Joint UAV Deployment and Space-Time-Frequency Resource Allocation for Low-Altitude Economy

This letter addresses a joint unmanned aerial vehicle (UAV) deployment and space-time-frequency resource allocation problem for air-ground networks. From the perspective of low altitude economy, we aim to maximize the economic profit by delivering high-quality services to different users with diverse and varying requirements, while reducing resource consumption and operational costs. To address the high-dimensional complexity of this combinatorial optimization problem, we decompose different categories of tasks to separate agents and utilize multi-agent deep reinforcement learning (MA-DRL) in a centralized training and distributed execution approach. Furthermore, we employ an attention mechanism to analyze heterogeneous effects of different local observations from individual UAVs and generate agent-specific global state augmentations so that the performance is further improved. Simulation results demonstrate that the proposed scheme is effective in enhancing the economic profit, and outperforms typical state-of-the-art DRL approaches.