Application-aware power management for IAB-enabled V2I networks in 6 G: a DRL-based framework for energy efficiency and learning stability

Roadside vehicular networks require dense access infrastructure and flexible wireless backhaul to support heterogeneous vehicle-to-infrastructure (V2I) services. Millimeter-wave (mmWave) integrated access and backhaul (IAB) networks provide a cost-effective architecture for such deployments by allowing roadside IAB nodes to aggregate access traffic and forward it through wireless multi-hop backhaul links. However, energy-efficient power control remains challenging because access and backhaul transmissions share limited radio resources, while different roadside nodes may carry traffic with different service priorities. To address this issue, this paper develops an application-aware deep reinforcement learning (DRL)-based power-control framework for roadside multi-hop mmWave IAB-assisted V2I networks, where application awareness is represented by node-level service-priority weights in the WSEE objective. We formulate a weighted sum of energy efficiency (WSEE) maximization problem that jointly accounts for node-level traffic aggregation, multi-hop backhaul forwarding, differentiated service-priority weights, and transmit-power constraints. The resulting nonlinear and non-convex problem is transformed into a continuous-control Markov decision process (MDP). To improve sample utilization and exploration stability, prioritized experience replay and parameter-space noise are incorporated into deep deterministic policy gradient (DDPG), resulting in prioritized-experience-replay DDPG (PER-DDPG) and parameter-noise DDPG (PN-DDPG), respectively. Simulation results show that the two customized DDPG variants achieve at least 4% higher WSEE than the baseline DDPG scheme. In the tested setting, PN-DDPG reaches a stable WSEE region earlier and maintains higher normalized WSEE as the vehicular user scale increases. These results indicate that application-aware DRL can provide an effective online power-control mechanism for energy-efficient roadside IAB deployment.