Stochastic Cluster Models for 5G Channels Based on Passive Sounding in In-Service Networks

In this paper, a passive channel sounding system specifically designed for the fifth generation communication systems (5G) was used to conduct channel measurements on Sub-6GHz frequency bands in four typical propagation scenarios including time-variant cases on high-speed-train. This passive channel sounder, built with dual-channel software-defined radio peripherals (SDRs), antenna array with 32 elements, parallel multiple-input multiple-output (MIMO) switches and other ancillaries, can be applied to collection the exact 5G downlink signals with center-frequencies of 2-4GHz, bandwidth 122.88MHz in in-service 5G new-radio (NR) networks. Based on post-processing of downlink signals of different types, multi-dimensional channel power spectra in delay, Doppler frequency and direction of arrival, as well as the estimates of parameters of multipath components (MPCs) are obtained. The distribution of MPCs in consecutive snapshots is applied to splitting the observations into wide-sense-stationary segments using Kolmogorov-Smirnov hypothesis-testing (KS-testing). Considering the fact that the passive sounding may result in more severe dispersion for MPCs, we compare the performance of two clustering approaches, i.e.\ the K-means and the Gaussian-Mixture-Model (GMM) and finally choose the former for its superiority in the aspects of cluster tightness, separation and exclusivity. A stochastic channel model in the same format as 3GPP standards is established for the four scenarios considered. A discussion of the difference between this new 5G-NR model and existing WINNER II model is provided and comments on the model applicability for system performance evaluation is made.