Beamformer design with sparse filters

Abstract

In designing broadband beamformers, when the number of microphone increases and the filters are long, the complexity can grow significantly. It is advantageous if many of the filter coefficients are zeroes so that the implementation can be executed with less computation. In this paper, the sparse design of beamformers is studied. We employ the penalty decomposition method to constrain the l0-norm of the filters and investigate the performance of the designed frequency responses. Numerical results show that sparsity of the designed beamformers can be reduced without affecting very much of the performance.