Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ComAgent: Multi-LLM based Agentic AI Empowered Intelligent Wireless Networks

Haoyun Li, Ming Xiao, Kezhi Wang, Robert Schober, Dong In Kim, Yong Liang Guan

TL;DR

The work addresses the bottleneck of translating high-level wireless intents into solver-ready, constraint-satisfying formulations in emergent 6G networks. It introduces ComAgent, a multi-LLM agentic AI framework that uses a closed-loop Perception–Planning–Action–Reflection cycle to coordinate Literature, Planning, Coding, and Scoring Agents, generating executable optimization code and verified simulations under constraints such as $P_{\max}$ and $E_{\min}$. Through case studies on MIMO SWIPT beamforming and a 25-task generalization set, ComAgent achieves expert-comparable performance and superior robustness compared with monolithic LLM baselines, demonstrating effective end-to-end automation from modeling to validation. The results indicate a promising direction for autonomous, constraint-aware network design that bridges intent understanding, mathematical rigor, and reproducible implementation, paving the way for self-evolving, zero-touch 6G networks.

Abstract

Emerging 6G networks rely on complex cross-layer optimization, yet manually translating high-level intents into mathematical formulations remains a bottleneck. While Large Language Models (LLMs) offer promise, monolithic approaches often lack sufficient domain grounding, constraint awareness, and verification capabilities. To address this, we present ComAgent, a multi-LLM agentic AI framework. ComAgent employs a closed-loop Perception-Planning-Action-Reflection cycle, coordinating specialized agents for literature search, coding, and scoring to autonomously generate solver-ready formulations and reproducible simulations. By iteratively decomposing problems and self-correcting errors, the framework effectively bridges the gap between user intent and execution. Evaluations demonstrate that ComAgent achieves expert-comparable performance in complex beamforming optimization and outperforms monolithic LLMs across diverse wireless tasks, highlighting its potential for automating design in emerging wireless networks.

ComAgent: Multi-LLM based Agentic AI Empowered Intelligent Wireless Networks

TL;DR

The work addresses the bottleneck of translating high-level wireless intents into solver-ready, constraint-satisfying formulations in emergent 6G networks. It introduces ComAgent, a multi-LLM agentic AI framework that uses a closed-loop Perception–Planning–Action–Reflection cycle to coordinate Literature, Planning, Coding, and Scoring Agents, generating executable optimization code and verified simulations under constraints such as and . Through case studies on MIMO SWIPT beamforming and a 25-task generalization set, ComAgent achieves expert-comparable performance and superior robustness compared with monolithic LLM baselines, demonstrating effective end-to-end automation from modeling to validation. The results indicate a promising direction for autonomous, constraint-aware network design that bridges intent understanding, mathematical rigor, and reproducible implementation, paving the way for self-evolving, zero-touch 6G networks.

Abstract

Emerging 6G networks rely on complex cross-layer optimization, yet manually translating high-level intents into mathematical formulations remains a bottleneck. While Large Language Models (LLMs) offer promise, monolithic approaches often lack sufficient domain grounding, constraint awareness, and verification capabilities. To address this, we present ComAgent, a multi-LLM agentic AI framework. ComAgent employs a closed-loop Perception-Planning-Action-Reflection cycle, coordinating specialized agents for literature search, coding, and scoring to autonomously generate solver-ready formulations and reproducible simulations. By iteratively decomposing problems and self-correcting errors, the framework effectively bridges the gap between user intent and execution. Evaluations demonstrate that ComAgent achieves expert-comparable performance in complex beamforming optimization and outperforms monolithic LLMs across diverse wireless tasks, highlighting its potential for automating design in emerging wireless networks.
Paper Structure (22 sections, 5 figures, 1 table)

This paper contains 22 sections, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. The Landscape of LLM-Driven Optimization: Progress and Gaps
  3. Recent advances in LLM-enabled wireless optimization
  4. Open challenges toward end-to-end agentic optimization copilots
  5. Agentic AI Framework and Methodology
  6. A communication-oriented agentic AI architecture
  7. End-to-end workflow instantiation for network optimization
  8. Knowledge acquisition
  9. Plan formulation
  10. Data preparation
  11. Simulation generation
  12. Performance evaluation: From autonomous design to robust generalization
  13. Case study 1: MIMO SWIPT beamforming optimization
  14. Autonomous end-to-end solution
  15. Performance verification
  16. ...and 7 more sections

Figures (5)

  • Figure 1: Framework of ComAgent: A multi-LLM based agentic AI system.
  • Figure 2: Operational Breakdown of the ComAgent Framework. A detailed example workflow that demonstrates the autonomous execution of a case study on MIMO SWIPT beamforming optimization across four distinct stages.
  • Figure 3: An example 2D channel gain map with the BS located at the origin.
  • Figure 4: Average sum-rate versus transmit power $P_{\mathrm{max}}$.
  • Figure 5: Average sum-rate versus the number of antennas $M$.