Exploring LLM in Semantic Communication for V2X Networks

Navdeep Singh; Nicola Marchetti; Sihem Bakri

arxiv: 2606.20576 · v1 · pith:5J5HXZTXnew · submitted 2026-05-02 · 💻 cs.NI

Exploring LLM in Semantic Communication for V2X Networks

Sihem Bakri , Navdeep Singh , Nicola Marchetti This is my paper

Pith reviewed 2026-07-01 00:08 UTC · model grok-4.3

classification 💻 cs.NI

keywords semantic communicationV2X networkslarge language modelsbandwidth reductionvehicle coordinationnatural language messages

0 comments

The pith

A semantic communication system using large language models reduces data transmission in vehicle-to-everything networks by an average of 33.54 percent.

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

The paper aims to establish that an LLM can convert raw sensor data into concise natural language messages for V2X networks, thereby cutting bandwidth use while still allowing effective coordination. This is tested in a multilane traffic simulation that compares semantic and traditional modes. The approach matters because traditional V2X systems transmit large amounts of redundant raw data, straining network resources as vehicle numbers grow. If the method works, it could enable more scalable connected vehicle systems by focusing communication on meaning rather than volume.

Core claim

The central claim is that integrating a large language model with graph-based knowledge representation allows semantic transformation of sensor inputs into natural language messages, resulting in an average 33.54% reduction in data transmission and supporting context-aware coordination in V2X scenarios.

What carries the argument

The LLM-based semantic communication framework that converts structured sensor inputs into concise natural language messages describing context and intent.

If this is right

V2X networks transmit less data while maintaining situational awareness for coordination.
High-level control decisions are generated from shared situational awareness.
Bandwidth usage is reduced in multilane traffic simulations.
Context-aware coordination is illustrated in representative scenarios.

Where Pith is reading between the lines

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

This could reduce network congestion in areas with high vehicle density.
The method might be applied to other real-time coordination systems like autonomous drone fleets.
Standardization of the message format would be needed for broad adoption across different vehicle systems.

Load-bearing premise

The assumption that LLM-generated natural language messages preserve all information necessary for safe real-time vehicle coordination decisions without introducing errors or omissions that the simulation does not detect.

What would settle it

A test scenario in which an LLM-generated message leads to a coordination error or collision that raw sensor data transmission would have prevented.

Figures

Figures reproduced from arXiv: 2606.20576 by Navdeep Singh, Nicola Marchetti, Sihem Bakri.

**Figure 2.** Figure 2: Total Bandwidth Usage in Semantic and Non [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Figure 3: Per-Stream Bandwidth Usage in Semantic and [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: Initial simulation setup with five cars placed [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

**Figure 8.** Figure 8: Car 3 now maintains a consistent distance from [PITH_FULL_IMAGE:figures/full_fig_p006_8.png] view at source ↗

**Figure 6.** Figure 6: Car 3 approaches an obstacle, while other lanes [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗

read the original abstract

The rapid growth of connected and autonomous vehicles has created a demand for more efficient and intelligent communication systems. Traditional Vehicle-to-Everything (V2X) networks rely on transmitting raw sensor data, leading to high bandwidth usage and redundant information exchange. To address this, we propose a semantic communication framework that integrates a Large Language Model (LLM) with graph-based knowledge representation, to transmit only high-level, meaningful messages instead of raw data. Within this framework, the LLM performs semantic transformation, converting structured sensor inputs into concise natural language messages that describe context and intent. It also generates high-level control decisions based on shared situational awareness across the V2X network. A multilane traffic simulation was developed to compare semantic and non-semantic modes in terms of bandwidth usage. Results show an average 33.54% reduction in data transmission and illustrate context-aware coordination in representative scenarios.

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 applies an LLM plus graph setup to turn V2X sensor data into natural-language messages and reports a 33.54% bandwidth drop in one multilane simulation, but the result hinges on untested claims that those messages keep every safety-critical detail.

read the letter

The core takeaway is straightforward: they built a semantic framework where an LLM converts structured sensor inputs into concise natural-language descriptions of context and intent, then used that for V2X coordination instead of raw data. A multilane traffic simulation compared the two modes and produced the 33.54% average reduction figure plus some example coordination scenarios.

What is actually new is the concrete combination for the V2X setting—LLM semantic transformation paired with graph-based knowledge representation to generate both messages and high-level control decisions. The paper does a clear job explaining the bandwidth problem in connected vehicles and why sending only high-level meaning could matter.

The simulation itself is the main empirical contribution, and it illustrates the intended context-aware behavior in representative cases. That part is useful as a proof-of-concept sketch.

The soft spots sit in the evaluation. The reported reduction comes from a single unspecified simulation run with no error bars, no explicit baseline details, and no description of how message size was measured or how fidelity was checked. There is no evidence that the LLM outputs were tested against ground-truth safety variables such as exact positions, velocities, or required maneuvers, nor any collision or omission metrics. If the natural-language messages drop or distort information the simulation never catches, the bandwidth saving is not comparable to the non-semantic case. No math or derivation is offered, so the result stands or falls on the simulation setup alone.

This is for people working on V2X efficiency or semantic communication who want to see an LLM integration tried in a traffic model. A reader already familiar with semantic comm papers will recognize the extension but may still want the full simulation parameters and validation steps.

I would send it to peer review. The application is timely enough that referees should see the full manuscript and push for the missing checks on information preservation and safety equivalence.

Referee Report

3 major / 0 minor

Summary. The manuscript proposes an LLM-integrated graph-based semantic communication framework for V2X networks in which the LLM converts structured sensor inputs into concise natural-language messages describing context and intent and also generates high-level control decisions. A multilane traffic simulation is used to compare semantic and non-semantic modes, with the central empirical claim being an average 33.54% reduction in data transmission together with illustrations of context-aware coordination.

Significance. If the bandwidth-reduction claim can be supported by reproducible simulation details, fidelity metrics, and safety validation, the work would address a timely problem in vehicular networks by showing how LLMs might enable semantic rather than raw-data exchange. The combination of graph knowledge representation with LLM semantic transformation is a plausible direction, but the absence of any quantitative validation of information preservation or coordination correctness currently prevents assessment of practical impact.

major comments (3)

[Abstract] Abstract: the reported 33.54% reduction is given as a single point estimate from an unspecified simulation with no error bars, no description of how message size is quantified for the semantic case, no baseline details, and no message-fidelity or safety metrics; this single number is the sole quantitative support for the central claim.
[Abstract] Abstract / simulation description: no ground-truth comparison is provided that measures information loss on coordination-critical variables (exact positions, velocities, intent) or that reports collision / safety metrics to confirm that LLM-generated messages do not omit or distort facts required for safe real-time decisions.
[Framework description] Framework description: the construction of the graph-based knowledge representation, its integration with the LLM, and the precise semantic-transformation procedure are not specified, leaving the mechanism that supposedly enables the bandwidth saving unreproducible and untestable.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the constructive and detailed comments, which highlight important aspects of reproducibility and validation. We address each major comment point-by-point below, indicating planned revisions to the manuscript where feasible.

read point-by-point responses

Referee: [Abstract] Abstract: the reported 33.54% reduction is given as a single point estimate from an unspecified simulation with no error bars, no description of how message size is quantified for the semantic case, no baseline details, and no message-fidelity or safety metrics; this single number is the sole quantitative support for the central claim.

Authors: We agree that the abstract presents the 33.54% figure without sufficient supporting context. In the revised manuscript we will expand the abstract to briefly describe the multilane traffic simulation setup, clarify that message size for the semantic case is quantified via token count of the LLM-generated natural-language descriptions (versus raw sensor data volume for the baseline), note the non-semantic baseline, and reference variability measures from the results section. The reported value is an average across simulation runs. revision: yes
Referee: [Abstract] Abstract / simulation description: no ground-truth comparison is provided that measures information loss on coordination-critical variables (exact positions, velocities, intent) or that reports collision / safety metrics to confirm that LLM-generated messages do not omit or distort facts required for safe real-time decisions.

Authors: The simulation was designed to demonstrate bandwidth savings and context-aware coordination through representative scenarios rather than exhaustive fidelity or safety benchmarking. We will revise the manuscript to explicitly discuss this scope limitation, add qualitative analysis of whether critical variables appear preserved in the illustrated cases, and include a dedicated future-work subsection outlining the need for ground-truth information-loss metrics and collision-rate evaluations. revision: partial
Referee: [Framework description] Framework description: the construction of the graph-based knowledge representation, its integration with the LLM, and the precise semantic-transformation procedure are not specified, leaving the mechanism that supposedly enables the bandwidth saving unreproducible and untestable.

Authors: We will substantially expand the framework section in the revision to include: (i) the exact procedure for constructing the graph from structured sensor inputs (node/edge definitions and attribute encoding), (ii) the prompt templates and integration points used with the LLM for semantic transformation, and (iii) pseudocode or a step-by-step algorithmic description of the end-to-end semantic pipeline. This will directly address reproducibility. revision: yes

standing simulated objections not resolved

Quantitative ground-truth measurements of information loss on coordination-critical variables and explicit safety metrics (e.g., collision rates) were not computed in the original simulation and cannot be supplied without conducting new experiments.

Circularity Check

0 steps flagged

No circularity: empirical simulation output with no derivation or self-citation reduction

full rationale

The paper reports an empirical result from a multilane traffic simulation comparing semantic (LLM-based) and non-semantic V2X modes, yielding a 33.54% average reduction in data transmission. No equations, fitted parameters, uniqueness theorems, or self-citations are invoked to derive this figure; it is presented as a direct simulation output. The framework description (LLM semantic transformation plus graph knowledge) is a proposed architecture whose performance is measured externally rather than defined into the result by construction. This is a standard non-circular empirical claim.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The framework rests on the untested premise that LLM outputs are both compact and information-preserving for safety-critical decisions; no free parameters are explicitly fitted in the abstract, but the simulation result itself functions as an implicit performance claim whose validity depends on unstated modeling choices.

axioms (1)

domain assumption LLM-generated natural-language messages retain all information required for correct real-time control decisions in V2X scenarios
This premise is required for the claimed bandwidth reduction to be useful rather than merely smaller.

invented entities (1)

LLM-plus-graph semantic communication framework for V2X no independent evidence
purpose: Convert raw sensor streams into concise intent-carrying messages and generate control decisions from shared summaries
The paper introduces this specific architecture as the proposed solution; no external falsifiable evidence for its correctness is supplied beyond the simulation result.

pith-pipeline@v0.9.1-grok · 5677 in / 1429 out tokens · 32097 ms · 2026-07-01T00:08:58.620454+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

21 extracted references · 9 canonical work pages · 1 internal anchor

[1]

Luoet al., ”Semantic Communications: Overview, Open Issues, and Future Research Directions,” in IEEE Wireless Communications, vol

X. Luoet al., ”Semantic Communications: Overview, Open Issues, and Future Research Directions,” in IEEE Wireless Communications, vol. 29, no. 1, pp. 210-219, February 2022, doi: 10.1109/MWC.101.2100269

work page doi:10.1109/mwc.101.2100269 2022
[2]

Shimizuet al., ”Comparative Analysis of DSRC and LTE- V2X PC5 Mode 4 with SAE Congestion Control,” 2020 IEEE Vehicular Networking Conference (VNC), New York, NY , USA, 2020, pp

T. Shimizuet al., ”Comparative Analysis of DSRC and LTE- V2X PC5 Mode 4 with SAE Congestion Control,” 2020 IEEE Vehicular Networking Conference (VNC), New York, NY , USA, 2020, pp. 1-8, doi: 10.1109/VNC51378.2020.9318353

work page doi:10.1109/vnc51378.2020.9318353 2020
[3]

6G for vehicle-to-everything (V2X) communications: Enabling technologies, challenges, and oppor- tunities,

M. N. A. Rahimet al., “6G for vehicle-to-everything (V2X) communications: Enabling technologies, challenges, and oppor- tunities,”IEEE Access, vol. 11, pp. 103538–103572, 2023

2023
[4]

Mode Collapse in Generative Adversarial Networks: An Overview,

H. Ouamnaet al., ”6G and V2X Communications: Applications, Features, and Challenges,” 2022 8th International Conference on Optimization and Applications (ICOA), Genoa, Italy, 2022, pp. 1-6, doi: 10.1109/ICOA55659.2022.9934407

work page doi:10.1109/icoa55659.2022.9934407 2022
[5]

A survey on semantic communications: Tech- nologies, solutions, applications and challenges,

Y . Liuet al., “A survey on semantic communications: Tech- nologies, solutions, applications and challenges,”Digital Com- munications and Networks, vol. 10, no. 3, pp. 528–545, 2024

2024
[6]

Semantic communication: A survey on research landscape, challenges, and future directions,

T. Getuet al., “Semantic communication: A survey on research landscape, challenges, and future directions,”TechRxiv preprint, 2023

2023
[7]

Semantic communication: A survey of its theoretical development,

G. Xinet al., “Semantic communication: A survey of its theoretical development,”Entropy, vol. 26, no. 2, p. 102, 2024

2024
[8]

A survey on semantic communication networks: Architecture, security, and privacy,

S. Guoet al., “A survey on semantic communication networks: Architecture, security, and privacy,”IEEE Communications Sur- veys & Tutorials, 2024

2024
[9]

Semantic vehicle-to-everything (V2X) commu- nications towards 6G,

T. Lyuet al., “Semantic vehicle-to-everything (V2X) commu- nications towards 6G,”arXiv preprint arXiv:2402.06473, 2024

work page arXiv 2024
[10]

Language models are few-shot learners,

T. B. Brownet al., “Language models are few-shot learners,” Advances in Neural Information Processing Systems, vol. 33, pp. 1877–1901, 2020

1901
[11]

GPT-4 Technical Report

OpenAI, “GPT-4 technical report,”arXiv preprint arXiv:2303.08774, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023
[12]

Large language model (LLM) for telecommunications: A comprehensive survey,

H. Zhouet al., “Large language model (LLM) for telecommunications: A comprehensive survey,”arXiv preprint arXiv:2401.00315, 2024

work page arXiv 2024
[13]

Leveraging LLMs to explain DRL de- cisions for transparent 6G network slicing,

M. Ameuret al., “Leveraging LLMs to explain DRL de- cisions for transparent 6G network slicing,”arXiv preprint arXiv:2312.08874, 2024

work page arXiv 2024
[14]

LLM-based edge intelligence: A comprehen- sive survey,

O. Frihaet al., “LLM-based edge intelligence: A comprehen- sive survey,”IEEE Communications Surveys & Tutorials, 2024

2024
[15]

The role of LLMs in sustainable smart cities,

A. Ullahet al., “The role of LLMs in sustainable smart cities,” IEEE Access, 2024

2024
[16]

6G comprehensive intelligence: Network operations and optimization based on large language models,

S. Longet al., “6G comprehensive intelligence: Network operations and optimization based on large language models,” IEEE Wireless Communications, 2024

2024
[17]

Large AI model empowered multimodal semantic communications,

F. Jianget al., “Large AI model empowered multimodal semantic communications,”IEEE Wireless Communications, 2024

2024
[18]

Large language models for semantic communi- cation in edge-based IoT networks,

A. Kalita, “Large language models for semantic communi- cation in edge-based IoT networks,”IEEE Internet of Things Magazine, vol. 7, no. 1, pp. 42–49, 2024

2024
[19]

Embodied AI-enhanced vehicular networks,

R. Zhanget al., “Embodied AI-enhanced vehicular networks,” arXiv preprint arXiv:2501.01141, 2025

work page arXiv 2025
[20]

Hybrid reasoning based on large language models for autonomous car driving,

M. Azarafza, C. Steinmetz, A. Rettberg, M. Nayyeri, and S. Staab, “Hybrid reasoning based on large language models for autonomous car driving,”arXiv preprint arXiv:2402.13606, 2024

work page arXiv 2024
[21]

Neo4j Graph Database,

Neo4j, “Neo4j Graph Database,” Neo4j Inc., 2024. [Online]. Available: https://neo4j.com

2024

[1] [1]

Luoet al., ”Semantic Communications: Overview, Open Issues, and Future Research Directions,” in IEEE Wireless Communications, vol

X. Luoet al., ”Semantic Communications: Overview, Open Issues, and Future Research Directions,” in IEEE Wireless Communications, vol. 29, no. 1, pp. 210-219, February 2022, doi: 10.1109/MWC.101.2100269

work page doi:10.1109/mwc.101.2100269 2022

[2] [2]

Shimizuet al., ”Comparative Analysis of DSRC and LTE- V2X PC5 Mode 4 with SAE Congestion Control,” 2020 IEEE Vehicular Networking Conference (VNC), New York, NY , USA, 2020, pp

T. Shimizuet al., ”Comparative Analysis of DSRC and LTE- V2X PC5 Mode 4 with SAE Congestion Control,” 2020 IEEE Vehicular Networking Conference (VNC), New York, NY , USA, 2020, pp. 1-8, doi: 10.1109/VNC51378.2020.9318353

work page doi:10.1109/vnc51378.2020.9318353 2020

[3] [3]

6G for vehicle-to-everything (V2X) communications: Enabling technologies, challenges, and oppor- tunities,

M. N. A. Rahimet al., “6G for vehicle-to-everything (V2X) communications: Enabling technologies, challenges, and oppor- tunities,”IEEE Access, vol. 11, pp. 103538–103572, 2023

2023

[4] [4]

Mode Collapse in Generative Adversarial Networks: An Overview,

H. Ouamnaet al., ”6G and V2X Communications: Applications, Features, and Challenges,” 2022 8th International Conference on Optimization and Applications (ICOA), Genoa, Italy, 2022, pp. 1-6, doi: 10.1109/ICOA55659.2022.9934407

work page doi:10.1109/icoa55659.2022.9934407 2022

[5] [5]

A survey on semantic communications: Tech- nologies, solutions, applications and challenges,

Y . Liuet al., “A survey on semantic communications: Tech- nologies, solutions, applications and challenges,”Digital Com- munications and Networks, vol. 10, no. 3, pp. 528–545, 2024

2024

[6] [6]

Semantic communication: A survey on research landscape, challenges, and future directions,

T. Getuet al., “Semantic communication: A survey on research landscape, challenges, and future directions,”TechRxiv preprint, 2023

2023

[7] [7]

Semantic communication: A survey of its theoretical development,

G. Xinet al., “Semantic communication: A survey of its theoretical development,”Entropy, vol. 26, no. 2, p. 102, 2024

2024

[8] [8]

A survey on semantic communication networks: Architecture, security, and privacy,

S. Guoet al., “A survey on semantic communication networks: Architecture, security, and privacy,”IEEE Communications Sur- veys & Tutorials, 2024

2024

[9] [9]

Semantic vehicle-to-everything (V2X) commu- nications towards 6G,

T. Lyuet al., “Semantic vehicle-to-everything (V2X) commu- nications towards 6G,”arXiv preprint arXiv:2402.06473, 2024

work page arXiv 2024

[10] [10]

Language models are few-shot learners,

T. B. Brownet al., “Language models are few-shot learners,” Advances in Neural Information Processing Systems, vol. 33, pp. 1877–1901, 2020

1901

[11] [11]

GPT-4 Technical Report

OpenAI, “GPT-4 technical report,”arXiv preprint arXiv:2303.08774, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023

[12] [12]

Large language model (LLM) for telecommunications: A comprehensive survey,

H. Zhouet al., “Large language model (LLM) for telecommunications: A comprehensive survey,”arXiv preprint arXiv:2401.00315, 2024

work page arXiv 2024

[13] [13]

Leveraging LLMs to explain DRL de- cisions for transparent 6G network slicing,

M. Ameuret al., “Leveraging LLMs to explain DRL de- cisions for transparent 6G network slicing,”arXiv preprint arXiv:2312.08874, 2024

work page arXiv 2024

[14] [14]

LLM-based edge intelligence: A comprehen- sive survey,

O. Frihaet al., “LLM-based edge intelligence: A comprehen- sive survey,”IEEE Communications Surveys & Tutorials, 2024

2024

[15] [15]

The role of LLMs in sustainable smart cities,

A. Ullahet al., “The role of LLMs in sustainable smart cities,” IEEE Access, 2024

2024

[16] [16]

6G comprehensive intelligence: Network operations and optimization based on large language models,

S. Longet al., “6G comprehensive intelligence: Network operations and optimization based on large language models,” IEEE Wireless Communications, 2024

2024

[17] [17]

Large AI model empowered multimodal semantic communications,

F. Jianget al., “Large AI model empowered multimodal semantic communications,”IEEE Wireless Communications, 2024

2024

[18] [18]

Large language models for semantic communi- cation in edge-based IoT networks,

A. Kalita, “Large language models for semantic communi- cation in edge-based IoT networks,”IEEE Internet of Things Magazine, vol. 7, no. 1, pp. 42–49, 2024

2024

[19] [19]

Embodied AI-enhanced vehicular networks,

R. Zhanget al., “Embodied AI-enhanced vehicular networks,” arXiv preprint arXiv:2501.01141, 2025

work page arXiv 2025

[20] [20]

Hybrid reasoning based on large language models for autonomous car driving,

M. Azarafza, C. Steinmetz, A. Rettberg, M. Nayyeri, and S. Staab, “Hybrid reasoning based on large language models for autonomous car driving,”arXiv preprint arXiv:2402.13606, 2024

work page arXiv 2024

[21] [21]

Neo4j Graph Database,

Neo4j, “Neo4j Graph Database,” Neo4j Inc., 2024. [Online]. Available: https://neo4j.com

2024