I built a bridge

[BradC]

So I've been playing with working towards building Conrads Mini-Metrology Lab, and I got sidetracked comparing resistors.
I work in a relatively noisy environment, and I was getting a lot of noise in the meter when getting close to a balance, so I built a dedicated bridge in an aluminium box.

My construction needs some work. I had it all laid out and prepared for the arrival of the resistors, and I then went and soldered half of them in, bent them up and realised I was trying to connect them incorrectly resulting in some resistor-lead-yoga trying to get them to fit without putting any stress on the resistor bodies.

I used a 10k/10k Rc/Rd & open Ra/Rb.
Rd is a 10k/50k in parallel (8.333k), in series with a 2k in parallel with a 1k pot & 9.53k in series (1.653k-1.680k). So Rd is variable from about 9.986k to 10.014k.

Both Rc & Rd are TE UPW50 series wirewounds (3ppm/C). The 50K & 2K are TE UPF50 metal film (5ppm/C) and the 10 turn pot is a Vishay 20ppm wirewound.
The capacitors on the input and output are 0.001uF MKT 100V and the 220k resistor is just a 600mW 1% Metal film. The output clipping diodes are 1n4148.
The toggle switch is a Apem unit with gold plated silver contacts rated to a .01 ohm maximum resistance.

The reason for this post is the toggle switch. I wired it such that it changes the input polarity to Rc/Rd, effectively swapping the resistors. The theory was once settled I could flick the switch and if I balanced the bridge such that the output didn't change when I flicked the switch, then in theory the Rc/Rd side of the bridge should be relatively equally balanced, meaning the differential is due to the ratio difference between Ra/Rb.

The idea was being able to relatively accurately center Rc/Rd without needing anything more accurate than the meter being used as the differential detector.

I tried it using a spare 0.1% 10K UPF50 resistor as Rb, and a Genrad 1434 Decade as Ra. I deliberately set the decade off to unbalance the bridge and then alternated the switch polarity while adjusting the pot to get equal magnitudes. Once that was stable, I adjusted the decade to get as close to a null as I could, then separately measured the resistor and the decade on a hp3457a. They agreed +/- a couple of counts, so I figured the concept was sound, and thought I'd ask about it here as I'd likely find people with experience using bridges to measure/compare resistors in an absolute sense.

Is this a sane way of balancing one side of the bridge? Aside from the ability to move the bridge about significantly for resistor matching, I wanted the ability to be able to know relatively accurately that the a/b side of the bridge was as equal as I could possibly make it without resorting to exotic resistors or witchcraft.

[guenthert]

Nice work there.

[..]
The toggle switch is a Apem unit with gold plated silver contacts rated to a .01 ohm maximum resistance.

The reason for this post is the toggle switch. I wired it such that it changes the input polarity to Rc/Rd, effectively swapping the resistors. The theory was once settled I could flick the switch and if I balanced the bridge such that the output didn't change when I flicked the switch, then in theory the Rc/Rd side of the bridge should be relatively equally balanced, meaning the differential is due to the ratio difference between Ra/Rb.

The idea was being able to relatively accurately center Rc/Rd without needing anything more accurate than the meter being used as the differential detector.
[..]

What are you using as null detector?  Isn't the circuit a bit ill-defined during the switching?  I wonder whether there's a time when only one of the resistors is connected.  A DVM might not care much, but a more traditional null meter might not take it lightly, if instead of the expected few mV or uV, it sees a couple of Volt, even for a brief moment.  Oh, I guess, that's what the diodes are there for.

Isn't it more common to toggle the polarity of the supply voltage?  It won't assure that the balance is achieved, but at least show you which influence lingering thermal EMF has.

[Conrad Hoffman]

Nice! A matching bridge is a handy thing to have, at least for a voltnutter. I've never built a dedicated one, but should. My preference would be to switch the DUTs, but it's probably OK to switch the reference side. Just switching voltage doesn't help- you have to reverse the relationship between the DUT side and the reference side, as I think you've done. People don't even recognize them today, but these are what used to be used for bridge work- https://www.worthpoint.com/worthopedia/vintage-leeds-northrup-electric-knife-535680819

The old engineering and metrology books show some commercial matching bridges, but I've never seen one in the wild.

[BradC]

What are you using as null detector?  Isn't the circuit a bit ill-defined during the switching?  I wonder whether there's a time when only one of the resistors is connected.  A DVM might not care much, but a more traditional null meter might not take it lightly, if instead of the expected few mV or uV, it sees a couple of Volt, even for a brief moment.  Oh, I guess, that's what the diodes are there for.

I'm using either a HP3478a or a HP3457a, so yeah the point of the diodes was to clamp the output at ~600mV. The meters don't mind that so much, but they do tend to report overrange on the odd switch. I don't think I'd do that with a proper null meter.

Isn't it more common to toggle the polarity of the supply voltage?  It won't assure that the balance is achieved, but at least show you which influence lingering thermal EMF has.

Yes, but I'm not really looking for the lingering EMF here, I'm more interested in balancing the bridge as best as possible.

Nice! A matching bridge is a handy thing to have, at least for a voltnutter. I've never built a dedicated one, but should. My preference would be to switch the DUTs, but it's probably OK to switch the reference side. Just switching voltage doesn't help- you have to reverse the relationship between the DUT side and the reference side, as I think you've done. People don't even recognize them today, but these are what used to be used for bridge work- https://www.worthpoint.com/worthopedia/vintage-leeds-northrup-electric-knife-535680819

The old engineering and metrology books show some commercial matching bridges, but I've never seen one in the wild.

Yes, I was aiming to only switch one side of the bridge to use that to set the reference side as closely as possible to dead center. I thought about switching the DUT side but figured at some point I might want to do something out of the ordinary with the 4 binding posts connected to some other piece of kit and figured it'd be more predictable to keep the polarity outside the box constant.

Now I've had some time to use it, I realise I've made a bit of a blunder. I set the pot up with the intention of roughly +/- 1000ppm for resistor matching. The reality is that isn't enough to get a balance on a set of 1% metal films, and its so wide that I'm trying to balance the pot between windings when centering the reference side for a comparison measurement. So it's too narrow to be useful for resistor matching and too wide to be accurate in centering the reference. I think I'll re-jig it to narrow it right down to allow accurate reference centering and use my decade box as the reference resistor for comparison matching. That'll get me +/- 10ppm on 10K metal films, which would mean I can narrow the adjustment down far enough to be useful for comparison measurements.

Still, that's why we do these things.

<edit>
So I re-worked the pot from near enough +/-1500ppm to +/- 118ppm. Much more usable for getting a good zero. I can probably wind it even lower, but I'll play with this for a while and get a feel for it.

In all reality with a 27V input I now can easily get centered within 1ppm on a 10K pair (so ~7uV). I only have to move around in my chair and I get more noise than that. I might have to try more of a low pass filter on the output.

This is actually pretty cool as it becomes frightfully easy to see (and to some degree measure) all the other factors that inject noise, instability and variability into the system.

[BradC]

Nothing I'm going to put here will come as any surprise to anyone versed in the art of sub-ppm measurements, and I've known about all of this, but knowing about it in theory and seeing it happen on the bench is a whole other world of "wow".

I've been using my HP3457a as a null detector on the 30mV range. While I was on a long flight recently I put a bit of a spreadsheet together to calculate the relative accuracy of each range, and noticed the 3V range is by far the most accurate and quiet. I already knew that, but it never really twigged (the other thing it demonstrated is the hp3457a just less than twice as accurate as the hp3478a, which was a bit of an eye opener when looking at the error budgets. Those 3478a are a really good buy for the accuracy!). I digress.

Looking at the specs, on the 3V range I get a 1uV resolution, which is all I can reliably deal with anyway given the noise in my environment and the stability of the components.
So I started measuring on the 3V range. A couple of advantages :
- I no longer get intermittent overrange when flipping the reference side polarity as the 600mV diode clip is well within the range.
- I get a much more stable reading without the distraction of a whole heap of fluctuating digits I don't need to pay attention to.

Winner!

I also re-worked the balance control again to further reduce its range, and put more of the resistance in the 10K wirewound which has helped the temperature stability and made it much easier to balance (I've now got something like 5ppm per turn of the pot).

I can now compare resistors to within about 1/4ppm, which is another issue itself as each time I breathe the resistors change value (or I see thermoelectric offsets in the cheap binding posts I used). So high resolution, but not necessarily accurate.

The other thing I've noticed is the triboelectric noise in the coax I'm using to connect the bridge to the meter is HUGE! Bump the cable and there is a spike of between 10's to 100's of uV before it settles back down.

This has started me on a good learning exercise.

[Vgkid]

Really nice job.

[BradC]

So, chasing stability issues. I've identified the BNC connector on the panel as a source of ~20uV EMF that varies and doesn't seem to go away even with trying to get things thermally settled. Switching to the Banana sockets solves that.

Now to the switch. It turns out the contact resistance varies significantly each time the switch is toggled. Not good, particularly when the opposing contacts seem to vary in opposite directions causing wild swings in the output. I'm weighing up my options, but I wonder if a mercury wetted DPDT relay would offer a more consistent contact resistance?

Does anyone have any experience with low and consistent contact resistance switching? When 10mOhm is one PPM and you're trying to balance the bridge to within 0.1PPM a couple of mOhm variance on each terminal makes things jump around madly.

[Vgkid]

For a toggle switch, the best way to drop the resistance is to use many in parallel. Not too familliar with mercury relays, though mercury analgamed switches were used in bridges. I did some testing on a very special rotary switch(less than 1milliohm of contact resistance) below:
https://www.eevblog.com/forum/metrology/4-wire-switch-selectable-resistance-standard/

[BradC]

Yes, I'll try and replace it with a 4PDT toggle and parallel up the contacts.
Unfortunately the switch I used has a 10mm collar and all the others are 1/4" so I'll need to turn up a bush.

I did put a 100nf cap in series with a 470 ohm resistor across the switch common contacts to give it a decent whack of cleaning current each time it is toggled. That actually helped, but maybe half at best.

It's all a learning experience.

[BradC]

Right, so now I get to talk about resistor drift and thermal (soldering) abuse.

Somewhere below there is a rough schematic of the current configuration.

Power in on the left. The null-detector connects across A & B. The switch is the sod I have trouble with. I've sourced a 6P2T rotary switch, so I'm going to have a go at using 3 & 3 to see how that looks for contact resistance. Wiped contacts should be easier to keep clean.

Resistors are :
2 x 10K. UPW50 wirewounds (3-5ppm/K).
1 x 1M. Holco metal film 5ppm/K.
1 x 100R Holco metal film 5ppm/K.
1 x 1K 10T wirewound pot. 20ppm/K.
3 x 15ppm/K metal film to make 5K.
1 x metal film trim resistor. ~50ppm/K

When I built this thing I didn't really need a trim resistor. Within days it quickly drifted way past my ability to use it and I ended up with the trim being 9.906K (made up of a combination of 5 & 15ppm metal films).
After a week or so it had drifted past that, and I ended up re-working it with a 6.06k trim (again, more metal films).
I gave up on the trim resistor and patched a Genrad decade in there to keep up with the drift, and the trim is now down to 4.55K having had to be lowered about 100 ohms a day in the last week.

If I re-solder any of the joints it shoots back up 20 or 30ppm and takes a couple of days to settle back down again. So each time I de-soldered the trim resistor and put the decade across it to determine the new value I had shifted things around with the soldering heat that took days to settle. So I was calculating the next trim value on a heat shifted set. It was never going to work.

I plugged all this in a spreadsheet, and the change between the top (fixed) and bottom (adj) sides of the bridge since I put it together has been well in excess of 1 ohm (100ppm). Now, none of this is out of spec for the resistors (they are 0.1% after all), but it has been a serious learning experience in resistor behaviour. The 10T pot has roughly a +/- 8ppm range, so it's as sensitive as all buggery making it much more obvious (which is what I am after to be honest).

I've put the trim resistor out on some 1mm copper arms now, so when I eventually solder a new one in I can heat clamp those arms and keep the heat away from the rest of the resistors. That should allow for much more rapid settling when I have to re-trim it in the future.

When it has settled and reached a thermal/power equilibrium it's pretty good and I can get reliable measurements out of it for a couple of hours, so it's slow drift I'm up against. I'm going to get some better resistors, but rather than refurb this one I'll build a second using the lessons I've learned from this one, and keep tabs on how much this unit settles as it ages.

Learning, learning, learning...

[Vgkid]

Did you heat sink the resistors while soldering?

[BradC]

Did you heat sink the resistors while soldering?

Yep. Each resistor lead had a goot clip on it.

I suspect it's all to do with stress relaxing and initial settling. I only had to drop the trim 10 ohms this morning.
These metal film resistors seem to have quite a bit of hysteresis when exposed to differential heat, as while the heat clips do a good job there's still several 10s of degrees across the resistors as the leads are soldered.

Remember, we're talking 10s of PPM on resistors that are rated way higher than that, so I'm really using the wrong tools for the job and the process is not particularly well managed (a steep part of the learning curve). As in "oops, shouldn't have done that, now I'll need to wait a week for things to settle again".

[BradC]

An update. The resistors continue to drift. The trim is down to 3.11k now.

I did a quick set of tests/calculations and this indicates the 2 reference resistors have moved 205ppm since I started this, and they continue to slowly diverge. On average it's about 2-3ppm per day. Some days it's none, some days it's 7. Still well within the 0.1% tolerance, but interesting in that it's relatively reliable and consistently in the same direction. It is also slowing with time. I suppose that validates the little of what I know about resistor ageing.

My biggest issue is I don't know if it's the 10K wirewound or the 1M metal film causing the drift as I can't separate them. The stand-alone wirewound doesn't seem to move much based on a relatively inaccurate absolute measurement (hp3457a).

I wish I'd started collecting and date-stamping the data as this would produce an interesting graph.