3D-ESP: An Efficient Subspace Pursuit Algorithm for MIMO-OTFS Channel Estimation

Multiple-input multiple-output orthogonal time frequency space (MIMO-OTFS) channels exhibit unique sparse properties in the delay- Doppler-angle (DDA) dimension. In this paper, we are dedicated to exploring how to utilize these special sparse properties to design an efficient channel estimation algorithm to balance the computational cost and estimation accuracy. To this end, we propose an iterative 3D efficient subspace pursuit (3D-ESP) channel estimation algorithm by exploiting these unique properties. Firstly, we invent a novel subspace pursuit technique to efficiently locate channel supports by detecting multiple dominant paths per iteration. Secondly, we adopt least squares twice in each iteration to optimize the estimation accuracy. Thirdly, we propose a noise-adaptive criterion to filter the estimated channels to promote the estimation accuracy in high-level noise cases. The performance verification results show that, compared to the classical 3D-SOMP, the 3D-ESP reduces the runtime by a range of 17.19% to 45.15%, and meanwhile decreases the normalized mean square error (NMSE) by 12.29% to 21.82%, when the signal-to-noise ratio (SNR) is 0 dB and the antenna number goes from 1 to 20.