The Use of Machine Learning for Remote Discrimination of Applications’s Classes in 6G Terahertz Systems with Directional Antennas

User micro-mobility is known to affect the performance of applications in systems with directional antennas, such as millimeter wave (mmWave) 5G New Radio (NR) cellular systems and future 6G systems. It has been shown that the type of application currently utilized by the user specifies the micro-mobility properties and thus affects the time between beam-tracking time instants. In this study, we utilize the real-time measurements of the received signal strength at 156GHz with user equipment subject to micro-mobility patterns to design machine-learning (ML)-based methods for application detection. Our numerical results show that even the simplest ML classifiers, such as trees and random forests, are capable of distinguishing between applications with slow and fast micro mobility with extremely high accuracy, reaching 97%. Differentiating between all the applications’ classes with high, moderate, and slow micromobility is more difficult, but is still feasible with an accuracy higher than 80%. To improve this accuracy, further deep learning methods that capture the time-dependence in the time-series structure are needed.