Clustering Automatico di Stime di Parametri di un Canale MIMO Usando MATLAB

This thesis addresses the problem of identifying clusters from an input data set consisting of several MIMO channel parameter estimates, achieved by a channel parameter estimator, such as SAGE. In many previous work visual inspection has been used for the cluster identification. Nevertheless this approach becomes impractical when one has to analyze large amounts of measurement data. Furthermore visual inspection is limited by human senses, since one is hardly able to identify clusters in more than three dimensions. In this work we introduced a novel scalable framework to automatically identify multi-path clusters, which requires only a very limited user interaction. The proposed framework is novel in five respects: (i) the framework algorithm enables to cluster MIMO channel parameter estimates automatically with a minimum of user input; (ii) as a distance metric we use the multi-path component distance (MCD), which makes the framework able to scale the data, solve the problem of the angular periodicity and hence enable for joint multidimensional clustering; (iii) by including power and the MCD into the clustering algorithm, we make it applicable to cluster identification in propagation research; (iv) the cluster validation package provides a trustworthy estimate of the correct number of clusters; (v) the cluster pruning package does not change the cluster behavior significantly, but improves visibility and future tracking performance. Furthermore, the clustering algorithm introduces a convincing, inherent definition of a cluster itself. We assess the performance of the framework by clustering both synthetic scenarios and real-world MIMO measurement data from an indoor big hall environment.