Comparison of the performance of acoustic beamformers under differing atmospheric conditions.

Abstract

It is well known that atmospheric turbulence can negatively impact the performance of acoustic beamformers. While many beamformers, especially adaptive ones such as minimum variance distortionless response (MVDR), may be robust when the turbulent fluctuations are mild to moderate, they fail when the fluctuations are large. Other methods, such as maximum likelihood estimation, may be used to mitigate the effects of turbulence by directly incorporating the physics of the propagation medium into the assumed model of the acoustic signal (through the covariance and mean). When using synthesized data, the previously developed maximum likelihood estimator (MLE) for the azimuthal angle of arrival was found to outperform classical beamformers. However, in reality the atmosphere does not exhibit the exact behavior of the assumed atmospheric model, or all the required input parameters, such as the meteorological data, for the atmospheric model are not available. Therefore, we compare the performance of methods such as the MLE to that of classical and parametric methods, such as MVDR, multiple signal classification, and matched subspace detector, for data collected during a variety of atmospheric conditions. We critically examine the expense of gained accuracy over computational speed.