Abstract
WiFi networks have achieved remarkable success in enabling seamless communication and data exchange worldwide. The IEEE 802.11be standard, known as WiFi 7, introduces Multi-Link Operation (MLO), a groundbreaking feature that enables devices to establish multiple simultaneous connections across different bands and channels. While MLO promises substantial improvements in network throughput and latency reduction, it presents significant challenges in channel allocation, particularly in dense network environments. This paper addresses the dynamic channel allocation problem in dense WiFi 7 networks with MLO capabilities. We formulate this challenge as a combinatorial optimization problem and model it within a Multi-Armed Bandit (MAB) framework to enable online learning of optimal channel allocations. Our proposed Best-Arm Identification-enabled Monte Carlo Tree Search (BAI-MCTS) algorithm includes rigorous theoretical analysis, providing upper bounds for both sample complexity and error probability. To further reduce sample complexity, we put forth LLM-BAI-MCTS, which integrates Large Language Models (LLMs) into the BAI-MCTS algorithm. Numerical results demonstrate that the BAI-MCTS algorithm achieves a convergence rate approximately 50.44% faster than state-of-the-art algorithms, while the LLM-BAI-MCTS algorithm increases the convergence rate by over 63.32% in dense networks.
Introduction
The evolution of wireless standards has reached a new milestone with IEEE 802.11be, or WiFi 7. The cornerstone of this standard is Multi-Link Operation (MLO), which allows Multi-Link Devices (MLDs) to transmit and receive data simultaneously across multiple frequency bands (2.4 GHz, 5 GHz, and 6 GHz). This capability is designed to meet the soaring demands for high throughput and ultra-low latency in modern applications like VR/AR and industrial automation.
However, the introduction of MLO significantly complicates channel allocation. In dense deployments—such as stadiums or large office buildings—the number of possible channel combinations for multiple Access Points (APs) and stations (STAs) grows exponentially. For instance, a network with just six devices and seven configurations per device yields over 117,000 possible global states. Traditional static optimization methods fail to adapt to the dynamic nature of these environments, necessitating a more intelligent, real-time approach.
System Model and Methodology
We consider a dense WiFi 7 network consisting of multiple Basic Service Sets (BSSs). Each BSS includes an AP-MLD and several STA-MLDs, as illustrated in Figure 1. To maximize network throughput, we must solve a combinatorial optimization problem: assigning the best channel configuration to each link of every device while accounting for interference and collisions.
BAI-MCTS Framework
To navigate the massive search space, we model the problem as a Multi-Armed Bandit (MAB) problem. We propose the Best-Arm Identification-enabled Monte Carlo Tree Search (BAI-MCTS) algorithm. This approach structures the channel allocation as a tree where each level corresponds to a device's configuration. By applying Best-Arm Identification (BAI) at each node, the algorithm focuses its exploration on the most promising branches, ensuring that the final allocation is within a specified margin ((\epsilon)) of the optimal value with high probability.
LLM-Enhanced Optimization
While BAI-MCTS is efficient, the initial "cold start" in a massive search space can still be time-consuming. We introduce LLM-BAI-MCTS, which leverages the reasoning capabilities of Large Language Models (LLMs) like GPT-4o. By providing the LLM with a spatial description of the network and using In-Context Learning (ICL) with Chain-of-Thought (CoT) prompting, the LLM acts as a "wireless expert."
As shown in Figure 4, the LLM helps prune the search space by identifying a "valuable arm area" or providing a high-quality initial configuration for a subset of devices. This "warm start" allows the MCTS algorithm to bypass sub-optimal regions of the tree, drastically reducing the number of samples required to reach convergence.
Performance Evaluation
We evaluated our algorithms using a discrete event simulator across 1,000 random network topologies. Key findings include:
- Convergence Speed: BAI-MCTS reached 98% of the optimal throughput 50.44% faster than the previous state-of-the-art (DNG-MCTS). When integrated with an LLM, this speedup increased to 63.32% in dense scenarios.
- Throughput and Latency: The optimized Simultaneous Transmit and Receive (STR) mode achieved a 36.83% reduction in latency compared to traditional single-link operations.
- LLM Comparison: GPT-4o demonstrated superior performance as an initializer compared to Claude-3-Sonnet and Gemini-1.5-Pro, showcasing its ability to transfer logic from small-scale examples to complex, dense networks.
Conclusion
This work demonstrates that the synergy between Multi-Armed Bandit frameworks and Large Language Models offers a powerful solution for the complex resource management tasks in WiFi 7. By using LLMs to prune the combinatorial search space, we can achieve near-optimal network performance with significantly reduced computational overhead. Future research will explore the application of this framework to Non-STR modes and the inclusion of fairness metrics in the optimization objective.
Summary updated