Source Localization by Multidimensional Steered Response Power Mapping with Sparse Bayesian Learning

We propose a method that combines Steered Response Power (SRP) with sparse optimization for localizing multiple sources. While conventional SRP is robust under adverse conditions, it struggles with scenarios involving neighboring sources, often resulting in ambiguous SRP maps. The current state-of-the-art approach optimizes observed SRP maps through group-sparse modeling, but its performance degrades in reverberant scenarios. To address this issue, we extend the framework by modeling SRP functions as a multidimensional matrix, thereby preserving time-frequency information. Additionally, we employ multi-dictionary sparse Bayesian learning as the sparse optimization method to identify source positions without prior knowledge of their quantity. We validate our method through practical experiments using a 16-channel planar microphone array and compare it against three other localization methods. Results demonstrate that our proposed method outperforms other methods, including the current state-of-the-art, in localizing closely spaced sources in reverberant environments.