Deep Reinforcement Learning for the Joint AoI and Throughput Optimization of the Random Access System

With the characteristics of low signaling overhead, the random access protocol emerges as an enabling technology for massive machine-type communication (mMTC). However, the data collisions resulted from the random transmission patterns causes inevitable transmission failures. If one user experiences multiple consecutive transmission failures, it will remain active for a long time and its Age of Information (AoI) will gradually increase. The resulting high standby delay and high power overhead cause an unbearable burden on devices. Based on the above considerations, we propose a reinforcement learning (RL)-based user random transmitting strategy where the AoI of users and the total throughput of the system are jointly considered to be optimized. In the proposed scheme, users are classified into two sets according to their AoI levels and allocated with differential access patterns. Then, a deep neural network is proposed to dynamically adapt the access patterns according to the environment. The simulation results show that the proposed scheme achieves a higher throughput and improves the communication fairness over the conventional static random access protocols.