A unified approach to three eigendecomposition methods for frequency estimation

Abstract

The authors present a unified approach to three eigendecomposition-based methods for frequency estimation in the presence of noise. These are the Tufts-Kumaresan (TK) method, the minimum-norm (MN) method, and the total least squares (TLS) method. It is shown that: (1) the MN method is a modified version of the TK method; (2) the TLS method is a generalization of the MN method; (3) the TLS solution vector can be expressed in matrix form, and an alternative way of computing it is presented; (4) the MN and the TLS methods exhibit some improvement over the TK method. >