Optimal Pilot Sequence Design for Machine Learning Based Channel Estimation in FDD Massive Mimo Systems

In this paper, we consider the problem of channel estimation for large scale Multiple-Input Multiple-Output (MIMO) systems, in which the main challenge that limits the functionality ofmassive MIMO is the acquisition of precise Channel State Information (CSI). We introduce an efficient channel estimation approach based on a block Sparse Bayesian Learning (SBL) that exploits the temporal common sparsity of channel coefficients. Furthermore, an optimal pilot approach to reduce the pilot overhead is derived. The optimal pilot is obtained by minimizing the Mean Square Error (MSE) of the proposed SBL estimator using Semi-Definite Programming (SDP). Simulation results demonstrate that the SBL-based approach is more robust than conventional methods when fewer training pilots are used.