Machine Learning-aided MmWave Beam Tracking

Millimeter-wave(mmWave)beamforming has recently attracted considerable interest as a critical element of the fifth-generation mobile communication systems and beyond.However,beamforming and directional communications raise various challenges related to beam management.Specifically,the base station(BS)needs to sequentially sweep its beams for the user equipment(UE)to measure their qualities,determine the best beam and report it.This directional procedure incurs significant delay and overhead,which lowers the link throughput and spectral efficiency.Furthermore,the mm Wave communication in the quasi-static scenario has gradually matured,while the research of mobile mm Wave communication systems in the dynamic scenario still faces many challenges,which is currently one of the academic issues with the most research potential.Therefore,this thesis proposed a deep learning model integrated in the mm Wave communications system for beam tracking.First,we designed the deep neural network.Then we trained the model to learn the hidden pattern between the mm Wave environment(via an uplink pilot)and the optimal beamforming vector using accurate ray-tracing scenarios of the DeepMIMO dataset[1].Furthermore,the performance of the developed solution has been evaluate using the metrics of average sum throughput of the system.Numerical results obtained show that the average sum rate of the proposed system approaches that of the exhaustive search base beamforming solution that knows the optimal beamforming with no training overhead.In addition,the proposed solution achieves a noticeable gain,especially when users are moving with high speed and when the BS deploy large antenna arrays.