Pith. sign in

REVIEW 5 major objections 5 minor 44 references

LLM-Aided Joint Secrecy Precoding and Trajectory for RSMA-Based Heterogeneous UAV Networks

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

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

Pith's one-line read A two-layer design pairs convex secrecy precoding with LLM-seeded reinforcement learning to improve both secrecy rate and energy efficiency in heterogeneous UAV swarms.

desk verdict Plausible LLM-guided RL + SDR/DC pipeline for secure UAV networks; the LLM's specific value is unproven because there is no scripted-expert control, and the D.C. derivation mislabels concave terms as convex. read the letter →

arxiv 2507.17188 v4 pith:C5XBVCBZ submitted 2025-07-23 cs.NI cs.AIcs.CR

Lijie Zheng , Ji He , Shih Yu Chang , Yulong Shen This is my paper · ORCID
classification cs.NIcs.AIcs.CR
keywords UAVnetworksrate-splittingmultipleaccessphysicallayersecuritymulti-objectiveoptimizationmulti-agentreinforcementlearninglargelanguagemodelsecrecyprecodingtrajectory
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 tackles a conflict: heterogeneous UAVs serving ground users over rate-splitting multiple access must fly to stay secure against eavesdroppers while spending as little propulsion energy as possible. It claims that the jointly coupled problem of trajectory design and secrecy precoding is best solved by splitting it into an inner convex-optimization layer for precoding and an outer multi-agent reinforcement-learning layer for trajectories, with the learning seeded offline by expert trajectories generated by a large language model. In the simulated scenarios, this LLM-seeded hierarchical method reports roughly a 25 percent improvement in episode reward over the strongest RL baseline with the same precoding inner layer, and at ten UAVs it consumes 7 to 16 percent less propulsion energy than that baseline. A sympathetic reader would take the contribution to be a template: LLM reasoning can be distilled into a real-time policy for wireless control, provided the expensive optimization steps are handled by a separate mathematical operator.

What carries the argument

The argument rides on two nested mechanisms. The inner mechanism is S2DC, which applies semidefinite relaxation to the RSMA secrecy precoding problem, rewrites the objective as a difference-of-convex function, pushes the non-convex rank-one constraints into the objective as an exact penalty, and iteratively solves convex programs; it acts as a fixed operator that turns positions into secrecy-rate rewards. The outer mechanism is LLM-HeMARL, in which a large language model, prompted with the network and secrecy models, rolls out expert trajectories that are parsed into a dataset, distilled by conservative Q-learning into a fast policy, and then adapted by online independent soft actor-critic, so the LLM never runs during real-time decisions.

What would settle it

Run the whole pipeline with Eve CSI corrupted by zero-mean Gaussian estimation error at several variances; if the reported ~25 percent reward advantage over ISAC-S2DC disappears or shrinks dramatically, then perfect Eve CSI is the load-bearing assumption.

Watch

Extended reading notes

Core claim

The central claim is that LLM-generated expert trajectories, distilled through conservative Q-learning and then fine-tuned by independent soft actor-critic, give heterogeneous UAV agents a heterogeneity-aware starting policy that learns faster and lands on a better secrecy-energy trade-off than any baseline that learns from scratch or shares experience across agents. The paper also establishes, on its own evidence, that the hierarchical split itself is the main performance driver: methods using S2DC for precoding with RL for trajectories outperform coupled RL baselines, and independent replay buffers outperform shared ones when UAV capabilities differ. The reported numbers are the 25 percent reward gain over ISAC-S2DC and the 7 to 16 percent energy saving at ten UAVs, with the advantage growing as the swarm grows.

Load-bearing premise

The load-bearing premise is that the UAVs hold perfect, instantaneous eavesdropper channel state information; if Eve CSI is imperfect or stale, the secrecy-rate numbers that drive both the S2DC precoder and the RL reward are no longer reliable, and the reported gains are not guaranteed. A second premise is that the hand-tuned LLM prompt is a valid expert source, since no non-LLM scripted expert is used as a control.

Editorial extensions

If this is right

  • The hierarchical split converts a hopelessly coupled non-convex problem into two tractable pieces, so the same S2DC-plus-RL pattern should transfer to other wireless problems where one subproblem admits convex optimization and the other is a long-horizon decision.
  • LLM expert distillation buys a cold-start advantage: agents start with higher reward and converge faster, meaning the LLM's cost is paid once offline rather than on every control step.
  • The energy advantage over ISAC-S2DC grows to 7–16 percent as the number of UAVs reaches ten, which supports scalability of the hierarchical architecture as the number of decision variables grows.
  • Runtime per decision stays at 15–16 ms regardless of swarm size, whereas S2DC alone grows polynomially and SCA-based trajectory design takes tens of seconds to minutes per slot, so the method is compatible with real-time operation if precoding is computed in parallel.
  • Independent replay buffers beat a shared buffer in these heterogeneous swarms, implying that experience-sharing schemes should group agents by capability rather than pooling all agents together.

Reading between the lines

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

  • Editorial inference: because the LLM prompt already spells out heterogeneity-aware and secrecy-aware rules, the experiment does not isolate the LLM's semantic reasoning from the information already encoded in the prompt; a deterministic rule-based expert with the same coverage-aware assignment logic would test whether the gain is really from LLM reasoning or merely from expert seeding.
  • Editorial inference: the perfect-eavesdropper-CSI assumption is untested; under estimated or delayed Eve channels the secrecy-rate gap over baselines would likely shrink, and the method's advantage may become an energy-only one.
  • Editorial inference: the same offline-distillation pattern could serve other real-time control tasks, such as jamming avoidance or handover management, wherever a cheap environment reward exists and an LLM is consulted only to produce demonstration data.
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

5 major / 5 minor

Summary. The paper proposes a hierarchical optimization framework for RSMA-enabled heterogeneous UAV networks, maximizing the worst-case secrecy rate while minimizing propulsion energy consumption. The inner layer (S2DC) uses semidefinite relaxation, exact penalty, and difference-of-convex iterations to compute secrecy precoding at fixed UAV positions; the outer layer (LLM-HeMARL) distills an LLM-generated expert policy into an independent soft actor-critic multi-agent reinforcement learning scheme for trajectory optimization. Experiments compare the proposed method against ISAC, MASAC, and SCA-based baselines under varying random seeds and UAV swarm sizes, reporting improvements in secrecy rate and energy consumption.

Significance. If the claims hold, the paper offers a practical integration of convex precoding with LLM-guided reinforcement learning for a strongly coupled nonconvex problem. The idea of using offline LLM rollouts to warm-start MARL policies, thereby avoiding real-time LLM inference, is timely and addresses a real latency concern. The paper's strengths include a detailed RSMA/heterogeneous-UAV system model, a clear hierarchical decomposition, and evaluations across multiple random seeds and swarm sizes. However, the current version does not establish the correctness of the inner-layer convergence or the feasibility of the reconstructed precoders, and it does not isolate the LLM's specific contribution. The reported empirical gains are therefore not yet convincingly attributed to the proposed mechanism.

major comments (5)
  1. [Section III, Eqs. (15)-(19)] The statement that \tilde F1,1 through \tilde F1,4 are convex is incorrect: each is the logarithm of an affine function of (P^c, P^p) and is therefore concave. Consequently, \tilde F1 in Eq. (15) is not a difference of convex functions in the standard sense, and the 'D.C. iteration' rationale for solving (28) is not valid as written. The later linearization of \tilde F1,3 and \tilde F1,4 could potentially be recast as a lower-bounding surrogate method, but the paper does not present that argument, and no monotonic-convergence proof is provided. Please correct the convexity claim, replace the D.C. terminology with the appropriate surrogate-based justification, and give a convergence guarantee for the iterates.
  2. [Section III, Eqs. (24e) and (27)-(32)] The algorithm drops the rank-one constraints (24e) and later reconstructs precoding vectors by taking the principal eigenvector of each matrix. No theorem or numerical evidence shows that the reconstructed precoders satisfy the secrecy rate constraints (13f)/(24c) or achieve the objective value computed from the relaxed solution. The rank-one approximation is therefore unquantified, and the S2DC output fed to the RL reward may not be a feasible or accurate solution of P2. Please provide an approximation bound or report the feasibility and performance gap of the reconstructed precoders in the simulated scenarios.
  3. [Section IV-A and Table II] The reward function depends on scalarization weights w_sr and w_ec, the collision penalty p_col, the CQL penalty beta in Eq. (38), and the S2DC penalty mu in Eq. (27), but none of these values is reported in Table II or the text. As a result, the episode-reward comparison in Fig. 4 and the 'approximately 25%' improvement over ISAC-S2DC are not reproducible and cannot be interpreted as a specific secrecy/energy trade-off. Please report all hyperparameters, and ideally present a Pareto-front analysis over the weight ratio rather than a single scalarization.
  4. [Section V-C] The comparison between LLM-HeMARL-S2DC and ISAC-S2DC changes two things simultaneously: the use of a distilled-policy initialization and the LLM as the source of that policy. No non-LLM scripted-expert warm-start or random offline-pretraining control is included, and the prompt in Section IV-B explicitly instructs the LLM to maximize GT coverage, enhance secrecy rate, minimize energy consumption, and follow heterogeneity rules. The reported 25% gain may therefore reflect the benefit of any competent heuristic initialization rather than LLM reasoning. Please add a scripted-expert warm-start baseline and report the decomposed secrecy rate and energy consumption for the LLM-warm-started versus scripted-warm-started agents.
  5. [Section II.B and Section V] The entire evaluation assumes perfect eavesdropper CSI, and both the S2DC inner layer and the RL reward use this CSI to compute secrecy rates. This assumption is load-bearing for the central claims of secrecy-rate gains, but no experiments with imperfect, delayed, or estimated Eve CSI are reported. Please either add a robustness study against CSI error or clearly delimit the paper's claims to the ideal-CSI regime.
minor comments (5)
  1. [Section II.E] In problem P1, the optimization variables are listed as omega, v, and P, while constraints (4) and (5) involve the binary scheduling matrix S_I; it is not specified whether S_I is a decision variable, a fixed heuristic, or a deterministic function of coverage and channel quality. Please clarify.
  2. [Section II.A, Eq. (3)] The definition of E_k(t) uses a summation over t in T inside a per-slot expression; this appears to be a typo and should be E_k = sum_{t in T} P_k(t) Delta t.
  3. [Section V-C, Figs. 7 and 8] The text and captions refer to comparisons with SCA-S2DC, but the legends of Figs. 7 and 8 do not include SCA-S2DC; please add the baseline to the plots or remove the corresponding text.
  4. [Section V-B, Fig. 3] The experiment in Section V-B deploys the LLM directly as a planner, but the text describes it as the LLM expert policy; please clarify that this is the untrained LLM planner, not the distilled policy used in the main evaluation.
  5. [Table I and Section I] The table row 'LLM ✓' for references [14]-[16] is not explained in the text; a sentence describing how those existing LLM-based methods differ from the proposed offline distillation would improve readability.

Circularity Check

1 steps flagged · score 3.0 of 10

The 'LLM naturally produces heterogeneity-aware trajectories' claim is written into the prompt rather than discovered; the downstream secrecy/energy numbers are still independently measured.

  1. self definitional [Section I (Introduction), Section IV.B / Fig. 2 prompt, and Section V.B]
    "To validate this intuition, a preliminary experiment (detailed in Section V-B) demonstrates that even a direct LLM planner, without domain-specific training, naturally produces heterogeneity-aware trajectories. ... The prompt: 'You are an expert in UAV physical layer network security. ... you are tasked with perform multi-UAV trajectory panning to maximize GT coverage, enhance secrecy rate, and minimize energy consumption. ... * Make full use of the heterogeneity of UAVs...'"

    The 'LLM expert policy' is generated with a hand-tuned prompt that explicitly instructs the model to 'maximize GT coverage, enhance secrecy rate, and minimize energy consumption' and to 'Make full use of the heterogeneity of UAVs.' Section V.B then validates this same policy as evidence that 'LLMs can reason about heterogeneous agent capabilities' and 'naturally produce heterogeneity-aware trajectories.' The behavior used as confirmation is therefore an input to the prompt, not an independent emergent property of the LLM; the expert label re-imports the target behavior. This is partial rather than total circularity because the final secrecy rate and energy results are computed from the channel model and measured independently, so the headline comparison does not reduce to this step.

full rationale

No formal circularity is found in the S2DC precoding derivation: it follows standard SDR, exact-penalty, and D.C. programming steps with external references, and the secrecy/energy quantities are evaluated from the channel model after the optimization, not defined into the objective. The MARL reward is aligned with the evaluation metrics, which is standard practice and not a fitted prediction. The self-citations that appear (e.g., [4], [5]) are contextual references to earlier related work and are not load-bearing. The main circularity concern is confined to the LLM expert-policy motivation: the prompt in Section IV.B explicitly contains the heterogeneity-aware, secrecy-aware, energy-aware behavior that Section V.B later presents as emergent LLM capability. This weakens the claim that the LLM itself contributes expert reasoning beyond the hand-specified prompt, and the 25% improvement over ISAC-S2DC is not isolated from a non-LLM scripted-expert control. Nevertheless, the central performance comparison is an independent empirical measurement, so the appropriate score is 3: one partial, non-derivation-collapsing circular step.

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

All reported results are simulation-based. The central claim depends on several unquantified weights in the RL reward, on the assumed availability of perfect Eve CSI, and on the effectiveness of a hand-tuned LLM prompt as an expert source; none of these are independently validated.

free parameters (5)
  • Reward weights w_sr, w_ec = not reported
    Scalarize the multi-objective problem P1; chosen to keep reward components on the same order of magnitude, but actual values are omitted. These weights directly shape the reported secrecy-energy trade-off.
  • Collision penalty p_col = not reported
    Used in the reward to penalize collision risks; value is not specified and affects learned trajectories.
  • CQL penalty coefficient beta = not reported
    Controls the conservatism penalty in offline distillation (Eq. 38); no numerical value is given.
  • S2DC penalty parameter mu = not reported
    Controls the rank-one penalty in Eq. (27); not specified, and no sensitivity study is provided.
  • LLM prompt content and tuning = hand-tuned
    The prompt text encodes heterogeneity-aware and secrecy-aware rules; manual tuning is a free design choice not independently benchmarked.
assumptions (5)
  • domain assumption Perfect passive eavesdropper CSI is available at UAVs
    Section II.B states that perfect eavesdropper CSI is assumed; all secrecy rate expressions and RL rewards rely on this.
  • domain assumption Rotary-wing propulsion energy model from Zeng et al. [17] holds
    The energy objective in Eqs. (2)-(3) is taken from prior literature and is not validated in this paper.
  • domain assumption Standard RSMA rate expressions and SIC decoding order apply
    Rates in Section II.D follow [23]-[25]; no protocol-level validation is provided.
  • ad hoc to paper D.C. iterations with linearized concave components converge
    Algorithm 1 iterates until a tolerance, but convergence conditions for the nonconvex problem (14) are not proven.
  • ad hoc to paper LLM (DeepSeek-R1) generates expert-level trajectories from the provided prompt
    Section V-B qualitatively asserts this; no comparison to scripted heuristics is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LLM-Aided Joint Secrecy Precoding and Trajectory for RSMA-Based Heterogeneous UAV Networks." pith.science (2026). https://pith.science/paper/C5XBVCBZ

@misc{pith2026250717188,
  author       = {Pith},
  title        = {Pith review of: LLM-Aided Joint Secrecy Precoding and Trajectory for RSMA-Based Heterogeneous UAV Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/C5XBVCBZ}},
  note         = {Machine review of arXiv:2507.17188}
}
read the original abstract

This paper investigates secure communications in rate-splitting multiple access (RSMA) enabled heterogeneous UAV networks, where multiple UAVs collaboratively serve ground terminals in the presence of eavesdroppers. By jointly considering secrecy rate maximization and propulsion energy consumption minimization, we formulate a multi-objective optimization problem involving UAV trajectory design, service association, power allocation, and secrecy precoding under mobility, collision-avoidance, service-capacity, and communication constraints. The formulated problem is highly non-convex due to the coupling among UAV trajectories, RSMA transmission variables, and secrecy constraints. To address the resulting non-convex and highly coupled optimization problem, we propose a hierarchical optimization framework. The inner layer uses a semidefinite relaxation (SDR)-based S2DC algorithm combining penalty functions and difference-of-convex (D.C.) programming to solve the secrecy precoding problem with fixed UAV positions. The outer layer introduces a Large Language Model (LLM)-guided heuristic multi-agent reinforcement learning approach (LLM-HeMARL) for trajectory optimization. LLM-HeMARL efficiently incorporates LLM-generated expert heuristic policy, enabling UAVs to learn energy-aware, security-driven trajectories without the inference overhead of real-time LLM calls. The simulation results show that our method outperforms existing baselines in secrecy rate and energy efficiency, with consistent robustness across varying UAV swarm sizes and random seeds.

Figures

Figures reproduced from arXiv: 2507.17188 by the authors.

Figure 1
Figure 1. Illustration of RSMA-enabled HetUAVNs method that integrates LLM-generated expert poli￾cies into MARL, guiding heterogeneity-aware trajectory learning and reducing blind exploration to improve con￾vergence. Notably, the LLM does not participate in real￾time decision-making. Instead, a combination of offline and online RL distills its expert policy into fast policies that meet the stringent latency requirements of wi… view at source ↗
Figure 2
Figure 2. Framework of the LLM-HeMARL-S2DC algorithm in secure HetUAVNs. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. UAV trajectories generated by the LLM expert policy [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Convergence comparison of the baselines over training. [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 6
Figure 6. Figure 6: UAV trajectories under different random seeds in one episode. [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: The cumulative secrecy rate under different numbers of UAVs in one episode. [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: The cumulative propulsion energy consumption under different numbers of UAVs in one episode. [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 39 canonical work pages

  1. [1]

    What will the future of UA V cellular communications be? A flight from 5G to 6G,

    G. Geraciet al., “What will the future of UA V cellular communications be? A flight from 5G to 6G,”IEEE Commun. Surveys Tuts., vol. 24, no. 3, pp. 1304–1335, Jul. 2022

    work page 2022

  2. [2]

    A tutorial on UA Vs for wireless networks: Applications, challenges, and open problems,

    M. Mozaffari, W. Saad, M. Bennis, Y .-H. Nam, and M. Debbah, “A tutorial on UA Vs for wireless networks: Applications, challenges, and open problems,”IEEE Commun. Surveys Tuts., vol. 21, no. 3, pp. 2334– 2360, Mar. 2019

    work page 2019

  3. [3]

    UA V-enabled secure multiuser backscatter communications with planar array,

    L. Bai, Q. Chen, T. Bai, and J. Wang, “UA V-enabled secure multiuser backscatter communications with planar array,”IEEE J. Sel. Areas Commun., vol. 40, no. 10, pp. 2946–2961, Oct. 2022

    work page 2022

  4. [4]

    Covert communication in hybrid microwave/mmwave UA V-enabled systems with transmission mode selection,

    W. Zhang, J. He, Y . Shen, and X. Jiang, “Covert communication in hybrid microwave/mmwave UA V-enabled systems with transmission mode selection,”Comput. Commun., vol. 219, pp. 216–228, 2024

    work page 2024

  5. [5]

    RSMA- enabled multi-UA V secure communication via MARL with multi-task attention DRNN,

    L. Zheng, J. He, X. Zhu, Y . Zhang, Y . Shen, and T. Taleb, “RSMA- enabled multi-UA V secure communication via MARL with multi-task attention DRNN,”IEEE Trans. Inf. Forensics Security, vol. 20, pp. 12 624–12 639, 2025

    work page 2025

  6. [6]

    On the physical layer security of the cooperative rate- splitting-aided downlink in UA V networks,

    H. Bastamiet al., “On the physical layer security of the cooperative rate- splitting-aided downlink in UA V networks,”IEEE Trans. Inf. Forensics Security, vol. 16, pp. 5018–5033, 2021

    work page 2021

  7. [7]

    Maximizing network throughput in heterogeneous UA V networks,

    S. Liet al., “Maximizing network throughput in heterogeneous UA V networks,”IEEE/ACM Trans. Netw., vol. 32, no. 3, pp. 2128–2142, Jun. 2024

    work page 2024

  8. [8]

    Multi-objective optimization approaches for physical layer secure communications based on collaborative beamforming in UA V networks,

    J. Liet al., “Multi-objective optimization approaches for physical layer secure communications based on collaborative beamforming in UA V networks,”IEEE/ACM Trans. Netw., vol. 31, no. 4, pp. 1902–1917, Aug. 2023

    work page 1902

Show all 44 references
  1. [9]

    Multi-objective optimization for UA V swarm-assisted iot with virtual antenna arrays,

    J. Li, G. Sun, L. Duan, and Q. Wu, “Multi-objective optimization for UA V swarm-assisted iot with virtual antenna arrays,”IEEE Trans. Mobile Comput., vol. 23, no. 5, pp. 4890–4907, May. 2023

    2023

  2. [10]

    Multi-objective optimization for multi-UA V-assisted mobile edge computing,

    G. Sunet al., “Multi-objective optimization for multi-UA V-assisted mobile edge computing,”IEEE Trans. Mobile Comput., vol. 23, no. 12, pp. 14 803–14 820, Dec. 2024

    2024

  3. [11]

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

    C. Zhanget al., “Multi-objective aerial collaborative secure com- munication optimization via generative diffusion model-enabled deep reinforcement learning,”IEEE Trans. Mobile Comput., vol. 24, no. 4, pp. 3041–3058, Apr. 2024

    2024

  4. [12]

    Evolutionary multi-objective reinforcement learning based trajectory control and task offloading in UA V-assisted mobile edge computing,

    F. Songet al., “Evolutionary multi-objective reinforcement learning based trajectory control and task offloading in UA V-assisted mobile edge computing,”IEEE Trans. Mobile Comput., vol. 22, no. 12, pp. 7387– 7405, Dec. 2022

    2022

  5. [13]

    Collaborative ground-space communications via evolution- ary multi-objective deep reinforcement learning,

    J. Liet al., “Collaborative ground-space communications via evolution- ary multi-objective deep reinforcement learning,”IEEE J. Sel. Areas Commun., vol. 42, no. 12, pp. 3395–3411, Dec. 2024

    2024

  6. [14]

    Large language model based multi-objective optimization for integrated sensing and communications in UA V networks,

    H. Li, M. Xiao, K. Wang, D. I. Kim, and M. Debbah, “Large language model based multi-objective optimization for integrated sensing and communications in UA V networks,”IEEE Wireless Commun. Lett., vol. 14, no. 4, pp. 979–983, Apr. 2025

    2025

  7. [15]

    Joint user association and beamforming design for ISAC networks with large language models,

    H. Li, M. Xiao, K. Wang, R. Schober, D. I. Kim, and Y . L. Guan, “Joint user association and beamforming design for ISAC networks with large language models,”arXiv:2506.05637

    arXiv

  8. [16]

    LLM-guided drl for multi-tier LEO satellite networks with hybrid FSO/RF links,

    J. Liet al., “LLM-guided drl for multi-tier LEO satellite networks with hybrid FSO/RF links,”arXiv:2505.11978

    arXiv

  9. [17]

    Energy minimization for wireless communication with rotary-wing UA V,

    Y . Zeng, J. Xu, and R. Zhang, “Energy minimization for wireless communication with rotary-wing UA V,”IEEE Trans. Wireless Commun., vol. 18, no. 4, pp. 2329–2345, Apr. 2019

    2019

  10. [18]

    Energy efficient UA V commu- nication with energy harvesting,

    Z. Yang, W. Xu, and M. Shikh-Bahaei, “Energy efficient UA V commu- nication with energy harvesting,”IEEE Trans. Veh. Technol., vol. 69, no. 2, pp. 1913–1927, Feb. 2019

    1913

  11. [19]

    Joint RIS and beamforming design for secure and energy-efficient two-way relay communications,

    S. Zhao, X. Zhu, Y . Zhang, Z. Zhang, and Y . Shen, “Joint RIS and beamforming design for secure and energy-efficient two-way relay communications,”IEEE Trans. Mobile Comput., vol. 24, no. 8, pp. 7440–7457, Aug. 2025

    2025

  12. [20]

    Detecting passive eavesdroppers in the MIMO wiretap channel,

    A. Mukherjee and A. L. Swindlehurst, “Detecting passive eavesdroppers in the MIMO wiretap channel,” in2012 IEEE Int. Conf. on Acoust., Speech and Signal Process. (ICASSP). IEEE, 2012, pp. 2809–2812

    2012

  13. [21]

    Deep reinforcement learning for RIS-aided multiuser full-duplex secure com- munications with hardware impairments,

    Z. Peng, Z. Zhang, L. Kong, C. Pan, L. Li, and J. Wang, “Deep reinforcement learning for RIS-aided multiuser full-duplex secure com- munications with hardware impairments,”IEEE Internet Things J., vol. 9, no. 21, pp. 21 121–21 135, Nov. 2022

    2022

  14. [22]

    Optimal LAP altitude for maximum coverage,

    A. Al-Hourani, S. Kandeepan, and S. Lardner, “Optimal LAP altitude for maximum coverage,”IEEE Wireless Commun. Lett., vol. 3, no. 6, pp. 569–572, Dec. 2014

    2014

  15. [23]

    Rate-splitting multiple access: Fundamentals, survey, and future research trends,

    Y . Mao, O. Dizdar, B. Clerckx, R. Schober, P. Popovski, and H. V . Poor, “Rate-splitting multiple access: Fundamentals, survey, and future research trends,”IEEE Commun. Surveys Tuts., vol. 24, no. 4, pp. 2073– 2126, Jul. 2022

    2022

  16. [24]

    Optimization of rate allocation and power control for rate splitting multiple access (RSMA),

    Z. Yang, M. Chen, W. Saad, and M. Shikh-Bahaei, “Optimization of rate allocation and power control for rate splitting multiple access (RSMA),” IEEE Trans. Commun., vol. 69, no. 9, pp. 5988–6002, Sep. 2021

    2021

  17. [25]

    Robust transmission in downlink multiuser MISO systems: A rate-splitting approach,

    H. Joudeh and B. Clerckx, “Robust transmission in downlink multiuser MISO systems: A rate-splitting approach,”IEEE Trans. Signal Process., vol. 64, no. 23, pp. 6227–6242, Dec. 2016

    2016

  18. [26]

    Secrecy rate beamforming for multicell networks with information and energy harvesting,

    A. A. Nasir, H. D. Tuan, T. Q. Duong, and H. V . Poor, “Secrecy rate beamforming for multicell networks with information and energy harvesting,”IEEE Trans. Signal Process., vol. 65, no. 3, pp. 677–689, Feb. 2016

    2016

  19. [27]

    Robust secure beamforming design for two-user downlink MISO rate-splitting systems,

    H. Fu, S. Feng, W. Tang, and D. W. K. Ng, “Robust secure beamforming design for two-user downlink MISO rate-splitting systems,”IEEE Trans. Wireless Commun., vol. 19, no. 12, pp. 8351–8365, Dec. 2020

    2020

  20. [28]

    SpaceRIS: LEO satellite coverage maximization in 6G sub-thz networks by MAPPO DRL and whale optimization,

    S. S. Hassan, Y . M. Park, Y . K. Tun, W. Saad, Z. Han, and C. S. Hong, “SpaceRIS: LEO satellite coverage maximization in 6G sub-thz networks by MAPPO DRL and whale optimization,”IEEE J. Sel. Areas Commun., vol. 42, no. 5, pp. 1262–1278, May. 2024

    2024

  21. [29]

    Joint deployment and resource allocation for multi-AeBS networks: A two-timescale optimization framework using MADRL,

    Y . Zhaoet al., “Joint deployment and resource allocation for multi-AeBS networks: A two-timescale optimization framework using MADRL,” IEEE Trans. Commun., vol. 73, no. 6, pp. 4272–4289, Jun. 2025

    2025

  22. [30]

    Dynamic role switching scheme with joint trajectory and power control for multi-UA V cooperative secure communication,

    A. Gao, Q. Wang, Y . Hu, W. Liang, and J. Zhang, “Dynamic role switching scheme with joint trajectory and power control for multi-UA V cooperative secure communication,”IEEE Trans. Wireless Commun., vol. 23, no. 2, pp. 1260–1275, Feb. 2023

    2023

  23. [31]

    Fast global optimal power allocation in wireless networks by local DC programming,

    H. H. Kha, H. D. Tuan, and H. H. Nguyen, “Fast global optimal power allocation in wireless networks by local DC programming,”IEEE Trans. Wireless Commun., vol. 11, no. 2, pp. 510–515, Feb. 2011

    2011

  24. [32]

    Nonsmooth optimization for efficient beamforming in cognitive radio multicast transmission,

    A. H. Phan, H. D. Tuan, H. H. Kha, and D. T. Ngo, “Nonsmooth optimization for efficient beamforming in cognitive radio multicast transmission,”IEEE Trans. Signal Process., vol. 60, no. 6, pp. 2941– 2951, Jun. 2012

    2012

  25. [33]

    UA V trajectory and beamforming optimization for integrated periodic sensing and communication,

    K. Meng, Q. Wu, S. Ma, W. Chen, and T. Q. S. Quek, “UA V trajectory and beamforming optimization for integrated periodic sensing and communication,”IEEE Wireless Commun. Lett., vol. 11, no. 6, pp. 1211– 1215, Jun. 2022

    2022

  26. [34]

    Joint 3D beamforming-and-trajectory design for UA V- satellite uplink covert communication,

    J. Yuet al., “Joint 3D beamforming-and-trajectory design for UA V- satellite uplink covert communication,”IEEE Trans. Commun., vol. 73, no. 5, pp. 3469–3481, May. 2025

    2025

  27. [35]

    Joint resource, trajectory, and artificial noise optimization in secure driven 3-D UA Vs with NOMA and imper- fect CSI,

    Y . Li, H. Zhang, and K. Long, “Joint resource, trajectory, and artificial noise optimization in secure driven 3-D UA Vs with NOMA and imper- fect CSI,”IEEE J. Sel. Areas Commun., vol. 39, no. 11, pp. 3363–3377, Nov. 2021

    2021

  28. [36]

    Power allocation in multi- user cellular networks: Deep reinforcement learning approaches,

    F. Meng, P. Chen, L. Wu, and J. Cheng, “Power allocation in multi- user cellular networks: Deep reinforcement learning approaches,”IEEE Trans. Wireless Commun., vol. 19, no. 10, pp. 6255–6267, Oct. 2020

    2020

  29. [37]

    A prompt pattern catalog to enhance prompt engineering with chatgpt,

    J. Whiteet al., “A prompt pattern catalog to enhance prompt engineering with chatgpt,”arXiv:2302.11382

    arXiv

  30. [38]

    Chain-of-thought prompting elicits reasoning in large language models,

    J. Weiet al., “Chain-of-thought prompting elicits reasoning in large language models,”Advances in Neural Inf. Process. Syst. 35 (NeurIPS 2022), vol. 35, pp. 24 824–24 837, 2022

    2022

  31. [39]

    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,” inProc. 35th Int. Conf. Mach. Learn. (ICML), vol. 80. PMLR, 2018, pp. 1861–1870

    2018

  32. [40]

    Stabilizing off- policy Q-learning via bootstrapping error reduction,

    A. Kumar, J. Fu, M. Soh, G. Tucker, and S. Levine, “Stabilizing off- policy Q-learning via bootstrapping error reduction,” inAdvances in Neural Inf. Process. Syst. 32 (NeurIPS 2019), vol. 32, Vancouver, BC, Canada, 2019

    2019

  33. [41]

    Conservative Q-learning for offline reinforcement learning,

    A. Kumar, A. Zhou, G. Tucker, and S. Levine, “Conservative Q-learning for offline reinforcement learning,” inAdvances in Neural Inf. Process. Syst. 33 (NeurIPS 2020), vol. 33, 2020, pp. 1179–1191

    2020

  34. [42]

    DeepSeek-R1: Incentivizing reasoning capability in llms via reinforcement learning,

    D. Guoet al., “DeepSeek-R1: Incentivizing reasoning capability in llms via reinforcement learning,”arXiv:2501.12948

    arXiv

  35. [43]

    Computation offloading and trajectory planning of multi-UA V-enabled MEC: A knowledge-assisted multiagent reinforcement learning approach,

    X. Li, Y . Qin, J. Huo, and W. Huangfu, “Computation offloading and trajectory planning of multi-UA V-enabled MEC: A knowledge-assisted multiagent reinforcement learning approach,”IEEE Trans. Veh. Technol., vol. 73, no. 5, pp. 7077–7088, May. 2023

    2023

  36. [44]

    Modelling the scaling properties of human mobility,

    C. Song, T. Koren, P. Wang, and A. Barab ´asi, “Modelling the scaling properties of human mobility,”Nature Physics, vol. 6, no. 10, pp. 818– 823, Oct. 2010

    2010

Pith tools

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