Energy-Efficient Aerial Base Station Enabled MBSFN: A Multi-Agent Reinforcement Learning Approach

Rapid expansion of the Internet of Unmanned Agents (IUAs) has led to a dramatic increase in demand for network capacity. Multicast-broadcast transmission, as an one-to-many communication paradigm, efficiently alleviates resource consumption by delivering common content to massive users over shared time-frequency resources. To address the flexible and large-scale networking requirements under IUA scenarios, this article introduces the multicast broadcast single frequency network (MBSFN) framework deploying aerial base stations (ABSs) based on autonomous aerial vehicles (AAVs). To deal with energy consumption challenges under AAVs’ coupled operational constraints, this article proposes a multiagent deep reinforcement learning (MADRL) approach, grounded in the multiagent deep deterministic policy gradient (MADDPG) algorithm, to optimize energy efficiency (EE) while ensure user Quality of Service (QoS) in ABS MBSFN systems. By modeling ABS as cooperative agents, the MADDPG algorithm dynamically adjusts discrete power levels, integrating straight-through estimators (STEs), reparameterization techniques, and Gumbel distribution sampling to resolve challenges in discrete action space optimization. The simulation results demonstrate that the proposed scheme achieves an average EE of 88.7% of the theoretical optimum and outperforms the single-agent DDPG method by up to 30%. Further analysis in diverse UE distributions validates the robustness and practicality of the proposed approach, highlighting its potential for energy-efficient multicast-broadcast deployments in AAV-enabled networks.