Signal subspace analysis and improvement of spectral estimation algorithms

The signal subspace approach was introduced by Schmidt [1] for the narrowband source location problem and generalized by Su and Morf ([2] - [5]) for wideband source location and multidimensional spectral estimation. Many authors have made contributions to related methods of spectral estimation and source location, generally using eigenstructure or singular value decomposition. In this paper three methods of spectral estimation are related to the signal subspace approach. Besides demonstrating the common thread in their development, the signal subspace approach suggests immediate modifications and generalizations of the methods. The methods are those of Prony [6], Tufts-Kumaresan [7] and Cadzow [8].