Dynamic Offloading Strategy for Delay Sensitive Task in Mobile Edge Computing Networks

With the rapid development of the Internet of Things (IoTs), the fifth-generation (5G) networks need to serve massive connection and accommodate ultra-low delay. In response to these challenges, mobile edge computing (MEC) and nonorthogonal multiple access (NOMA) have been considered as the promising solutions. In this paper, we investigate the joint optimization problem of computation offloading and resource allocation in NOMA MEC networks to minimize the delay to complete tasks of all users. Different from the conventional optimization approach, we propose and develop an online solution based on deep reinforcement learning (DRL) algorithm, which can fit with dynamic networks with time-varying channels. In particular, we employ deep neural networks (DNNs) to process the raw state inputs and then output the computation offloading decision and resource allocation at different times. The weights of DNNs are continuously trained with the observed data via interactions with the environment. Simulation results reveal that our proposed algorithm achieves higher delay reduction compared to the existing strategies.