QoS Guaranteed Dual-Scale Dual-Stage Resource Allocation in RAN for Mixed Traffic Scenarios

In mixed traffic scenarios, human-driven, autonomous, and assisted vehicles have distinct service requirements, necessitating a more granular resource allocation strategy. However, conventional resource allocation approaches for the Internet of Vehicles (IoVs) typically classify all vehicle services into a single slice type. To address the challenge of radio access network (RAN) resource allocation in mixed traffic scenarios with finer-grained segmentation, we propose a dual-scale, dual-stage resource allocation technique based on neural network and reinforcement learning. First, the RAN resource allocation problem is decomposed into two time scales, which are the long and short time scales. At the long time scale, resource blocks (RBs) are allocated at the slice level, while at the short time scale, power and bandwidth are distributed within each slice. Second, resource allocation is performed in two stages. At the long time scale, the first stage predicts service volume trends, while the second stage optimizes inter-slice resource allocation based on these predictions. At the short time scale, intra-slice resource allocation is adjusted dynamically based on the user's channel state and available slice resources. The simulation results show that the proposed scheme significantly enhances overall system performance. Specifically, the scheme leads to higher average SINR, lower average power consumption, and a 5%-28% improvement in the average transmission rate across all slice types.