In practical environments, recent massive MIMO measurements demonstrate that user channels can be correlated. In this paper, we study the user channel correlation induced by shared angles-of-departure. We first derive the user correlation distribution in the large array regime, and then examine the user correlation using actual measurements from a large array. As a data-driven observation, we discover that the correlation of all close-by users is higher than $0.4$ and barely reduces as the number of base-station antennas $M$ increases beyond $36$ antennas. Furthermore, nearly one-third of users, even when they are tens of wavelengths apart, have a correlation that is more than twice the correlation of an i.i.d. Rayleigh fading model. Lastly, we characterize the impact of user correlation on system performance. As $M$ increases, conjugate beamforming systems suffer a linearly growing inter-user interference due to correlated channels. However, for zero-forcing beamforming systems, the inter-user interference is a constant that does not increase with M. In particular, zero-forcing beamforming systems can serve a linearly increasing number of correlated users and achieve a linear growth in the system achievable rate as $M$ increases. Hence, spatial multiplexing correlated users can be an attractive massive MIMO design.
Download