Throughput Based Adaptive Beamforming in 5G Millimeter Wave Massive MIMO Cellular Networks via Machine Learning

In this paper the performance of an adaptive beamforming framework is evaluated, when deployed in fifth-generation massive multiple-input multiple-output millimeter wave cellular networks. To this end, active beams are formed dynamically according to traffic demands, in order to maximize spectral and energy efficiency (SE, EE) with reduced hardware and algorithmic complexity. In the same context, a machine learning (ML) approach is considered as well, where the configuration of the active beams per cell is directly related to the requested throughput in the cell's angular space. According to the presented results, the ML-assisted beamforming framework can improve EE with reduced algorithmic complexity compared to the non-ML case, depending on the tolerable amount of blocking probability.