Least-Squares Multiple Frequency Estimation using Recursive Regression Sum of Squares

An accurate frequency estimation method from a signal consisting of multiple complex sinusoids is presented. The method avoids inversion of data dependent matrices of large dimensions, and hence do not suffer from rank deficiency of the data covariance matrix. Using recursion, a procedure to reduce mw/t/-dimensional maximum likelihood estimation to a series of one-dimensional optimization while achieving Cramer-Rao bounds for estimation is proposed.