Generalization of the Total Variance Approach to the Modified Allan Variance

摘要

The Total variance approach involves periodically extending a data sequence beyond its normal measurement duration and in such a way that a particular time statistic is expected to have the same value with extended data as without. For those statistics which estimate components of broadband noise processes, the approach can significantly reduce the spread or uncertainty in the result. We describe a Total variance approach for improving the confidence of the estimation of the modified Allan variance (Mvar) for the five common integer power-low noises and which simultaneously has low, easily removable bias. We have found in simulation studies that if a relection-only extension procedure is applied to Mvar's individual estimates, we obtain a new estimate of Mvar which exhibits an increase in equivalent degrees of freedom at mid- and long-term integration times.