Maximizing throughput in NOMA-enable industrial IoT networks using digital twin and reinforcement learning

Increased deployment of heterogeneous and complex Industrial Internet of Things (IIoT) applications such as predictive maintenance and asset tracking places a substantial strain on the limited computational and communication resources. To cater to the rigorous demands of these applications, it is imperative to devise an adaptive online resource allocation method to enhance the efficiency of the current network operations. Multiaccess edge computing (MEC) and digital twins (DTs) are promising solutions that facilitate the realization of edge intelligence and find applications in various industrial applications. Yet, little is known about the advantage the two technologies offer to IIoT networks.

This study presents a joint optimization of offloading and resource allocation approach where MEC-server DT is created at the edge, and nonorthogonal multiple access (NOMA) communication is considered between IIoT devices and the industrial gateways (IGWs) for spectral efficiency. Our proposed framework is tailored to reduce mean task completion latency and enhance overall IIoT network throughput.

To achieve our objective, we jointly optimize the computation resource allocation (RA), subchannel assignment (SA), and offloading decisions (OD). Given the inherent complexity of the problem, we further divide it into RA and SA/OD sub-problems. Employing Deep Reinforcement Learning (DRL), we have formulated a solution delineating the most efficient RA strategy and leveraged DT for optimal SA/OD strategies.

Simulation results demonstrate the superior efficiency of our framework, realizing up to 92 % of the efficiency of the exhaustive search method while reducing computation and action decision time.

In light of system dynamics considered for our work, the proposed framework perfomance showcase its robustness and potential application in real-world IIoT networks.