A novel cyclic algorithm for maximum likelihood frequency estimation

Abstract

An algorithm for estimation of frequencies of narrowband sources from noisy observation data is presented. For Gaussianly distributed noise, the algorithm produces maximum likelihood estimates, otherwise least-squares estimates, are obtained. The proposed algorithm is iterative, and at each step of iteration the optimization is with respect to a single frequency only, and hence simple hardware/software is sufficient for implementation. The performance of the algorithm has been compared with theoretical Cramer-Rao bounds.<>