Autonomous Connectivity Control in UAV-Assisted Adhoc Networks Using Limited Network Sensing

Abstract

Unmanned Aerial Vehicle (UAV) based wireless networks have attracted significant attention due to their flexibility and efficiency in establishing ad hoc communication systems. In such networks, UAVs are typically tasked with following user groups on the ground to ensure consistent wireless connectivity with a high quality of service. However, most existing works assume access to precise user location data, which is often impractical in real-world scenarios. This paper instead investigates a UAV-based wireless network where only the Received Signal Strength Indicator (RSSI) feedback from users is available. Relying solely on RSSI introduces challenges due to rapid fluctuations arising from inherent wireless channel characteristics. To address this, this paper proposes a two-phase Deep Q-Network (DQN) approach with a carefully designed reward mechanism. The first phase handles UAV deployment, while the second phase tracks dynamic user cluster. Simulation results demonstrate that the RSSI-only model achieves better performance compared to a location-based baseline, while requiring a state space only 25% the size of the baseline and a replay buffer that consumes just 34% of the memory. By eliminating the dependency on user location and leveraging universally available RSSI information, this work presents a more practical and robust solution for UAVbased wireless networking. Further, experimental results using a DJI Tello drone demonstrate the practical feasibility of the proposed approach in a real-world setting.