AI for 5G SON

With large growth of global cellular traffic, and projections for further increase, interference has become a prominent problem for network operators, often limiting the quality of service that can be provided to users. Additionally, rapid and large-scale densification is expected in 5th Generation networks, through the introduction of small cells. This is expected to further complicate the interference problem by adding additional sources of interference. Densification also makes the extensive level network planning currently experienced undesirable when deploying new cells. Self-Organising functionality aims to reduce the impact of minimal cell planning on the operation of the network by performing installation and management tasks autonomously. While interference mitigation techniques seek to reduce the impact of interference on the user experience. Techniques to mitigate interference are broad and diverse, with a large body of literature dedicated to the problem.

With so many methods of mitigating interference, and the need for reduced network planning, the 5G network will be subject to increased unplanned interaction between diverse methods. In this project, we focus on the interaction of scheduling-based interference coordination techniques; where interference is mitigated by the efficient and interference aware scheduling of spectral resources. The goals are to investigate the effects of unplanned interactions between scheduling methods and providing a method for the dynamic coordination of these techniques. This method will look to increase the quality of service provided to users that experience a high level of interference, by reducing the impact of unplanned interaction between scheduling techniques.

To achieve this, we propose intelligence based dynamic configuration of diverse distributed scheduling methods within the network according to reinforcement based deep learning techniques. By doing so the method aims to reduce the impact of unplanned interactions while increasing the quality of service offered to the most interfered users. Additionally, such a method would facilitate the rapid deployment of network cells by reducing the need for intensive network planning.