Unmanned Aerial Vehicle-Enabled Mobile Edge Computing for 5G and Beyond

The technological evolution of the fifth generation (5G) and beyond wireless networks not only enables the ubiquitous connectivity of massive user equipments (UEs), i.e., smartphones, laptops, tablets, but also boosts the development of various kinds of emerging applications, such as smart navigation, augmented reality (AR), virtual reality (VR) and online gaming. However, due to the limited battery capacity and computational capability such as central processing unit (CPU), storage, memory of UEs, running these computationally intensive applications is challenging for UEs in terms of latency and energy consumption. In order to realize the metrics of 5G, such as higher data rate and reliability, lower latency, energy reduction, etc, mobile edge computing (MEC) and unmanned aerial vehicles (UAVs) are developed as the key technologies of 5G. Essentially, the combination of MEC and UAV is becoming more and more important in current communication systems. Precisely, as the MEC server is deployed at the edge network, more and more applications can benefit from task offloading, which could save more energy and reduce round trip latency. Additionally, the implementation of UAV in 5G and beyond networks could play various roles, such as relaying, data collection, delivery, SWIFT, which can flexibly enhance the QoS of customers and reduce the load of network. In this regard, the main objective of this thesis is to investigate the UAV-enabled MEC system, and propose novel artificial intelligence (AI)-based algorithms for optimizing some challenging variables like the computation resource, the offloading strategy (user association) and UAVs’ trajectory. To achieve this, some of existing research challenges in UAV-enabled MEC can be tackled by some proposed AI or DRL based approaches in this thesis. First of all, a multi-UAV enabled MEC (UAVE) is studied, where several UAVs are deployed as flying MEC platform to provide computing resource to ground UEs. In this context, the user association between multiple UEs and UAVs, the resource allocation from UAVs to UEs are optimized by the proposed reinforcement learning-based user association and resource allocation (RLAA) algorithm, which is based on the well known Q-learning method and aims at minimizing the overall energy consumption of UEs. Note that in the architecture of Q-learning, a Q-table is implemented to restore the information of all state and action pairs, which will be kept updating until the convergence is obtained. The proposed RLAA algorithm is shown to achieve the optimal performance with comparison to the exhaustive search in small scale and have considerable performance gain over typical algorithms in large-scale cases. Then, in order to tackle the more complicated problems in UAV-enabled MEC system, we first propose a convex optimization based trajectory control algorithm (CAT), which jointly optimizes the user association, resource allocation and trajectory of UAVs in the iterative way, aiming at minimizing the overall energy consumption of UEs. Considering the dynamics of communication environment, we further propose a deep reinforcement learning based trajectory control algorithm (RAT), which deploys deep neural network (DNN) and reinforcement learning (RL) techniques. Precisely, we apply DNN to optimize the UAV trajectory with continuous manner and optimize the user association and resource allocation based on matching algorithm. It performs more stable during the training procedure. The simulation results prove that the proposed CAT and RAT algorithms both achieve considerable performance and outperform other traditional benckmarks. Next, another metric named geographical fairness in UAV enabled MEC system is considered. In order to make the DRL based approaches more practical and easy to be implemented in real world, we further consider the multi agent reinforcement learning system. To this end, a multi-agent deep reinforcement learning based trajectory control algorithm (MAT) is proposed to optimize the UAV trajectory, in which each of UAV is instructed by its dedicated agent. The experimental results prove that it has considerable performance benefits over other traditional algorithms and can flexibly adjusts according to the change of environment. Finally, the integration of UAV in emergence situation is studied, where an UAV is deployed to support ground UEs for emergence communications. A deep Q network (DQN) based algorithm is proposed to optimize the UAV trajectory, the power control of each UE, while considering the number of UEs served, the fairness, and the overall uplink data rate. The numerical simulations demonstrate that the proposed DQN based algorithm outperforms the existing benchmark algorithms