Expectation-Maximization-Based Information Geometry Approach for Massive MIMO-OFDM Channel Estimation

In this paper, we investigate the channel estimation in massive multiple-input multiple-output orthogonal frequency division multiplexing (MIMO-OFDM) systems. Many channel estimation algorithms, such as the recently proposed original information geometry (OIG) and simplified information geometry (SIG) algorithms, require statistical channel state information (CSI) when performing instantaneous channel estimation. However, in practice, obtaining statistical CSI may require a substantial amount of additional overhead, especially when the number of users is large. To address this issue, we propose an expectation-maximization (EM) based IG approach for massive MIMO-OFDM channel estimation, which does not need to know the statistical CSI beforehand and can estimate the instantaneous channel and statistical CSI simultaneously. Specifically, the EM-based procedure is embedded within the IG algorithms’ iteration loop to learn the statistical CSI. Moreover, all of the quantities needed for the EM updates can be obtained by the IG procedure, making the overall process computationally efficient. Furthermore, we propose a low complexity implementation for the proposed EM-SIG algorithm with the adjustable phase shift pilots of multiple roots, which can significantly reduce the computational complexity of EM-SIG. Simulations confirm the excellent performance of the proposed EM-OIG and EM-SIG algorithms both in terms of channel estimation and statistical CSI estimation.