Scholarlink

Due to the limitation of computing resources in end nodes, computation tasks in Internet of Things (IoT) can be offloaded to mobile edge servers. To measure the effectiveness and timeliness of data, Age of Information (AoI) is proposed as a timeliness metric. In this paper, we consider the AoI minimization problem for computation-intensive status update. Our former work has designed optimal policy for Peak AoI (PAoI) optimization in single source case. This paper further studies the average AoI optimization in multi-source case, which is challenging. We focus on a multi-source Mobile Edge Computing (MEC) system and study when to generate a new packet. The system operates in two modes: preemptive and non-preemptive computing server. Firstly, we formulate this problem as a Markov Decision Process (MDP). Then, we adopt Deep Reinforcement Learning (DRL) algorithm to optimize average AoI and PAoI respectively, considering both random transmission and computation time. Simulation results demonstrate that preemptive server gets better performance than non-preemptive server. We further extend the PAoI-optimal policy in single source case to the multi-source case, which is shown to be optimal for minimizing average PAoI with the DRL algorithm approaching its performance. When minimizing average AoI, the DRL algorithm achieves better performance than the PAoI-optimal policy.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Joint Optimization of Transmission and Computation for Multi-source MEC System Based on Deep Reinforcement Learning

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Qi Zhang,
  • Jianhang Zhu,
  • Jie Gong

摘要

Due to the limitation of computing resources in end nodes, computation tasks in Internet of Things (IoT) can be offloaded to mobile edge servers. To measure the effectiveness and timeliness of data, Age of Information (AoI) is proposed as a timeliness metric. In this paper, we consider the AoI minimization problem for computation-intensive status update. Our former work has designed optimal policy for Peak AoI (PAoI) optimization in single source case. This paper further studies the average AoI optimization in multi-source case, which is challenging. We focus on a multi-source Mobile Edge Computing (MEC) system and study when to generate a new packet. The system operates in two modes: preemptive and non-preemptive computing server. Firstly, we formulate this problem as a Markov Decision Process (MDP). Then, we adopt Deep Reinforcement Learning (DRL) algorithm to optimize average AoI and PAoI respectively, considering both random transmission and computation time. Simulation results demonstrate that preemptive server gets better performance than non-preemptive server. We further extend the PAoI-optimal policy in single source case to the multi-source case, which is shown to be optimal for minimizing average PAoI with the DRL algorithm approaching its performance. When minimizing average AoI, the DRL algorithm achieves better performance than the PAoI-optimal policy.