Toward Routing in Low-Altitude Drone Networks: A Physical Sensing-Aided Intelligent Forwarding Mechanism With Deep Learning

In recent years, self-organizing networks composed of drones have received more attention due to their ability to expand coverage and improve mission efficiency. However, in global positioning-denied complex low-altitude environments, typical routing protocols, as the cornerstone of drone communications, are greatly restricted or even in failures by numerous obstacles around, which can cause frequent none line of sight (NLOS) links leading to sharp declines in communication performance or even interruptions. Therefore, in this work, we propose a physical sensing-aided intelligent forwarding (PSIF) mechanism for low-altitude drone network (LDNET), which could enhance the forwarding capability between drones by integrating a long-short-term memory (LSTM)-based multifeature link prediction with a deep Q-network (DQN) enabled forwarding decision. Simulation results indicate that PSIF can efficiently facilitate packet forwarding in LDNET, resulting in enhanced system performance with regards to delay, packet loss ratio, throughput, and power consumption.