SVD-Based Estimation and Rank Detection for Reduced-Rank MIMO Channel

Abstract

Multi-input-multi-output (MIMO) system, due to its large channel capacity and high reliability, has been widely studied in numerous existing work. It has a lot of applications in reality, e.g., 3GPP and LTE standards. These advantages owes to the understanding of the MIMO channel, which shows the necessity of the obtainment of channel state information (CSI). In this thesis, channel estimation schemes based on singular value decomposition (SVD) are proposed for MIMO systems, where instead of estimating each entry of the channel matrix, the singular spaces and singular values are estimated, respectively. When the channel rank is fixed and known, the maximum-likelihood (ML) estimator is derived. When the channel rank is unknown, by using the singular values of the observation matrix as the test statistics, three threshold-based rank detection algorithms are proposed. In finding the thresholds, lower bounds on the correct detection probability are derived and the thresholds are chosen to maximize the lower bounds. To evaluate the performance, mean square error (MSE) and achievable beamforming capacity are introduced as the criteria. Compared with entry-based ML estimation, simulations show that the combination of the proposed rank detection and SVD-based channel estimation achieves lower MSE and higher capacity.