Sleep-Scheduling and Joint Computation-Communication Resource Allocation in MEC Networks for 5G IIoT

Abstract

Industrial 4.0 will be supported by Internet of Things (IIoT), which will bring profound revolutions to the industrial manufacturing. The fifth generation wireless communication system (5G) will be one of the key technologies to support IIoT. However, the connectivity-massive, computation-intensive and time-critical features of IIoT pose great challenges to the spectrum and computation resource in 5G IIoT networks. Non-orthogonal multiple access (NOMA) and mobile edge computing (MEC) are regarded as promising paradigms to tackle these problems, called NOMA-based MEC. To enhance computing performance of MEC system, we consider that devices can also offload their computation tasks to some idle devices with rich computation resources through machine-to-machine (M2M) communication, called M2M-assisted NOMA-based MEC scheme. We formulate an optimization problem under tasks delay constraints to minimize the system energy consumption through sleep-scheduling and joint computation-communication resource allocation. Specifically, we propose a deep reinforcement learning (DRL) based sleep-scheduling scheme to arrange some idle devices to work at sleep-mode for saving energy while satisfies the system computation requirements. Furthermore, we design an iterative algorithm for the joint computation-communication resource allocation problem. Numerical results demonstrate our proposed scheme and algorithm achieve significantly reduction of system energy consumption, while satisfying network computation requirements.