An OTFS Channel Estimation Scheme Based on Efficient Sparse Bayesian Learning

Abstract

Orthogonal time frequency space (OTFS) modulation is a delay-Doppler (DD) domain modulation scheme that can adapt to the fast time-varying channels. Since OTFS is a two-dimensional modulation scheme in DD domain, effective channel in DD domain can be represented by a few delay and Doppler parameters. Due to the sparsity of effective channel in DD domain, we model the output signal matrix as the product of pilot symbol matrix and sparse channel vector and introduce the Doppler gird segmentation factor to subdivide the Doppler taps, which solves the issue of fractional Doppler. Then, we model the channel estimation problem as a sparse signal recovery problem, and propose a channel estimation method based on an efficient sparse Bayesian learning (ESBL) in this paper. Specifically, we design a pilot pattern with no guard interval, and the pilot power is equal to the symbol power, and then use a low-complexity sparse Bayesian learning based on Gaussian-scale mixtures to directly estimate the effective channel. Compared to traditional channel estimation based on sparse Bayesian learning (SBL), our proposed ESBL-based algorithm has superior normalized mean squared error (NMSE) and bit error rate (BER) performance.