Optimization of WiFi Network Resource Allocation Based on Graph Neural Networks and Deep Reinforcement Learning

A study combining Graph Neural Network (GNN) and Deep Reinforcement Learning (DRL) technologies aims to address the complex challenges of WiFi network resource allocation. It improves wireless network performance and user experience through intelligent decision-making, providing an innovative solution for automated and intelligent network management.

The explosive growth of IoT devices and rising wireless demand have posed resource management challenges for WiFi. Traditional methods rely on fixed rules or simple heuristics, ignoring the global structure of network topology and struggling to adapt to dynamic environments. The time-varying characteristics of wireless channels and the diversity of user demands further increase complexity.

In WiFi networks, the relationships between APs, users, and interference form a graph structure. GNN can process non-Euclidean data: nodes (APs/users) contain features such as location and channel status. Through message passing, it aggregates neighbor information, perceives the global state, understands interference patterns and spatial relationships, and provides context for decision-making.

DRL provides end-to-end decision optimization: the agent observes the network state (GNN embedding representation), selects resource allocation actions (channel/power adjustment), and the environment feeds back rewards (throughput, latency, fairness). It learns the optimal strategy by maximizing cumulative rewards without manual rule design.

The core architecture uses GNN as a state encoder to convert raw observations into a compact representation, which is input into the DRL policy network for decision-making. Training uses a simulated environment to generate diverse scenarios (topology, traffic, interference), and algorithms such as PPO or SAC are used to learn robust strategies for practical deployment.

Compared with traditional methods, this solution can adaptively learn dynamic networks, discover optimization opportunities end-to-end, and make fast decisions suitable for real-time scheduling. In high-density scenarios (office buildings, venues), it can optimize spectrum usage, reduce interference, increase capacity, lower operation and maintenance costs, and improve user satisfaction.

This research is an important progress in the interdisciplinary field of wireless networks and AI, providing a general framework for solving resource management using GNN+DRL. In the future, it can be extended to 5G/6G and millimeter-wave communication, explore multi-agent RL, transfer learning/meta-learning to improve adaptability, and develop efficient training algorithms to shorten convergence time.