Experimental study of robust beamforming techniques for acoustic applications

Abstract

In this paper, we investigate robust beamforming techniques for wideband signal processing in noisy and reverberant environments. In such environments, steering vector estimation errors are inevitable, leading to a degradation of the beamformer performance. Here, we study two types of beamformers that are robust against steering vector estimation errors. The first type includes robust Capon beamformers, where the underlying principle is to add a steering vector uncertainty constraint and/or a norm constraint to the optimization problem to improve the beamformer's robustness. The second type is the amplitude and phase estimation method, which utilizes both temporal and spatial smoothing. Experiments are presented to demonstrate the performance of the considered robust beamformers in acoustic environments. The results show that the robust beamformers outperform the non-robust beamformers in terms of predicted speech quality and intelligibility for different steering vector and covariance matrix estimation errors.