On supervised channel estimation techniques for very large MIMO communication systems

This paper compares and assesses the performance of different channel estimation techniques for multicell multiuser multiple-input multiple-output (MIMO) systems in a very large (VL) MIMO scenario. Some of these techniques exploit properties of large random matrices and are less affected by pilot contamination, which is the case of the technique based on the eingenvalue decomposition (EVD) of the output covariance matrix. Other techniques do not exploit such properties and are more sensitive to errors as the number of antenna sensors grow large, which is the case of the classical least squares (LS) method. Our main goal is to investigate the use of those techniques in a VL-MIMO system under different scenarios. This paper also proposes a new method to solve the multiplicative matrix ambiguity of the EVD-based method by using a simple Khatri-Rao product. Numerical results are shown to atest the increased effectiveness of the EVD-based channel estimation in a VL-MIMO environment.