Fast, accurate subspace tracking using operator restriction analysis
Craig S. MacInnes · 2002
A new noniterative subspace tracking method is presented. This method is called the operator restriction algorithm (OPERA) and it can be used whenever an update to the principal components of an EVD or SVD of a rank-one update of a given matrix is needed. The updating algorithms are based on the technique of restricting a linear operator to a subspace and the concept of an invariant subspace and its generalization, a pair of singular subspaces. The accuracy of the algorithm is comparable to an EVD or SVD. An application is made to bearing estimation of highly nonstationary sources. Flop counts, tracking accuracy and subspace accuracy for OPERA are compared with other fast algorithms and with the EVD.