FREQUENCY ESTIMATION PERFORMANCE BY EIGENVECTOR METHOD

Abstract

We study the theoretical performance of the MUSIC (Multiple Signal Classification) and MN (Minimum Norm) algorithms in estimating the hidden periodicities in the presence of white noise. Both algorithms are based on the principal components of signal correlation matrix of size $m$. An asymptotical analysis of frequency distribution is given under conditions that the number of observation $N\to \infty$ and $m$ is fixed or it increases with $N$. Surprisingly, the frequency accuracy of order $o(N^{-1})$ is impossible for $m\simeq N- c$.

Back to
Computational Seismology, Vol. 2.