Information Geometry Approach for Massive MIMO Channel Estimation

In this paper, we investigate the channel estimation for massive multi-input multi-output orthogonal frequency division multiplexing (MIMO-OFDM) systems. Using the sampled steering vectors in the space and frequency domains, we first establish a space-frequency (SF) beam based statistical channel model. Based on the channel model, the channel estimation is formulated as obtaining the a posteriori information of the beam domain channel. We solve this problem by calculating an approximation of the a posteriori distribution's marginals within the information geometry framework. Specifically, by viewing the set of Gaussian distributions and the set of the marginals as a manifold and its submanifold, respectively, we turn the calculation of the marginals into an iterative projection process between sub manifolds with different constraints. We derive the information geometry approach (IGA) for channel estimation by calculating the solutions of projections. We prove that the mean of the approximate marginals at the fixed point of IGA is equal to that of the a posteriori distribution. Simulations demonstrate that the proposed IGA can accurately estimate the beam domain channel within limited iterations.