.png){kind=icon}
.png){kind=icon}
Return
Task execution latency minimization for energy-sensitive IoTs in wireless powered mobile edge computing: A DRL-based method
.png){kind=icon}
PRE
AI
DOI：10.1016/j.comnet.2024.110633.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Wireless powered mobile edge computing (MEC) has become a vital component of future 6G networks, offering efficient computational capabilities to internet of things (IoT) devices and ensuring a stable energy supply. In this paper, we propose an innovative design for a wireless powered MEC-assisted energy-sensitive IoT systems. The system integrates wireless power technology (WPT), allowing IoT devices to efficiently perform tasks under wireless power supply without consuming their battery energy. Therefore, we formulate a dynamic computational task execution latency minimization (DCTELM) problem to tackle the impacts of unpredictable energy requirements and potential communication latency on resource allocation. We also analyze the differences between time division multiple access and non-orthogonal multiple access communication protocols. Given the complexity and dynamic nature of this problem, we define it as a mixed integer nonlinear programming problem. To address this problem, we introduced an enhanced deep reinforcement learning (DRL) algorithm. First, the DCTELM problem is transformed into a Markov decision process (MDP), which simplifies the structure of the problem. Then, after determining the offloading decision, we reveal the convex relationship between resource allocation and task latency. Furthermore, the action space of MDP is further reduced, and a multihead proximal policy optimization (PPO) algorithm is proposed to deal with the simplified MDP problem. This process reduces the complexity of problem solving, decreases the consumption of computational resources. Simulation results demonstrate that our method outperforms other baselines in terms of computation latency, particularly with the introduction of an exploration strategy during the training phase, which significantly reduces task latency.
Keywords:
Mobile edge computing
Wireless power technology
Deep reinforcement learning
Markov decision process
Non-orthogonal multiple access
Journal
IF:
4.6
Papers:
1.5K
Citations:
1.6W