Addressing UAV Relay Challenges Using Deep Reinforcement Learning
Ref: CISTER-TR-191205 Publication Date: 17, Dec, 2019
Addressing UAV Relay Challenges Using Deep Reinforcement Learning
Ref: CISTER-TR-191205 Publication Date: 17, Dec, 2019Abstract:
The use of unmanned aerial vehicles (UAVs) as
ying wireless communication platforms
has attracted the attention of researchers over the past few years due to their autonomy,
exibility, and a broad range of application domains. Indeed, UAVs enjoy having a broad
range of applications including military, surveillance and monitoring, telecommunications,
delivery of medical supplies, and rescue operations. UAVs can be utilized as wireless relays
for improving connectivity of ground wireless devices and extending network coverage.
Also, UAVs can act as mobile aerial base stations to provide reliable communications for
ground users and extend the capacity of wireless networks. The high altitude of UAVs
can enable line-of-sight (LoS) communication links to the ground users. Due to the adjustable
altitude and mobility, UAVs can move toward potential ground users and establish
reliable connections with a low transmit power. Hence, they can provide a cost eective
solution to collect data from ground mobile users that are spread around a geographical
area with limited terrestrial infrastructure.In this dissertation, we focus on UAV relaying
as one use case of UAV-assisted wireless communication, and propose to address three of
its main challenges. Scheduling algorithms are important components in the provision of
guaranteed quality of service parameters such as delay,jitter,packet loss rate, or throughput.
Scheduling has been studied in wireless networks,however the unique characteristics of UAV
assisted wireless communications such as, mobility of UAV,timing-varying channel conditions
and multi-user diversity,motivate us for the development of new scheduling solution
that are specically tailored for this environment. Trajectory optimization is an important
design aspect of UAV. For UAV relay systems,trajectory optimization can increase network
throughput and substantially shorten the communication distance and hence is critical for
high-capacity performance. Optimizing the
ight path of UAVs is challenging as it requires
considering many physical constraints and parameters.Furthermore, solving a continuous
UAV trajectory optimization problem is challenging from analytical point of view since it
involves nding an innite number of optimization variables such as UAV's locations. In
the realm of security for UAV, GPS spoong is one of the most important cyber attacks. In
this attack, the GPS receiver on the UAV is feed with fake signals generated by a malicious
user so that UAV is unable to distinguish between the fake signals and the ones receiving
from GPS. The fake signals can mislead not only the aircraft but also air trac controllers,
leading to serious problems. These problems range from aircraft hijacking to collisions and
human casualties. On the other hand, deep reinforcement learning(DRL) has capability
to address these emerging issues in UAV-assisted wireless communication through interaction
with environment. Indeed, DRL provides an autonomous decision-making mechanism
for the network entities to solve non-convex, complex model-free problems. Hence, in this
dissertation, we utilize DRL to address the mentioned challenges.
Notes: Comissão de acompanhamento: Comissão Científica PDEEC: Luis Almeida Orientador: Kai Li Coorientador: Eduardo Tovar Elemento da FEUP: Luis Almeida Elemento externo: Xiaoming Fu (University of Göttingen, Germany)
Record Date: 12, Dec, 2019