Building a better clock predictor

17 Jun 2002

For those who operate extremely precise clocks and oscillators, a principal task is characterizing the frequency stability of the devices. One recommended method hasbeen to use the two-sample frequency variance known as the Allan variance statistic.

Recent NIST analysis has shown that the Allan vairiance statistic does not adequately characterize certain types of oscillator behavior and measurement protocols. Anew data manipulation, dubbed "total variance," has been devised that time-shifts the sampling window. Instead of a limited two-sample measurement, themultisample approach estimates all possible observed frequency deviations over a set period of time. Previously seen estimation errors are reduced by as much as100-fold. Use of the "total variance" method results in more accurate predictions of the time errors that evolve after a precision clock is synchronized, researchers say.