TLS based methods for exponential parameter estimation

Many spectral estimation algorithms have been developed for resolving closely-spaced sinusoids in the presence of noise. Some well-known subspace based techniques are the forward and backward linear prediction methods and Kung et al.'s method based on state-space modeling. The accuracy of these methods is improved by making use of the TLS technique. The performance can be further improved by incorporating prior knowledge of known signal poles corresponding to some of the closely-spaced sinusoids. In this paper, we review these TLS based parameter estimators with or without prior knowledge and illustrate their performance with examples.