Pilot Contamination in Massive MIMO: A Measurement-Based Analysis Using 2D-MUSIC

When a base station (BS) sees desired and interfering users at different angles, it results in non-overlapping angle-of-arrival (AoA) regions for those users. This is important for reducing pilot contamination (PC) of massive MIMO systems. Most state of the art studies utilize a simple non-line-of-sight (NLoS) one-ring model which assumes sparse support and can reasonably schedule users with different AoA to minimize PC. However, it is not confirmed with measurements that the one-ring model is accurate to characterize the real performance of PC. In this paper, we aim to validate these assumptions by analyzing channel measurements for an NLoS indoor to outdoor massive MIMO channel. Our results show a significant overlap in AoA when users and BS are separated by a wall. First, we analyze how the AoAs depend on the user locations by transforming the measured channels to the angular domain. Then, we analyze the dependency of AoA overlap of two users and the impact on PC, and our results show that there is a very high correlation. This finding indicates that angular information is useful in pilot scheduling to reduce PC. To be precise, instead of employing a virtual antenna array, a channel measurement from a 64-antenna rectangular array operated at 2.6GHz in the BS is considered. Furthermore, two deployment topologies for the BS array are considered, namely, collocated and distributed sub-arrays are included for quantifying the benefits from improved channel diversity when distributing the arrays.