Channel Estimation With Pilot Decontamination Based on a Multistep Kalman Filter in mMIMO Communications

Owing to limited radio resources (RRs), nonorthogonal pilot sequences are usually used in time-division duplex massive multiple-input multiple-output communications, resulting in pilot contamination (PC) problems that are detrimental to practical maximum-likelihood channel estimation (CE). Minimum mean-square error (MMSE) CE, which requires a priori covariance matrices, has become preferable. Conventional studies exploited multicell cooperation to overcome PC problems and calculated sample covariance matrices on extra RRs to replace the covariance matrices required for MMSE CE. This paper proposes a novel CE technique based on multistep Kalman filtering (KF) to effectively overcome PC problems without the need for multicell cooperation or extra RRs. To account for both spatial and temporal correlations, the wide-sense stationary uncorrelated scattering channel model is statistically characterized to develop signal and observation models of the KF. The proposed KF-based CE technique works iteratively in pilot and data-aided (DA) modes to not only achieve MMSE CE but also address the outdated-CE problem. The covariance matrix of the channel vector between the UE of interest and the serving base station can be replaced by a sample covariance matrix that is iteratively evaluated during the DA mode. The sum of the covariance matrices of the channels through which all cochannel interference (CCI) UEs propagate is replaced by a single sample covariance matrix that is iteratively evaluated during the pilot mode. The simulation results show that the proposed technique achieves pilot decontamination without extra requirements or computational power for individual sample covariance matrices. This results in a low mean-square error, close to that obtained with complete a priori knowledge of all covariance matrices of the desired channel and all CCI channels.