Multi-Path Grouping Using a Novel Clustering Algorithm for Stochastic Channel Modeling

In this contribution, a novel clustering algorithm is proposed which is applicable for generating accurate parameters for the Clustered-Delay-Line (CDL) channel models. In this algorithm, we first use a Kolmogorov-Smirnov testing method to split multiple channel impulse responses into individual segments, each containing the observations of a wide-sense stationary channel. Then, the path estimates returned by the Space-Alternating Generalized Expectation-maximization (SAGE) algorithm are grouped into clusters by using a modified K-means iterative method based on multipath component distance (MCD). The dispersion dimensions considered for path grouping include delay, azimuth and elevation of arrival, as well as azimuth and elevation of departure. Different from the conventional clustering algorithms that rely on heuristic settings for weighting the MCD in dispersion dimensions, we cluster the paths in a specific sequence to different dimensions, which is predefined to reflect the priorities of the dimensions for the usage of the CDL models. The performance of the proposed clustering algorithm is compared with the conventional clustering algorithm by using measurement data in three indoor scenarios.