Is a Solartron 7081 worth the money?

[e61_phil]

Hi,

I have the chance to get a 7081 well below 1000€. But I'm not sure if it is worth the money.

Everything I saw from these meters was very noisy and slow. It will take 52s for 8.5 digits. And the other modes are noisier than my 34401A and 3456A. From the datasheet the meter is much better than the 34401A, but how is it in real life? The 34401A will also give 8.5 digits over GPIB and it can take more than 12 measurements until the 7081 will take one. I haven't tried it, but it could be less noisy than the 7081 (average 12 of NPLC100).

The main point for me is linearity. My 34401A seems to be able to do a 1:10 Ratio within ~0.1ppm of FS (measured against two 3458A at work). What about real life linearity of the 7081?

I found this: https://www.eevblog.com/forum/testgear/8-5-digit-dmm/msg270530/#msg270530
It seems to be anyone has measured one (?) unit to 0.1ppm INL. Does anyone have more experience on this topic?


The reference isn't a big argument, I think. In both cases a LTZ1000-mod would be possible.

So is it worth to spend some money to a Solartron 7081?

Thanks
Philipp

[guenthert]

The 34401A will also give 8.5 digits over GPIB and it can take more than 12 measurements until the 7081 will take one.

Can you elaborate on that?  I found earlier only that an extra digit (for a total of 7.5digits) can be read from the HP34401a (that unit was meant, right?) GPIB (or 7 digits using its math operations).  Either way, that doesn't mean that the maximum discrimination is 1 part in 20 (or even 200) million.

[ap]

IMO, the 7081 is a great meter. It is slow, true, but it has its values. Have a look at the data sheet. And if its properly working, it easily exceeds its spec. And if so, no need to change the reference. Why?
E.g. the 7081 has the ohms transfer accuracy specified, not so e.g. the 3458A.
If you dont want it, send me a PM.

[e61_phil]

The 34401A will also give 8.5 digits over GPIB and it can take more than 12 measurements until the 7081 will take one.

Can you elaborate on that?  I found earlier only that an extra digit (for a total of 7.5digits) can be read from the HP34401a (that unit was meant, right?) GPIB (or 7 digits using its math operations).  Either way, that doesn't mean that the maximum discrimination is 1 part in 20 (or even 200) million.

I never verified if all of the digits are fully used, but I did it now

I found a drift measurement with the HP 34401A and the Fluke 5440B:

14994 Samples
min. value: 9,9999505
max. value: 9,9999637

So there should be 132 counts between the values. Therefore, I made a histogram of the data with 132 bins. Only a few counts are missing. It would be interesting to know if these values will never appear..

[e61_phil]

And if its properly working, it easily exceeds its spec. And if so, no need to change the reference. Why?

I hoped one can reduce the noise with a LTZ1000.

For example I would like to calibrate my F5450A with the Solartron 7081 in volt mode (a calibrator is used as a current/voltage source). Therefore, most important is the INL I think. The specification of the 7081 is 0,3ppm of Range in 10V mode for transfer. Hopefully it is better. But it seems to me, that the noise is higher than 0,3ppm. Therefore I have to measure many times to reach a good transfer. Isn't it true?

[mimmus78]

You guys seem you always forget that a multimeter is a "multi" meter non just a 10V ADC.
You may be able to stick 4 LTZ reference inside a 34401A and get it very good at 10V, but what is the use of it if you than get the other ranges "going all over the places" exactly as it was before with the LM399?

[TiN]

For example I would like to calibrate my F5450A with the Solartron 7081 in volt mode (a calibrator is used as a current/voltage source). Therefore, most important is the INL I think.

The most important would be to have ability to measure current output from MFC accurately first, which in term require known resistance standard. Otherwise how you know what current flowing thru the DUT 5450A?

I'd say 7081 leaves nothing left from 34401A in case of stability/noise/resolution, but that's based on internet data. I never touched one before. Mickle.T have done amazing job reviving/modding his 7081's while ago.

[e61_phil]

You guys seem you always forget that a multimeter is a "multi" meter non just a 10V ADC.
You may be able to stick 4 LTZ reference inside a 34401A and get it very good at 10V, but what is the use of it if you than get the other ranges "going all over the places" exactly as it was before with the LM399?

You're absolutely right, but I only need a good 10VDC voltmeter for calibrations.

For example I would like to calibrate my F5450A with the Solartron 7081 in volt mode (a calibrator is used as a current/voltage source). Therefore, most important is the INL I think.

The most important would be to have ability to measure current output from MFC accurately first, which in term require known resistance standard. Otherwise how you know what current flowing thru the DUT 5450A?

It doesn't matter which currents flows. The 5450A has an "AutoCAl" feature. Therefore, you only need a short time stable current/voltage source and a short time stable meter (even calibration of the meter doesn't matter) but you will need a meter with very good linearity.
Starting point of calibration is the transfer from my VHP101 10k "standards" to the F5450-10k. All measurements are only ratios.

I'd say 7081 leaves nothing left from 34401A in case of stability/noise/resolution, but that's based on internet data. I never touched one before. Mickle.T have done amazing job reviving/modding his 7081's while ago.

I will give it a try. But as already said: stability isn't so important for me, resolution is the same as the 34401A and noise seems to be worse (I will check that). What really matters is the INL and that is better specified, but the specification is worse than the measured INL of the 34401A.

[e61_phil]

I did a short noise test with the Solatron 7081 and my Agilent 34401A. Both meters are in parallel on 10V out of the Fluke 5440B.

Solartron 7081: 7.5 digit mode ~3,2s sample time
Agilent 34401A: 100NPLC ~4s sample time (with AZ)

It isn't much, but it seems that the 7081 is noisier than the 34401A

[ap]

It would be interesting to see the noise at 8 digits in your setup (34401 NPLC adjusted). This is where the value of the instrument is, even though its slow.

[Henrik_V]

e61_phil,  thank you for the test and sharing results.
How did you wire the guard and ground?

[tggzzz]

My recently acquired Solartron 7081 and 10V "Trancell-II" Zener diode source, which I'm slowly resurrecting as I slowly descent into voltnuttery.

The cable is homemade, the 7081 has "baked itself" as described in the manual, and been left on for several weeks. One internal connector has been cleaned and firmly reseated to avoid measurement lockups.

Guard connected to the Trancell's "GND", i.e. case. Trancell powered from its internal battery.

[ap]

So looks like your combined noise, including Trancell, is 0.5ppm pp. I have not found anything about the Trancell reference, but would think that it contributes at least 50% of that. Any info?

[e61_phil]

e61_phil,  thank you for the test and sharing results.
How did you wire the guard and ground?

The wiring isn't ideal I think. The guard is connected to LO at the front of the Solartron (within the LEMO). The shield isn't connected to anything, but it is connected to PE anyway.

It would be interesting to see the noise at 8 digits in your setup (34401 NPLC adjusted). This is where the value of the instrument is, even though its slow.

I started this measurement now. But 100 samples will take some time

[tggzzz]

So looks like your combined noise, including Trancell, is 0.5ppm pp. I have not found anything about the Trancell reference, but would think that it contributes at least 50% of that. Any info?

I'm still "bootstrapping" myself in voltnuttery. Previously I was in the "not sure which is less wrong" territory, but I'm slowly moving to the "two clocks problem" territory.

The best info I have is https://www.eevblog.com/forum/metrology/standard-reference-labs-trancells-any-information-available/ but that is far from comprehensive.

Ignoring drift of a few ppm, the standard deviation for 8 nines measurements (calculated in OpenOffice Calc, not the 7081) is of the order of:

• 10V Trancell, 10V range: 1.5uV
• 10V Trancell, 100V range: 10.5uV
• 1.0186V saturated Weston standard cell, guard not connected:0.27uV
• 0V, guard connected to positive and negative: 0.07uV

Make of those what you will! They are certainly much better than when I received the devices and started measuring.

Edit: added the 100V range reading, which tends to suggest that 1uV of the 1.5uV is due to the meter.

[e61_phil]

new data..

Solartron: 8,5 digit mode ~52s sample time
34401A: 12 times 100NPLC averaged ~48s sample time

The difference in the noise isn't that big. What is more confusing is this > 1ppm jump from the solartron. I think it is due its auto zero. But it shouldn't generate such a big jump.

[mimmus78]

This 1ppm jump is not new to me ... I read about it some year ago on this forum somewhere. If I remember someone speculated it can be a firmware bug.

[e61_phil]

This 1ppm jump is not new to me ... I read about it some year ago on this forum somewhere. If I remember someone speculated it can be a firmware bug.

Interesting! I can try the software (whole board with EPROMS) from my broken 7071.


Does anyone know how many NPLC the 34401A can do? Somewhere in the manuals of the HP 3456A I read the 3456A is capable doing 10NPLC at maximum. If you set the instrument to 100NPLC it will take 10 samples with 10NPLC and average them. Does anyone know if the 34401A will do the same?

edit: First results with 1000 samples 10x10NPLC

[Kleinstein]

Due to internal 1/f noise, there is a limit to extending the time for a single ADC conversion. It is well possible that a 100 PLC time is already past the optimum for the 34401. So it does not really matter how long the meter internal integration time can be, one can always use later averaging - this is what many modern DMMs do internally when using longer NPLC setting. I am not sure the 34401 actually does 100 PLC as a single integration or as averaging. Comparing the noise (with shorted input to exclude reference noise) could give a hint.

The 7081 (like the Datron 1281) is different in this respect, as it uses an chopper stabilized input amplifier and AZ OP in the ADCs integrator. So here the longer integration time is still a real improvement as there is very little 1/f noise. This is also why it works quite well without autozero mode and does AZ only at long intervals.

There is a good chance that reference noise of the 7081 shows significant scattering. So not all the 7081 are equally good or bad. To some extend this is also true for other meters.

[e61_phil]

Thanks, Kleinstein!
At the moment I'm running some measurements on the 34401A with 10x10NPLC. By now, I think the ADC resolution of the 34401A is only 7.5 digits. The rest ist averaging. This could also explain why the 34401 will cut one digit above 10V, but you can get it back if you set an offset value (thanks Martin!). And that can also explain the missing data points in my histogram abvoe. If the 34401A is only capable of doing 10NPLC it is properly better to set up the instrument for measuring 10 times 10NPLC and do the averaging on the computer.


Next part in testing the Solartron 7081 should be a linearity test. But unfortunately this is much harder than the other tests

[Andreas]

Hello,

one question is also what floating point or math resolution is used within the instrument.
A single float (4 byte in "C") has only 24 bit mantissa resolution. (~7 digits)
So subtraction of a small number (zero offset) near full range can be a problem.
Double float has 6 Bytes mantissa. But I do not know if it was used when having no Co-Processor.

But at that time also programming languages (non "C") with 5 byte floats where popular.
Early Turbo-Pascal versions in CP/M had 6 byte floats.
Double Floats where used later with the 8087 Co-processor.

So what resolution is really usable (without math artefacts) within the instruments?

With best regards

Andreas

[e61_phil]

Hello,

one question is also what floating point or math resolution is used within the instrument.
A single float (4 byte in "C") has only 24 bit mantissa resolution. (~7 digits)
So subtraction of a small number (zero offset) near full range can be a problem.
Double float has 6 Bytes mantissa. But I do not know if it was used when having no Co-Processor.

But at that time also programming languages (non "C") with 5 byte floats where popular.
Early Turbo-Pascal versions in CP/M had 6 byte floats.
Double Floats where used later with the 8087 Co-processor.

So what resolution is really usable (without math artefacts) within the instruments?

With best regards

Andreas

It is also possible that the 34401A is using fixed point math. It will drop to 7.5 digit resolution if you apply voltages above 10V.

As shown in https://www.eevblog.com/forum/metrology/is-a-solartron-7081-worth-the-money/msg1178755/#msg1178755 8.5 digits seems to lead to some small artefacts. I would say that doesn't matter, if it is only 10 times averaged 10NPLC. You will need a computer with an interface to the 34401 to get more than 6.5 digits, anyway. Therefore, one can easily run the 34401A with 10 samples of 10NPLC and do all the math on the computer.

Andreas, I read somewhere you are using a resistor chain for measuring the linearity of your LTC2400 setup. Is it possible to measure linearity down to 0.1ppm FS with such a setup?

[guenthert]

It might be worth clarifying how exactly you're reading the data from your HP 34401A.  Those extraneous digits seem to be an artifact of the programming interface.   From the remote interface reference (part of the User's Guide, p.114):
--8<--
Use the resolution parameter to specify the desired resolution for
the measurement. Specify the resolution in the same units as the
measurement function, not in number of digits. For example, for
dc volts, specify the resolution in volts. For frequency, specify the
resolution in hertz.
-->8--
Which I read, as that you can specify any resolution you like.  The instrument will then scale the measured value to the desired resolution.  Try to specify nV for the 10V range, perhaps you get even 9.5 digits! 

[e61_phil]

It might be worth clarifying how exactly you're reading the data from your HP 34401A.  Those extraneous digits seem to be an artifact of the programming interface.   From the remote interface reference (part of the User's Guide, p.114):
--8<--
Use the resolution parameter to specify the desired resolution for
the measurement. Specify the resolution in the same units as the
measurement function, not in number of digits. For example, for
dc volts, specify the resolution in volts. For frequency, specify the
resolution in hertz.
-->8--
Which I read, as that you can specify any resolution you like.  The instrument will then scale the measured value to the desired resolution.  Try to specify nV for the 10V range, perhaps you get even 9.5 digits! 

You can't increase the resolution above 7.5 (8.5) Digits with this function. This is only usefull if you want reduce the resolution. Perhaps, measure 10V with only 1mV resolution. This will increase the measurement speed.

[Kleinstein]

Especially with the Solartron there is also a Quantization limit of the ADC. They use just a kind of PWM ratio with a relative low frequency. The ADCs resolution is relatively limited and thus scaling for the gain and than conversion to a decimal number can introduce some kind of rounding errors in the decimal format. AFIK there is a way to get the result in a floating point format with plenty of digits, however only a limited number of values that actually occur. This might be an issue in the 8.5 digit mode as the integration time is not that much longer there and the internal resolution of the ADC should only increase linear in time (if done the simple way).

A similar effect might happen with the 34401 too. There is also a limited internal (binary) resolution - though things could be a little more tricky with counting time slots for integration and the additional use of an extra ADC for residual charge. So there will be an extra scaling factor involved and thus an slightly more complicated situation of possible raw results from the ADC. Using averaging of several readings (internal of at the PC) could be a good idea here.

The math part can be important if one wants to check DNL from histogram data.

A possible partial test of the INL is by measuring two voltages and the sum of the two voltages. In the simplest form this is reversing the voltage source, knowing that the sum in 0, but it can be done with other test points too. In principle this should work down to the limits set by thermal EMF (and similar) when switching between the voltage. Noise may require quite a few repeats. Usually this needs a well isolated (e.g. battery powered) low noise source for two test voltages in series. A single source and a resistive divider is the most convenient setup.