核心概念
The paper proposes an optimized offloading decision and resource allocation solution based on generative AI and deep reinforcement learning algorithms to enhance the integration of communication and computing in 802.11ax Wi-Fi networks.
要約
The paper presents an integrated communication and computing scheme for Wi-Fi networks based on generative AI and reinforcement learning. The key highlights are:
-
System Model:
- Constructed an MEC edge-end architecture based on 802.11ax Wi-Fi, with a single AP and multiple STAs.
- Divided STAs into computing STAs and communication STAs, with computing STAs offloading tasks to MEC servers.
- Formulated the optimization problem to minimize the weighted sum of system latency and energy consumption.
-
Offloading Decision Algorithm:
- Proposed a deep diffusion learning model called Diffusion Twin Delayed DDPG (DTD3) to solve the offloading decision problem.
- Utilized a Generative Diffusion Model (GDM) as the policy network for the Twin Delayed DDPG (TD3) algorithm.
- The GDM-based approach significantly reduces the convergence time and training cost compared to traditional DRL algorithms.
-
Resource Allocation Scheme:
- Designed a resource allocation scheme based on the Hungarian algorithm to allocate communication resources (RUs) to STAs.
- Considered the resource allocation characteristics of 802.11ax OFDMA and prioritized tasks based on factors like data size, CPU requirement, and channel condition.
-
Performance Evaluation:
- Simulation results demonstrated that the proposed solution outperforms baseline approaches in terms of reducing system latency, energy consumption, and enhancing QoS and communication success rate.
- The DTD3 algorithm also exhibited superior convergence performance compared to DQN and SAC algorithms.
The paper presents a comprehensive solution to address the challenges in communication-computing integration for 802.11ax Wi-Fi networks, leveraging the strengths of generative AI and reinforcement learning.
統計
The data size of computing tasks is uniformly distributed between 2.4Mbits and 4Mbits.
The data size of communication tasks is uniformly distributed between 10Mbits and 20Mbits.
The number of CPU cycles required for computing tasks is uniformly distributed between 900 Megacycles and 1100 Megacycles.
The computational capacities of MEC and STA are 10GHz and 1GHz, respectively.
The transmission power of STAs is 500mW.
引用
"The introduction of Generative AI significantly reduces model training costs, and the proposed solution exhibits significant reductions in system task processing latency and total energy consumption costs."
"The DTD3 algorithm converges around 400 episodes with optimal convergence performance, attributed to the introduced generative diffusion model, which effectively reduces the convergence time and training cost by working in coordination with the RL framework."