Statistical Engineering Division
Seminar Series

Lara S. Schmidt
U.S. Naval Observatory

The world's official timescale, Coordinated Universal Time (UTC), is calculated from the outputs of high performance atomic clocks keeping time to better than 10e-9 seconds per day. Physicists have traditionally described the statistical behavior of these atomic standards in terms of their spectral properties. Namely, that the power spectral density of a suitably normalized time series of clock outputs has a so-called power-law structure. It is shown in this research that a more convenient description of the statistical behavior of atomic clocks is found by employing long-memory models.

Long-memory, or fractionally integrated, models have been used extensively in econometrics and other application areas to describe time series exhibiting serial correlations persisting over long time spans. Here, their application to atomic timescales yields a new approach to clock modeling based upon a fractional difference prewhitening strategy. The fractional difference prewhitening (FDPW) strategy is developed and tested via simulations and live data tests. Estimators of atomic clock parameters made via this technique are shown to be both more efficient and more powerful than traditional estimators. Application of this technique to the world's largest ensemble of atomic clocks validates the use of long-memory time series models in atomic timekeeping.