Abstract
In this article a robust parameter estimation method for two-channel sinusoidal model has been proposed. In a two-channel sinusoidal frequency model, the frequencies at different channels are same but they may have different amplitudes. The noise random variables in each channel are assumed to follow a stationary linear process, and the errors from the different channels are assumed to be dependent. The generalized least squares (GLS) estimators may be used in this case as they are most efficient estimators and in presence of the bivariate Gaussian errors, the asymptotic variances of the GLSEs attend the Cramer–Rao lower bound. It is well known that the GLS estimators are very sensitive to the outliers. In this article a very efficient weighted GLS (WGLS) estimators has been proposed, which are quite robust to the outliers. The asymptotic properties of the WGLS estimators have been provided. In simulation studies, it is observed that in presence of outliers the WGLS estimators perform better than the GLS estimators. The implementation of the GWLS estimators is quite straight forward. The performances of the WGLS estimators depend on the weight function. The proper choice of the weight function has been discussed and it has been illustrated using a synthetic data set.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call