Maximizing coverage in UAV-based emergency communication networks using deep reinforcement learning

In the optimization of traditional Unmanned Aerial Vehicle (UAV) emergency communication systems in response to natural disasters, existing studies often overlook the simultaneous optimization of time-domain and frequency-domain resources, which leads to low communication efficiency and limited coverage. To address these issues, we propose a UAV-based emergency communication system with a HybridComm architecture. This architecture optimizes the UAV's aerial position, uplink and downlink time slot ratio, and bandwidth allocation based on feedback from transmission rates and channel losses, ensuring optimal resource allocation. Additionally, while ensuring full-duplex communication and a minimum data transmission rate for all nodes, we designed a communication priority mechanism to ensure the communication quality of rescue nodes. Simulation results using deep reinforcement learning show that with a 1.5 Mb/s threshold, the number of ground nodes covered by communication converges to approximately 820 per session, an increase of 170 nodes compared to an amorphous reward function. Furthermore, in systems without priority settings, the number of rescue personnel devices covered per event is nearly zero, while systems with priority settings cover approximately 175 devices per event, thereby ensuring high-quality communication for rescue operations. This study offers an innovative approach to enhancing the efficiency and reliability of disaster response communications.