Distinguishing Micromobility and Blockage in 6G Sub-THz Systems: A Machine Learning Approach

Abstract

Dynamic human body blockage and user equipment (UE) micromobility in the hands of a user are two impairments that may lead to connectivity losses in 5G millimeter wave (mmWave, 30-100 GHz) and 6G sub-terahertz (sub-THz, 100-300 GHz) systems. The efficient handling of these events calls for fundamentally different actions from the system, with the former requiring a search for a new base station (BS), while the latter involves a beam tracking procedure. Failing to differentiate the reason for an outage may lead to a prolonged time in outage conditions when blockage is misinterpreted as micromobility and/or waste of system resources when micromobility is mistaken for blockage. This work presents an algorithm that distinguishes between blockage and micromobility events at 156 GHz, developed using machine learning (ML) techniques based on measurements of both phenomena. Our numerical results show that among all the considered algorithms, ensemble-based techniques such as Random Forest achieve the highest accuracy of around 95% in distinguishing between micromobility and blockage impairments. Random Forest is also characterized by a very small fraction of false negative decisions falsifying blockage events for its absence and thus misinterpreting blockage for micromobility which is costly in terms of the amount of time spent in outage conditions. These properties make Random Forest a preferable choice, allowing for a simple and robust online implementation.