[PhD defence] 28/11/2024 - Antoine DEJONGHE : "Self-organising networks and learning algorithms for post-5G networks" (UPR LIA)

Antoine Dejonghe will submit his thesis on 28 November 2024 on the subject of "self-organising networks and learning algorithms for post-5G networks".

Laboratory

UPR 4128 LIA - Avignon Computing Laboratory

Composition of the jury

MR FRANCESCO DE PELLEGRINI	Avignon University	Thesis supervisor
MR ZWI ALTMAN	Orange Labs	Thesis co-director
Ms INBAR FIJALKOW	ENSEA	Rapporteur
Ms MARIE-LINE ALBERI MOREL	Nokia Paris-Saclay	Examiner
MR SALAH EDDINE EL AYOUBI	Centrale-Supélec	Examiner
MR MARCO DI RENZO	Centrale-Supélec	Rapporteur

Summary

The concept of self-organising networks (or SON) was introduced in 4G and extended to 5G in order to simplify the operation of mobile networks. It is based on intelligent algorithms capable of automating complex operations. 5G and post-5G networks are introducing technological developments that will have a significant impact on SON. Some SON algorithms need to be adapted to take account of new 5G functionalities such as massive antenna systems (or mMIMO). New SON algorithms must also be designed to meet new needs. Artificial Intelligence (AI) is gaining importance in 5G and post-5G networks, and can be seen as a natural evolution of SON. This trend is particularly important in O-RAN networks where AI is natively supported. This thesis aims to exploit optimisation, AI and control theory techniques to develop SON algorithms for the optimisation of 5G and post-5G O-RAN networks. These algorithms are designed to offer strong performance guarantees, converge quickly, have low complexity, and be robust to a certain level of noise.

Firstly, this thesis aims to adapt 4G load balancing (MLB) algorithms to 5G and post-5G networks. In mobile networks, uneven load distribution between cells is often the cause of congestion problems, which in turn can lead to significant degradation in network performance. In order to solve these problems, MLB algorithms aim to balance traffic between the different cells of a network by optimising mobility parameters. To date, few works in the literature have proposed implementations of these algorithms in the context of 5G. However, their application to this technology requires several adaptations, in particular to mMIMO systems and multi-user scheduling. Furthermore, the introduction of mMIMO systems in 5G networks allows the exploitation of performance metrics evaluated at the beam resolution, introducing the challenge of load balancing at this scale. Thus, using techniques from control theory, this thesis aims to show how 4G MLB algorithms can be efficiently adapted to 5G networks. This thesis also aims to optimise the Energy Efficiency (EE) of mMIMO networks. The deployment of mMIMO systems considerably improves the performance of mobile networks. However, the improved performance achieved by increasing the number of antennas equipping these systems is generally accompanied by an increase in energy consumption due to the use of additional RF chains. In mMIMO networks, an effective energy-saving mechanism is to switch RF channels and associated antennas on and off according to network load. Therefore, based on AI techniques, this thesis aims to design an efficient and low-complexity solution for RF channel activation and deactivation. This is to maximise EE under QoS constraints.

Finally, this thesis addresses the problem of optimising mobile networks assisted by intelligent reflective surfaces (or RISs). The deployment of RISs in future 6G networks should considerably improve the coverage of these networks. However, this deployment presents a number of challenges, including the configuration of base stations and RISs, and the fair allocation of radio resources to users. Thus, by exploiting techniques from control theory and convex optimisation, this thesis aims to address these challenges jointly by introducing a new low-complexity solution.

Key words : self-organising networks, learning algorithms, post-5G networks