Direct Localization of Emitters Based on Sparse Bayesian Learning

Abstract

In this paper, a novel perspective for the problem of direct localization of emitters is presented based on sparse Bayesian learning (SBL). The localization problem is addressed in three scenarios, where we derive the angle-of-arrival (AOA) based models for two kinds of narrowband conditions and the model jointly taking use of time-difference-of-arrival (TDOA) and AOA information. Unlike the existing $\ell _1$ -penalization based direct localization methods, we expand the localization models to the sparsity-based framework with the scheme of SBL, which makes the partly unknown dictionary based problems solvable. On the basis of the bound-optimization SBL algorithm, modifications are provided to make its solving procedures adapt to the localization models. For one thing, the learning rule of channel attenuation factors (i.e., the unknown part of the dictionaries) is added in the solving procedures. For another, we expand the original single measurement vector (SMV) based model to the multiple measurement vector (MMV) scenario. Furthermore, we modify the original procedures by updating the parameters with an alternating minimization strategy, which guarantees the convergence of the algorithms. Simulation results show the features of three SBL-based models and the performance superiority of the proposed methods.