Time-domain subsampling and reconstruction for microphone array

For low-cost sound monitoring of machinery, we propose a novel method for time-domain subsampling and reconstructing of the multi-channel audio recorded by a far- field microphone array. A conventional subsampler performs an analog mixing process, so it needs amplitude calibration between microphones. However, the proposed subsampler is based on random time-division multiplexing without analog mixing, so it does not need amplitude calibration. Also, the performance of conventional reconstruction methods deteriorates severely when the sparsity in the frequency domain is not sufficiently high. In this paper, we focus on multi-microphones redundancy, and the proposed reconstruction uses group sparsity regularization derived from the redundancy. The experimental results show that the proposed combination of subsampling and reconstruction is suitable for audio reconstruction in microphone-array-based sound monitoring.