Interference-Aware Trajectory Design for Fair Data Collection in UAV-Assisted IoT Networks by Deep Reinforcement Learning

Abstract

Unmanned Aerial Vehicle (UAV)-assisted data collection for Internet-of-Things (IoT) systems has drawn increasing attention, which is pivotal for providing seamless coverage and promoting system performance in wireless networks. In this paper, we study a UAV-aided data collection scenario where a UAV takes off from an initial location and flies to multiple abutting ground sensor nodes (SNs) arbitrarily scattered in the physical environment to collect sensor data. Specifically, the UAV synchronously communicates with SNs when passing through the collective communication area. Considering the signal-to-interference-plus-noise ratio (SINR) and fair performance among SNs, we maximize the fair throughput among all SNs and minimize energy consumption by optimizing the trajectory of the UAV and power allocation. The multi-optimization problem is modeled as a Markov Decision Process due to the system dynamic. To handle the continuous state and action space in this problem, we propose a deep reinforcement learning-based algorithm, named as multi-optimization trajectory design and power allocation (MOTDPA), which chooses the state-of-the-art methods, soft actor-critic with prioritized experience replay to find the efficient policy. Meanwhile, we use min-max state normalization (MMSN) to stabilize the training process. Simulation results demonstrate the better performance of the proposed approach than other commonly used baselines.