Millimeter-wave massive MIMO channel estimation based on majorization-minimization approach

Abstract

Abstract Millimeter-wave massive MIMO systems acquire a high dimensional distorted channel matrix to perform channel estimation. The captured channel matrix is sparse and rank deficient due to low scattering at millimeter-wave bands. The reduced number of RF chains enable the indirect access of signal at each antenna. This makes it difficult to estimate the channel accurately. To address this issue, researchers have focussed on compressive sensing and convex regularization methods. The first one performs well in noiseless scenario and while the latter suffers from over shrinking of all eigenvalues equally and is not suitable for large dimensional matrices. In this paper, a majorization minimization based fast non-convex (MM-FNC) algorithm is proposed to solve non-convex regularization based channel estimation.\xa0The proposed algorithm estimates the low-rank channel with fast convergence properties for large dimensional arrays. Simulation results show that the proposed approach exhibits superior performance over the CS-based orthogonal matching pursuit (CS-OMP) method, nuclear norm minimization methods (NNM) and matrix factorization (MF) methods in terms of normalized mean square error and achievable spectral efficiency with respect to different signal-to-noise ratio level.