Implementation of a robustness constraint in adaptive beamforming

In highly nonstationary environments, the time available to estimate the correlation matrix used in the MVDR (minimum variance distortionless response) beamformer is severely limited. This requires that the beamformer use very short averaging times, which produces poor estimates of the correlation matrix and thus poor suppression of spatially white noise. A robustness constraint is incorporated to correct the poor noise suppression. The MVDR algorithm with the robustness constraint is implemented effectively utilizing a modified version of stabilized hyperbolic Householder transformations, and updates the adaptive weights every FFT frame. For an n element array forming m beams where m>n, the resulting dominant computational load is proportional to n-m. If triangular backsubstitution is used to form beams instead of this update method the load is proportional to n ²m. The algorithm is demonstrated on in-water acoustic data and shows the efficacy of the robustness constraint