A Study of Clustering Algorithms for Time-Varying Multipath Components in Wireless Channels

Extensive channel measurements have shown that multipath components (MPCs) generally exist as clusters, and cluster-based channel models are therefore widely used for system design and assessment. Since the dynamic behavior, i.e., the time evolution, of the channels plays an important role for many applications, an accurate algorithm for the clustering of time-varying MPCs is required; a variety of algorithms have been proposed, yet little attention has been given to a fair comparison of their relative advantages and drawbacks. In this paper, we review and investigate the typical clustering methods for MPCs in wireless channels. Three popular classes of algorithms, namely distance-based (K-Power-Means), density-based (K-power-density), and evolution-based clustering methods, are analyzed and compared based on both synthetic and measured channel data. The F-measure is used to quantify and evaluate the clustering performance of the three algorithms, and also investigate their performance when only static snapshots of the channel are available. From the comparison, the evolution-based clustering method shows great potential to address the dynamic clustering problem for wireless time-varying channels.