Robust frequency estimation of sinusoids in the presence of outliers and noise

The authors consider robust methods for frequency estimation of sinusoids. The sinusoids are disturbed by additive white noise and patches of outliers or missing data. The location of these patches is unknown. Standard methods for frequency estimation like the periodogram, the modified covariance and an eigendecomposition method underlie their approach. They provide algorithms for enhancing these estimates and studies of the performance of procedures which are known to be well suited for the problem at hand (e.g. iterated weighted least squares). The results shown provide some information about the reliability of frequency estimates under such conditions as well as support for selecting an appropriate estimation procedure in other situations.<>