On universal diagonally loaded beamformer under steering vector mismatch

We explore universal adaptive beamforming (UABF) under the non-ideal conditions of steering vector mismatch and signal contamination of data. We focus on the optimal choice of diagonal loading level as the regularization parameter. Assuming a statistical characterization for the steering vector mismatch, we use the average output signal-to-interference plus noise ratio (SINR) loss that accounts for signal mismatch as the loss function metric used by the UABF to determine the best coefficients for linear combination in the formation of the UABF weight vector. The SINR loss metric requires an accurate estimate of the interference-only covariance matrix in practice. Our initial goal here, however, is to focus on the ability of the UABF to choose good coefficients for linear combinations that provide improved performance with respect to the choice of diagonal loading level. Thus, we consider a simpler version of the universal beamformer, where it is assumed that interference-only training samples are available to compute this loss function metric. Numerical results showing the usefulness of this simplified beamformer will be presented that provide insight and motivation to consider more practical versions of the universal beamformer where signal-free training data may not be as available.