A Framework of Multipath Clustering Based on Space-Transformed Fuzzy c-Means and Data Fusion for Radio Channel Modeling

In the radio channels of cellular networks, signals generally propagate through multiple paths in scattering environments. Multipath Components (MPCs) are shown to be distributed in different groups, known as clusters, in Channel Impulse Responses (CIRs). Clustering MPCs is a critical step in channel measurement and modeling. Many clustering algorithms have been proposed but few of them are able to handle noise effectively. In this paper, we propose a de-noising MPC-clustering framework based on a new Space-Transformed Fuzzy c-Means (ST-FCM) algorithm and the fusion of channel measurement snapshots. ST-FCM solves the issue that the Multipath Component Distance that quantifies the similarity among MPCs cannot be adopted in the conventional FCM algorithm. Then we apply the Dempster-Shafer evidence theory to fuse the clustering results of multiple snapshots, which can detect and remove noise by making a full use of all the measurement data. Furthermore, we design a censoring process for hard partition and a validation process to determine the optimal number of clusters. We have performed extensive simulations on MPC clustering using the CIRs generated by the Third Generation Partnership Project 3-dimensional channel models. We also have developed a space-time channel sounder and have performed experiments in a typical rural macrocell scenario. The simulation and experiment results have shown that the proposed framework has a better performance in clustering accuracy than the current methods.