Master of Science (MS)・
通訊工程學系 2018 - 2020
Description
Thesis of Master Degree: Joint Trajectory Design and BS Association for Cellular-Connected UAV: An Imitation Augmented Deep Reinforcement Learning Approach, 2020.
• Require: UAV trajectory should be designed to meet the following items.
o The flight duration of UAV is limited by the onboard battery capacity, so that length of UAV trajectory should be minimized to reduce energy consumption.
o Receive reliability control and command (C2) signals from the GBS for flying status monitoring.
• Motivation: UAV-BS association should be taken into account when designing UAV trajectory in order to reflect the realistic link performance of aerial users. However, aforementioned works did not consider its issue.
• Goal: Present a joint design of UAV trajectory and BS association with the objective to minimize the mission completion.
• Approach: Propose an imitation augmented deep reinforcement learning (DRL)-based method to minimize UAV trajectory length and thereby achieve fast convergence to the optimal policy. Also, utilize deep neural networks (DNN) to approximate the nonlinear mapping from UAV’s position to the optimal BS selection.
• Results: Proposed DRL based approach achieves faster convergence speed and shorter trajectory compared to the standard DRL. Besides, justify the superiority of DNN-based association strategy over the conventional nearest and max-SINR association strategies.