Importance-Aware Optimization for Age of Information in Multi-Cluster IoT Systems With Random Delays

Abstract

Age of information (AoI) is a newly developed real-time metric that is frequently employed to assess the timeliness of received data. In this letter, we consider a multi-cluster Internet of Things system with random delays and develop scheduling policy to minimize the ratio of long-term average AoI to the importance of received data. Due to the strong coupling between inter-cluster channel allocation and intra-cluster link selection, traditional methods may be difficult to implement due to excessive action space. To overcome this issue, we develop a virtual queue-based sub-policy to make link selection decision, and then utilize deep reinforcement learning to design a master policy for channel allocation. The scheduling policy is obtained by embedding the sub-policy into the master policy for training. Simulation results show up to 55.7% performance improvement compared to the existing algorithms.