A Novel Power Weighted Multipath Component Clustering Algorithm Based on Spectral Clustering

Abstract

In the real propagation environments, multipath components (MPCs) in wireless channel usually exist as clusters. Cluster based structure of MPCs has been widely used in wireless channel modeling. In this paper, a novel MPC clustering algorithm is proposed based on spectral clustering. Considering that MPCs having strong power should usually be grouped into different clusters, the algorithm introduces a power-weighted processing to identify the similarity of each MPC. The process of weighting the power is conducted by generating similarity matrix using the full link method of Gaussian kernel function in the traditional spectral clustering algorithm. In order to achieve high clustering accuracy, the dimensionality reduction method of Laplacian Eigenmap is decomposed by using the normalized cut method, the obtained eigenvectors are re-clustered to cut the generated similarity matrix for MPC clustering. In the simulation results, it is found that this algorithm can well separate MPCs with high powers into different clusters and achieves better clustering performance compared with KPowerMeans, Kmeans, and the traditional spectral clustering algorithms.