Comparison of adaptive and compressed sensing beamformers

Abstract

We will review the capabilities and requirements of the Minimum Variance Distortionless Response (MVDR) and sparsity-seeking beamformers, from a system design perspective. We will focus on the problem of detecting a quiet source in the presence of loud interferers. These methods are solutions to particular and different optimization problems, requiring particular combinations of inputs. Thus, for example, MVDR does not require us to model the interference, whereas the sparsity-seeking methods need both interferer and quiet source models to be put into their dictionary. MVDR is notorious for requiring more precise calibration than is needed for conventional beamformers. We will present demonstrations on simulated and experiment data, illustrating key differences between the two methods.