Causal System Identification based Compensation for Reverberation-Robust DOA Estimation

One big challenge of direction of arrival (DOA) estimation is reverberation, since the spatial cues of the source and its reflections cannot be well distinguished. In this paper we propose a DOA compensation factor based on the mean squared error (MSE) of system identification (SI), and used it to adjust the DOA cost function values of conventional methods, such that the robustness of DOA estimation to reverberation can be improved. By analysing the relationship between the beamforming outputs towards different DOAs and the signal captured by the microphone, we show that the observed signal can be more accurately predicted by the signal from beamforming towards the direct-path, than from other directions. This is indicated by a small MSE of SI, and is exploited to calculate a DOA compensation factor. The proposed compensation factor can be easily applied to conventional DOA estimation methods, regardless of the number of microphones and the type of algorithms. Experiments show that the proposed method outperforms baseline methods in various reverberant conditions.