Sparse Bayesian Learning Assisted CFO Estimation Using Nonnegative Laplace Priors

This correspondence paper aims at addressing the estimation of carrier frequency offset (CFO) for the uplink orthogonal frequency-division multiple access systems. Since the CFOs of the signals from different active users are sparsely distributed in the frequency domain, a sparse Bayesian learning (SBL) is tailored to determine the CFO in this paper, ending up with the SBL assisted CFO (SBL-CFO) estimator. In particular, the CFO estimation problem is first formulated as a sparse nonnegative least squares (S-NNLS) problem. Meanwhile, background noise and sampling errors are mitigated utilizing a selection matrix and a whitening filter, respectively. This enables us to exploit the SBL with nonnegative Laplace prior (SBL-NLP) to solve the S-NNLS problem. Furthermore, in order to make the convergence of the SBL-NLP algorithm faster and its estimation more accurate, the hyperprior inherent in the SBL-NLP algorithm is initialized by the traditional SBL with nonnegative Gaussian prior. Simulation results show that our proposed SBL-CFO estimator significantly outperforms the state-of-the-art estimators in terms of estimation accuracy, especially when the CFOs and the number of active users are large.