Microphone Subset Selection for MVDR Beamformer Based Noise Reduction

摘要

In large-scale wireless acoustic sensor networks (WASNs), many of the sensorswill only have a marginal contribution to a certain estimation task. Involvingall sensors increases the energy budget unnecessarily and decreases thelifetime of the WASN. Using microphone subset selection, also termed as sensorselection, the most informative sensors can be chosen from a set of candidatesensors to achieve a prescribed inference performance. In this paper, weconsider microphone subset selection for minimum variance distortionlessresponse (MVDR) beamformer based noise reduction. The best subset of sensors isdetermined by minimizing the transmission cost while constraining the outputnoise power (or signal-to-noise ratio). Assuming the statistical information oncorrelation matrices of the sensor measurements is available, the sensorselection problem for this model-driven scheme is first solved by utilizingconvex optimization techniques. In addition, to avoid estimating the statisticsrelated to all the candidate sensors beforehand, we also propose a data-drivenapproach to select the best subset using a greedy strategy. The performance ofthe greedy algorithm converges to that of the model-driven method, while itdisplays advantages in dynamic scenarios as well as on computationalcomplexity. Compared to a sparse MVDR or radius-based beamformer, experimentsshow that the proposed methods can guarantee the desired performance withsignificantly less transmission costs.