A revisited three-cornered hat method for estimating frequency standard instability

The three-cornered hat method used for evaluating the individual noise contributions of three frequency standards in a stability measurement is reexamined. The hypothesis of uncorrelation of the standards is not assumed a priori, as it is considered a cause of negative estimated variances. The introduction of covariances of measured data, together with the proposal of a suitable optimization criterion, allows one to estimate the complete covariance matrix of the three standards, ensuring its positive definiteness.>