Learning age of information-oriented energy-efficient scheduling with non-orthogonal multiple access

We propose a deep Q-network (DQN)-based user scheduling scheme aimed at low network-wide age of information (AoI) in a non-orthogonal multiple access (NOMA) uplink status update system. In our system, end users can harvest energy from wireless signals received on the radio-frequency band to extend the lifespan of their batteries. Given the battery replacement cost in many practical scenarios, our objective is to design a scheduling scheme to maintain a low AoI while efficiently utilizing energy harvesting and reducing the number of battery replacements. Considering the instantaneous AoI, remaining energy, and channel condition of end users, we model the user scheduling problem as a Markov decision process (MDP), which is solved by DQN. In particular, the choice of transmission power to save energy while enabling a high NOMA decoding rate is incorporated in the design of DQN. Experiments indicate that our DQN-based scheduling scheme achieves a low average AoI of the system with significantly fewer battery replacements compared to various benchmarking schemes. Copyright © 2024 IEEE.