Performance of time domain and time-frequency domain adaptive beamformers with moving sound sources

The need to extract a single audio signal of interest from a multi-source and noisy environment is common across many disciplines. Adaptive beamforming, due to its superior interference rejection and noise suppression, is a preferred processing technique for obtaining high quality audio in noisy environments. In a previous study, we have examined the performance of two classes of adaptive beamforming, namely, time domain and time-frequency domain adaptive beamforming, under the conditions that the sound sources were stationary and the signal model was accurate. In this paper, we extend the study to the situations where the sources are moving and certain amount of signal mismatch is allowed. Four different types of adaptive beamformers are considered. The robust Capon beamformer is used for time-frequency domain beamforming. The tapped delay line (TDL) structure is adopted for time domain adaptive beamforming. Three different adaptive algorithms are used for obtaining the optimal TDL filters. These are the sample matrix inversion method, the recursive least squares method with sliding window, and the block constrained least mean square method with diagonal loading. In the paper, the performances of those four adaptive beamformers are evaluated in terms of fidelity of the beamformer output, robustness of the system, and the computational complexity of the algorithm. It has been found that the robust Capon beamformer provides better performance than the time domain beamformers.