Automatic data segmentation based on statistical hypothesis testing for stochastic channel modeling

In order to extract the statistical characteristics of propagation channels, multiple impulse responses of the channels are measured and saved e.g. in terms of cycles or bursts. In the case where measurements are performed in time-variant environments, the received signals need to be divided into multiple segments. It is essential to perform the data segmentation reasonably in order to guarantee that each segment contains the observations of the same stationary process. In this contribution, we first show experimentally the impact of the number of data bursts per segment on the clustering results. Then a novel Kolmogorov-Smirnov hypothesis-testing-based approach is proposed for automatically determining the number of bursts in each segment. The applicability of this approach is evaluated using indoor channel measurement data. The results obtained show that the number of bursts in a segment follows lognormal distributions with parameters dependent on the environments and the mobilities of the transmitter, receiver and the scatterers during the measurement campaigns.