Robust Multichannel Microphone Beamforming
In this thesis, a method for the design and implementation of a spatially robust multichannel microphone beamforming system is presented. A set of spatial correlation functions are derived for 2D and 3D far-field/near-field scenarios based on von Mises(-Fisher), Gaussian, and uniform source location distributions. These correlation functions are used to design spatially robust beamformers and blocking beamformers (nullformers) designed to enhance or suppress a known source, where the target source location is not perfectly known due to either an incorrect location estimate or movement of the target while the beamformers are active. The spatially robust beam/null-formers form signal and interferer plus noise references which can be further processed via a blind source separation algorithm to remove mutual components - removing the interference and sensor noise from the signal path and vice versa. The noise reduction performance of the combined beamforming and blind source separation system approaches that of a perfect information MVDR beamformer under reverberant conditions. It is demonstrated that the proposed algorithm can be implemented on low-power hardware with good performance on hardware similar to current mobile platforms using a four-element microphone array.