The NIST Watt Balance : Progress Toward Monitoring the Kilogram

The National Institute of Standards and Technology (NIST) watt balance is an experiment to compare measurements of the watt using electrical references (volt, ohm) to those using mechanical references (length, time, mass). A coil within a radial magnetic field has a dual use of: 1) generating a voltage by moving at some velocity to calibrate the magnetic flux density, and, 2) generating a force with electrical current to balance the gravitational force of a mass. This experiment has had several improvements made to it in the last year. These include the incorporation of three-laser interferometry and a refractometer to improve the velocity measurements, temperature control and coil rotation damping to reduce drifts and stabilize laser and mechanical alignments, and a gravimeter to determine local gravity. Systematic errors and scatter in long-term measurements have been greatly reduced in the last year, but statistically significant deviations relative to within-run uncertainty still persist. The source of these deviations has not yet been identified. Recent within-run standard deviations are generally near 0.1 W/W, which is the target precision of this present design.

The NIST Watt Balance: Progress Toward Monitoring the Kilogram

I. INTRODUCTION
T HE NIST watt balance [1] has been developed to compare the watt derived from the ohm and the volt to the watt derived from the meter, kilogram, and second.The practical realization of the ohm and volt are based on the quantum Hall effect and the Josephson effect, and thus it is possible to monitor the kilogram artifact against these invariant quantum standards.To briefly summarize our experiment [2], the force on a current-carrying induction coil in a radial magnetic flux is measured, as is the voltage induced in that same coil when it moves at some velocity through that same magnetic flux.The same quantity of power is calculated from the measured electrical and mechanical quantities, by comparing the force times the velocity to the voltage times the current.As an indication of its earlier accuracy, the 1988 NIST value for the watt derived from this basic design had a relative combined standard uncertainty of 1.3 W/W [3].
To reduce this uncertainty, a series of improvements to the experiment were needed.By 1991, the radial magnetic field was produced by two opposing high-field superconducting solenoids which replaced the 1988 room temperature solenoids, and an immediate improvement in signal-to-noise was seen.Incremental improvements were made over the next three years, but within this last year, significant changes to the system, both physical and procedural, have allowed us to approach our desired goal of one part in 10 .The best withinrun (nine readings taking a total of 17 h) experimental standard deviation is 0.059 W/W.(The standard deviation of the mean for this particular data set was 0.02 W/W.)However, the potential of an accurate new watt determination from this high-precision data has not yet been realized.A number of improvements are still necessary to reduce the experimental standard deviation over the long term (weeks) below its present level of 3 to 4 parts in 10 .This paper gives a brief description of these recent improvements, recent data, estimates for some sources of uncertainty, and changes currently being designed.

II. SYSTEM IMPROVEMENTS
The most significant improvements to the experiment have been: 1) using three laser interferometers for velocity measurement, 2) minimizing induction-coil rotation, 3) reducing filter time delays and electrical noise in the voltage measurements, and 4) improving the control of the balance at room temperature.Additionally, several subsystems have been added to ascertain some correction factors including, 5) a gravimeter for determining local gravity, and 6) a refractometer for monitoring the refractive index of air.Furthermore, 7) procedures and instruments for aligning the electromagnetic, optical, and mass centers of the inductive coil with the axis of the superconducting solenoids and gravity have been incorporated, greatly reducing the overall alignment errors, and finally, 8) five subsystem computers have been added, networked, and programmed in a graphics-based instrument control language, with the replacement of old computers nearing completion.Details of items 1 to 5 are discussed here, items 6 and 7 discussed in greater detail in companion papers [4], [5], and though it's one of the most encompassing enhancements, item 8 will be described in a later paper, after completion.U.S. Government work not protected by U.S. Copyright.

A. Volt-Velocity (V/v) Quotient
Improvements 1 to 3 were initiated to decrease the uncertainties associated with simultaneously measuring the velocity and induced voltage of the induction coil as it passes through the radial magnetic flux density.To allow an accurate measurement of velocity and voltage, the experiment has been devised as a differential measurement.The voltage difference and the velocity difference between the moving induction coil and a twin fixed coil are measured.This arrangement ensures that unwanted motions of the superconducting solenoid with respect to the two coils cancel in the quotient, as well as reducing the noise from ac powerline pickup and other time-dependent magnetic fields.To measure the velocity of the center of the coil, a single, axially central interferometer was originally used to record coil position.However, this interferometer was vertically offset from the induction coil center by about With the interferometers located on the rim, the inherent coil rotation resulting from a suspension wire-band winding on or off the balance wheel required active angular control of the coil about the -axis, .This control helps keep the lasers from veering off the detectors, and reduces errors associated with laser alignment.A separate set of optics to record motions of the coil in the , and three rotational coordinates was already present.To serve as force actuators for this subsystem, gold-coated, glass plates were installed on each of the three coil support rods.Bipolar electrodes on both sides of the grounded plates provide an electrostatic force.Coil rotation during measurement is now controlled to within 0.3 mrad.Small forces associated with the verticality of the plates are discussed in an accompanying paper on alignment [4].
We have improved the time synchronization of the voltage and position measurements by using three high-speed, digital voltmeters (DVM's) instead of one.This permits using the DVM autozero function, while reducing the measurement deadtime to less than 200 ns, substantially lower than the 22 s between measurements when using one DVM.Deadtime can result in the accumulation of phase error between powerline cycles and DVM integration.The autozero function greatly reduces the DVM's drift and noise, which somewhat compensates for the removal of a chopper-stabilized preamplifier that had been used in earlier versions of this experiment.The preamp required substantial lowpass filtering, but data recorded with the three DVM's revealed a 20 ms phase shift in voltage with respect to the velocity of the 29 Hz signal.The preamp and filtering were discarded since the DVM's provided the wide dynamic range to accept the large amplitude frequencies.This trade-off has led to highly synchronized voltage-to-velocity readings, but with increased noise of about 100 nV in DVM readings and zero offset drifts of about 300 nV over 20 min.Reincorporating a quieter preamp into the experiment is desirable, if the filtering signal delay can be digitally compensated.
Further improvements in the voltage measurement chain are planned.Presently, voltages are generated at the 1.018 V level, and measured in series opposition with an op-amp current source (biased with a Hg battery) driving a standard 100 resistor.This reference voltage is in turn calibrated during experimental runs against a very stable Zener reference standard.The Zener is regularly calibrated in place against a Josephson array voltage standard system maintained in the watt balance building, and has a history of no measurable environmentallydependent drift.The relative standard uncertainty of this volt chain is about 50 nV/V.We are studying ways to reduce the high-frequency noise associated with using a Zener, and the uncertainty associated with nonlinear drifts of the Hg battery.Leakage resistance from the induction coil to ground is also larger than desired, contributing a relative standard uncertainty of about 20 nW/W.New induction coils are being wound to address this problem along with the induction-matching considerations.
Corrections to the displacement measurements for the refractive index of air, , are applied from measurements of temperature, humidity, carbon dioxide, and pressure.A highprecision refractometer was built [5] to check the uncertainty of the corrections.Early results from this refractometer showed differences of between the measured value and the modified Edlén formulae [6] due to temperature drift, leading to the installation of the temperature control system described below.Persistent differences in sometimes increase to as much as lower than the Edlén formulae.The most probable reason is that a small amount of either helium or a light organic gas is contaminating the room air.We plan to use real-time corrections from the refractometer results in the watt data.

B. Force-Current (F/I) Quotient
The implementation of definitive coil alignment procedures significantly decreased the previously largest component of uncertainty due to systematic effects in the force measurements to about 20 nW/W [4].Systematic effects from the hysteresis of the knife edge balance point have also been reduced by the replacement of an old, damaged sapphire knife-edge with a composite tantalum/tungsten one.To decrease a potential cause of hysteresis, balance rotations due to our method of placing or removing the mass onto the pan have been reduced to 0.6 mrad (about 0.3 mm -motion of the mass).Tests have also shown that a single oscillation of 3 mrad helps to erase material memory of earlier positions of prolonged stay.
Experience has shown that, not surprisingly, we get a better result from a data set when we average the quotient over as large a -axis displacement as possible.However, since we only have time to measure the force at one point, this averaging requires an accurate knowledge of the profile of magnetic flux density deviations along the displacement -axis from a perfectly uniform radial field.The magnetic flux density profile is far from insignificant, effectively varying by 360 W/W over the 7 cm of the velocity sweep.Fortunately, we have two ways to measure this profile, from the quotient versus displacement or from the quotient versus displacement.In our previous work [3] the quotient profile was used exclusively, but the measurement is time consuming and subject to hysteretic effects in the knife edge.Alternatively, measuring the quotient over 0.1 mm intervals produces an 800 point profile with typically 100 W/W noise.By fitting the average of all the velocity sets from each run, typically about 180 sets, a profile is obtained which can be compared to the profile taken with quotient data.At present the difference between these methods is within the statistical scatter of the measurements.This quotient averaged profile data is proving very useful, allowing daily records of the profile, for example.We are working to improve the techniques used for evaluating this profile function, since our results for either method have (0.5 to 2) W/W localized variations between different days' profile measurements.Data from the gravimeter installed in the watt balance building has indicated that deviations from calculated tidal effects correspond to a maximum of about 12 nW/W.A relative transfer from the gravimeter site to the level of the mass, some 4 m higher, needs to be remeasured, so the relative standard uncertainty in the assignment of gravity is currently about 40 nW/W.
To reduce the deviations in the measured refractive index of air and drifts in the balance due to temperature differences across the balance wheel, a subsystem for regulating the temperature of the balance room itself has also been added.The temperature is controlled to within 20 mK/day variations (greater after a liquid helium transfer).We presently have a controlled thermal vertical gradient of 1 C from the second to the first floor level of the apparatus, with the capability to further optimize this gradient.
A recognized limitation to our system concerns the myriad effects of air on the measurement process, thus our stated goal for this experiment is one part in 10 .Placing the entire coil and balance system into vacuum enclosure, as Kibble has done [7], addresses this problem, and we enter a major new phase of development with enclosure installation tentatively scheduled to begin in early 1997.Although measurements in a vacuum would entirely eliminate index of refraction corrections, we plan to begin with a rarefied, single-constituent atmosphere to shorten the intervals of two effects: bringing the system to thermal equilibrium, and dispersing the induction coil's self-heating.

III. DISCUSSION
Progress has been considerable in this last year, but each change was time consuming, and not all of these improvements have yet been optimized.The changes reported here have reduced our long term scatter (weeks, as opposed to day-to-day between-run) from several W/W over a year ago to that of our most recent data, which is shown in Fig. 1.The unweighted experimental standard deviation of these data sets is 0.43 W/W (standard deviation of the mean is 0.07 W/W).Since our within-run experimental standard deviation has generally less scatter, with the potential for much less scatter (e.g., 0.059 W/W as mentioned earlier), the shifts within this figure do not appear to be random.Thus our present standard deviation of the mean is of little value.We will report a value for the watt when we better understand the sources for these shifts.Table I lists our best estimates for the uncertainties arising from many possible error sources.Most components appear insignificant to the level of scatter.The three lettered (a) components represent the latest focus of our investigations and are probably smaller than stated, but we have not had sufficient time to gather specific measurement data.They are expected to be reduced or more exactly ascertained in the near future.
Richard L. Steiner, Aaron D. Gillespie, Ken-ichi Fujii, Edwin R. Williams, David B. Newell, A. Picard, Gerard N. Stenbakken, and Paul T. Olsen Abstract-The National Institute of Standards and Technology (NIST) watt balance is an experiment to compare measurements of the watt using electrical references (volt, ohm) to those using mechanical references (length, time, mass).A coil within a radial magnetic field has a dual use of: 1) generating a voltage by moving at some velocity to calibrate the magnetic flux density, and, 2) generating a force with electrical current to balance the gravitational force of a mass.This experiment has had several improvements made to it in the last year.These include the incorporation of three-laser interferometry and a refractometer to improve the velocity measurements, temperature control and coil rotation damping to reduce drifts and stabilize laser and mechanical alignments, and a gravimeter to determine local gravity.Systematic errors and scatter in long-term measurements have been greatly reduced in the last year, but statistically significant deviations relative to within-run uncertainty still persist.The source of these deviations has not yet been identified.Recent within-run standard deviations are generally near 0.1 W/W, which is the target precision of this present design.Index Terms-Absolute ampere, balance, fundamental constants, kilogram, mass standard, watt experiment.

Fig. 1 .
Fig. 1.Recent experimental results relative to W=W 90 .Some corrections of several tenths have not been included.Each set consists of typically 8-12 determinations.The outer error bars are the experimental standard deviations of each set, the inner bars are the standard deviation of the mean.
1.5 m, as our coil surrounds the superconducting solenoid and liquid helium Dewar.Vibrations in the coil support rods induce large 29 Hz signals in the velocity and voltage, so a displacement noise existed between the two signals.The positions of the coil's outer edges are now measured with three laser interferometers mounted at equally spaced points on the twin coils.Three pairs of position and time signals are measured via three time interval analyzers.These three signals are mathematically weighted to minimize the error caused by the difference between the optical center and the electrical center of the coils (Abbe offset).This three laser arrangement provides a more precise record of the coil motion.

TABLE I SUMMARY
OF RELATIVE STANDARD UNCERTAINTIES FOR SEVERAL MAJOR FACTORS IN THE WATT BALANCE EXPERIMENT.LETTERED (A) ITEMS ARE EXCEPTIONALLY LARGE ESTIMATES WHERE MORE TESTING IS NEEDED TO ASCERTAIN THEIR ACTUAL CONTRIBUTION