Learning-Based UE Classification in Millimeter-Wave Cellular Systems With Mobility

The authors in [7] train a six-layer fully efficient beam tracking it is advantageous to classify users according connected network on real-time observations at sub-6 GHz to their traffic and mobility patterns. Research to date bands to predict beamforming weight coefficients and blockages, has demonstrated efficient ways of machine learning based while the study of [8] demonstrates the use of a simple, UE classification. Although different machine learning approaches feed-forward neural network for band assignment to different have shown success, most of them are based on UEs. While [9] surveys an extensive list of related literature, physical layer attributes of the received signal. This, however, the vast majority of the works in the literature only consider imposes additional complexity and requires access to physical layer (PHY) properties of the transmitted/received those lower layer signals. In this paper, we show that traditional signal and do not capture the interaction of the PHY with the supervised and even unsupervised machine learning data link and media access control layers of the system. In reality, methods can successfully be applied on higher layer channel these higher system layers greatly manipulate the PHY measurement reports in order to perform UE classification, signals seen to/from the phased array ports which capture the thereby reducing the complexity of the classification process.