Unsupervised Clustering for Millimeter-Wave Channel Propagation Modeling

Abstract

To date, we have designed and assembled millimeter-wave channel sounders at 60 GHz and 83 GHz. They can estimate the angle-of-departure and angle-of-arrival of channel multipath components as well as their delay and Doppler frequency shift. In addition, due to the fast acquisition time and because the receiver is mounted on a mobile robot, the systems can collect measurements for hundreds of different transmitter-receiver configurations in just minutes. It follows that channel-model reduction, including the multipath- component clustering process, must be reliable, consistent, and unsupervised. In this paper, we describe a simple clustering process tailored to the properties of millimeter-wave channels that fully exploits the multi-dimensionality of the extracted multipath components and requires only a few tunable parameters. Through extensive experimentation, we have verified that the process is robust and delivers consistent results across five different environments and across both frequency bands investigated. Illustrative examples are provided.