Novel pilot decontamination methods for Massive MIMO systems under practical scenarios

Accurate and efficient channel estimation methods have the ability to realize the theoretical gain in multi-input multi-output (Massive MIMO) systems which have a massive number of antennas. However, the pilot contamination in Massive MIMO channel estimation process, rooted from the pilot reuse, is a critical problem that severely degrades the performance of the system. This work aims to address the problem of pilot contamination in covariance-aided channel estimation methods while considering practical scenarios where the channel covariance matrices change due to a new user arrival and users mobility. To that end, we first design a method to track the channel covariance matrices and then use these estimated values in Bayesian estimation. Simulation results indicate that the normalized mean square error (NMSE) for both channel covariance matrices and the CSI itself of our proposed methods are much lower than those of classical methods based on least square (LS) and Bayesian estimation. Additionally, for the case that users move slowly (e.g., at walking speed), our proposed method can provide satisfactory performance for more than three times as much as classical Bayesian estimation before system re-train channel covariance matrices. In other words, compared with classical Bayesian methods, our proposed methods are able to get good system performance with less overhead and complexity by a lower frequency of re-training process.