The Random-cluster Model – a Stochastic Mimo Channel Model for Broadband Wireless Communication Systems of the 3rd Generation and Beyond

This thesis presents the Random-Cluster Model (RCM), a novel geometry-based stochastic model for frequency-selective and smoothly time-variant MIMO radio channels. The RCM uses the concept of clusters, i.e. groups of multipath components (MPCs), to model the propagation environment. In the RCM, the environment is solely specified by the multivariate distribution of the cluster parameters, such as the cluster positions, the cluster movement, and the cluster spreads. In this way, even correlations between the cluster parameters are easily reflected. The most significant feature of the RCM is that it is parametrised directly from channel measurements by an automatic procedure. In this way, the RCM is specific to the environment. It closes the gap between channel measurements and channel modelling. By using clusters, the model parametrisation complexity becomes quite low. Compared to a single propagation path, needing 6 parameters per modelled time instant, a cluster is described by as few as 21 parameters for the whole cluster lifetime (over many time instants). Since a cluster usually consists of 6 to 20 propagation paths, the reduction in the number of parameters is significant. Of course, the cluster distribution needs to be parametrised accurately. Identifying clusters from timevariant MIMO channel measurements constitutes the basis to parametrise the RCM consistently. Since visual clustering algorithms are highly subjective and cumbersome to use, I concentrated on the long cherished goal of automatic clustering. This thesis compares different clustering approaches regarding their suitability for multipath clustering, i.e. visual clustering, hierarchical clustering, K-means clustering, and Gaussian Mixture Model clustering. Eventually, I present a new and complete framework that provides a solution for automatic clustering and tracking of MPCs, which bases on (i) an initial-guess estimator choosing clusters to be as separate as possible, (ii) the KPowerMeans algorithm, an extension of the K-means algorithm, that takes the power of MPCs into account and handles the ambiguity of the angular domain, and (iii) a Kalman filter for cluster tracking. Using this framework, the RCM is automatically parametrised from measurements. To account for non-discrete contributions in the MIMO channel, the RCM is the first channel model to include the concept of diffuse multipath (DMP) modelling. I will argue that validation is an important task when completing a channel model. I validate the RCM by comparing its fit to measurements using the following validation metrics: (i) mutual information, which will turn out to be no distinctive validation metric, (ii) channel diversity, (iii) the Demmel condition number of the MIMO channel matrices, and (iv) my Environment Characterisation Metric (ECM) comparing directly the discrete propagation paths in the channel. It turns out that the RCM shows a very close fit to measurements, making it well suitable to simulate channels in the kind of measured ones. To satisfy the never-ending need for measurements, I conducted a MIMO channel sounding campaign at the University of Oulu, Finland. I measured a total number of 28 scenarios in three different indoor environments: offices, larger rooms, and a big hall. To compare the frequency dependence of the model parameters, each measurement route was sounded at 2.55 GHz and at 5.25 GHz. The thesis contains a comprehensive documentation of the measurement campaign.