A generalization of weighted subspace fitting to full-rank models |
| |
Authors: | Bengtsson M Ottersten B |
| |
Affiliation: | Dept. of Signals, Sensors & Syst., R. Inst. of Technol., Stockholm; |
| |
Abstract: | The idea of subspace fitting provides a popular framework for different applications of parameter estimation and system identification. Previously, some algorithms have been suggested based on similar ideas, for a sensor array processing problem where the underlying data model is not low rank. We show that two of these algorithms (DSPE and DISPARE) fail to give consistent estimates and introduce a general class of subspace fitting-like algorithms for consistent estimation of parameters from a possibly full-rank data model. The asymptotic performance is analyzed, and an optimally weighted algorithm is derived. The result gives a lower bound on the estimation performance for any estimator based on a low-rank approximation of the linear space spanned by the sample data. We show that in general, for full-rank data models, no subspace-based method can reach the Cramer-Rao lower bound (CRB) |
| |
Keywords: | |
|
|