An improved three-dimensional auto-clustering algorithm for indoor millimeter-wave multipath propagation analysis

This paper develops an improved three-dimensional (3D) automatic KPowerMeans clustering algorithm using Silhouette validation index for millimeter-wave (mmWave) channel analysis. A simplified multipath components (MPCs) estimation method for direction-scan sounding is applied before clustering. Considering the cluster characteristics, we redesign an initial centroid selection algorithm, including path power and multipath component distance (MCD). When the cluster number is less than 6 for each clustering, the improved method effectively reduces the time of iteration compared with the random method. Meanwhile, we give a principle of choosing delay scaling factor when calculating MCD. Thanks to the twice KPowerMeans clustering and redesigned outlier pruning algorithm, the accuracy for the delay statistics in different MPC sets is significantly improved, which enhances the precision of clustering results in visual inspection. Based on the 25.5 GHz channel measurement data in corridor, we evaluate the performance of the improved algorithm and compare several cluster-level channel parameters with existing works at 28 GHz in similar environment. The investigations here demonstrate the cluster characteristics of mmWave propagation and can be used to model mmWave channel in future communication system.