ON MODIFIED EVLP AND ML METHODS FOR ESTIMATING SUPERIMPOSED EXPONENTIAL SIGNALS

We consider the estimation procedure of the multiple sinusoidal model for signals, when the damping factor is not present. The solution in t he general case depends on the roots of a polynomial, whose coefficien ts are estimated from the observed data. When the damping factor is ab sent, the coefficients exhibit a certain symmetry. It reduces in estim ating almost half of the total number of unknown parameters. Under the se symmetric constraints modified methods have been developed to estim ate the coefficients. It is observed that the standard errors for the modified methods are closer to the Cramer-Rao lower bound than before in almost all the situations. It is also observed that the computation al cost of the modified maximum likelihood method is lower than the or dinary one. The modified maximum likelihood estimates can be obtained by an iterative process. Theoretical justification has been provided f or the convergence of the iterative process.