Measuring thermal EMF's

[essele]

I'm slowly working towards getting my volt-nut setup more, er, volt-nutty ... I now have a few seemingly good meters and I'd like to start characterising some of the references I've bought and built, but the first thing is to minimise the errors in my setup.

I'm just looking at how I can measure the thermal EMF's in my setup so that I can track improvements as I move from cheap and DIY test leads to low-thermal stuff and I just wanted to make sure I'm doing this in a sensible way.

So first thing is that I have three meters in parallel measuring a reference (or a short if I'm looking at emf's) ... I've seen something before about them impacting each other, is that a real issue? The alternative is switching them which feels like it would add significantly more error sources??

At the moment I've got my 732A connected using both LO terminals (which hopefully is a short, but keeps any lead-to-terminal impact) and then they run through a variety of home made or bought test leads to get to the three meters, and then on to a K155 to measure thermal emf's.

I'm seeing a fairly consistent movement around +/- 100-200nV with periods of zero, and then the very occasional swing for brief periods (small numbers of minutes) up to 3uV. These numbers seem to be reasonably consistent with what I'm seeing on the meters (albeit with different offsets.) ... and intuitively feels related to temperature variations in the room (at least the bigger swing.)

Does this sound like a sensible way to measure this?

I'm hoping that I can track improvements as I replace leads and move to spades where I can.

[MegaVolt]

A few books.

[Conrad Hoffman]

Most commercial meter leads are trouble. I use lengths of untinned copper wire, "bell wire", or pulls from multi-line phone cable. Ends cleaned and clamped directly under the banana posts. If I have to use plugs for the safety type recessed meters, I just try to be sure they're at the same temperature with no drafts. When in doubt, reverse connections and see if the offset changes. Also, don't trust anything you've touched with your fingers until it has a chance to stabilize.

[razvan784]

I can't say exactly how you could measure thermal emfs precisely in your application, but I can share my experience with various test leads and my 34401A.
First of all, if you're working at the 10V level, these probably won't matter unless you want to go sub-ppm. My understanding is that's one of the reasons high-end references are 10V.
All my data is taken with the meter on the 100mV scale at 10 PLC, read out over RS-232, optionally averaged, and plotted (volts / seconds).
First, a warmup plot. This is averaged 10 times (100 PLC effectively). The inputs are shorted with a short, multi-stranded copper wire, bent backwards and stuck directly inside the jacks.



You can see my meter needs about 3 hours to stabilize, and that's with the window closed and AC off IIRC.
With the window open, it looks like this:



So my first point is: if you want sub-uV stability you probably need climate control or at least a stable environment, and you have to let your equipment stabilize for a few hours, or at least take a long series of readings so you can gain some confidence that the readings are stable.
Perhaps an obvious point, but the meter may not stabilize to exactly zero with shorted inputs, and it may stabilize to a different reading if you're shorting the inputs with the cables you're later going to use for your measurements - you may need to take that into account.

Now on to thermal emf "measurements". My procedure is as follows:
* insert two identical leads in the meter jacks
* short the leads at the far end, keeping them together (minimizing loop area - some of my leads are twisted)
* let things stabilize for a few minutes
* grab the positive lead between my fingers right where it enters the meter (that would be the body of the banana jack in most cases), with a reasonably strong grip, and trying to avoid touching the other lead or the meter itself
* watch the graph and hold tight until it starts to level off
* release the lead and wait until the graph returns to the baseline
* repeat for the negative lead.
Example graph, this time taken without averaging (10PLC):



Of course you can do this with different test leads and learn about their thermoelectric behavior. Except for bare copper wire, I couldn't find any that are significantly better than the data above. Within my procedure I find that CuTe Pomonas, CuBe MultiContacts and brass MultiContacts behave about the same, at least in conjuntion with my 34401A input terminals. Remember, these are "low-emf" connectors, not "zero-emf"
So here my advice would be to wait at least 5 minutes after handling the equipment in order for the thermal transients to fade away. Combined with the others' advice to also take measurements with reversed polarity, I think this issue is manageable.
Note that at first I tried to take some measurements off the DMM screen; it is MUCH easier and clearer to use a graphing software.

[razvan784]

Here is another 34401a warmup plot, this time with the rear inputs shorted with a copper wire.
The plots are taken in three different days, not more than a week apart, but probably with different room temperatures.



My point, which I didn't state clearly enough in the previous post, is that (at least in my examples) thermal emfs generated inside instruments may be more important than those generated by external interconnect.
Of course these are evaluated by the manufacturer and included in the total uncertainty specification (which for the 34401A on the 100mV range is significantly higher than for the other ranges, perhaps in part due to thermal emf), but we were talking about voltnuttery therefore about what can be obtained beyond datasheet specs.

[razvan784]

About whether measuring with multiple meters in parallel might cause issues: maybe, but again it depends a lot on what you're measuring.
As a last example, here is another plot



I had access to a brand-new 34461A for a few hours. I wanted to do a cross-check with my ebay 34401A, so I quickly bodged a "reference voltage source" from an LM329 I had in my parts bin and a series resistor. Top trace is the '401, bottom is the '461. The source is the obvious limiting factor here, and any crosstalk is probably drowned in the noise. This crude setup was good enough for my purposes.
Others however are using 3458As and 2002s with properly-built LTZ1000-based references, and they reported seeing sub-ppm effects - see the 2020 Metrology Meeting thread for example.

[Dr. Frank]

I'm slowly working towards getting my volt-nut setup more, er, volt-nutty ... I now have a few seemingly good meters and I'd like to start characterising some of the references I've bought and built, but the first thing is to minimise the errors in my setup.

I'm just looking at how I can measure the thermal EMF's in my setup so that I can track improvements as I move from cheap and DIY test leads to low-thermal stuff and I just wanted to make sure I'm doing this in a sensible way.

So first thing is that I have three meters in parallel measuring a reference (or a short if I'm looking at emf's) ... I've seen something before about them impacting each other, is that a real issue? The alternative is switching them which feels like it would add significantly more error sources??

At the moment I've got my 732A connected using both LO terminals (which hopefully is a short, but keeps any lead-to-terminal impact) and then they run through a variety of home made or bought test leads to get to the three meters, and then on to a K155 to measure thermal emf's.

I'm seeing a fairly consistent movement around +/- 100-200nV with periods of zero, and then the very occasional swing for brief periods (small numbers of minutes) up to 3uV. These numbers seem to be reasonably consistent with what I'm seeing on the meters (albeit with different offsets.) ... and intuitively feels related to temperature variations in the room (at least the bigger swing.)

Does this sound like a sensible way to measure this?

I'm hoping that I can track improvements as I replace leads and move to spades where I can.

Good evening,
I think you've posed a related question a few months ago.. and here was my answer: https://www.eevblog.com/forum/metrology/what-kind-of-wire-and-connect-do-you-use-for-3458-measure-of-7-10v-references/msg3088969/#msg3088969

At first, anyhow, you have to identify and separate timely and temperature instabilities ("errors") in your measurements, as the e.m.f. is probably the smallest and hardest to evaluate, and to be mitigated disturbance variable. Most of the examples which the other colleagues have shown here, have nothing to do with e.mf., but with such other instabilities on the order of several ppm.

Paralleling several DMM is not a good idea, because you might superimpose sources of disturbance. We had a similar problem on our recent Metrology Meeting, and speculated that the current spikes from the 3458As affected the other one in parallel, but later it was obviously a ground loop, combined with the Guard connection.. I don't remember correctly any more, but anyhow, the easier your setup, the less convoluted is your signal path and error possibilities.

If you have constant room and equipment temperature, no external EMC, voltage references w/o popcorn noise, and so on, and achieve stable measurements, over an extended period of time (e.g. 24h), on the order of +/- 0.1 ppm, only then you prepared your experimental setup for an analysis of these residual e.m.f. errors.
I suppose that you compare a 3458A vs. your 732A, or vs. a DIY LTZ reference. Therefore, typical e.m.f. values are about 1... 0.1 µV, or on the order of < 0.1ppm, relative to 10V. Please judge for yourself, if that is really the determining factor of the stability ("accuracy") of your measurements.

In the above link, I already gave some practical e.mf. numbers for cable / connector combinations, which are measured indirectly by the reversing method.
You could also use your 155 to directly characterise your cable /  spade / plug combinations by shorting the contact e.m.f.
Also take care of temperature equilibrium everywhere, thermal guarding, no draught, and so on.

In the end, the basic problem with measuring voltage references is, that you can't short their e.m.f. voltages to null them, you only have the chance to mitigate or to determine them as best as possible.


Frank

[dietert1]

Thermal EMF is proportional to a temperature difference. In order to measure it, you need an oven (maybe some cheap incubator thingy) that lets you change the temperature of one of the "solder points" in a well defined way, while keeping the other one in more or less constant ambient temperature. I didn't have time to do that, but i also did those finger warming tests to see the size of the effect.
From those simple tests i learned how to do zero EMF measurements with low EMF material like copper lugs, copper cabling etc. Essentially a zero EMF measurement is the result of EMF cancellation. When you have the inputs of a meter wired up with two cables, you have two Seebeck circuits, one for each wire. If you are using proper material, a low EMF will be the result of the temperature difference at the two ends of each cable. The EMF won't be zero, because the meter and the reference may be at slightly different temperatures. To get cancellation, the two EMFs need to be the same. Assuming that both circuits are made of the same materials (wires, connectors, lugs) the two EMFs will be the same it the two input terminals of the meter are at exactly the same temperature, including the plugs in them, and if the two output terminals of the reference are at the same temperature, again including the plugs in them.

You need to protect the terminals of the meter from temperature differences. In the case of our Keithley nano-voltmeter, the concept of EMF cancellation by thermal isolation works down to 1 or 2 nV. The input connector got a 3D printed thermal shield meant to keep air movements away from the connector.

Regards, Dieter

[VNUTDENYER]

My gear is not at this level (yet - I hope) but I plan on using pliers (stored near VM for stability) to handle all plugs.  Looking for wine cooler to isolate from A/C and heater action.  DVM and std cells powered > 3k hours and seem stable.

[dl1640]

Thermal emf can be well controlled if temperature difference is insignificant between terminals e.g. Hi and Lo, critical applications may need thermal shield to against wind and thermal shock, normally I use cat.5e lan cable for low thermal measurements, just hook up freshly stripped solid copper wire onto meter binding posts, I read somewhere that NIST uses tinned stranded copper wires which gives better low thermal behavior in their application.

[VNUTDENYER]

So, moving plugs with a thermally stable tool to eliminate the 5 minute wait each time is not useful??  Just wondering.  Maybe just wrapping plugs in good thermal cover would reduce handling delay AND draft problems. 

[SilverSolder]

Wear gloves?

[Kleinstein]

Using a tool to reduce heating when handling the plugs can help a little. However it only reduces the starting values - the time constant for the equilibration stays similar.
It is not just the heat from handling the plugs, but with some gears also the terminals running a little warmer than the environment. Using a thermal cover may even make this temperature rise larger.
The main purpose of the cover is to avoid variable air currents - so there is also some advantage to a metal cover.

It there are screw terminals using bare wire is a relatively easy way to avoid the plugs.

Just for measuring the thermal EMF effect one does not need a stable reference - a short is usually more stable than a 7 / 10 V reference.