Initial access in 5G mmWave networks with different base station parameters

Abstract: Nowadays in the fifth generation (5G) communication systems, millimeter wave (mmWave) has aroused interest to not only industrial use but also network operators due to the massive amount of bandwidth available at mmWave frequencies. Initial access in cellular systems is an essential procedure in which new mobile user equipment (UE) establishes a connection with a base station (BS). However, mmWave relies on highly directional beamforming (BF) to overcome its severe path loss, while the initial access requires a wide beam to obtain sufficient information for beamforming. So the challenge is to handle the balance between highly directional mmWave and fast and reliable initial access. The high path loss of millimetre wave transmission dictates that multiple BSs may be closer and interfere more with each other. We focus our study on two BS parameters under the random search method. In our study, the beamwidth can be different for each BS, but a uniform number of slot limits needs to be chosen for all BSs. Our objective is to obtain the best parameters for each BS in a reasonable period of time. We build a systemlevel simulation in MATLAB and explored a variety of methods to select the best parameters, including reinforcement learning, supervised learning, and genetic algorithms. It is identified that the main challenge of applying reinforcement learning and supervised learning is the exponentially growing variety of BS parameters. A genetic algorithm is able to derive approximate best values in complex relational species. Therefore the genetic algorithm is considered to be able to be applied in scenarios with a high number of BSs. The result shows that reinforcement learning has great performance in a few BS cases, and the genetic algorithm is able to provide a large improvement over most of the BS methods with the same parameters.