A novel two-stage DOA estimation of sound sources based on hierarchical sparse Bayesian inference

Abstract

To efficiently achieve off-grid sparse Bayesian direction-of-arrival (DOA) estimation for sound sources, this paper proposes a two-stage off-grid DOA estimation method based on a hierarchical sparse Bayesian inference (HSBI) framework. In the first stage, the proposed HSBI model utilizes a Gaussian distribution to form a sparse representation of sound sources in space, and a Laplace distribution to enhance sparsity in intermediate parameters during Bayesian inference. This stage provides initial estimates of the unknown sound source positions, serving as the foundation for the second stage. In the second stage, a novel particle swarm optimization (PSO)-based high-precision DOA solver is proposed. This solver leverages spatial beamforming theory to efficiently search for the maximum energy distribution in the physical space, yielding accurate sound source position estimates at high speed. The HSBI model exhibits superior sparsity performance, allowing for precise estimation of unknown sound sources while preventing the PSO solver from becoming trapped in local optima. Simulation and experimental results show that the PSO-HSBI algorithm achieves efficient sound source DOA estimation with low computational complexity, while significantly reducing amplitude and localization errors. The source code for the proposed PSO-HSBI algorithm is available at https://github.com/Xiaob0-Zhang/PSO-HSBI.