Performance of a Novel Automatic Identification Algorithm for the Clustering of Radio Channel Parameters

Abstract

A multipath component distance (MCD)-based automatic clustering identification algorithm is proposed to group multipath components (MPCs) obtained from radio channels. The developed algorithm iteratively and dynamically assigns the MPCs to the best cluster thanks to the MCD metric. Its performance and robustness are compared with the K-means MCD algorithm using cluster data simulated with four reference scenarios of the WINNER II channel model. The results indicate that K-means MCD is outperformed for all investigated scenarios in spite of its having a lower computational complexity and faster convergence. Moreover, a by-product of the algorithm is an optimal MCD threshold, that is, the characteristic of the cluster statistical properties for a given propagation scenario. This parameter provides a stronger physical link between the MPCs distribution and the propagation scenario. Therefore, it could be introduced in radio channel models with clusterlike features.