Research Article
An LLM-Based Approach to Improving Throughput and Reducing Collisions in IEEE 802.11 Wireless Networks
@INPROCEEDINGS{10.1007/978-3-032-34048-1_10, author={Guan-Sheng Huang and Chih-Heng Ke and Chong-Yi Yang and Yeong-Sheng Chen}, title={An LLM-Based Approach to Improving Throughput and Reducing Collisions in IEEE 802.11 Wireless Networks}, proceedings={Smart Grid and Internet of Things. 9th EAI International Conference, SGIoT 2025, Taichung, Taiwan, December 6--7, 2025, Proceedings}, proceedings_a={SGIOT}, year={2026}, month={10}, keywords={IEEE 802.11 Wireless network optimization LLM (Large Language Model) RL (Reinforcement Learning) QoS Intelligent network management}, doi={10.1007/978-3-032-34048-1_10} }- Guan-Sheng Huang
Chih-Heng Ke
Chong-Yi Yang
Yeong-Sheng Chen
Year: 2026
An LLM-Based Approach to Improving Throughput and Reducing Collisions in IEEE 802.11 Wireless Networks
SGIOT
Springer
DOI: 10.1007/978-3-032-34048-1_10
Abstract
We propose DCFLLM (Distributed Coordination Function Large Language Model), a novel large language model-based framework designed for intelligent contention window (CW) optimization in IEEE 802.11 wireless LANs. Unlike reinforcement learning (RL) methods that require retraining across different environments, DCFLLM leverages pretrained language models to infer network dynamics and recommend CW adjustments directly. By integrating protocol knowledge from technical documents and empirical data, the model achieves adaptive, zero-shot decision-making without relying on trial-and-error learning. DCFLLM supports multi-objective optimization including collision rate minimization and throughput enhancement and adapts to varying node densities and traffic loads. It can also interface with federated or hierarchical control systems for scalable deployment. Experimental simulations confirm that DCFLLM significantly improves quality of service (QoS), reduces collision rates, and eliminates the overhead of RL training. This work pioneers a new class of intelligent, training-free and interpretable solutions for proactive wireless network control.