NEW EFFICIENT LS AND SVD BASED TECHNIQUES FOR HIGH-RESOLUTION FREQUENCY ESTIMATION

New least squares and singular value decomposition based methods for t he estimation of the frequencies of complex sinusoids in white noise a re presented. The methods are based on a new symmetric prediction prob lem that has some very useful properties leading to algorithms that ha ve considerably reduced complexity. The new symmetric predictor is sup erior in performance as compared to the well known symmetric Smoother and has a performance comparable to other well known methods. Finally a new LS based method, which combines the new prediction technique wit h the FBLP method is proposed. This method performs slightly better th an the FBLP offering at the same time a significant computational savi ng. As a by-product in the derivation of the new methods is the establ ishment of some useful properties concerning the eigenstructure of Her mitian and Persymmetric matrices.