Recursive and knowledge-aided implementations of the multistage Wiener filter

The multistage Wiener filter (MWF) is a reduced rank adaptive signal processing algorithm suitable for applications such as adaptive beamforming. This paper discusses a technique for augmenting the MWF in order to exploit pre-existing environmental knowledge. We consider a priori knowledge such as known fixed interferers as well as the information contained in previously computed adaptive solutions. It has recently been shown that the MWF can be implemented via the method of conjugate gradients. In this implementation, referred to as CG-MWF, the filter is initialized with a starting weight vector. We use this initial weight vector to introduce the prior knowledge into the adaptive filtering solution. We evaluate the performance under ideal conditions such as perfect knowledge and stationary data. Additionally, we investigate the sensitivity of the performance to a relaxation in the quality of the weight vector initialization.