Semi-Blind Channel Estimation for Frequency-Selective Massive MIMO Systems

Channel state information (CSI) is essential in massive multiple-input mulitple-output (MIMO) systems. However, most of the literature focuses on frequency-flat channel estimation; in a realistic broadband setting this implies that the estimation must be repeated for each subcarrier. In this paper, we propose a new channel estimation technique for frequency-selective channels in the time-domain. Time-domain CSI acquisition requires the estimation of fewer parameters. Importantly, since the receiver samples at the symbol rate, time-domain based estimation yields many samples of the received signal. Our approach depends on the estimation of the subspace spanned by users' channels, requiring a large number of antennas to ensure that the required rank constraints are met. Unlike previous published results, in this paper, we do not make any assumption on orthogonality of users' channel vectors or knowledge of large-scale fading coefficients. Additionally, our channel estimation, does not require orthogonality between the training symbols of the users in all cells.