Interpreting the environmental information embodied in universal adaptive beamformers

Abstract

Practical adaptive beamformers often include regularization parameters to mitigate the impact of limited training data. These parameters may include the dominant subspace dimension for projection beamformers and the averaging window for the sample covariance matrix. Choosing the optimal value for these regularization parameters requires environmental knowledge not available to the algorithm in real time. Universal adaptive beamformers (UABFs) avoid the need for this knowledge. UABF outputs are a performance-weighted blend of the outputs of a competing family of beamformers. These blend weights embody operationalized knowledge about the array's environment. This environmental knowledge may be explicitly or implicitly represented, depending on the underlying beamformers within the universal framework. Previously proposed UABF implementations were universal over the dominant subspace dimension [Buck et al., ASA (2018)] and the sample covariance matrix averaging window [Buck, ASA (2019)]. The former explicitly provides information about the number of interferers in the environment, while the latter implicitly captures information about the bearing rates of the interferers. Data analysis from an ocean acoustic experiment will demonstrate an example of the environmental information available from the blend weights. [Work funded by ONR 321US.]