

    C. Zhang, G. Sun, J. Li, Q. Wu, J. Wang, D. Niyato, and Y . Liu, “Multi- objective aerial collaborative secure communication optimization via generative diffusion model-enabled deep reinforcement learning,” IEEE Trans. Mob. Comput. , vol. 24, no. 4, pp. 3041–3058, 2025

  34. [43]

    Joint trajectory and passive beamforming design for intelligent reflecting surface-aided UA V communications: A deep reinforcement learning approach,

    L. Wang, K. Wang, C. Pan, and N. Aslam, “Joint trajectory and passive beamforming design for intelligent reflecting surface-aided UA V communications: A deep reinforcement learning approach,” IEEE Trans. Mob. Comput., vol. 22, no. 11, pp. 6543–6553, 2023. IEEE INTERNET OF THIN...

  35. [44]

    Continuous control with deep reinforcement learning,

    T. P. Lillicrap, J. J. Hunt, A. Pritzel, N. Heess, T. Erez, Y . Tassa, D. Silver, and D. Wierstra, “Continuous control with deep reinforcement learning,” arXiv preprint arXiv:1509.02971 , 2015

  36. [45]

    Human-level control through deep reinforcement learning,

    V . Mnih, K. Kavukcuoglu, D. Silver, A. A. Rusu, J. Veness, M. G. Bellemare, A. Graves, M. Riedmiller, A. K. Fidjeland, G. Ostrovski et al., “Human-level control through deep reinforcement learning,” Nature, vol. 518, no. 7540, pp. 529–533, 2015

  37. [46]

    Soft actor-critic: Off- policy maximum entropy deep reinforcement learning with a stochastic actor,

    T. Haarnoja, A. Zhou, P. Abbeel, and S. Levine, “Soft actor-critic: Off- policy maximum entropy deep reinforcement learning with a stochastic actor,” arXiv preprint arXiv:1801.01290 , 2018

  38. [47]

    Prox- imal policy optimization algorithms,

    J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov, “Prox- imal policy optimization algorithms,” arXiv preprint arXiv:1707.06347 , 2017

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.