Robust rank selection for the Multistage Wiener Filter

The Multistage Wiener Filter (MWF) is an adaptive processing technique that has the ability to outperform the full rank Wiener solution in a low sample support environment. However, this requires knowledge of the appropriate rank in order to enjoy these advantages. In this paper we show that the addition of a diagonal loading-like quantity to the Multistage Wiener Filter will add significant robustness to the rank selection process. Since the MWF is most often implemented without the formation of a sample covariance matrix, diagonal loading cannot be added in the usual sense. In this paper we describe an alternate technique called “error loading” for augmenting the filter. The incorporation of error loading is shown to provide the rank selection robustness that we desire.