Automatic Clustering of Nonstationary MIMO Channel Parameter Estimates

Many geometrical channel models with stochastically placed clusters of scatterers have been proposed in literature. A major practical problem related to the parametrization of such models is the identification of scattering clusters from channel measurement data, which is typically multidimensional and nonstationary. Conventionally, visual inspection has been used for the cluster identification. Such an approach may be suitable for short data records, but becomes impractical when a large amount of measurement data has to be analyzed. In this paper, we propose an automatic procedure for finding clusters from an output of a channel parameter estimator, such as SAGE. The algorithm is based on sequential clustering of windowed multipath estimates, and tracking of cluster centroids in consecutive data windows. Visual inspection of the automatically identified multipath clusters is usually still required when processing measurement data. The practical benefit of the present method is that it significantly speeds up the process of cluster extraction with large data records.