Automatic clustering of multipath arrivals in radio-frequency channels using kurtosis

In wireless channel propagation modeling, the multipath arrivals of a transmitted signal appear in clusters at the receiver. Because the notion of clusters tends to be intuitive rather than well-defined, cluster identification has traditionally been carried out through human visual inspection. Besides time-consuming for large-scale measurement campaigns, this approach is subjective and will vary from person to person, leading to arbitrary selection of clusters. To address these concerns, automatic clustering algorithms have emerged in the past decade. Most, however, are laden with settings which are very sensitive to different radio-frequency environments, again leading to arbitrary selection. In this paper, we propose a novel clustering algorithm based on the kurtosis metric which, in related work, has been used precisely for its channel independence. We compare ours to two recent algorithms through a standard validation method on simulated channel impulse responses from five different environments. The proposed algorithm delivers better results and, because it has no channel-specific settings, is inherently robust to varying channel conditions.