.png){kind=icon}
.png){kind=icon}
Return
Agentic AI for Network Coverage Optimization via Reasoning-Enhanced RAG-Based Large and Small AI Model Collaboration
.png){kind=icon}
PRE
AI
DOI：10.1109/tccn.2026.3706093.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
As wireless networks evolve toward highly complex, dynamic, and autonomous systems, agentic intelligence has become a fundamental paradigm for next-generation network management, enabling perception-driven reasoning, self-directed decision-making, and continuous self-optimization. Existing static configurations and standalone small AI model (SAM) agents are difficult to adapt across changing scenarios, while large AI model (LAM) agents may suffer from incomplete domain knowledge and hallucination-prone decisions. To address these limitations, this paper develops a reasoning-enhanced retrieval-augmented generation-based large-small AI model collaboration (RRAG-based LSMC) agentic AI paradigm for network coverage optimization. The proposed framework combines chain-of-thought (CoT) prompting and retrieval-augmented generation (RAG) to inject domain-specific knowledge into LAM reasoning, and uses the LAM to generate knowledge-enhanced policy guidance for the SAM. A LSMC collaborative alignment degree based on Kullback-Leibler (KL) divergence is further established to regularize the SAM toward the LAM’s policy behavior while preserving coverage-oriented optimization. Extensive simulation results validate the effectiveness of the proposed agentic AI framework, demonstrating superior convergence and decision-making performance compared to baseline standalone methods and collaborative methods without external knowledge.
Keywords:
Agentic AI
large and small AI model collaboration
retrieval-augmented generation (RAG)
wireless network optimization
Journal
I
IF:
7
Papers:
1.5K
Citations:
5.5K