Machine Learning Technique for Uplink Link Adaptation in 5G NR RAN at Millimeter Wave Frequencies

Abstract: The demands on wireless communications are continuously growing, due to the fact that when higher network capabilities are delivered, new features and applications are created, calling for even higher requirements. To keep pace with these demands and to allow new applications to rise, the limits of mobile networks must be pushed regularly. Therefore, the International Telecommunication Union targets on achieving new milestones almost every decade. 5G is the fifth-generation standard for wireless cellular networks, which was planned to push the limits once more to a new level. To achieve the standards of 5G, a new frequency spectrum of mmWave was introduced. This spectrum was unutilized in earlier generations due to its complex environment relatively to sub 6 GHz spectrum. However, since then, new techniques were introduced helped to overcome these challenges. This thesis is investigating on the possibility of improving UL Link Adaptation using ML technique on mmWave frequency environment. After studying previous related work and different ML techniques, a reinforcement learning algorithm was suggested. The algorithm uses the feedback of previous actions in consideration when taking future decisions. The system was implemented on a professional simulation tool provided by Ericsson. The results showed an improvement on both throughput and BLER performance when compared to a non-ML system.