Estimating the Number of Stages of the Multistage Wiener Filter Using Mutual Information

In airborne radar, adaptive processing requires estimation of a clutter-plus-noise covariance matrix that is often poorly estimated due to insufficient training data. This is particularly true in the maritime domain, where the data varies both temporally and spatially with non-Gaussian statistics. The multistage Wiener filter is a reduced rank adaptive detector that works well in this environment by reducing the required sample support while lowering the computational burden. However, the challenge in implementing this algorithm is determining the number of stages that provides the greatest detection performance. This paper presents a new method for determining the number of stages based on mutual information. We demonstrate that it outperforms other estimates, specifically in situations where sample support is low or the target is close to the clutter.