Sequential Sparse Bayesian Learning For Doa
Yongsung Park, Florian Meyer, Peter Gerstoft · 2020
Sparse Bayesian learning (SBL) can effectively and accurately solve the direction-of-arrival (DOA) estimation problem. In this paper, we introduce a sequential SBL method for time-varying DOA estimation. Statistical information provided from previous time steps is modeled by a zero-mean multivariate Gaussian that is characterized by variance parameters. The presented method propagates statistical information across time by means of a prediction and an update step. The prediction step computes the prior distribution of current variance parameters from previous variance parameters and the update step incorporates current observations. A performance evaluation based on simulated and experimental data demonstrates that the proposed sequential SBL method can provide the capability of tracking time-varying sources with a high resolution.