Capon with steering vector errors
Per Zetterberg · OCEANS 2022 - Chennai · 2022
The Capon (a.k.a MVDR) beamformer (CB) has a reputation of being very sensitive to steering vector errors - to the point of performing worse than the delay-and-sum method (DAS) (a.k.a conventional beamforming). On a closer look, this weakness applies more to Capon as a method for estimating the power and/or signal waveform of impinging far-field sources and less when used as direction of arrival (DOA) estimator. In the latter case we find that the method is remarkably robust against steering vector errors - almost always performing better than DAS except in cases with high levels of spatially white noise. This motivates a fresh look at the choice of diagonal loading with focus on DOA estimation performance. We do this by Monte-Carlo simulations where we integrate the extraction and classification of peaks into the framework. We consider cases with very few samples available for the estimation of the sample covariance matrices, i.e. the rank deficient case. We also consider low-rank noise contributions and steering vector errors.