Variational Bayesian inference for beamforming
Yongsung Park, Florian Meyer, Peter Gerstoft · The Journal of the Acoustical Society of America · 2021
A variational Bayesian method for beamforming is presented. The proposed method aims at estimating the time-varying directions of arrivals (DOAs) of source signals. Sequential estimation is performed based on a random process representation of the unknown DOAs. Since calculating the posterior probability density functions (pdfs) of the individual DOA variables is intractable, we utilize variational Bayesian inference to analytically approximate these posterior pdfs. A von Mises representation enables closed-form integration over approximate DOA pdfs. The proposed algorithm iteratively estimates DOAs, source amplitudes, number of sources, and model parameters. Furthermore, it is grid-less, promotes sparse solutions, and provides an uncertainty characterization for estimated DOAs. For an improved time-varying DOA tracking performance, inference results from previous time steps can be used as prior information for sequential processing. We evaluate the proposed method using simulated data and acoustic data from an underwater source localization experiment.