Resource management of space, frequency and power in 5G networks using machine learning

With the onset of the 5th generation of wireless communications, new requirements have formed for various types of users. The New Radio systems are required to serve users of diverse needs such as personal mobile devices, autonomous driving vehicles, industrial machines, dark factories and household appliances. The demand on data rate, reliability and traffic volume have increased immensely. To accommodate the much higher user traffic and data rates, new frequency ranges have been introduced. The higher frequencies have made it essential to use beamforming as a way to increase the Quality of Service by improving signal integrity at the User Equipment, making beam management an important point to optimize. To allocate the available resources of a 5G network, Radio Resource Management is conducted, allocating power and frequency resources, handling user associations and handovers etc. The management of beams, power and resource blocks can be formulized as an optimization problem, where we aim to maximize the CQI of each user, as an indicator of quality of the downlink connection. In this thesis, we investigate the use of reinforcement learning to allocate space, power and frequency resources jointly. We aim to achieve better performance with a Deep Q-Network than classical optimization methods or an exhaustive search in the resource space.