Joint Offloading, Communication and Collaborative Computation Using Deep Reinforcement Learning in MEC Networks
Xuefang Nie, Xingbang Chen, DingDing Zhang, Tianqing Zhou, Jiliang Zhang · 2023
To meet the low-delay demands of delay-sensitive and computation-intensive applications of user equipments (UEs), we present a multi-UE end-edge-edge collaborative paradigm in mobile edge computing (MEC) networks by incorporating the other adjacent MEC servers with idle computation resources. To adaptively scheduling end-edge-edge spectrum and computation resources in different workload scenarios, we present a deep reinforcement learning (DRL) based joint offloading, communication and collaborative computation strategy to enhance the system efficiency of multi-UE and multi-MEC server networks. We aim at maximizing the system utility for a total processing delay and energy consumption tradeoff under delay constraints. We formulate the optimization problem as an optimal decision problem and then solve it using a DRL-based double deep Q-learning network (DDQN) algorithm. Numerical results demonstrate that our proposed algorithm is superior to conventional algorithms and can achieve over 11% performance improvement.