Generative Artificial Intelligence for Beamforming in Low-Altitude Economy

Chuang Zhang; Dong In Kim; Dusit Niyato; Geng Sun; Jiacheng Wang; Jia Qi; Xuejie Liu; Yuanwei Liu

REVIEW 4 major objections 3 minor 16 references

Generative Artificial Intelligence for Beamforming in Low-Altitude Economy

T0 review · 4 major / 3 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read This paper claims that adding a generative diffusion model to a TD3 reinforcement-learning agent improves collaborative UAV beamforming, raising the secrecy rate by 23% and lowering energy use by 18% in simulation.

desk verdict A useful survey of GenAI for beamforming in low-altitude networks, but the case study's 23%/18% gains rest on an opaque simulator with no equations, no code, and no variance. read the letter →

arxiv 2504.15079 v3 pith:UTTKORDD submitted 2025-04-21 cs.NI

Geng Sun , Jia Qi , Chuang Zhang , Xuejie Liu , Jiacheng Wang , Dusit Niyato , Yuanwei Liu , Dong In Kim This is my paper · ORCID

classification cs.NI

keywords generativediffusionmodelbeamforminglow-altitudeeconomyUAVswarmsecrecyrateenergyefficiencydeepreinforcementlearningTD3

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 article argues that generative AI, and in particular generative diffusion models, can solve beamforming problems that conventional optimization handles poorly in low-altitude economy networks. After a survey of how VAEs, GANs, GDMs, and Transformers support channel estimation, resource allocation, security, and energy efficiency, it presents a case study: GDMTD3, a TD3 reinforcement-learning agent whose actor network is replaced by a diffusion model, controls four UAVs forming a virtual antenna array for secure remote communication. The paper claims GDMTD3 achieves 23% higher secrecy rate and 18% lower energy consumption than DDPG, SAC, and standard TD3 in simulation, and that the diffusion actor yields more stable training and better adaptation to dynamic environments. If right, this makes GDM-enhanced DRL a practical recipe for secure, energy-aware collaborative beamforming in aerial networks.

What carries the argument

The load-bearing object is GDMTD3, a variant of the twin delayed deep deterministic policy gradient (TD3) algorithm in which the standard actor network is replaced by a generative diffusion model. The GDM learns the probability distribution of optimal actions and refines noisy inputs into high-quality actions through a reverse denoising process, while twin critic networks evaluate candidate actions in terms of secrecy rate and energy consumption. A replay buffer stores state-action-reward tuples that drive iterative policy updates, and the closed loop lets the swarm adapt to environmental changes, including perceived faults in individual UAVs. This machinery is what the paper credits for the reported 23% secrecy-rate gain and 18% energy saving.

What would settle it

Re-implement the four-UAV scenario with an explicit, published channel and secrecy model, including path loss, antenna pattern, eavesdropper channel, and per-UAV energy draw, then rerun GDMTD3 against DDPG, SAC, and TD3 with identical random seeds; if the 23% secrecy-rate gain and 18% energy reduction shrink to statistical noise, the claimed advantage does not survive a specified physical model.

Watch

Extended reading notes

Core claim

The central claim is that injecting a generative diffusion model into a deep reinforcement learning actor converts beamforming for a UAV swarm into a problem the agent can solve stably under dynamic eavesdropping. The proposed GDMTD3 models the swarm as a Markov decision process: the state is the UAV positions plus the estimated eavesdropper location; the action consists of selecting excitation current weights and adjusting UAV positions; and the reward combines secrecy rate and energy consumption with penalties for speed-limit violations and collisions. Because the diffusion actor learns the distribution of good actions and denoises noisy candidates, the authors report that the agent reaches higher and stabler rewards over 1000 training iterations and lands on a better secrecy-energy trade-off than DDPG, SAC, and TD3.

Load-bearing premise

The simulation's physical model is not specified, so the reported gains could be properties of the simulator rather than of the algorithm.

Editorial extensions

If this is right

If the reported results hold, GDMTD3 offers a single DRL recipe that keeps secure links and battery use in balance for small UAV swarms.
A diffusion-based actor can replace a standard actor in TD3 without disturbing the twin-critic update structure, so the improvement is portable to other continuous-control wireless tasks.
The framework's claimed robustness to UAV faults, such as re-routing traffic and reallocating tasks after a perceived failure, would make it useful for missions where individual platforms can drop out.
Decoupled offline training with deployment on resource-constrained UAVs would let the learned policy run without carrying the training compute onboard.

Reading between the lines

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

Editorial inference: because the paper gives no channel equations, array geometry, path-loss law, secrecy-rate formula, or energy model, the quantitative 23% and 18% figures should be read as simulator-specific until reproduced with a disclosed physical model.
Editorial inference: the same diffusion-actor design could in principle extend to larger swarms or to metrics like latency and coverage, but the paper does not test those regimes.
Editorial inference: a direct comparison against a conventional model-based beamforming optimizer, rather than only DRL baselines, would show whether the generative actor earns its added complexity.

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.

Desk Editor's Note

read the letter

Quick read: this is a survey with a case study bolted on. The survey part is a competent, readable map of GenAI techniques for beamforming in low-altitude economy. The case study proposes GDMTD3, a TD3 variant with a diffusion model in the actor, and claims a 23% secrecy-rate gain and 18% energy reduction over DDPG, SAC, and TD3. That numerical claim is the load-bearing part of the paper, and the text does not support it.

What is actually new: GDMTD3 itself is a plausible, incremental extension of diffusion-policy work — replacing the TD3 actor with a conditional diffusion model that samples actions is a reasonable thing to try, and I don't see this exact combination in the literature. What the paper does well: Sections II and III are a decent primer. The LAE-specific challenges (heterogeneous demands, dynamic topologies, complex environments) are laid out clearly, and the mapping of VAEs, GDMs, GANs, and transformers to channel estimation, resource allocation, security, and energy is useful for someone entering the area. The figures help.

Soft spots, in order of severity. First, the case study is unverifiable. Section IV-B gives platform details (4 UAVs, 0.1 W, 40x40m, Gauss-Markov eavesdropper) and Section IV-C describes state, action, and reward in one sentence each. There is no channel equation, no antenna array model, no path-loss law, no secrecy-rate formula, no energy model. The reward combines secrecy rate and energy consumption, and the reported gains are evaluated on those same metrics; there are no confidence intervals, seeds, or code. So the 23%/18% numbers are asserted, not demonstrated. That's a load-bearing flaw, not a cosmetic omission. Second, the survey sections are not novel, but that's normal for a review. Third, the future-directions section is thin — more slogans than a research agenda.

The paper leans on the authors' own prior work [11], which is relevant and not a problem by itself. The citation pattern looks okay.

Who this is for: someone who wants a first orientation on GenAI for beamforming in LAE will get value from the survey. Someone evaluating GDMTD3 should not trust the numbers as-is. I'd send this to peer review, because the topic is timely and the survey is competently done, but a serious referee should make the authors release the full simulator, add baselines and variance, and either present the case study as illustrative or provide the missing equations. If they can't, the numeric claims should be dropped. Recommend: engage, but conditionally.

Referee Report

4 major / 3 minor

Summary. The paper is a survey-plus-case-study article on applying generative AI to beamforming in low-altitude economy (LAE) networks. It surveys LAE communication demands, beamforming concepts and applications, and how VAE/GDM/GAN/Transformer models can assist channel estimation, resource allocation, communication security, and energy efficiency. The central novel contribution is a case study in Section IV: GDMTD3, a twin delayed DDPG variant whose actor is replaced by a diffusion model, applied to a four-UAV collaborative beamforming task with a mobile eavesdropper. The abstract and Section IV-D claim a 23% improvement in secrecy rate and an 18% reduction in energy consumption over DDPG, SAC, and TD3 baselines, as shown in Fig. 5.

Significance. If the case-study claims were fully supported, the paper would offer a useful tutorial synthesis and a plausibly novel combination of diffusion models with TD3 for physical-layer security in UAV networks. The survey organization around GenAI model families and beamforming tasks is informative, and the lessons-learned section is a helpful summary. However, as presented, the quantitative claims are not verifiable: the simulation is specified only at the level of platform parameters, with no equations for the channel, array response, secrecy rate, or energy model, and the reported metric is the same as the training reward. The paper's archival value currently rests mainly on the survey portions; the case study is not yet a self-contained demonstration.

major comments (4)

[Section IV-B and IV-C, Fig. 5] The physical model underlying the simulation is never stated. There is no equation for the wireless channel between the UAV swarm and the remote base station or the eavesdropper, no antenna-array response model for collaborative beamforming, no secrecy-rate formula, and no energy-consumption model; Section IV-B lists only four UAVs, 0.1 W transmit power, a 40 m by 40 m area, and Gauss-Markov mobility parameters. Because the claimed 23% secrecy-rate improvement and 18% energy reduction are outputs of this unspecified simulator, the central quantitative result cannot be reproduced, checked, or attributed to the algorithm rather than to hidden simulation assumptions.
[Section IV-C and IV-D] The evaluation uses the same objective as the training reward. The reward combines the secrecy rate and energy consumption (Section IV-C), and Fig. 5(b) reports average secrecy rate and energy consumption per step; the algorithm is therefore scored on the very quantity it was trained to maximize. This is not a formal circularity, but it leaves open whether GDMTD3 generalizes to a held-out evaluation objective or to unseen environment configurations; the paper should evaluate on a distinct metric or on out-of-distribution scenarios.
[Section IV-D] No statistical confidence is reported. Fig. 5 shows single learning curves without error bars, confidence intervals, the number of random seeds, or variance across runs, and no baseline hyperparameter settings are provided. With a single trajectory per algorithm, the 23%/18% figures cannot be distinguished from stochastic variation; the authors should report means and variances over multiple seeds and specify the exact definition of the improvement, including the baseline to which it is relative.
[Section IV-C] The GDMTD3 algorithm is underspecified. The paper does not provide the diffusion actor's architecture, the forward and reverse diffusion equations, the number of denoising steps, the action sampling procedure, the critic update rules, the replay-buffer size, learning rates, or exploration noise parameters. Since the proposed method is the key contribution, these details are necessary for reproducibility and for understanding what GDM-enhanced means beyond replacing the actor network.

minor comments (3)

[Section III-A] There are formatting issues in the text, such as 'V ariational' split with a space and 'UA V' appearing with a space; the manuscript should be carefully copyedited.
[Section IV-D] The claim that GDMTD3 achieves 'higher and more stable rewards' is based on visual inspection of Fig. 5(a); the paper should quantify convergence and stability, for example by reporting final reward means and variances over seeds.
[Section II and III] The survey would benefit from explicit mathematical definitions or system models for the beamforming and GenAI components; several statements, such as the claim that GenAI models use self-supervised learning, are verbal and would be clearer with concrete formalism.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the case-study improvement is an empirical simulation result, not derived from its own inputs; self-citations are background only.

full rationale

The paper's central quantitative claim (23% secrecy-rate improvement and 18% energy reduction in Section IV-D, Fig. 5) is an empirical comparison of GDMTD3 against DDPG, SAC, and TD3 in a simulated collaborative-beamforming task. The algorithm is trained to maximize a reward that combines secrecy rate and energy consumption (Section IV-C), and the evaluation in Fig. 5(b) reports the same metrics; this is a standard RL setup, not a circular reduction, because the baselines are trained on the same reward and the outcome is not logically forced—GDMTD3 could have failed to outperform them. No parameter is fitted from the reported results, and the improvement is not a renamed input. The survey sections cite prior works by overlapping authors ([10], [11], [15]) as supporting examples of GenAI-based beamforming, but these citations are illustrative background for the tutorial portions and are not used to prove the case-study simulation results. The absence of channel equations, array models, secrecy-rate formulas, and energy models in Section IV-B is a serious reproducibility gap and a correctness risk, but it does not make the derivation circular: the missing equations are not shown to be equivalent to the output by construction. Therefore no circular step can be exhibited with the paper's own text, and the circularity score is 0.

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

The case study depends on several unstated simulation models and an ad hoc reward. The review sections depend on prior literature. No new physical entities are introduced.

free parameters (5)

Reward weighting coefficient between secrecy rate and energy consumption = not reported
The reward 'combines the secrecy rate and energy consumption' but the relative weight is not stated; this weight sets the Pareto trade-off and therefore the reported 23%/18% numbers.
UAV transmit power = 0.1 W
Stated in Section IV-B; affects both secrecy and energy results, with no sensitivity analysis.
Number of UAVs and deployment area = 4 UAVs, 40 m x 40 m
Chosen for the simulation; scaling behavior to realistic swarms is untested.
Eavesdropper mobility parameters = speed 5.0 m/s, correlation 0.1, variance 1.0
Section IV-B defines the threat scenario; no robustness check across mobility regimes.
Diffusion model and TD3 hyperparameters = not reported
Network sizes, denoising steps, noise schedule, learning rates, and replay buffer details are omitted; these choices materially affect the comparison.

assumptions (4)

domain assumption Secrecy rate and energy consumption can be computed from UAV positions and the estimated eavesdropper location.
Section IV-C builds the state and reward from these quantities without giving the underlying communication and energy models.
domain assumption Diffusion-model reverse sampling can serve as a policy network in TD3 and converge to good actions.
Section IV-C replaces the TD3 actor with a diffusion model but provides no convergence argument, ablation, or formal support.
domain assumption Collaborative beamforming among synchronized UAVs creates a virtual antenna array with standard gain properties.
Section IV-A invokes collaborative beamforming for long-distance secure links without specifying array geometry, synchronization error, or excitation model.
domain assumption The Gauss-Markov mobility model represents a realistic mobile eavesdropper.
Section IV-B uses this model with specific parameters; no validation against real mobility data.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Generative Artificial Intelligence for Beamforming in Low-Altitude Economy." pith.science (2026). https://pith.science/paper/UTTKORDD

@misc{pith2026250415079,
  author       = {Pith},
  title        = {Pith review of: Generative Artificial Intelligence for Beamforming in Low-Altitude Economy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UTTKORDD}},
  note         = {Machine review of arXiv:2504.15079}
}

read the original abstract

The growth of low-altitude economy (LAE) has driven a rising demand for efficient and secure communication. However, conventional beamforming optimization techniques struggle in the complex LAE environments. In this context, generative artificial intelligence (GenAI) methods provide a promising solution. In this article, we first introduce the core concepts of LAE and the roles of beamforming in advanced communication technologies for LAE. We then examine their interrelation, followed by an analysis of the limitations of conventional beamforming methods. Next, we provide an overview of how GenAI methods enhance the process of beamforming, with a focus on its applications in LAE. Furthermore, we present a case study using a generative diffusion model (GDM)-based algorithm to enhance the performance of aerial collaborative beamforming-enabled remote secure communications in LAE and simulation results verified the effectiveness of the proposed algorithms. Finally, promising research opportunities are identified.

Figures

Figures reproduced from arXiv: 2504.15079 by the authors.

**Figure 1.** This figure presents the two key technologies in beamforming: beam management and beam tracking, each of which illustrates how it contributes to [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** Beamforming enables efficient communications and sensing applications in the LAE, such as interference avoidance, spectrum reuse, target tracking [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** An overview of the timeline presenting the developments in low-altitude networking, beamforming techniques, and applications of beamforming in [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: The workflow of the GDMTD3 algorithm includes environmental observation, action generation via a diffusion-based actor network, evaluation by [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

**Figure 5.** Figure 5: Performance evaluation of the GDMTD3 algorithm compared with [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

16 extracted references · 8 canonical work pages

[11]

Generative AI-enabled wireless communications for robust low-altitude economy networking

C. Zhao, J. Wang, R. Zhang, D. Niyato, G. Sun, H. Du, D. I. Kim, and A. Jamalipour, “Generative AI-enabled wireless communications for robust low-altitude economy networking.” arXiv:2502.18118 [eess],

arXiv
[1]

6G non-terrestrial networks enabled low-altitude econ- omy: Opportunities and challenges

Y . Jiang, X. Li, G. Zhu, H. Li, J. Deng, K. Han, C. Shen, Q. Shi, and R. Zhang, “6G non-terrestrial networks enabled low-altitude econ- omy: Opportunities and challenges.” arXiv:2311.09047 [cs], 2023. doi:10.48550/ARXIV .2311.09047

arXiv doi:10.48550/arxiv.2311.09047 2023
[2]

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:2506.12308 [eess], 2025. doi: 10.48550/ARXIV .2506.12308

arXiv doi:10.48550/arxiv.2506.12308 2025
[3]

Integrated sensing and communication for low altitude economy: Opportunities and challenges,

Y . Jiang, X. Li, G. Zhu, H. Li, J. Deng, K. Han, C. Shen, Q. Shi, and R. Zhang, “Integrated sensing and communication for low altitude economy: Opportunities and challenges,”IEEE Commun. Mag., Early Access, 2025. doi: 10.1109/mcom.001.2400685

doi:10.1109/mcom.001.2400685 2025
[4]

Multi-UA V-enabled energy- efficient data delivery for low-altitude economy: Joint coded caching, user grouping, and UA V deployment,

Q. Wei, R. Li, W. Bai, and Z. Han, “Multi-UA V-enabled energy- efficient data delivery for low-altitude economy: Joint coded caching, user grouping, and UA V deployment,”IEEE Internet Things J., Early Access, 2025. doi: 10.1109/jiot.2025.3562872

arXiv 2025
[5]

Distributed transmit beamforming: Design and demonstration from the lab to UA Vs,

S. Hanna and D. Cabric, “Distributed transmit beamforming: Design and demonstration from the lab to UA Vs,”IEEE Trans. Wirel. Commun., vol. 22, no. 2, pp. 778–792, Feb. 2023

work page 2023
[6]

Deployment and robust hybrid beamforming for UA V mmWave communications,

K. Liu, Y . Liu, P. Yi, Z. Xiao, and X. Xia, “Deployment and robust hybrid beamforming for UA V mmWave communications,”IEEE Trans. Commun., vol. 71, no. 5, pp. 3073–3086, May 2023

work page 2023
[7]

Channel estimation in underdetermined systems utilizing variational autoencoders,

M. Baur, N. Turan, B. Fesl, and W. Utschick, “Channel estimation in underdetermined systems utilizing variational autoencoders,” inProc. IEEE Int. Conf. Acoust., Speech Signal Process., pp. 9031–9035, Seoul, Republic of Korea, Apr. 14-19 2024

work page 2024

Show all 16 references

[8]

ICDM: Interference cancellation diffusion models for wireless semantic communications

T. Wu, Z. Chen, D. He, F. Yang, M. Tao, X. Xu, W. Zhang, and P. Zhang, “ICDM: Interference cancellation diffusion models for wireless semantic communications.” arXiv:2505.19983 [cs], 2025. doi:10.48550/ARXIV.2505.19983

2025 doi arXiv
[9]

Gen- erative semantic communications with foundation models: Perception- error analysis and semantic-aware power allocation,

C. Xu, M. B. Mashhadi, Y . Ma, R. Tafazolli, and J. Wang, “Gen- erative semantic communications with foundation models: Perception- error analysis and semantic-aware power allocation,”IEEE J. Sel. Areas Commun., vol. 43, no. 7, pp. 2493–2505, Jul. 2025

2025
[10]

Denoising diffusion model-based channel estimation in IRS-assisted ISAC sys- tems,

L. Yi, J. Miao, T. Zhang, Z. Yuan, J. Kang, and D. Niyato, “Denoising diffusion model-based channel estimation in IRS-assisted ISAC sys- tems,” inProc. Int. Conf. Wireless Commun. Signal Process., pp. 425– 430, 2024

2024
[12]

Wireless sig- nal denoising using conditional generative adversarial networks,

H. Tang, Y . Zhao, G. Wang, C. Luo, and W. Wang, “Wireless sig- nal denoising using conditional generative adversarial networks,” in Proc. IEEE Conf. Comput. Commun. Workshops (INFOCOM WKSHPS), pp. 1–6, Hoboken, NJ, USA, May 20 2023

2023
[13]

Transformer-based predictive channel estimation for mmwave massive MIMO systems,

H. Ju, S. Jeong, B. Lee, and B. Shim, “Transformer-based predictive channel estimation for mmwave massive MIMO systems,” inProc. IEEE V eh. Technol. Conf., pp. 1–5, 2024

2024
[14]

Deep reinforcement learning with communication transformer for adaptive live streaming in wireless edge networks,

S. Wang, S. Bi, and Y . A. Zhang, “Deep reinforcement learning with communication transformer for adaptive live streaming in wireless edge networks,”IEEE J. Sel. Areas Commun., vol. 40, no. 1, pp. 308–322, Jan. 2022

2022
[15]

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:2506.23493 [cs], 2025. doi: 10.48550/ARXIV .2506.23493

2025 doi arXiv
[2025]

doi: 10.48550/ARXIV .2502.18118

doi arXiv

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.

[1] [11]

Generative AI-enabled wireless communications for robust low-altitude economy networking

C. Zhao, J. Wang, R. Zhang, D. Niyato, G. Sun, H. Du, D. I. Kim, and A. Jamalipour, “Generative AI-enabled wireless communications for robust low-altitude economy networking.” arXiv:2502.18118 [eess],

arXiv

[2] [1]

6G non-terrestrial networks enabled low-altitude econ- omy: Opportunities and challenges

Y . Jiang, X. Li, G. Zhu, H. Li, J. Deng, K. Han, C. Shen, Q. Shi, and R. Zhang, “6G non-terrestrial networks enabled low-altitude econ- omy: Opportunities and challenges.” arXiv:2311.09047 [cs], 2023. doi:10.48550/ARXIV .2311.09047

arXiv doi:10.48550/arxiv.2311.09047 2023

[3] [2]

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:2506.12308 [eess], 2025. doi: 10.48550/ARXIV .2506.12308

arXiv doi:10.48550/arxiv.2506.12308 2025

[4] [3]

Integrated sensing and communication for low altitude economy: Opportunities and challenges,

Y . Jiang, X. Li, G. Zhu, H. Li, J. Deng, K. Han, C. Shen, Q. Shi, and R. Zhang, “Integrated sensing and communication for low altitude economy: Opportunities and challenges,”IEEE Commun. Mag., Early Access, 2025. doi: 10.1109/mcom.001.2400685

doi:10.1109/mcom.001.2400685 2025

[5] [4]

Multi-UA V-enabled energy- efficient data delivery for low-altitude economy: Joint coded caching, user grouping, and UA V deployment,

Q. Wei, R. Li, W. Bai, and Z. Han, “Multi-UA V-enabled energy- efficient data delivery for low-altitude economy: Joint coded caching, user grouping, and UA V deployment,”IEEE Internet Things J., Early Access, 2025. doi: 10.1109/jiot.2025.3562872

arXiv 2025

[6] [5]

Distributed transmit beamforming: Design and demonstration from the lab to UA Vs,

S. Hanna and D. Cabric, “Distributed transmit beamforming: Design and demonstration from the lab to UA Vs,”IEEE Trans. Wirel. Commun., vol. 22, no. 2, pp. 778–792, Feb. 2023

work page 2023

[7] [6]

Deployment and robust hybrid beamforming for UA V mmWave communications,

K. Liu, Y . Liu, P. Yi, Z. Xiao, and X. Xia, “Deployment and robust hybrid beamforming for UA V mmWave communications,”IEEE Trans. Commun., vol. 71, no. 5, pp. 3073–3086, May 2023

work page 2023

[8] [7]

Channel estimation in underdetermined systems utilizing variational autoencoders,

M. Baur, N. Turan, B. Fesl, and W. Utschick, “Channel estimation in underdetermined systems utilizing variational autoencoders,” inProc. IEEE Int. Conf. Acoust., Speech Signal Process., pp. 9031–9035, Seoul, Republic of Korea, Apr. 14-19 2024

work page 2024