Multi-Agent Deep Reinforcement Learning Based Joint Power Control and Resource Allocation for Uavs Networks

This paper investigates the challenge of limited wireless network resources in collaborative traffic monitoring scenarios involving multiple unmanned aerial vehicles (UAVs). We propose an efficient communication resource allocation strategy that enables multiple UAV-to-UAV (U2U) links to share the spectrum allocated to Air-to-Ground (A2G) links while dynamically adjusting transmission power. The rapidly changing environment in multi-UAV networks poses significant challenges in centralizing Channel State Information (CSI) for effective resource management. To tackle this issue, we introduce a Deep Reinforcement Learning (DRL) approach aimed at maximizing the total capacity of A2G links and enhancing the transmission reliability of U2U links. The resource sharing problem is formulated as a DRL task, where each U2U link is treated as an individual agent that interacts with the environment to iteratively update the network, optimizing both spectrum and power allocation. Moreover, we design an enhanced DRL algorithm tailored for distributed implementation to solve this problem. Experimental results demonstrate that, compared to baseline algorithms, the proposed approach effectively improves both the total capacity of A2G links and the transmission rate of U2U links.