Spatial Robustness of MVDR Beamformer in Acoustic Environment
The beamforming technique is widely used in audio systems for spatial processing. When designed for audio applications, its uniqueness lies in the properties of room acoustics and speech signals. While many important results for robustness considerations in beamforming are derived for narrow-band signals in free-space propagation, the design of robust beamformers that account for the properties of audio systems is an open research question. In this work we focus on the spatial robustness of beamformers in acoustic environment, that is, the immunity of the array to uncertainty in the position of the desired speaker. We adopt a statistical approach for characterizing the acoustic transfer functions between the speaker and each of the microphones, which allows us to propose a statistical model of an acoustical bempattern. By considering its first- and second-order statistics, we analyze the spatial properties of a beamformer designed for acoustic environment and propose a practical technique for improving the beamformer spatial robustness. The statistical model, as well as the proposed technique, are verified in an extensive simulation study.
* This work was carried out towards the M.Sc. degree in the Faculty of Engineering, Bar-Ilan University, under the supervision of Prof. Sharon Gannot.