Sparse Bayesian learning for time-varying DOA estimation
Yongsung Park, Florian Meyer, Peter Gerstoft · The Journal of the Acoustical Society of America · 2020
Sparse Bayesian learning (SBL) provides sparse direction-of-arrival (DOA) estimation performance, and an SBL scheme with sequential processing is proposed for time-varying DOA estimation. SBL employs a Gaussian prior for the source signals and models variances of the source amplitudes as a hyperparameter. For sequential processing, the Gaussian-distributed source signals are modeled as Gaussian processes, and we consider the variances of the source amplitudes as the parameter of the covariance function in the Gaussian process. The sequential SBL estimates the variance parameter that evolves sequentially over time based on a state-space model. The suggested SBL with the sequential processing provides high-resolution capabilities for time-varying DOAs with varying source strengths or moving sources over time. The present method is evaluated by using simulated and real data (SWellEx-96 Event S5).