Project Abstract

Wireless communications have played an essential role in realising the world as we experience it. Key services will continue to proliferate and become more mobile, whilst on-demand information and entertainment will be delivered over wireless networks, which will result in our ‘always-on’ society. Future wireless systems must support an increased traffic volume by increasing capacity, extending coverage, and improving efficiency with respect to cost, energy, and spectrum utilisation. Deep Learning (DL) algorithms are inherently data-driven and, rather than depending upon model assumptions and approximations and hence they implicitly capture hidden environmental effects.

Furthermore, the future wireless networks also promise to provide ubiquitous connectivity to a multitude of devices with diversified traffic patterns wherever and whenever needed. For the sake of boosting resilience against faults, natural disasters, and unexpected traffic, both the Unmanned Aerial Vehicle (UAV) and Low Earth Orbit (LEO) satellite-based non-terrestrial wireless communication systems can provide a unique opportunity to cater for such demands in a timely fashion without relying on the overly engineered cellular network. However, for UAV- and LEO satellite-assisted communication, issues of capacity, coverage, vertical network design, relevant parameter optimisation, and energy efficiency are considered of paramount importance.

By leveraging the great potential of deep reinforcement learning, the best point of harmonising those key factors can be investigated and found. Based on the above-mentioned, we can design a novel intelligent adaptive UAV and LEO satellite communication system, which is optimised for Hierarchically Layered Non-Terrestrial Communications for the ‘Always-On’ Intelligent Information Society.