Wi-Fi Optimization with Deep Diffusion Deterministic Policy (2404.15684v2)

Abstract: Generative Diffusion Models (GDMs), have made significant strides in modeling complex data distributions across diverse domains. Meanwhile, Deep Reinforcement Learning (DRL) has demonstrated substantial improvements in optimizing Wi-Fi network performance. Wi-Fi optimization problems are highly challenging to model mathematically, and DRL methods can bypass complex mathematical modeling, while GDMs excel in handling complex data modeling. Therefore, combining DRL with GDMs can mutually enhance their capabilities. The current MAC layer access mechanism in Wi-Fi networks is the Distributed Coordination Function (DCF), which dramatically declines in performance with a high number of terminals. In this paper, we apply diffusion models to deep deterministic policy gradient (DDPG), namely the Deep Diffusion Deterministic Policy (D3PG) algorithm to optimize the Wi-Fi performance. Although similar integrations of reinforcement learning with generative diffusion models have been explored previously, we are the first to apply this approach to Wi-Fi network performance optimization. We propose an access mechanism that jointly adjusts the contention window and aggregation frame length based on the D3PG algorithm. Through simulations, we have demonstrated that this mechanism significantly outperforms existing Wi-Fi standards in dense Wi-Fi scenarios, maintaining performance even as the number of users sharply increases.