Title :
Lowering the threshold SNR of singular value decomposition based methods
Author_Institution :
AMES Dept.-Syst. Sci., California Univ., San Diego, La Jolla, CA, USA
Abstract :
The author examines the performance of singular value decomposition (SVD) based methods for estimating the frequencies of multiple sinusoids. The concept of angle between subspaces is used to derive a criterion for determining when SVD-based procedures fail. The signal-to-noise ratio (SNR) at which a method breaks down is termed the threshold SNR of the method. It is shown that existing SVD based methods have a higher threshold SNR than predicted by this criterion. A method that utilizes the singular vectors and directly minimizes the angle between subspaces is developed. The method is shown to have better performance at low SNRs. The procedure lowers the threshold SNR thereby extending the range of SNR for which SVD can be used
Keywords :
signal processing; frequency estimation; multiple sinusoids; performance; signal processing; signal-to-noise ratio; singular value decomposition; singular vectors; state space approach; subspaces angle; threshold SNR; Eigenvalues and eigenfunctions; Frequency estimation; Hafnium; Parameter estimation; Signal processing; Signal to noise ratio; Singular value decomposition; State-space methods; Vectors; White noise;
Conference_Titel :
Acoustics, Speech, and Signal Processing, 1988. ICASSP-88., 1988 International Conference on
Conference_Location :
New York, NY
DOI :
10.1109/ICASSP.1988.197144
