A Novel Wireless Channel Clustering Algorithm Based on Robust Mean-Shift

Abstract

Clustering characterizes the grouping of multipath components (MPCs) in radio channels. Accurate clustering is a prerequisite for cluster-based channel characterization and sensing in the beyond fifth-generation (B5G) and sixth-generation (6G) communication. However, existing clustering algorithms commonly depend on thresholds and initializations, and are not fully consistent with the characteristics of MPC distributions in the radio channel. Additionally, clustering based on power spectrum has not been thoroughly researched. In this paper, we propose a unified clustering method named power-weighted nearest-neighbor robust mean-shift (MP-NN-RMS) algorithm, which is a kernel density estimation (KDE)-based method. The K-nearest neighbor (KNN) kernel is utilized to adapt to the changes in local density. Two variants of this clustering method for the power spectrum and MPCs are provided. Both simulation and measurement-based verifications demonstrate the effectiveness of the proposed algorithms. Compared with traditional clustering methods, the proposed algorithm can achieve more accurate and robust clustering results without requiring prior information of predefined parameters or models. Moreover, mathematical proof on the convergence guarantees the rationality of the proposed algorithm. This advancement is beneficial for the development of future wireless communication systems.