Statistical Engineering Division, ITL
Clark A. Hamilton
Electromagnetic Technology Division, EEEL
In the 1997 Josephson Voltage Standard (JVS) Interlaboratory Comparison (ILC), a traveling set of four 10 V Zener reference standards was circulated among the 15 participating standards laboratories. Each laboratory made 16 measurements of each of the four Zener standards over a period of 2-6 days. All of the laboratories used 10 V Josephson standards to measure the output voltage of the four standards. A comparison of the measurements of all laboratories yielded data on the uncertainty of each laboratory's measurement procedure, the performance of the traveling standards, and any offsets that may exist between the reference and testing laboratories.
In some previous JVS interlaboratory comparisons, one laboratory has served as a pivot with the traveling standards returning to the pivot after each set of measurements by a satellite laboratory. This is a big effort and in 1997 no laboratory volunteered to be the pivot laboratory. Thus, in this ILC, the standards traveled from one laboratory to the next until all 15 laboratories had participated. NIST-Boulder contributed three sets of measurements near the beginning, middle, and end of the ILC.
It is well known that Zener standards drift with time, and their voltage
has a small and unknown dependence on pressure and other factors.
A pivot laboratory allows the drift with time of the standard to be
Without a pivot laboratory, we used the following model to describe
where Vlsk represents the kth measured voltage for the sth standard of the lth laboratory, tlsk is the elapsed time (in days), and Plis the laboratory pressure (in kilopascals). In addition, , , and . The quantity Lls is the laboratory effect and assumed distributed as Gaussian with mean 0 and variance . The random errors are all independent and assumed distributed as Gaussian with mean 0 and variance .
Maximum likelihood was used to estimate and after adjusting for the drift and pressure effects. Least-median-of-squares regression was used to identify multiple outliers. The between- and the within-laboratory standard deviations, averaging over the four standards, were found to be 80 nV and 176 nV, respectively. The uncertainty of the ILC is obtained by combining the between- and the within-laboratory variations for the 64 measurements made at each laboratory. Specifically, the uncertainty () is nV, or 8.3 parts in 109.
This work will appear in the February or May 1998 issue of Metrologia.
Figure 10: Voltage measurements of all participants. The horizontal and vertical axes are, respectively, the time in days from December 27, 1996 and the deviation from 10 V (in nV) of the measured voltage. The plotting symbols ``'', ``'', ``+'', and ``'' are for standards 1, 2, 3, and 4, respectively.
Date created: 7/20/2001