Pith. sign in

REVIEW 3 major objections 6 minor 2 cited by

Joint User Association and Beamforming Design for ISAC Networks with Large Language Models

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper reports that a prompted large language model can solve the binary user-association subproblem of a multi-BS ISAC network exactly for small networks, matching the brute-force upper bound while convex methods optimize the…

desk verdict A plausible LLM-plus-convex hybrid for small ISAC networks whose empirical claims are interesting but whose guaranteed-convergence proof has an unverified monotonicity assumption. read the letter →

arxiv 2506.05637 v1 pith:YR66FHGK submitted 2025-06-05 cs.IT eess.SPmath.IT

Haoyun Li , Ming Xiao , Kezhi Wang , Robert Schober , Dong In Kim , Yong Liang Guan This is my paper
classification cs.ITeess.SPmath.IT
keywords integratedsensingandcommunicationISAClargelanguagemodelsuserassociationbeamformingalternatingoptimizationpromptengineeringfractionalprogramming
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

Integrated sensing and communication (ISAC) networks must serve communication users and sense targets at the same time, which forces a joint choice of discrete user associations and continuous multi-antenna beamformers. The paper attacks this mixed-integer problem by alternating between a binary user-association subproblem and a beamforming subproblem, and it proposes a large language model as the solver for the integer part while convex methods handle the continuous part. Its central claim, supported by simulations with the GPT-o1 model, is that the LLM-based user association converges within about ten iterations and reaches exactly the same sum rate as exhaustive brute-force search on the tested small networks, so the overall algorithm matches a performance upper bound. The result suggests a division of labor in which LLMs handle combinatorial blocks without retraining while convex solvers handle the smooth blocks.

What carries the argument

The machinery is an alternating optimization that decomposes the mixed-integer problem (P1) into a binary user-association subproblem (P2) and a beamforming subproblem (P3). On the integer side, the load-bearing device is the prompt-engineering pipeline that converts the LLM into a black-box optimizer: it provides in-context wireless expert knowledge, the SINR matrix and objective, a chain-of-thought instruction, and an iterative self-enhancement step where the previous association and its sum rate are fed back until improvement or convergence. On the continuous side, the beamforming subproblem is reformulated through fractional programming (a Lagrangian dual transform that introduces auxiliary variables $\upsilon$ and $b$ to pull the beamformer out of the logarithm), majorization-minimization surrogates for the non-convex CRB-related penalty terms, and an ADMM augmented Lagrangian that enforces the coupling constraints $q_i = Q_i(\mathbf{W}_k)$, producing convex QCQP subproblems solved by standard tools. The convergence argument rests on inequality (61), which asserts that each alternating step does not decrease the objective function.

What would settle it

Run the GPT-o1 prompt pipeline on 100 random channel realizations for the paper's K=3, N=10 setup and for a larger instance (e.g., K=4, N=15), comparing the LLM-chosen user association's sum rate against exhaustive brute-force enumeration: any realization where the LLM is strictly worse than brute force, or where it returns an association worse than the previous iteration, would refute the equality and monotonicity claims.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that a reasoning-oriented large language model can act as an exact-enough black-box optimizer for the binary user-association subproblem when guided by a structured prompt: background description, explicit optimization problem, chain-of-thought instructions, and a self-enhancement loop that feeds back the previous association and its objective value. Wrapped in an alternating optimization with an ADMM-based beamforming solver built from fractional programming and majorization-minimization, the GPT-o1 model is reported to converge within ten iterations and to achieve exactly the sum rate of brute-force search over all user associations for the tested network with three base stations and ten users, a value the paper uses as the upper bound. The same algorithm, fed with conversation-style LLMs such as GPT-4-Turbo, Claude 3.5, and Gemini 2.0, is reported to be unreliable and often worse than fixed-association beamforming, so the paper's performance claim is specifically about reasoning-optimized LLMs.

Load-bearing premise

The convergence proof assumes the language model always returns a user association whose objective value is at least as good as the previous one, but the model is a stochastic black box with no such guarantee and the algorithm provides no fallback if it fails to improve.

Editorial extensions

If this is right

  • The user-association subproblem, nominally combinatorial, is solved to optimality on the tested small networks by a reasoning LLM given the prompt pipeline, so brute-force enumeration is not needed in those regimes.
  • The full algorithm reaches the same sum rate as the convex-plus-brute-force upper bound while satisfying radar SNR and CRB thresholds, meaning the sensing-communication trade-off is handled near-optimally at the tested scale.
  • Convergence within roughly ten iterations makes the hybrid practical for re-optimization as channels or target locations change.
  • Performance depends heavily on the LLM: reasoning-optimized models produce reliable associations, while general conversation models can produce erratic associations that underperform a fixed-association beamforming baseline.

Reading between the lines

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

  • A fallback that keeps the previous association when the LLM fails to improve would turn the monotonicity assumption in inequality (61) into a guaranteed property, making the convergence proof hold for any stochastic LLM; the paper does not propose such a fallback.
  • The reported equality with brute force is demonstrated for a single network size (K=3, N=10); the natural next experiment is to scale N and K and measure how often the LLM still matches exhaustive search, since the search space grows exponentially.
  • The prompt requires the current SINR matrix as input, so the scheme assumes the LLM can be called with fresh numerical data every iteration; the latency and cost of repeated LLM queries in a real-time system are not addressed.
  • If the pattern holds, the framework generalizes to any joint discrete-continuous wireless optimization—power allocation, RIS phase control, spectrum assignment—where an LLM as a combinatorial oracle can be wrapped by convex methods.
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

3 major / 6 minor

Summary. The paper studies a multi-base-station ISAC network and formulates a mixed-integer non-convex problem that jointly optimizes user association (UA) and transmit beamforming to maximize the sum communication rate subject to radar SNR and Cramér-Rao bound constraints. The authors decompose the problem via alternating optimization, propose a prompt-engineering-based LLM black-box solver for the binary UA subproblem, and develop an FP/MM/ADMM-based algorithm for the beamforming subproblem. Simulations compare several LLMs and report that GPT-o1 achieves fast convergence and matches a brute-force-search benchmark.

Significance. If the results hold, the paper offers a modular framework for integrating LLMs with convex optimization for ISAC, with a clearly specified prompt design and a standard convex derivation for the beamforming part. The sensing metric derivations (radar SNR and CRB) are consistent with prior work, and the empirical comparison across four LLMs is useful. The paper is also explicit about the stochasticity of LLM outputs, which is an important caveat. However, the theoretical convergence guarantee is not fully established because the LLM solver lacks a monotonicity guarantee, and the 'upper bound' benchmark is not a true upper bound.

major comments (3)
  1. [Section V.C, Eq. (61)] The proof of monotonic convergence of Algorithm 3 relies on inequality (a), F(W^(j),U^(j)) <= F(W^(j),U^(j+1)), which presumes that Algorithm 1 returns a non-worse UA than the previous iteration. However, Algorithm 1 is a stochastic black-box solver: the while loop (lines 7-9) only instructs the LLM to self-enhance and stops on an undefined 'no convergence' condition, without evaluating the new UA against the previous objective or providing a fallback. The prompt's statement 'The new user association should result in a higher total sum-rate' is a request, not a verification step. Therefore inequality (a) is not guaranteed, and the claim 'Algorithm 3 is guaranteed to converge' is not proven. The authors should modify Algorithm 1 to accept the new UA only if it improves the objective (or keep the previous UA otherwise) and state this explicitly in the proof.
  2. [Section VI.A and VI.B, Fig. 5] The benchmark 'Convex plus BF' is described as using brute-force search to find the optimal UA while optimizing the beamforming with Algorithm 2. Since Algorithm 2 converges to a stationary point of a non-convex problem, the resulting objective value is a feasible solution of (P1) and hence a lower bound on the global optimum, not an upper bound. Calling this benchmark 'a performance upper bound' and reporting that the proposed algorithm 'achieves exactly the same performance' as this bound overstates the near-optimality claim. The authors should either compute a genuine upper bound (e.g., via a relaxation) or rename this benchmark as a feasible reference and adjust the wording in the abstract and Section VI.B accordingly.
  3. [Section V.B.3, Algorithm 2, Step 5] The penalty factor rho is decreased as rho^(j+1)=0.9*rho^(j), and the convergence argument states that the algorithm converges to a stationary point because the search region is bounded and rho shrinks. In augmented Lagrangian and ADMM methods, a decreasing penalty parameter typically weakens the enforcement of the equality constraints (43f) and is not a standard mechanism for guaranteeing convergence to a stationary point of the constrained problem. No proof is given that a limit point satisfies Q_i(W_k)=q_i. Since the monotonicity of the beamforming step is used for inequality (b) in Eq. (61), this gap also affects the overall convergence guarantee. The authors should either justify the decreasing-rho rule with a formal convergence proof or adopt a standard non-decreasing penalty schedule.
minor comments (6)
  1. [Section I] The paragraph 'The remainder of this paper is organized as follows' states that 'Section V provides numerical results' and 'Section VI concludes the paper', but the simulation and conclusion sections are numbered VI and VII, respectively. Please correct the cross-references.
  2. [Equations (31), (47), (48)] The SINR denominators in these equations are missing the squared magnitude notation; for example, (31) writes |h_{k,i}^H w_{k,n}| instead of |h_{k,i}^H w_{k,n}|^2. The same typographical issue appears in (47) and (48).
  3. [Equation (33)] The text says that delta(upsilon,b) 'collects all forms that do not depend on upsilon and b', but the notation suggests a dependence on both variables. The intended meaning is likely 'do not depend on w_k'.
  4. [Section V.A] The paragraph beginning 'Fine-tuning Pre-trained Foundation Models (PFMs)...' appears to be a generic statement about hierarchical 6G coordination that is not well integrated with the surrounding prompt-engineering discussion; consider removing or rewriting it for coherence.
  5. [Section VI.B] The number of independent simulation runs for LLM-based schemes is not reported. Given the paper's own discussion of the stochasticity of LLM sampling, please report the number of trials and include variance or confidence intervals for the reported averages.
  6. [Figure 5] The axes of the convergence plot are not labeled. Adding axis labels (e.g., iteration number and sum rate) would make the figure interpretable without reference to the caption alone.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the beamforming subproblem and the brute-force upper bound are independent of the LLM outputs, and the only self-citation is contextual.

full rationale

The paper's derivation chain is an AO decomposition plus an LLM heuristic for UA and an FP/MM/ADMM ascent for beamforming. Inequality (b) of eq. (61) is supported by Algorithm 2's monotone surrogate construction following standard independent references [52]-[57]. The 'Convex plus BF' upper bound is an exhaustive brute-force search (3^10 associations for the default 10-CU, 3-BS case), not a fitted or self-referential quantity. The GPT-o1 performance result is empirical and externally checkable; the chain-of-thought prompt is a hand-designed heuristic rather than a parameter fitted to the target output. The sole self-citation, [41], appears in the related-work paragraph and is not load-bearing for any theorem or construction in this paper. The genuine weakness is eq. (61)(a): the proof asserts that Algorithm 1 returns a non-worse UA because it 'solves' P2, while Algorithm 1's self-enhancement loop contains no verification or fallback. This is an unproven convergence premise and a correctness risk, but it is not circularity: it does not make the claimed result equal to its input by construction. No quoted equation reduces to its own input, so the circularity score is 0.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The algorithm introduces no new physical entities. It relies on standard domain simplifications (orthogonal BS frequency bands, full-duplex, LoS sensing, point targets) and a few hand-chosen algorithm hyperparameters (ADMM penalty factor, convergence thresholds, LLM sampling and self-refinement settings). The chain-of-thought hint in the prompt is a hand-crafted heuristic rather than a fitted parameter.

free parameters (3)
  • ADMM penalty factor rho = not specified (rho^(0) >= 0, rho^(j+1) = 0.9 rho^(j))
    Chosen by hand; governs the augmented Lagrangian term and affects the convergence path of the beamforming subproblem; no initial value or sensitivity study is provided.
  • Convergence thresholds epsilon and xi = not reported
    Termination thresholds for the inner ADMM loop (epsilon) and the outer AO loop (xi); chosen by hand, no values given.
  • LLM sampling parameters (temperature, top-p) and number of self-refinement iterations = not reported
    These control the stochastic output of the LLM and thus directly influence whether the self-enhancement loop produces an improving UA; none are specified, so the exact behavior is not reproducible.
assumptions (5)
  • domain assumption Each ISAC BS employs a different orthogonal frequency band to eliminate inter-BS interference.
    Section II.A and used in (8) where the SINR contains only one BS's signals; this removes inter-cell interference and decouples the beamforming subproblems across BSs.
  • domain assumption Full-duplex radio allows simultaneous sensing echo reception and downlink communication without self-interference.
    Section II, stated without modeling self-interference; the sensing model in (9) assumes the BS can receive its own echoes cleanly.
  • standard math For large M, 1/M a^H(phi_i) a(phi_i') is approximately 0 for i != i' and 1/M a^H(phi_i) a(phi_i) is approximately 1.
    Section III.A, used to reduce the detection problem to a scalar chi-square test; requires sufficiently large antenna arrays.
  • standard math 1/L sum_{l=1}^L s_k[l] s_k[l]^H is approximately I_{N+M} for large L.
    Appendix A, after eq. (64); used to simplify the FIM expressions; valid for independent unit-energy symbols over many samples.
  • standard math The CRB constraint (27b) can be transformed via the matrix C and q auxiliary variables into the PSD constraint (43d) following [54].
    Section V.B.2, eqs. (39)-(43); the equivalence requires the FIM block structure and is borrowed from [15]/[54].

how reviews work

0 comments
Cite this review

Pith. "Pith review of Joint User Association and Beamforming Design for ISAC Networks with Large Language Models." pith.science (2026). https://pith.science/paper/YR66FHGK

@misc{pith2026250605637,
  author       = {Pith},
  title        = {Pith review of: Joint User Association and Beamforming Design for ISAC Networks with Large Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YR66FHGK}},
  note         = {Machine review of arXiv:2506.05637}
}
read the original abstract

Integrated sensing and communication (ISAC) has been envisioned to play a more important role in future wireless networks. However, the design of ISAC networks is challenging, especially when there are multiple communication and sensing (C\&S) nodes and multiple sensing targets. We investigate a multi-base station (BS) ISAC network in which multiple BSs equipped with multiple antennas simultaneously provide C\&S services for multiple ground communication users (CUs) and targets. To enhance the overall performance of C\&S, we formulate a joint user association (UA) and multi-BS transmit beamforming optimization problem with the objective of maximizing the total sum rate of all CUs while ensuring both the minimum target detection and parameter estimation requirements. To efficiently solve the highly non-convex mixed integer nonlinear programming (MINLP) optimization problem, we propose an alternating optimization (AO)-based algorithm that decomposes the problem into two sub-problems, i.e., UA optimization and multi-BS transmit beamforming optimization. Inspired by large language models (LLMs) for prediction and inference, we propose a unified framework integrating LLMs with convex-based optimization methods. First, we propose a comprehensive design of prompt engineering, including few-shot, chain of thought, and self-reflection techniques to guide LLMs in solving the binary integer programming UA optimization problem. Second, we utilize convex-based optimization methods to handle the non-convex beamforming optimization problem based on fractional programming (FP), majorization minimization (MM), and the alternating direction method of multipliers (ADMM) with an optimized UA from LLMs. Numerical results demonstrate that our proposed LLM-enabled AO-based algorithm achieves fast convergence and near upper-bound performance with the GPT-o1 model, outperforming various benchmark schemes.

Figures

Figures reproduced from arXiv: 2506.05637 by the authors.

Figure 1
Figure 1. FIGURE 1 [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. FIGURE 2 [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. FIGURE 3 [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: FIGURE 4 [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]
Figure 5
Figure 5. Figure 5: FIGURE 5 [PITH_FULL_IMAGE:figures/full_fig_p017_5.png]
Figure 8
Figure 8. Figure 8: FIGURE 8 [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]
Figure 7
Figure 7. Figure 7: FIGURE 7 [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]
Figure 9
Figure 9. Figure 9: FIGURE 9 [PITH_FULL_IMAGE:figures/full_fig_p018_9.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

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

    cs.NI 2025-07 conditional novelty 5.0 of 10

    A hierarchical framework combining semidefinite relaxation for secure precoding with LLM-guided multi-agent reinforcement learning for drone trajectories is claimed to outperform existing baselines in secrecy rate and...

  2. AI Reasoning for Wireless Communications and Networking: A Survey and Perspectives

    cs.NI 2025-09 conditional novelty 4.0 of 10

    A survey that organizes LLM and AI reasoning methods into a taxonomy and maps them onto the physical, link, network, transport, and application layers of wireless networks.

Reference graph

Works this paper leans on

59 extracted references · 50 canonical work pages · cited by 2 Pith papers

  1. [1]

    Joint radar and communication design: Applications, state-of-the-art, and the road ahead,

    F. Liu, C. Masouros, A. P. Petropulu, H. Griffiths, and L. Hanzo, “Joint radar and communication design: Applications, state-of-the-art, and the road ahead,”IEEE Trans. Commun., vol. 68, no. 6, pp. 3834–3862, Jun. 2020

    2020

  2. [2]

    Space-air-ground integrated wireless networks for 6G: Basics, key technologies, and future trends,

    Y . Xiao, Z. Ye, M. Wu, H. Li, M. Xiao, M. S. Alouini, A. Al- Hourani, and S. Cioni, “Space-air-ground integrated wireless networks for 6G: Basics, key technologies, and future trends,”IEEE J. Sel. Areas Commun., vol. 42, no. 12, pp. 3327–3354, Dec. 2024. VOLUME , 19 H. Liet al.: Joint User Association and Beamforming Design for ISAC Networks with Large L...

    work page 2024

  3. [3]

    Integrated Sensing and Communication: Towards dual- functional wireless networks for 6G and beyond,

    F. Liu, Y . Cui, C. Masouros, J. Xu, T. Han, Y . Eldar, and S. Buzzi, “Integrated Sensing and Communication: Towards dual- functional wireless networks for 6G and beyond,”IEEE J. Sel. Areas Commun., vol. 40, no. 6, pp. 1728–1767, Jun. 2022

    work page 2022

  4. [4]

    Waveform design and signal processing aspects for fusion of wireless communications and radar sensing,

    C. Sturm and W. Wiesbeck, “Waveform design and signal processing aspects for fusion of wireless communications and radar sensing,”Proc. IEEE, vol. 99, no. 7, pp. 1236–1259, Jul. 2011

    work page 2011

  5. [5]

    JCR70: A low-complexity millimeter-wave proof-of-concept platform for a fully-digital SIMO joint communication-radar,

    P. Kumari, A. Mezghani, and R. W. Heath, “JCR70: A low-complexity millimeter-wave proof-of-concept platform for a fully-digital SIMO joint communication-radar,”IEEE Open J. Veh. Technol., vol. 2, pp. 218–234, 2021

    work page 2021

  6. [6]

    Increasing capacity in wireless broadcast systems using distributed transmission/directional reception (DTDR),

    A. J. Paulraj and T. Kailath, “Increasing capacity in wireless broadcast systems using distributed transmission/directional reception (DTDR),” U.S. Patent 5 345 599, Sep. 6, 1994

    work page 1994

  7. [7]

    High resolution FDMA MIMO radar,

    D. Cohen, D. Cohen, and Y . C. Eldar, “High resolution FDMA MIMO radar,”IEEE Trans. Aerosp. Electron. Syst., vol. 56, no. 4, pp. 2806– 2822, Aug. 2020

    work page 2020

  8. [8]

    Extremely large aperture array (ELAA) communications: Foundations, research advances and challenges,

    S. Ye, M. Xiao, M.W. Kwan, Z. Ma, Y . Huang, G. Karagiannidis, and P. Fan, “Extremely large aperture array (ELAA) communications: Foundations, research advances and challenges,”IEEE Open Journal of the Commun. Soc., vol. 5, pp. 7075–7120, Oct. 2024

    work page 2024

Show all 59 references
  1. [10]

    Joint maneuver and beamforming design for UA V-enabled integrated sensing and communication,

    Z. Lyu, G. Zhu, and J. Xu, “Joint maneuver and beamforming design for UA V-enabled integrated sensing and communication,”IEEE Trans. Wireless Commun., vol. 22, no. 4, pp. 2424–2440, Apr. 2023

    2023

  2. [11]

    Joint beamforming and UA V trajectory optimization for covert communications in ISAC networks

    D. Deng, W. Zhou, X. Li, D. B. da Costa, D. W. K. Ng, and A. Nallanathan, “Joint beamforming and UA V trajectory optimization for covert communications in ISAC networks”,IEEE Trans. Wireless Commun., vol.24, no.2, pp. 1016–1030, Feb. 2025

    2025

  3. [12]

    MU-MIMO communications with MIMO radar: From co-existence to joint transmis- sion,

    F. Liu, C. Masouros, A. Li, H. Sun, and L. Hanzo, “MU-MIMO communications with MIMO radar: From co-existence to joint transmis- sion,”IEEE Trans. Wireless Commun., vol. 17, no. 4, pp. 2755–2770, Apr. 2018

    2018

  4. [13]

    Joint transmit beamforming for multiuser MIMO communications and MIMO radar,

    X. Liu, T. Huang, N. Shlezinger, Y . Liu, J. Zhou, and Y . C. Eldar, “Joint transmit beamforming for multiuser MIMO communications and MIMO radar,”IEEE Trans. Signal Process., vol. 68, pp. 3929–3944, 2020

    2020

  5. [14]

    Communication-sensing region for cell-free massive MIMO ISAC systems,

    W. Mao, Y . Lu, C. Y . Chi, B. Ai, Z. Zhong, and Z. Ding, “Communication-sensing region for cell-free massive MIMO ISAC systems,”IEEE Trans. Wireless Commun., vol. 23, no. 9, pp. 12396– 12411, Sept. 2024

    2024

  6. [15]

    SNR/CRB- constrained joint beamforming and reflection designs for RIS-ISAC systems,

    R. Liu, M. Li, Q. Liu, and A. Lee Swindlehurst, “SNR/CRB- constrained joint beamforming and reflection designs for RIS-ISAC systems,”IEEE Trans. Wireless Commun., vol. 23, no. 7, pp. 7456– 7470, Jul. 2024

    2024

  7. [16]

    Cram ´er-Rao bound optimiza- tion for active RIS-empowered ISAC systems,

    Q. Zhu, M. Li, R. Liu, and Q. Liu, “Cram ´er-Rao bound optimiza- tion for active RIS-empowered ISAC systems,”IEEE Trans. Wireless Commun., vol. 23, no. 9, pp. 11723–11736, Sept. 2024

    2024

  8. [17]

    Learning- based predictive beamforming for integrated sensing and communica- tion in vehicular networks,

    C. Liu, W. Yuan, S. Li, X. Liu, H. Li, and D. W. K. Ng, “Learning- based predictive beamforming for integrated sensing and communica- tion in vehicular networks,”IEEE J. Sel. Areas Commun., vol. 40, no. 8, pp. 2317–2334, Aug. 2022

    2022

  9. [18]

    Predictive beamforming for vehicles with complex behaviors in ISAC systems: A deep learning approach,

    X. Zhang, W. Yuan, C. Liu, J. Wu, and D. W. K. Ng, “Predictive beamforming for vehicles with complex behaviors in ISAC systems: A deep learning approach,”IEEE J. Sel. Top. Signal Process., vol. 18, no. 5, pp. 828–841, Jul. 2024

    2024

  10. [19]

    ISAC- NET: Model-driven deep learning for integrated passive sensing and communication,

    W. Jiang, D. Ma, Z. Wei, Z. Feng, P. Zhang, and J. Peng, “ISAC- NET: Model-driven deep learning for integrated passive sensing and communication,”IEEE Trans. Wireless Commun.vol. 72, no. 8, pp. 4692–4707, Aug. 2024

    2024

  11. [20]

    Fluid antenna system liberating multiuser MIMO for ISAC via deep reinforcement learning,

    C. Wang, G. Li, H. Zhang, K. K. Wong, Z. Li, and D, W. K. Ng, “Fluid antenna system liberating multiuser MIMO for ISAC via deep reinforcement learning,”IEEE Trans. Wireless Commun., vol. 23, no. 9, pp. 10879–10894, Sept. 2024

    2024

  12. [21]

    Hybrid deep reinforcement learning for enhancing localization and communication efficiency in RIS-aided cooperative ISAC systems,

    P. Saikia, K. Singh, W. J. Huang, and T. Q. Duong, “Hybrid deep reinforcement learning for enhancing localization and communication efficiency in RIS-aided cooperative ISAC systems,”IEEE Internet Things J, vol. 11, no. 18, pp. 29494–29510, Sept. 2024

    2024

  13. [22]

    Deep reinforcement learning-based resource allocation for integrated sensing, communi- cation, and computation in vehicular network,

    L. Yang, Y . Wei, Z. Feng, Q. Zhang, and Z. Han, “Deep reinforcement learning-based resource allocation for integrated sensing, communi- cation, and computation in vehicular network,”IEEE Trans. Wireless Commun., vol. 23, no. 12, pp. 18608–18622, Dec. 2024

    2024

  14. [23]

    Semidefinite relaxation of quadratic optimization problems,

    Z. Q. Luo, W. K. Ma, A. M. So, Y . Ye, and S. Zhang, “Semidefinite relaxation of quadratic optimization problems,”IEEE Signal Process. Mag., vol. 27, no. 3, pp. 20–34, May 2010

    2010

  15. [24]

    Deep learning,

    Y . LeCun, Y . Bengio, and G. Hinton, “Deep learning,”Nature, vol. 521, no. 7553, pp. 436–444, Nov. 2015

    2015

  16. [25]

    A survey of multi-objective optimization in wireless sensor networks: Metrics, algorithms, and open problems,

    Z. Fei, B. Li, S. Yang, C. Xing, H. Chen, and L. Hanzo, “A survey of multi-objective optimization in wireless sensor networks: Metrics, algorithms, and open problems,”IEEE Commun. Surveys Tuts., vol. 19, no. 1, pp. 550–586, 1st Quart., 2017

    2017

  17. [26]

    Knowledge-driven deep learning paradigms for wireless network optimization in 6G,

    R. Sun, N. Cheng, C. Li, F. Chen, and W. Chen, “Knowledge-driven deep learning paradigms for wireless network optimization in 6G,”IEEE Netw., vol. 38, no. 2, pp. 70–78, Mar. 2024

    2024

  18. [27]

    ChatGLM-Math: Improving math problem-solving in large language models with a self-critique pipeline

    Y . Xu et al., “ChatGLM-Math: Improving math problem-solving in large language models with a self-critique pipeline”, inProc. Conf. Empir. Methods Nat. Lang. Process. (EMNLP), pp. 9733–9760, Nov. 2024

    2024

  19. [28]

    Large language models as optimizers

    C. Yang, X. Wang, Y . Lu, H. Liu, Q. V . Le, D. Zhou, and X. Chen, “Large language models as optimizers”, inProc. Int. Conf. Learn. Representations (ICLR), pp. 1–22, Jan. 2024

    2024

  20. [29]

    Large language model for multiobjective evolutionary optimization

    F. Liu, X. Lin, S. Yao, Z. Wang, X. Tong, M. Yuan, and Q. Zhang, “Large language model for multiobjective evolutionary optimization”, inProc. Int. Conf. Evolutionary Multi-Criterion Optimization (EMO), pp. 178–191, Feb. 2025

    2025

  21. [30]

    Large language models to enhance bayesian optimization

    T. Liu, N. Astorga, N. Seedat, and M. Schaar, “Large language models to enhance bayesian optimization”, inProc. Int. Conf. Learn. Representations (ICLR), pp. 1–33, Jan. 2024

    2024

  22. [31]

    Leveraging large language models to develop heuristics for emerging optimization problems

    T. B ¨omer, N. Koltermann, M. Disselnmeyer, L. D ¨orr, and A. Meyer, “Leveraging large language models to develop heuristics for emerging optimization problems”, 2025. [Online]. Available: https://arxiv.org/abs/2503.03350

    2025 arXiv

  23. [32]

    Leveraging large language models for the generation of novel meta- heuristic optimization algorithms,

    M. Pluhacek, A. Kazikova, T. Kadavy, A. Viktorin, and R. Senkerik, “Leveraging large language models for the generation of novel meta- heuristic optimization algorithms,” inProc. Companion Conf. Genetic and Evolutionary Computation Conf. (GECCO), pp. 1812–1820, Jul. 2023

    2023

  24. [33]

    A survey of large language models,

    W. X. Zhao et al., “A survey of large language models,” 2023. [Online]. Available: https://arxiv.org/abs/2303.18223

    2023 arXiv

  25. [34]

    Large language model (LLM) for telecommunications: A comprehensive survey on principles, key techniques, and opportuni- ties,

    H. Zhou et al., “Large language model (LLM) for telecommunications: A comprehensive survey on principles, key techniques, and opportuni- ties,”IEEE Commun. Surv. Tutorials., Early Access, 2024

    2024

  26. [35]

    WirelessLLM: Empowering large language models towards wireless intelligence,

    J. Shao et al., “WirelessLLM: Empowering large language models towards wireless intelligence,”J. Commun. Inf. Netw., vol. 9, no. 2, pp. 99–112, Jun. 2024

    2024

  27. [36]

    WirelessAgent: Large language model agents for intelligent wireless networks,

    J. Tong, J. Shao, Q. Wu, W. Guo, Z. Li, Z. Lin, and J. Zhang, “WirelessAgent: Large language model agents for intelligent wireless networks,” 2024, [Online]. Available: https://arxiv.org/abs/2409.07964

    2024 arXiv

  28. [37]

    Large language model agents for radio map generation and wireless network planning,

    H. Quan, W. Ni, T. Zhang, X. Ye, Z. Xie, S. Wang, Y . Liu, and H. Song, “Large language model agents for radio map generation and wireless network planning,”IEEE Networking Letters, Early Access, 2025

    2025

  29. [38]

    Large language model-based wireless network design,

    K. Qiu, S. Bakirtzis, I. Wassell, H. Song, J. Zhang, and K. Wang, “Large language model-based wireless network design,”IEEE Wireless Commun. Lett., vol. 13, no. 12, pp. 3340–3344, Dec. 2024

    2024

  30. [39]

    Toward intelligent commu- nications: Large model empowered semantic communications,

    H. Xie, Z. Qin, X. Tao, and Z. Han, “Toward intelligent commu- nications: Large model empowered semantic communications,”IEEE Commun. Mag., vol. 63, no. 1, pp. 69–75, Jan. 2025

    2025

  31. [40]

    Large language models empowered autonomous edge AI for connected intelligence,

    Y . Shen, J. Shao, X. Zhang, Z. Lin, H. Pan, D. Li, J. Zhang, and KB. Letaief, “Large language models empowered autonomous edge AI for connected intelligence,”IEEE Commun. Mag., vol. 62, no. 10, pp. 140–146, Oct. 2024

    2024

  32. [41]

    Large language model based multi-objective optimization for integrated sensing and communication 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 communication in UA V networks,”IEEE Wireless Commun. Lett., vol. 14, no. 4, pp. 979–983, Apr. 2025

    2025

  33. [42]

    Radar-assisted predictive beamforming for vehicular links: Communication served by sensing,

    F. Liu, W. Yuan, C. Masouros, and J. Yuan, “Radar-assisted predictive beamforming for vehicular links: Communication served by sensing,” IEEE Trans. Wireless Commun., vol. 19, no. 11, pp. 7704–7719, Nov. 2020

    2020

  34. [43]

    Full duplex radio/radar technology: The enabler for advanced joint communication and sensing,

    C. B. Barneto, S. D. Liyanaar achchi, M. Heino, T. Riihonen, and M. Valkama, “Full duplex radio/radar technology: The enabler for advanced joint communication and sensing,”IEEE Wireless Commun., vol. 28, no. 1, pp. 82–88, Feb. 2021

    2021

  35. [44]

    Joint power allocation and user association optimization for IRS-assisted mmWave systems,

    D. Zhao, H. Lu, Y . Wang, H. Sun, and Y . Gui, “Joint power allocation and user association optimization for IRS-assisted mmWave systems,” IEEE Wireless Commun., vol. 21, no. 1, pp. 577–590, Jan. 2022. 20 VOLUME ,

    2022

  36. [45]

    LTE: the Evolution of Mobile Broadband,

    D. Astely, E. Dahlman, A. Furuskr, Y . Jading, M. Lindstrm, and S. Parkvall, “LTE: the Evolution of Mobile Broadband,”IEEE Commun. Mag., vol. 47, no. 4, pp. 44–51, Apr. 2009

    2009

  37. [46]

    Target detection and localization using MIMO radars and sonars,

    I. Bekkerman and J. Tabrikian, “Target detection and localization using MIMO radars and sonars,”IEEE Trans. Signal Process., vol. 54, no. 10, pp. 3873–3883, Oct. 2006

    2006

  38. [47]

    T. L. Marzetta,Fundamentals of Massive MIMO. Cambridge, U.K.: Cambridge Univ. Press, 2016

    2016

  39. [48]

    S. M. Kay,Fundamentals of Statistical Signal Processing: Detection Theory, vol. 1. Englewood Cliffs, NJ, USA: Prentice–Hall, 1998

    1998

  40. [49]

    Language models are few-shot learners,

    T. B. Brown et al., “Language models are few-shot learners,” inProc. Adv. Neural Inf. Process. Syst., 2020, pp. 1877–1901

    2020

  41. [50]

    Chain of thought prompting elicits reasoning in large language models,

    J. Wei, et al., “Chain of thought prompting elicits reasoning in large language models,”in Proc. Adv. Neural Inf. Process. Syst., 2022, vol. 35, pp. 24824–24837

    2022

  42. [51]

    Reflexion: Language agents with verbal reinforcement learning,

    N. Shinn, F. Cassano, B. Labash, A. Gopinath, K. Narasimhan, and S. Yao, “Reflexion: Language agents with verbal reinforcement learning,”

  43. [52]

    Fractional programming for communication systems–Part I: Power control and beamforming,

    K. Shen and W. Yu, “Fractional programming for communication systems–Part I: Power control and beamforming,”IEEE Trans. Signal Process., vol. 66, no. 10, pp. 2616–2630, May 2018

    2018

  44. [53]

    Fractional programming for communication systems–Part II: Uplink scheduling via matching,

    K. Shen and W. Yu, “Fractional programming for communication systems–Part II: Uplink scheduling via matching,”IEEE Trans. Signal Process., vol. 66, no. 10, pp. 2631–2644, May 2018

    2018

  45. [54]

    STARS enabled integrated sensing and communication,

    Z. Wang, X. Mu, and Y . Liu, “STARS enabled integrated sensing and communication,”IEEE Trans. Wireless Commun., vol. 22, no. 10, pp. 6750–6765, Oct. 2023

    2023

  46. [55]

    On augmented Lagrangian methods with general lower-level constraints,

    R. Andreani, E. G. Birgin, J. M. Martinez, and M. L. Schuverdt, “On augmented Lagrangian methods with general lower-level constraints,” SIAM J. Optim., vol. 18, no. 4, pp. 1286–1309, Nov. 2007

    2007

  47. [56]

    CVX: Matlab software for disciplined convex programming, version 2.1,

    M. Grant and S. Boyd, “CVX: Matlab software for disciplined convex programming, version 2.1,” http://cvxr.com/cvx, Mar. 2014

    2014

  48. [57]

    Majorization–minimization al- gorithms in signal processing, communications, and machine learning,

    Y . Sun, P. Babu, and D. P. Palomar, “Majorization–minimization al- gorithms in signal processing, communications, and machine learning,” IEEE Trans. Signal Process., vol. 65, no. 3, pp. 794–816, Feb. 2017

    2017

  49. [58]

    Machiavelli and the gale-shapley algorithm,

    L. E. Dubins and D. A. Freedman, “Machiavelli and the gale-shapley algorithm,”The American Mathematical Monthly, vol. 88, no. 7, pp. 485–494, 1981

    1981

  50. [59]

    User association and power control for UA V-enabled cellular networks,

    M. Sami and J. N. Daigle, “User association and power control for UA V-enabled cellular networks,”IEEE Wireless Commun. Lett., vol. 9, no. 3, pp. 267–270, Mar. 2020. VOLUME , 21

    2020

  51. [2023]

    Available: https://arxiv.org/abs/2303.11366

    [Online]. Available: https://arxiv.org/abs/2303.11366

    arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.