Two-Fold Sampling-Based Super-Resolution Estimation of Low-Rank MIMO-OFDM Channels

This paper studies the estimation of low-rank multiple-input multiple-output (MIMO) wideband channels in orthogonal frequency division multiplexing (OFDM) systems which are commonly considered for high-frequency wireless communications, e.g., at millimeter wave (mmWave) and Terahertz (THz) bands. To reduce the overhead for channel estimation, we propose a novel solution based on two-fold sampling of the original channel matrix. A subchannel associated with a subset of subcarriers and antennas is first selected by deterministic sampling. Low-rank matrix completion (LRMC) based on random sampling is then used to further reduce the training overhead. Utilizing the Toeplitz structure of the covariance matrices of uniform linear arrays, the angles and delays are then estimated separately at low complexities using super-resolution algorithms. Finally, the path gains are estimated and associated with the path angles and delays, followed by the extrapolation of the full-dimensional channel. As only a small number of antennas and subcarriers are required in the training, the overall training overhead and computational complexity are very low. Numerical results demonstrate that the proposed two-fold sampling-based estimator can achieve high-accuracy channel estimation. Besides, the sampling patterns and complexity of the proposed estimator are analysed, which shows that the proposed solution can be configured to provide different performance-complexity tradeoffs.