Noniterative subspace updating

In this paper, we introduce a rank-one spherical subspace update that is appropriate for tracking the dominant (signal) and/or subdominant (noise) subspaces associated with a slowly time-varying correlation matrix. This non-iterative, highly parallel, numerically stabilized, subspace update is closely related to rank-one eigenstructure updating. However, a rank-one subspace update involves less computation than simple rank-one correlation accumulation. Moreover, the frequency tracking capabilities of the non-iterative subspace update are virtually identical to and in some cases more robust than the more computationally expensive eigen- based methods.