Structured Channel Covariance Estimation for Dual-Polarized Massive MIMO Arrays

Abstract:
Channel covariance estimation is a necessity in various MIMO applications such as minimum mean squared error channel estimation and user grouping. It is a challenging task, especially in massive MIMO systems, where the number of antennas employed at the base station (BS) is large (M >> 1). The reason is that channel dimension is high and the affordable pilot dimension, dedicated to channel state information (CSI) acquisition, is relatively low. A natural question is posed: how can one achieve reliable covariance estimates given a limited number of noisy pilot samples? We study this question for the case of dualpolarized massive MIMO arrays, which are widely employed in today’s wireless systems. Our proposed solution is based on two steps: (1) we exploit the structure of the dual-polarized Gaussian channel to suggest a parametric representation of the covariance, and (2) we propose a Maximum-Likelihood (ML) estimator to obtain the parameters. The novelty of our work lies in both of these steps. Unlike previous literature, we consider a more general channel angular spread function (ASF) with both discrete and continuous components. We obtain the discrete components using the well-known MUSIC method, and represent the continuous components in terms of limited-support density functions. Furthermore, we solve the ML parameter estimation problem via the concave-convex procedure, which achieves a stationary point of the objective. With proper initialization, we empirically show that in most cases this method achieves highly accurate covariance estimates for various sampling ratios (number of samples to number of antennas).