An Improved KPD Algorithm for Clustering of Wireless Channel Multipath Components

Multipath components (MPCs) of wireless channels can be described as clusters of scatterers and reflectors with specific delay and angular power spectrum densities (PSDs). Hence, cluster-based wireless channel modeling has been widely adopted to characterize propagation environments for a good tradeoff between model accuracy and complexity. To enhance the performance of channel modeling, accurate clustering algorithms for MPCs are required. However, the accuracy of the existing kernel-power-density (KPD) algorithm for MPC clustering is limited by user-specific parameters, such as the nearest MPCs in the density calculation and the threshold in the cluster merging. In this paper, an improved KPD clustering algorithm is proposed without relying on priori knowledge. Specifically, the densities of MPCs and the minimum distance from one MPC to other MPCs with higher densities are jointly considered. A two-stage clustering method is used to mitigate the impact of the user-specific parameters. The key MPCs are found in the first stage. A novel criterion for determining cluster centroids is implemented in the second stage to merge neighbouring clusters generated in the first stage. To evaluate the performance of the proposed method, both simulation and measurement validations are performed. Results support the superiority of the improved KPD algorithm over the existing KPD algorithm for MPC clustering.