Sparse Bayesian Learning for Near-Field Narrowband Source Localization

Sparse Bayesian learning (SBL) method for direction of arrival (DOA) estimation of near-field sources is implemented, and subspace decomposition is used to estimate range based on the estimated DOAs. Given the prior for sources and the likelihood Gaussian, the posterior distribution is derived. Hyperparameters (i.e., the variance of source power) are updated by maximizing the evidence to estimate DOAs. The number of sources is estimated simultaneously by the model-order selection criterion, which reduces the demand of a priori of the number of sources. Theoretical analysis and simulation experiments show better performance compared with two dimensional multiple signal classification (2D-MUSIC). The method reduces the dependence on high signal-to-noise ratio (SNR) and multiple snapshots.