AOI-Aware Self-Powered Sensors in HSR Communication Networks: Intelligent Decision-Making and Resource Allocation

This paper investigates intelligent decision-making and resource allocation for self-powered sensors with the age of information (AOI) orientation in high-speed railway (HSR) communication networks. The sensors innovatively access the network via non-orthogonal multiple access (NOMA), thereby alleviating the train’s burden in terms of both energy and communication resources. For the sensors’ energy management and data offloading, we formulate a resource allocation problem aimed at maximizing the data offloading volume within a single base station (BS) while ensuring the average AOI requirement. As a complex cumulative reward problem, directly applying deep reinforcement learning (DRL) for solution is neither elegant nor efficient. Therefore, we propose a convex constrained soft actorcritic (CCSAC) scheme, decomposing the optimization problem into long-term and short-term subproblems. Specifically, at the long-term problem, a DRL-based policy is used to make intelligent decisions on energy variations. In each time slot, the energy variation derived from DRL serves as input to a convex optimization model that determines the optimal resources allocation for that slot. In the simulation section, we compare multiple baseline algorithms and conduct extensive tests under various conditions, including different train speeds, sensor positions, numbers of sensors, and channel estimation error. The simulation results demonstrate the effectiveness and robustness of the algorithm, and indicate the important trade-off between the volume of data transmission and AOI in HSR scenarios.