1V Reference @ 1ppm

[beanflying]

As you already have the same reference I jotted down some figures using my matching one this afternoon against my 34401A over a few Temps. Both meter and Reference were in the same environment and you will get an idea that the combined drift is +1ppm/C @ 10V the bulk of this based on other testing will be the Vref as the Agilent is well under 1ppm over this sort of environmental range.

Interestingly the buffer used on it is not effected by loading at either 10M or 1G more than 1 least significant digit.

Mainly done for me to upgrade the data for it after a years use. These cheapies are far from terrible for3.5 and 4.5 digit meters. You might even do reasonably putting a divider across one of these too just keep the values fairly high to minimize loading but still make your own as well

                                                                               25C
Nominal       25C               28C             32C          old data
2.5           2.49905        2.49908       2.49912       2.49905
5.0           5.00024        5.00027       5.00032       5.00021
7.5           7.49961        7.49964       7.49968       7.49955
10          10.00036      10.00039     10.00045     10.00028

[beanflying]

The answer is simple: there is not such an 1V/1ppm source available in form you would expect (not talking volnuttery capable calibrators). As I can see on the pictures from your link you are already volnut-positive, so you are asking in a wrong section

More than that without a base standard you have nothing to reference it to so you need to do the best you can and are happy with unlike volt nuttery. You can make all the LTZ references you like and still not know what a volt is.

[Rafael]

Accuracy and stability really make me happy. Having an equipment and believing in your measurements fascinates me. This antique equipment, Russian, is in excellent condition (at least cosmetic) and I want to get to a point in having vintage equipment with nixie displays and precision.

I've invested a few hundred dollars in parts and calibration devices.    But I still do not feel safe to calibrate my other equipment! I have older nixie-based devices clamoring for calibration, such as Meratroniks, Fluke, etc ... They leave any bench happy. 

[beanflying]

Not sure if this is widely known around the world but there is an adage "A man with a watch knows what time it is. A man with two watches is never sure." and that is clearly why a third and a forth and a fifth.. Meter is needed. Welcome to TEA. 

While I am comfortable with what standards and calibration equipment I have available is more than what I need my journey down the voltnut rabbit hole  is heading toward a 3458A and paying keep it in Calibration so I can start with a proper traceable voltage among others. If the budget doesn't allow it this year then my 34401A will get re calibrated by Keysight to get me a little closer to the truth and allow me to go over all of my other gear with good certainty.

I have a few Nixie meters in my collection there is just something about the glow

[spec]

To get the divider of 6 and 4k (assumed dead accurate) you would actually get 0.9996 volts read assuming the nominal 2.5V 10Meg meter.

which is -240ppm

So it would need the compensation pot fitting to the circuit to tweak it accordingly if it's purpose is for calibration and if not the input impedance of the driven circuit needs to be taken into account regardless.

This seems to defeat the whole purpose of a precision voltage source, or have I missed something?

The switch from virtually no load, to a relatively heavy load of 10M has changed the whole aspect of the design.

[Kleinstein]

To calibrate a meter one would likely need both a stable and accurate reference. So just building a reference could result in something like 1 V with < 1 ppm/K TC , but the absolute accuracy would be poor (e.g. like 0.1% or maybe even worse).
For a good absolute accuracy chances are a used Hg Cell (e.g. Weston) is probably the best chance if you don't want to pay for a used Fluke 732 or similar.  Depending on the meter it may not need exactly 1 V the slightly higher voltage mercury cell may be acceptable too. It would still need a precision buffer

[beanflying]

which is -240ppm

This seems to defeat the whole purpose of a precision voltage source, or have I missed something?

The switch from virtually no load, to a relatively heavy load of 10M has changed the whole aspect of the design.

It depends on what you want the divider to show and to what particular DUT. What Raphael needs effectively is a transfer standard using a cost effective method. Not sure of the nixie meters input impedance but as it is a 6.5 digit meter likely to be a lot higher than 10M?

My 735A Transfer Standard is sensitive to loading but it has a compensation pot available on the front panel to tweak this nominal 1V if needed. Normal procedure with this device is to Null it against a Weston Standard Cell (known voltage) or similar like I showed previously to set it's internal reference then via the internal divider network set a 1V output. Null meters in common use at the time in theory have an infinite resistance at the Null point so don't load the cell or dividers.  Manual is here if anyone is interested and it goes into the theory of operation and calculations for lower impedance meters using this transfer standard http://literature.cdn.keysight.com/litweb/pdf/00735-90002.pdf?id=1870825

To do this without a known volt source other than the 'claimed' accurate voltages on the 584K and or any other reference he may makes the 'initial accuracy' is the best guide to how close that may be and then test one against the other. So the MAX 6350 will have an initial accuracy of 5V +- 0.5mV in practice it will be better than this spec. Part of the issue is we are trying to make wine from pig swill 

You will not get the 'perfect' resistors to divide up the 2.5 or 5V source to 1V so compensation will be needed be that adjusting the input voltage or providing compensation on one or both legs of the divider. The reason I suggested tweaking the Reference is the trimpot (generally terrible tempco) will be fairly unaffected by temperature variation due to it's already narrowed range. The alternate is to do something similar to the 735A circuit and minimize the value of the trimmers in the circuit.

If he is to build a new reference I would use the 6350 as a start point as the +-0.5mV initial accuracy is the same for the 2.5V version which would make it twice the error.

[spec]

which is -240ppm

This seems to defeat the whole purpose of a precision voltage source, or have I missed something?

The switch from virtually no load, to a relatively heavy load of 10M has changed the whole aspect of the design.

It depends on what you want the divider to show and to what particular DUT. What Raphael needs effectively is a transfer standard using a cost effective method. Not sure of the nixie meters input impedance but as it is a 6.5 digit meter likely to be a lot higher than 10M?

My 735A Transfer Standard is sensitive to loading but it has a compensation pot available on the front panel to tweak this nominal 1V if needed. Normal procedure with this device is to Null it against a Weston Standard Cell (known voltage) or similar like I showed previously to set it's internal reference then via the internal divider network set a 1V output. Null meters in common use at the time in theory have an infinite resistance at the Null point so don't load the cell or dividers.  Manual is here if anyone is interested and it goes into the theory of operation and calculations for lower impedance meters using this transfer standard http://literature.cdn.keysight.com/litweb/pdf/00735-90002.pdf?id=1870825

To do this without a known volt source other than the 'claimed' accurate voltages on the 584K and or any other reference he may makes the 'initial accuracy' is the best guide to how close that may be and then test one against the other. So the MAX 6350 will have an initial accuracy of 5V +- 0.5mV in practice it will be better than this spec. Part of the issue is we are trying to make wine from pig swill 

You will not get the 'perfect' resistors to divide up the 2.5 or 5V source to 1V so compensation will be needed be that adjusting the input voltage or providing compensation on one or both legs of the divider. The reason I suggested tweaking the Reference is the trimpot (generally terrible tempco) will be fairly unaffected by temperature variation due to it's already narrowed range. The alternate is to do something similar to the 735A circuit and minimize the value of the trimmers in the circuit.

If he is to build a new reference I would use the 6350 as a start point as the +-0.5mV initial accuracy is the same for the 2.5V version which would make it twice the error.

Thanks for reply

(I do appreciate that most approaches would require an initial adjustment to get the absolute voltage spot on.)

I see, a transfer standard- that puts a whole new complexion on the design. What I see is an issue though is that relying on the input resistance/characteristics of the meter to be calibrated may wreck even the performance of a transfer standard as the characteristics of the meter input are not known (to the 1ppm level that is). Or are you proposing adjusting the transfer standard every time it is used? Perhaps that is the normal procedure.

Interesting stuff about your 735A transfer standard.

I wonder if a ref voltage of 1V25 would be acceptable because, if so, an LTC6655  based design could be used and no potential divider would be necessary. Admittedly the LTC6655 is 2ppm but, maybe, keep the LTC6655 at a constant temperature to fix that in terms of the overall performance of the voltage reference unit.

On the other hand, if a 1V reference was essential a voltage divider for the LTC6655 could be made to have a lower output resistance and that would reduce the effect of loading delta/unknowns.

It is a shame this is not a volume production job, because then we could just phone up the manufacturer, Maxim/LT/AD etc, and get them program a 1V Vref version for us.

https://www.analog.com/media/en/technical-documentation/data-sheets/6655ff.pdf

[beanflying]

Just to complete the sort of usage method here is what I did today.
Agilent 34401A set to 10Meg Impedance.
Adjust the HP 735A Transfer using the meter to 1V (actually ran out of cal range on the front panel so 0.999993)
Take that know 'known' (as known as I can make it) voltage and apply it to the 10Meg inputs of the DUT in this case the 121GW.
Result the 121GW is horrid and should never be trusted to read anything for reasons at least according to some ..... 

For interest the 735A is well under 1ppm/C (done by longer term logging) and the tests were done at about 28C as per the121GW.

The procedure using a homebrew solution would be fairly similar prepare the source as best you can for the test you plan for the  DUT in question.

[beanflying]

I wonder if a ref voltage of 1V25 would be acceptable because, if so, an LTC6655  based design could be used and no potential divider would be necessary. Admittedly the LTC6655 is 2ppm but, maybe, keep the LTC6655 at a constant temperature to fix that in terms of the overall performance of the voltage reference unit.

On the other hand, if a 1V reference was essential a voltage divider for the LTC6655 could be made to have a lower output resistance and that would reduce the effect of loading delta/unknowns.

It is a shame this is not a volume production job, because then we could just phone up the manufacturer, Maxim/LT/AD etc, and get them program a 1V Vref version for us.

https://www.analog.com/media/en/technical-documentation/data-sheets/6655ff.pdf

Generally for Cal points they are fairly specific. So Apply this then tweak that scenario is normal. So you would need the 1V.

Most meters would take the form of 1V, 10V and maybe 100V and 1kV (34401a needs all of these for a full cal) depending on how it is designed to be calibrated just for DCV. The 121GW for example needs 5,50,500 & 600 for the DCV and a couple of others for the mV range for a full Cal of just the DC volt side of the meter. This is why in part you see DC Calibrators to 1000V.

The thing to bear in mind is you were to use say the higher accuracy version of the LTC6655 at 1.25V then it's initial accuracy (0.025%) is 1.25 +-0.3mV so divided down the best 'known' accuracy will be higher than the MAX6350 not a deal breaker but if you don't have known volt it is an issue. A little care needs to be taken regards the long term stability figures of both devices and both will tend to settle with use to value X and then need an additional tweak but to what is the question 

Now we have scared or driven off everyone to or away from Metrology for ever come on in the Rabbit Hole is large and very very deep

[Kleinstein]

The Mercury reference cells are just 2% higher than 1 V. So it might still work. With such an old meter this might have well been the reference of choice at that time. Even if one had to divide down to exactly 1 V, it would help starting at just 1.019 V - the divider resistors would be responsible for only 2% and thus something like 12 times less critical than in a 1.25 V to 1 V divider.

An accurate 10 V reference and Hamon type divider (accurate 1:10) could be an option too.

[Wolfgang]

Hello,

Does anyone have any suggestions on how to find / build a 1v @ 1ppm reference?

Thanks a lot!

If you are serious there is but one choice: Make a Josephson standard. All other stuff is a crippled compromise. 1ppm is below everything you can buy from renowned companies. And dont forget to join the TEA and the voltnuttery forums.   

[beanflying]

The Mercury reference cells are just 2% higher than 1 V. So it might still work. With such an old meter this might have well been the reference of choice at that time. Even if one had to divide down to exactly 1 V, it would help starting at just 1.019 V - the divider resistors would be responsible for only 2% and thus something like 12 times less critical than in a 1.25 V to 1 V divider.

An accurate 10 V reference and Hamon type divider (accurate 1:10) could be an option too.

Can you still buy Mercury batteries in the EU or only from suspect sources?

Building a Hamon Divider for myself too when Mr Pettis catches up a bit 

[ArthurDent]

Here is a photo of the first of my two LTZ1000ACH 10 volt buffered references. It has been powered on and aging for months and is pretty damn stable. When in use it will only be run on battery power to avoid any possible offsets caused by any stray coupling caused by the power lines. Even with all the care I've used to construct this standard it still has to be adjusted and measured and no matter what I do it will fall short of 1 ppm.

If you can get 10 volts it is simple to divide it down to 1 volt.

[Kleinstein]

The Mercury reference cells are just 2% higher than 1 V. So it might still work. With such an old meter this might have well been the reference of choice at that time. Even if one had to divide down to exactly 1 V, it would help starting at just 1.019 V - the divider resistors would be responsible for only 2% and thus something like 12 times less critical than in a 1.25 V to 1 V divider.

An accurate 10 V reference and Hamon type divider (accurate 1:10) could be an option too.

Can you still buy Mercury batteries in the EU or only from suspect sources?

Building a Hamon Divider for myself too when Mr Pettis catches up a bit 

A mercury battery (like those old coin cells) would not help that much, they are more like 1.45 V and relatively high TC. There is essentially no way to get them anymore. AFAIK they are not manufactured anymore and thus would be old and likely bad by now. The point are used mercury reference cells. These are larger and last longer and still sometimes come up on ebay or similar. However they are difficult to transport - not all of them are like transport per parcel and the transport may not be allowed anyway, because of the mercury and acid inside. The cell may still need an extra high impedance buffer and temperature compensation or stabilization.

For a Hamon divider one does not need that super stable resistors. The idea usually is to first adjust the resistors chains to be equal and than do the extra connections to go from 3:3 to 1/3:3 resistor ratio. So it needs an extra adjustment step before use. The final accuracy is like the square of the resistor ratio accuracy. So starting with 0.1 % resistors is enough to get to the ppm range. It still needs care to avoid errors from self heating, thermal EMF and leakage.

[spec]

Doesn't a Harmon divider just provide an accurate, and stable potential divider? If that is the case, doesn't that still leave the inaccuracies cause by the 10M Ohm load of the meter under test?

Doesn't the same problem of loading apply to all reference cells too? Presumably, they have a significant output resistance (in terms of 1ppm).

[spec]

Can I ask a question of the experts in this field: When you are checking the precision of an instrument, what is the shortest time that the instrument could be connected to the reference source? I have a reason for asking.

[Kleinstein]

A mercury reference cell would need a butter for a 10 M input impedance meter. This usually also applies to a Hamon divider, unless rather low impedance, which would call for self heating problems.  However 6 Digit DMMs often have more than 10 M input impedance in the low voltage ranges.

[tggzzz]

Can I ask a question of the experts in this field: When you are checking the precision of an instrument, what is the shortest time that the instrument could be connected to the reference source? I have a reason for asking.

I would suggest "until the reading is stable and repeatable", where you have to supply the two obvious definitions.

Are you thinking of a mechanical multimeter and a Weston standard cell?

[spec]

Can I ask a question of the experts in this field: When you are checking the precision of an instrument, what is the shortest time that the instrument could be connected to the reference source? I have a reason for asking.

I would suggest "until the reading is stable and repeatable", where you have to supply the two obvious definitions.

Are you thinking of a mechanical multimeter and a Weston standard cell?

My thought was to use a capacitor to lower the effective output impedance of  a precision voltage divider but, in view of your and K's, remarks it is clear that a fair amount of time would  be required to do the calibration so that idea goes down the tubes as only a film capacitor could be used. A supercap, which could allow sufficient time, would be too leaky, I think

[beanflying]

This can just go around in circles until you fall deep into the rabbit hole. When you go beyond 4 1/2 digits of resolution each extra one is 10 times the work to do it right. Until you have a traceable stable known reference to work from you will never truly know what you are reading in particular at 6 digits and greater.

So at some point failing that YOU need to create YOUR Volt preferably as close as you can to the 'true' value.

For a basic home lab this can be as simple as pick your best meter and what it says rules the others. Worst option so buy/make a Vref.

Buy one of the AD584 based references making sure it has the 'known' voltages recorded on it (most seem not to have this currently). Works well offers a range of outputs and is stable against 10M and GOhm meters. Suffers some initial drift from stated figures over time and drifts in the order of 1PPM/C. Should handle a divider to 1V and maintain ok stability and accuracy of its voltage with a small load.

Standard Cells of various sorts. See Rocking Horse Poo Requires a GOhm meter or preferably a Null Meter and several cells. Not practical but a stable over time now curio and a sanity check device for me - if you can buy one they are Cool. NIST is back from being shutdown  https://www.nist.gov/sites/default/files/documents/calibrations/mn84.pdf

LM399 and LTX1000 based devices are POINTLESS unless you have a 'known' accurate standard or meter to set them with their initial accuracy varies widely. They will only be as 'accurate' as the device YOU measure them with but generally they are more stable than other References due to Temp but can take weeks to years to settle. They are a can of worms in the Rabbit Hole with you in free fall but always just out of reach 

Roll your own using the best initial accuracy Reference you can buy sensibly back to the Max series (Best Initial Accuracy) again unless someone else knows of a $10USD one that is better? I have Two in aluminum boxes running from LiPo batteries tweaked to give me an accurate 10V I am happy with for outside use. I do also own a Fluke 515A portable Calibrator which is great to 4.5 digits meters but not so good for 6.5 as it drifts more than the Max based References.

Use your best 'known' Reference to 'Null' out a separate supply with more grunt if you need it to avoid loading issues and test it's performance as you load it to see if it remains 'stable' as you load it. Feed that into a Hamon or whatever divider or op amp you like and test away being happy in the knowledge it is YOUR Volt

[tggzzz]

Can I ask a question of the experts in this field: When you are checking the precision of an instrument, what is the shortest time that the instrument could be connected to the reference source? I have a reason for asking.

I would suggest "until the reading is stable and repeatable", where you have to supply the two obvious definitions.

Are you thinking of a mechanical multimeter and a Weston standard cell?

My thought was to use a capacitor to lower the effective output impedance of  a precision voltage divider but, in view of your and K's, remarks it is clear that a fair amount of time would  be required to do the calibration so that idea goes down the tubes as only a film capacitor could be used. A supercap, which could allow sufficient time, would be too leaky, I think

Capacitors are crap components compared with resistors. Leakage, tempcos, non-linear voltage dependence, dielectric absorbtion, and more.

[tggzzz]

This can just go around in circles until you fall deep into the rabbit hole. When you go beyond 4 1/2 digits of resolution each extra one is 10 times the work to do it right. Until you have a traceable stable known reference to work from you will never truly know what you are reading in particular at 6 digits and greater.

So at some point failing that YOU need to create YOUR Volt preferably as close as you can to the 'true' value.

Getting a meter to agree with a voltage source is easy: just turn the knob.

Getting one meter to agree with another meter is a little more difficult: open one up and recalibrate it.

Getting one of your meters to agree with other people's meters is easy: just empty your wallet.

Standard Cells of various sorts. See Rocking Horse Poo Requires a GOhm meter or preferably a Null Meter and several cells. Not practical but a stable over time now curio and a sanity check device for me - if you can buy one they are Cool.

Cells also have another advantage: no popcorn noise. That enables you to infer the noise in a meter's reference.

[spec]

Can I ask a question of the experts in this field: When you are checking the precision of an instrument, what is the shortest time that the instrument could be connected to the reference source? I have a reason for asking.

I would suggest "until the reading is stable and repeatable", where you have to supply the two obvious definitions.

Are you thinking of a mechanical multimeter and a Weston standard cell?

My thought was to use a capacitor to lower the effective output impedance of  a precision voltage divider but, in view of your and K's, remarks it is clear that a fair amount of time would  be required to do the calibration so that idea goes down the tubes as only a film capacitor could be used. A supercap, which could allow sufficient time, would be too leaky, I think

Capacitors are crap components compared with resistors. Leakage, tempcos, non-linear voltage dependence, dielectric absorbtion, and more.

Oh yes I do know that, but there are capacitors and capacitors. Still overall you are obviously right. It was just a thought.

UPDATE: just noticed that the LTZ1000 does use a capacitor internally.

[spec]

Roll your own using the best initial accuracy Reference you can buy sensibly back to the Max series (Best Initial Accuracy) again unless someone else knows of a $10USD one that is better? I have Two in aluminum boxes running from LiPo batteries tweaked to give me an accurate 10V I am happy with for outside use. I do also own a Fluke 515A portable Calibrator which is great to 4.5 digits meters but not so good for 6.5 as it drifts more than the Max based References.

Are the electronics in the aluminum boxes in a controlled temperature?