Text preview for : 9010085_eng_a_w.pdf part of Fluke 9010085 eng a w Fluke 5790A 9010085_eng_a_w.pdf
Back to : 9010085_eng_a_w.pdf | Home
A New Method for the Calibration of the mV Ranges
of an AC Measurement Standard
Speaker/Author
Neil Faulkner
Fluke Corporation
PO Box 9090, Everett, WA 98206
Phone: (425) 446-5538
FAX: (425) 446-5649
E-mail: [email protected]
Abstract
Presented are different methods for the calibration of an AC Measurement Standard from 20 mV
to 2 mV, 10 Hz to 1 MHz. The two methods currently used for this calibration are the
comparison to an AC/DC Transfer Standard and the use of the bootstrap method given in the
manufacturer's service manual. A new method has been developed using the AC/DC Transfer
Standard with a resistive voltage divider in a way that gives better Test Uncertainty Ratios
(TURs) then the current methods. This paper describes these methods, sources of error and ways
of reducing these errors. Also shown are test results comparing the methods [1].
1. Introduction
The Fluke 5790A AC Measurement Standard [2] is calibrated in the Fluke Labs by comparison
to a Fluke 792A AC/DC Transfer Standard on all ranges from 1 kV down to the 70 mV range but
is not used to verify the 20 mV range and below. The reason for this is low test uncertainty ratios
(TURs) and noise. Instead the bootstrap method as given in the service manual is used. Tests run
during the development of the 5790A showed this method to work with adequate TURs but some
labs and accrediting bodies have been hesitant to accept this method without further verification.
So a new method was developed which gives excellent results that can be used instead. This
method uses a resistive voltage divider in conjunction with the 792A. This paper presents the
challenges encountered in developing the divider method and how they were overcome. Once
developed the divider method was used to evaluate the bootstrap method. This testing is still
ongoing but the results so far indicate that the bootstrap method does work as originally intended
The testing has also revealed ways of improving the method to get better results if so desired.
Lastly test results with the different methods are presented.
2. Direct Measurements with 792A
The 5790A is verified on the 70 mV range and above by connecting the input of a 792A to the
input of the 5790A under test using a type N TEE and driving both with the output of a calibrator
such as the 5700A. The calibrator supplies both the AC and DC Voltage needed for the test. The
most important requirement for the 792A is that its uncertainties be low enough to give a good
TUR at all the test points. At the time this product was developed the uncertainties for the 792A
on its 20 mV range resulted in low TURs so it could not be used to do these ranges. The better
uncertainties available today on the 20 mV range does result in acceptable TURs at 20 mV and
may also at 6 mV and 2 mV depending on where the 792A is calibrated.
2004 NCSL International Workshop and Symposium
Another consideration when using the 792A is noise in the measurement, both low frequency
and high frequency. Since the 792A is an AC/DC device, the calibrator must supply a rather low
level of DC Voltage. Since most calibrators have a 100 mV to 300 mV as their lowest DCV
range, supplying DCV at 20 mV and below has a significant floor error since the voltage is at
tenth scale to one hundredth scale. Also any DC offset voltage drift during the test is not entirely
cancelled by the polarity reversal that is done. These offsets comes from the Thermal EMF
voltages of the connections and offset voltages in the instruments. There is also a problem with
high frequency noise. Most calibrators output some high frequency noise on the DC output and
when operating near the bottom of the range this noise can add significantly to the RMS value of
the DC Voltage. The 792A operates to 1 MHz and so it will respond to noise up to several MHz.
This places an error in the DC Measurement. There are ways of overcoming the noise problems
so a direct measurement with the 792A is possible down to 2 mV but the details of this are not
part of this paper.
3. Measurement with 792A and Resistive Voltage Divider
Using a 792A with a resistive voltage divider can provide a way to measure a 5790A down to the
2.2 mV range with good TURs even when the 792A doesn't have a calibration with low enough
uncertainties to do a direct measurement. Figure 1 shows a simplified diagram of how such a
divider is used. It is connected between the output of the calibrator and the input of the 5790A
under test (UUT). The 792A is connected to the input of the divider. An appropriate division
ratio is used to allow the calibrator to operate at a high enough voltage that it performs well and
also where the uncertainties of the 792A are low. At the same time it provides the mV signal
with a good signal to noise ratio. Before the divider is used, its division ratio at each test
frequency is determined using the 792A. The voltage applied to the UUT is found by multiplying
the input voltage as measured by the 792A times the division ratio for that frequency.
SOURCE
DIVIDER 5790A
HI I
HI
792A
LO LO
Figure 1. Simplified diagram of divider connections and current flow.
2004 NCSL International Workshop and Symposium
The divider is calibrated with the 792A before it is used. The 792A is first used to characterize
the UUT at the voltage level the divider will be putting out when calibrated. Then the divider is
connected as shown in Figure 1 and the 792A measures the input voltage to the divider while the
characterized UUT measures the output voltage. From these two voltages the division ratio is
found. This process is repeated for all the test frequencies. It was desired to have one division
ratio work over the 20 mV to 2 mV range and still be able to calibrate it at a high enough voltage
to get low uncertainties. A division ratio of 100:1 was picked as the best choice. This allows the
input voltage to be between 2 V and 0.2 V which works well for both the calibrator and the
792A. The voltage at which the divider is calibrated is a compromise between two factors. The
higher the voltage the better to minimize the errors when characterizing the UUT but if it is too
high then there is a problem with the power dissipation within the divider. It was found that an
input of 10 V with an output of 100 mV or an input of 6 V with an output of 60 mV was the best
choice. Of these two choices the 6 V in is the best if the uncertainties of the 792A are low
enough at 60 mV; if not then use 10 V in and 100 mV out.
4. Design problems to overcome with the divider method
There were several problems to be overcome designing a divider that would perform to the
desired uncertainty[3]. Figure 1 shows the flow of current between the devices. Most of the
current that flows through the divider and 792A input impedance returns through the Source LO
connection back to the calibrator but some of it can also flow through the UUT LO path due to
the capacitance between the LO and ground. At low frequencies this current is very small and the
common mode rejection of the UUT is very high so it contributes no significant error. As the
frequency goes up the current increases through the capacitance and the common mode rejection
of the UUT decreases so that above 100 kHz or so a significant error can occur. For a typical
divider setup this error will be very significant at 1MHz so something must be done to reduce
this current. Figure 2, on the next page, shows what was done to accomplish this.
First the cable from the divider to the source is kept as short as possible and uses large gauge
wire, one foot of #18 wire in our case. Secondly a short heavy braid strap is connected between
the source Ground terminal and the 5790A Guard terminal and the 5790A is set to EXT GRD.
Figure 2 shows the connections for a 5790A that has the Guard Modification. For a 5790A that
doesn't have the modification this strap should be connected to the Ground terminal on the
5790A and the 5790A set to INT GRD. The 5700A was also set to EXT GRD and its Guard
connection brought to the divider and connected to the divider output low after the common
mode choke. Tests showed that this was the best connection scheme and almost any deviation
from this gave poor results at 1 MHz.
Next a common mode choke was placed on the output of the divider. This choke places
inductance in series with the common mode current while not affecting the normal mode current
flowing through the HI and LO leads of the UUT. Care should be taken when building the choke
to minimize the normal mode inductance introduced by the windings. A good way to do this is to
use coax. Our unit was built by winding a few turns of miniature coax through a very high mu
ferrite core. The inductance in series with the common mode current needs to be about 150