SVD-Assisted FI Sparse Bayesian Learning for OTFS Channel Estimation

Abstract

The sparse Bayesian learning (SBL) is an efficient channel estimation approach for orthogonal time-frequency-space (OTFS) systems. However, the ability of SBL to identify nonzero elements is limited by the sensing model structure, and its computational efficiency is limited by the computationally intensive covariance matrix inversion. In this letter, we propose a singular value decomposition-assisted fast inversion SBL (SVD-FI-SBL) algorithm for OTFS channel estimation. First, the SVD-assisted preprocessing framework is used to reconstruct the sensing model, thereby improving the accuracy of SBL in identifying nonzero elements. Next, we use the FI scheme to convert the matrix inversion in covariance computation into two inversions of diagonal matrices. Although the SVD preprocessing introduces additional overhead, the proposed scheme achieves a better trade-off between channel estimation accuracy and computational complexity. Simulation results show that the proposed SVD-FI-SBL algorithm can achieve more accurate channel estimation compared to existing benchmark schemes.