arxiv: 2604.10372 · v1 · submitted 2026-04-11 · 📡 eess.SP

Recognition: unknown

LLM-enabled Antenna Partitioning and Beamforming Optimization for Segmented Pinching

Qian Gao , Ruikang Zhong , Hyundong Shin , Yuanwei Liu

Authors on Pith no claims yet

Pith reviewed 2026-05-10 15:08 UTC · model grok-4.3

classification 📡 eess.SP

keywords ISACpinching antennaLLMbeamforming optimizationantenna partitioningself-graphintegrated sensing and communicationtransfer learning

0 comments

The pith

An LLM framework with self-graphs jointly optimizes antenna deployment, partitioning, and beamforming for adaptable ISAC using segmented pinching antennas.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a learning method to handle the complex joint optimization in segmented pinching antenna systems for integrated sensing and communication, where user and target numbers can vary. It builds a CSI-induced self-graph to create representations of interactions that stay the same regardless of order, then uses a large language model backbone with two heads to predict the needed configurations and beamforming vectors. A transfer mechanism tests if the deployment policy works across different user counts. Results indicate better communication rates and steady sensing performance, with the policy holding up well enough to avoid retraining the whole model.

Core claim

We propose a general learning framework for segmented pinching antenna-assisted ISAC systems. A channel state information (CSI)-induced self-graph is constructed to produce permutation-invariant representations of user-target interactions, and the resulting features are processed by a large language model (LLM) backbone with two task-specific heads for jointly predicting antenna deployment, segment partitioning, and ISAC beamforming. A user count transfer mechanism examines whether the learned deployment policy is site-specific and reusable under changed user configurations.

What carries the argument

CSI-induced self-graph for permutation-invariant user-target representations, processed by an LLM backbone with task-specific heads for joint prediction of deployment, partitioning, and beamforming.

If this is right

Simulation results demonstrate higher communication rates while maintaining reliable sensing accuracy.
The learned deployment policy remains highly stable when transferring to other user counts.
This stability reduces the training cost from full model retraining to only beamforming head adaptation.
The framework adapts effectively to changing numbers of communication users and sensing targets.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

This method could extend to other dynamic wireless environments where user numbers fluctuate frequently.
Reducing retraining needs might make ISAC systems more practical for real-world deployment in varying conditions.
Similar graph-based LLM approaches might apply to other optimization problems in signal processing involving variable numbers of entities.

Load-bearing premise

The CSI-induced self-graph produces permutation-invariant representations of user-target interactions that enable the LLM to generalize effectively across varying numbers of users and targets.

What would settle it

Observing in simulations that performance degrades significantly or the policy transfer fails when the number of users changes substantially from the training configuration.

Figures

Figures reproduced from arXiv: 2604.10372 by Hyundong Shin, Qian Gao, Ruikang Zhong, Yuanwei Liu.

**Figure 1.** Figure 1: Illustration of the SWAN-ISAC system.. producing permutation-invariant scenario representations for variable user settings. The graph-enhanced representation is then processed by an LLM backbone, followed by two taskspecific output heads for antenna deployment and beamforming prediction, respectively. In this way, the proposed framework jointly predicts antenna deployment, segment partitioning, and comm… view at source ↗

**Figure 2.** Figure 2: The proposed LLM-enabled general learning framewor [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: Convergence behavior and main benchmark comparison [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

**Figure 4.** Figure 4: Example of the predicted deployment and segment [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗

**Figure 6.** Figure 6: Achievable rate under different communication powe [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗

**Figure 7.** Figure 7: Achievable rate under imperfect CSI with different [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗

read the original abstract

Integrated sensing and communication (ISAC) requires spatial architectures that can flexibly balance data transmission and environment sensing. Segmented pinching antenna-assisted ISAC provides such flexibility by allowing different waveguide segments to be dynamically configured for transmission and reception. However, its design involves the joint optimization of antenna deployment, segment partitioning, and beamforming under coupled communication and sensing constraints, which becomes particularly challenging when the numbers of communication users and sensing targets vary across scenarios. To endow the system with stronger adaptability to changing user and target configurations, we propose a general learning framework for segmented pinching antenna-assisted ISAC systems. Specifically, a channel state information (CSI)-induced self-graph is constructed to produce permutation-invariant representations of user-target interactions, and the resulting features are processed by a large language model (LLM) backbone with two task-specific heads for jointly predicting antenna deployment, segment partitioning, and ISAC beamforming. In addition, a user count transfer mechanism is developed to examine whether the learned deployment policy is site-specific and reusable under changed user configurations. Simulation results show that the proposed framework achieves higher communication rates while maintaining reliable sensing accuracy. Moreover, the learned deployment policy remains highly stable when transferring to other user counts, which reduces the training cost from full model retraining to beamforming head adaption.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper puts forward a CSI self-graph plus LLM setup for joint antenna deployment, partitioning, and beamforming in segmented pinching ISAC, plus a transfer trick for new user counts, but the simulation backing for the claimed stability and cost savings is thin.

read the letter

The core idea is to build a CSI-induced self-graph that feeds an LLM backbone with separate heads for antenna deployment, segment partitioning, and beamforming, then add a user-count transfer step so the deployment policy stays usable and only the beamforming head needs retraining when the number of users or targets changes. That combination is new for this specific segmented pinching ISAC setting and directly targets the practical problem of varying user-target counts without starting from scratch each time. The self-graph construction for permutation-invariant features is a sensible choice to keep the input structure flexible, and splitting the LLM into task-specific heads gives a clean way to handle the coupled objectives. The reported simulations indicate better communication rates alongside stable sensing performance, and the deployment policy holds up across different user counts, which would cut training cost if it works as described. Those are the parts worth paying attention to if you work on ML-driven ISAC or reconfigurable antennas. The main soft spot is that the simulation results lack any description of baselines, statistical tests, error bars, or exact channel and noise models, so it is hard to tell how much of the gain comes from the graph-LLM design versus other modeling choices. The transfer claim rests on the graph actually delivering count-independent representations; without explicit checks like ablation on graph construction or side-by-side prediction comparisons before and after a count change, it is possible the LLM still encodes count-specific patterns and the head-only adaptation would not deliver the promised savings. This is aimed at researchers already working on graph or LLM methods for wireless optimization who want a concrete example for ISAC adaptability. It deserves a serious referee because it offers a full pipeline with empirical results on a relevant problem, even though the evidence on the transfer mechanism needs tightening. I would send it to review and ask the authors to add those checks on invariance and simulation rigor.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes a learning framework for segmented pinching antenna-assisted ISAC systems. A CSI-induced self-graph is constructed to yield permutation-invariant representations of user-target interactions; these features are fed to an LLM backbone with separate heads that jointly predict antenna deployment, segment partitioning, and beamforming. A user-count transfer mechanism is introduced that reuses the learned deployment policy across different user counts by adapting only the beamforming head. Simulation results are reported to show higher communication rates, maintained sensing accuracy, and stable transfer performance that reduces retraining cost.

Significance. If the simulation results prove reproducible and the transfer mechanism is verified to hold without count-dependent artifacts, the framework could provide a practical, lower-cost approach for adapting ISAC systems to varying user and target counts. The combination of graph-based permutation invariance with an LLM backbone for joint ISAC optimization is a novel direction worth exploring, though its advantages over conventional optimization or simpler ML baselines remain to be quantified.

major comments (2)

[Abstract] Abstract: The performance claims rest entirely on simulation results, yet no description is given of the simulation setup, channel models, number of users/targets, baselines (e.g., conventional joint optimization or other learning methods), Monte Carlo trial count, or error bars. Without these details it is impossible to evaluate whether the reported gains in communication rate and sensing accuracy are statistically meaningful or robust.
[Abstract] Abstract: The user-count transfer claim—that the deployment policy remains stable so that only the beamforming head needs adaptation—depends on the CSI self-graph producing representations whose structure and content are independent of the exact number of users and targets. The abstract provides no concrete description of the graph construction (e.g., local pairwise CSI processing, fixed-size pooling, or masking), leaving open the possibility that count-specific patterns are encoded and thereby invalidating the asserted reduction from full retraining to head-only adaptation.

minor comments (1)

[Abstract] The final sentence contains the typo 'adaption'; it should read 'adaptation'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We agree that the abstract would benefit from additional details on the evaluation and the graph construction to better support the performance and transfer claims. We address each point below and will revise the abstract in the next version.

read point-by-point responses

Referee: [Abstract] Abstract: The performance claims rest entirely on simulation results, yet no description is given of the simulation setup, channel models, number of users/targets, baselines (e.g., conventional joint optimization or other learning methods), Monte Carlo trial count, or error bars. Without these details it is impossible to evaluate whether the reported gains in communication rate and sensing accuracy are statistically meaningful or robust.

Authors: We agree that the abstract does not contain these specifics. The full manuscript provides them in Section IV (simulation setup with 3GPP channel models, user counts from 4 to 8 and targets from 2 to 4, baselines including alternating optimization and GNN predictors, 1000 Monte Carlo trials, and error bars as standard deviations). We will revise the abstract to include a concise summary of these elements so that the statistical robustness of the reported gains can be assessed directly from the abstract. revision: yes
Referee: [Abstract] Abstract: The user-count transfer claim—that the deployment policy remains stable so that only the beamforming head needs adaptation—depends on the CSI self-graph producing representations whose structure and content are independent of the exact number of users and targets. The abstract provides no concrete description of the graph construction (e.g., local pairwise CSI processing, fixed-size pooling, or masking), leaving open the possibility that count-specific patterns are encoded and thereby invalidating the asserted reduction from full retraining to head-only adaptation.

Authors: The manuscript details the CSI-induced self-graph in Section III-A: edges are formed via local pairwise CSI processing between users and targets, followed by fixed-size pooling that enforces permutation invariance without dependence on the exact user or target count. This design avoids encoding count-specific patterns and thereby supports the stability of the deployment policy. We will add a brief clause to the abstract describing this construction method to clarify the basis for the transfer claim. revision: yes

Circularity Check

0 steps flagged

No circularity: framework uses external CSI inputs and standard ML components for empirical predictions

full rationale

The paper describes a learning framework that constructs a CSI-induced self-graph from external channel inputs, feeds permutation-invariant features to an LLM backbone, and uses task-specific heads to output antenna deployment, partitioning, and beamforming. These outputs are model predictions trained and validated on simulations, not derivations that reduce to the inputs by construction. The user-count transfer claim is presented as an empirical observation from simulations (deployment policy stability reduces retraining to head adaptation), not a mathematical equivalence or fitted parameter renamed as prediction. No self-citations, uniqueness theorems, or ansatzes are invoked as load-bearing; the graph construction is a deliberate design for invariance rather than a self-referential loop. The derivation chain remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract does not identify any free parameters, axioms, or invented entities; the approach relies on standard machine-learning components applied to wireless optimization.

pith-pipeline@v0.9.0 · 5532 in / 1100 out tokens · 43431 ms · 2026-05-10T15:08:17.172865+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

31 extracted references · 3 canonical work pages

[1]

Integrated sensing and communications: Toward dual-func tional wire- less networks for 6G and beyond,

F. Liu, Y . Cui, C. Masouros, J. Xu, T. X. Han, Y . C. Eldar, an d S. Buzzi, “Integrated sensing and communications: Toward dual-func tional wire- less networks for 6G and beyond,” IEEE J. Sel. Areas Commun. , vol. 40, no. 6, pp. 1728–1767, 2022

2022
[2]

MARL-based UA V traj ectory and beamforming optimization for ISAC system,

Q. Gao, R. Zhong, H. Shin, and Y . Liu, “MARL-based UA V traj ectory and beamforming optimization for ISAC system,” IEEE Internet Things J., vol. 11, no. 24, pp. 40 492–40 505, 2024

2024
[3]

Mult i- beam object-localization for millimeter-wave ISAC-aided connected autonomous vehicles,

J. Singh, A. Gupta, A. K. Jagannatham, and L. Hanzo, “Mult i- beam object-localization for millimeter-wave ISAC-aided connected autonomous vehicles,” IEEE Trans. V eh. Technol. , vol. 74, no. 1, pp. 1725–1729, 2025

2025
[4]

Integrated sensin g and communications (ISAC) for vehicular communication networ ks (VCN),

X. Cheng, D. Duan, S. Gao, and L. Y ang, “Integrated sensin g and communications (ISAC) for vehicular communication networ ks (VCN),” IEEE Internet Things J. , vol. 9, no. 23, pp. 23 441–23 451, 2022

2022
[5]

DOA estimat ion-oriented joint array partitioning and beamforming designs for ISAC s ystems,

R. Liu, M. Li, Q. Liu, and A. L. Swindlehurst, “DOA estimat ion-oriented joint array partitioning and beamforming designs for ISAC s ystems,” IEEE Trans. Wirel. Commun. , vol. 24, no. 3, pp. 2052–2066, 2025

2052
[6]

Intelligent reﬂecting surface enhan ced wireless network via joint active and passive beamforming,

Q. Wu and R. Zhang, “Intelligent reﬂecting surface enhan ced wireless network via joint active and passive beamforming,” IEEE Trans. Wirel. Commun., vol. 18, no. 11, pp. 5394–5409, 2019

2019
[7]

Modeling and performance ana lysis for movable antenna enabled wireless communications,

L. Zhu, W. Ma, and R. Zhang, “Modeling and performance ana lysis for movable antenna enabled wireless communications,” IEEE Trans. Wirel. Commun., vol. 23, no. 6, pp. 6234–6250, 2024

2024
[8]

Fl uid antenna systems,

K.-K. Wong, A. Shojaeifard, K.-F. Tong, and Y . Zhang, “Fl uid antenna systems,” IEEE Trans. Wirel. Commun. , vol. 20, no. 3, pp. 1950–1962, 2021

1950
[9]

Uplink a nd downlink communications in segmented waveguide-enabled p inching- antenna systems (SW ANs),

C. Ouyang, H. Jiang, Z. Wang, Y . Liu, and Z. Ding, “Uplink a nd downlink communications in segmented waveguide-enabled p inching- antenna systems (SW ANs),” arXiv:2509.10666, 2025

work page arXiv 2025
[10]

Integrated sensin g and commu- nication for segmented waveguide-enabled pinching antenn a systems,

Q. Gao, R. Zhong, H. Shin, and Y . Liu, “Integrated sensin g and commu- nication for segmented waveguide-enabled pinching antenn a systems,” arXiv:2601.20658, 2025

work page arXiv 2025
[11]

Flexible-ant enna systems: A pinching-antenna perspective,

Z. Ding, R. Schober, and H. Vincent Poor, “Flexible-ant enna systems: A pinching-antenna perspective,” IEEE Trans. Commun. , pp. 1–1, 2025

2025
[12]

Dist ributed antenna selection and beamforming for ELAA based downlink I SAC systems,

Q. Gao, R. Zhong, X. Mu, M. Peng, H. Shin, and Y . Liu, “Dist ributed antenna selection and beamforming for ELAA based downlink I SAC systems,” IEEE Trans. V eh. Technol., vol. 74, no. 11, pp. 17 865–17 878, 2025

2025
[13]

An tenna selection with beam squint compensation for integrated sen sing and communications,

A. M. Elbir, A. Abdallah, A. Celik, and A. M. Eltawil, “An tenna selection with beam squint compensation for integrated sen sing and communications,” IEEE J. Sel. Top. Signal Process. , vol. 18, no. 5, pp. 857–870, 2024

2024
[14]

Joint transmit/receive anten na selection and beamforming design for ISAC systems,

R. Liu, M. Li, and Q. Liu, “Joint transmit/receive anten na selection and beamforming design for ISAC systems,” in Proc. IEEE Global Commun. Conf. (GLOBECOM), Kuala Lumpur , Malaysia , 2023, pp. 3118–3123

2023
[15]

Rate maximizat ion for down- link pinching-antenna systems,

Y . Xu, Z. Ding, and G. K. Karagiannidis, “Rate maximizat ion for down- link pinching-antenna systems,” IEEE Wirel. Commun. Lett. , vol. 14, no. 5, pp. 1431–1435, 2025

2025
[16]

Minimum data rate maximization for uplink pinching-anten na systems,

S. A. Tegos, P . D. Diamantoulakis, Z. Ding, and G. K. Kara giannidis, “Minimum data rate maximization for uplink pinching-anten na systems,” IEEE Wirel. Commun. Lett. , vol. 14, no. 5, pp. 1516–1520, 2025

2025
[17]

Joint transmit and pinching beamforming for pinching antenna systems (PAS S): Optimization-based or learning-based?

X. Xu, X. Mu, Y . Liu, and A. Nallanathan, “Joint transmit and pinching beamforming for pinching antenna systems (PASS): Optimiza tion-based or learning-based?” arxiv:2502.08637, 2025

work page arXiv 2025
[18]

Pinch ing antenna systems for integrated sensing and communications,

H. Li, R. Zhong, Z. Pan, C. Dong, J. Lei, and Y . Liu, “Pinch ing antenna systems for integrated sensing and communications,” IEEE Trans. Wirel. Commun., vol. 25, pp. 13 416–13 429, 2026. 13

2026
[19]

Joint beamformin g for noma assisted pinching antenna systems (PASS),

D. Gan, X. Xu, J. Zuo, X. Ge, and Y . Liu, “Joint beamformin g for noma assisted pinching antenna systems (PASS),” IEEE Trans. Commun. , vol. 74, pp. 2450–2465, 2026

2026
[20]

Deep learni ng-based beamforming optimization for ISAC systems: A low-complexi ty and transferable framework,

Q. Gao, R. Zhong, Y . Zou, H. Shin, and Y . Liu, “Deep learni ng-based beamforming optimization for ISAC systems: A low-complexi ty and transferable framework,” IEEE Trans. Wirel. Commun. , vol. 25, pp. 8754–8768, 2026

2026
[21]

Transfer learning and meta learning-based fast downlink beamformin g adapta- tion,

Y . Y uan, G. Zheng, K.-K. Wong, B. Ottersten, and Z.-Q. Lu o, “Transfer learning and meta learning-based fast downlink beamformin g adapta- tion,” IEEE Trans. Wirel. Commun., vol. 20, no. 3, pp. 1742–1755, 2021

2021
[22]

Deep reinforcement learning-based optimization fo r IRS-assisted cognitive radio systems,

C. Zhong, M. Cui, G. Zhang, Q. Wu, X. Guan, X. Chu, and H. V . Poor, “Deep reinforcement learning-based optimization fo r IRS-assisted cognitive radio systems,” IEEE Trans. Commun. , vol. 70, no. 6, pp. 3849–3864, 2022

2022
[23]

Semi-supervised classiﬁcat ion with graph convolutional networks,

T. N. Kipf and M. Welling, “Semi-supervised classiﬁcat ion with graph convolutional networks,” 2016

2016
[24]

IC GNN: Graph neural network enabled scalable beamforming for MISO interfer- ence channels,

C. He, Y . Lu, B. Ai, O. A. Dobre, Z. Ding, and D. Niyato, “IC GNN: Graph neural network enabled scalable beamforming for MISO interfer- ence channels,” IEEE Trans. Mob. Comput. , vol. 24, no. 10, pp. 10 778– 10 791, 2025

2025
[25]

Graph neural network enabled pinching antennas,

X. Xie, Y . Lu, and Z. Ding, “Graph neural network enabled pinching antennas,” IEEE Wirel. Commun. Lette. , pp. 1–1, 2025

2025
[26]

Graph transformer f or tri- beamforming in pinching antenna systems (PASS),

J. Guo, Y . Liu, and A. Nallanathan, “Graph transformer f or tri- beamforming in pinching antenna systems (PASS),” IEEE J. Sel. Areas Commun., vol. 44, pp. 2872–2888, 2026

2026
[27]

Channel-agnostic predictive beamforming for crow dsourced bistatic satellite ISAC with LLM,

W. D. Lukito, W. Xiang, C. Liu, P . Lai, P . Cheng, W. Y uan, a nd G. Mao, “Channel-agnostic predictive beamforming for crow dsourced bistatic satellite ISAC with LLM,” IEEE J. Sel. Areas Commun. , vol. 44, pp. 2348–2364, 2026

2026
[28]

LLM-empowered near-ﬁeld co mmunica- tions for low-altitude economy,

Z. Xu, T. Zheng, and L. Dai, “LLM-empowered near-ﬁeld co mmunica- tions for low-altitude economy,” IEEE Trans. Commun. , vol. 73, no. 11, pp. 11 186–11 196, 2025

2025
[29]

LLM in V2I: A data-driven predictive beamforming f rame- work for vehicle tracking in near-ﬁeld ISAC systems,

H. Huang, W. Y uan, C. Liu, L. Liu, F. Liu, W. Xiang, and D. W ing Kwan Ng, “LLM in V2I: A data-driven predictive beamforming f rame- work for vehicle tracking in near-ﬁeld ISAC systems,” IEEE J. Sel. Areas Commun., vol. 44, pp. 2510–2527, 2026

2026
[30]

LoRA: Low-rank adaptation of large language mo dels,

E. J. Hu, Y . Shen, P . Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, and W. Chen, “LoRA: Low-rank adaptation of large language mo dels,” in 10th Int. Conf. Learn. Represent. (ICLR) , 2021, pp. 1–13

2021
[31]

Near-ﬁeld communications: A comprehensive survey,

Y . Liu, C. Ouyang, Z. Wang, J. Xu, X. Mu, and A. L. Swindleh urst, “Near-ﬁeld communications: A comprehensive survey,” EEE Commun. Surveys Tuts., vol. 27, no. 3, pp. 1687–1728, 2025

2025