Cheap voltage reference ICs

[WaveyDipole]

About a couple of years ago, after reading about voltage references on this forum, I purchased an LT1021 CCN8-10 10v voltage reference IC and built a small circuit to use as a voltage reference to periodically checking my DMMs.

Initially, the voltage read 10.0004v on the best DMM I had at the time which was a HP3478A, which seemed pretty accurate I no longer have the HP3478A, but I do currently have a calibrated Solartron Schlumberger 7150, an 18 month old Brymen 869s and a recently acquired Keithley 2000. The Keithley has recently been to a calibration lab but is known to have failed on the high ohms range, but apparently was otherwise OK. All three read 9.96xx. That's a 40mV (or 0.4% assuming my math is correct) drift from the original reference. I appreciate that even a reference can change slightly over time, usually due to burn-in, but I paid a little more for the 'C' part because at between 9.995 to 10.005v it had a better accuracy than the standard tolerance of +/-0.05% for the 'B' variant. The reading I am now getting is clearly outside of that range. Does that sound like an excessive drift for a voltage reference over 2 years?

There is a provision to add a limited or 'full' trim to pin 5 of the IC. Since the limited trim arrangement only gives +/-10mv adjustment, it is now insufficient to adjust to within the expected range. The full trim arrangement is required and does bring the reading back to the 10.0000v, but then even with the adjustment, is further drift likely to ocurr?

I'm guessing that my reference is stuffed and I will to get another one. Any suggestions as to what to get? I am willing to spend a little more to ensure reliability.

But what is drifting? Get NPL to measure your reference, they are not that expensive (in the order of a hundred pounds) and if you get an enthusiast to do it you get a guided tour of all the interesting bits. A few years ago I went to calibrate some rather esoteric ultrasonic sensors, it think it cost £500 for half a day with me there. The lads (an I) had lunch paid for (part of the deal, a rather good canteen) and as they didn't have much to do in the afternoon I did get to see/hear an anechoic chamber and an echoic one too, lots more as well.

[WaveyDipole]

Originally Posted by merlinmaxwell

But what is drifting? Get NPL to measure your reference, they are not that expensive (in the order of a hundred pounds) and if you get an enthusiast to do it you get a guided tour of all the interesting bits.

The evidence would suggest that the reference voltage shifted sometime between January 2018 and October 2018.

I have been noting down the readings for all of the meters each time I check them. I also ensure that everything has had at least 30min to warm up and stabilise.

In January 2018, all meters readings started with 10.00xxx with minor variations of no more than 0.00020V between them. In October I acquired the Solatron 7150 and checked against the same voltage reference again. All DMM readings were now starting with 9.96xxx. The variation of the last 3 digits was a little more than previously at around 0.00130V. It was at this point when I became puzzled by the shift in the second digit after the decimal point. I was not sure whether all of the DMMs had really drifted by almost exactly that much, but I put this down to the cold weather at the time. Today, I checked the readings again because I had recently acquired the Keithley and wanted a comparison, and found that all meter readings still start with 9.96xxx and the readings have no more than 0.00040V of variation between them.

While the results show some small variations between each of the meters, I cannot account for the shift of 0.04V across all readings. Either all of the meters are wrong by the same offset, or the relatively cheap LT1021C is wrong by 0.04V. My suspicion falls on the latter. I checked all of soldering joints on my little board just in case but found no problems. I don't think that its worth spending over £100 to check a £5 reference? I think that I would rather put that towards a decent and reliable reference instead.

I don't think that its worth spending over £100 to check a £5 reference? I think that I would rather put that towards a decent and reliable reference instead.

Say £50, even then you don't know how good it is. Have you a local calibration firm that will do a "back door" measurement for a pint or two?

Your results do point to the reference, how about getting 10 fairly good ones for a quid or so?

[WaveyDipole]

Well there's no shortage of cheap AD584 units on eBay. Sadly, I don't have a "back door" option at the moment.

As a matter of curiosity, what do they use in calibration references such as the Fluke 343 or similar?

[MrBungle]

AD584 is hardly worth it at 5-10ppm/oC. You can build a really really really good reference for about £20.

LM399 + LT1001. See datasheet app notes: https://www.analog.com/media/en/tech...s/199399fc.pdf

Both LM399 + LT1001 are 0.5ppm/oC. Add 5ppm/oC resistors. RS sell both parts and the resistors.

If you have 1 calibrated 6.5 digit you can measure it and then use it as a transfer standard for the other ones. This is what I'm doing

Edit: However you need to leave it powered up for about 3 months to age as there is some initial drift on the LM399 I understand. That might be semi-religious rubbish I've read however so don't take my word for it.

[WaveyDipole]

Anyone know much about the Time electronics 2003s DC voltage calibrator? I am wondering whether this is a better investment that one of these cheaper ICs? It obviously would cost me more, but it is selling at less than £100. Is this a reasonable price? Looks like it should be good to calibrate to three or 4 decimal points?

[MrBungle]

You'd have to get the calibrator calibrated. I got quoted I think £400 to have a calibrator calibrated so YMMV there...

[WaveyDipole]

Well yes, unless its already calibrated (and I am awaiting a reply to that question), then I guess that calibration would be required as a starting reference point. I didn't realise it would cost quite that much to get it calibrated though. I could compare it to my calibrated DMM to get a reasonable "transfer standard" or point of reference, but as time goes on, I guess it starts to come down to stability of the reference device and how prone it is to drift over time. I was just hoping that something like this calibrator would be much less prone to drift than these cheap AD584 or LT1021 chips.

BTW, I am also interested in your suggestion about the the LM399 + LT1001 solution and am keeping this in mind as an option, its just that this reference came up recently.

[MrBungle]

Yes indeed. I got a bit of a shock with the calibrator. This was a Fluke 5720A which I nearly bought but decided not to after the calibration cost was revealed. That was also if I lugged it down to the calibration company (not far from me fortunately) and back again. With courier return that was another £200 on top! If you want a basic voltage reference calibrated

You can work around the drift issue here by making three assumptions:

1. The absolute value of the reference doesn't actually matter (as you're not entering calibration routines)
2. The drift over an hour is minimal (observed by the long term drift being slow)
3. You use one device as a reference (backed up by the keithley being most stable on paper)

I'd pick the Keithley 2000 as the master reference as it's got a nice LM399 in it, 6.5 digits and has been calibrated. Get that calibrated once a year. Every N months, sit down, let the LT1021 reference warm up for half an hour, then measure it with the keithley. Assume that is gospel for the next hour. Then measure the remaining meters on the reference.

Thoughts?

That measurement procedure eliminates the drift problem.

[WaveyDipole]

I guess that means that the Keithley then becomes the de facto reference. I hadn't realised it used an LM399 as the master reference.

Given this and what you mentioned earlier, what do you think to one of these to replace my dodgy LT1021?

https://www.ebay.co.uk/itm/LM399H-10...9/254141535407

https://www.ebay.co.uk/itm/LM399AH-1...9/254141534028

It uses ther LM399 and the LT1001 OP amp. The second one uses the AH variant and seems a bit more expensive but I couldn't see any difference in the specs between the two on the datasheet?

I notice that LM399AH's in the ceramic casing cost typically anywhere from around £20 and up to £30, (£15.12 from the US + £163.51 shipping!) but can be had from China for as little as £3.99 although I would have to question the source. If one looks carefully, it is possible to differentiate these packages from the more expensive ones, so I have my doubts as to whether they are the genuine article.

I digress, but the above linked boards do seem to be using a genuine part.

[MrBungle]

I would go for one of them. They're respectable based on the things I have read (worth looking around on EEVblog) . The A variety is actually on my watch list but I have a deficit of cash for voltnuttery at the moment

LM399A is 0.3-1.0ppm/oC ... LM399 (non A) is 0.3-2.0ppm/oC. Depends on what you need.

The ones from China are probably from old equipment and might actually be genuine parts. What's more is they will have probably been aged nicely. Might be worth more. Might be junk you never know. I've had a lot of genuine stuff from that direction.

[evingar]

Originally Posted by MrBungle

Edit: However you need to leave it powered up for about 3 months to age as there is some initial drift on the LM399 I understand. That might be semi-religious rubbish I've read however so don't take my word for it.

Most of them have a burn in period. For our critical stuff at work we accelerate this by baking them before fitting to the PCB (the PCB's also have cut out "fingers" around the reference to reduce physical stresses and increase accuracy).

[WaveyDipole]

Yes, I am aware that such references have a burn-in period and it is just possible that the heat of the last summer (my "workshop" is just a shed) might have artificially 'aged' the LT1021 a bit. I just didn't expect it to go beyond tolerance.

Regarding the LM399, I noticed that the part used on the board that I had linked in #11 had a ceramic shield and the actual metal cased LM399 appears to sit inside this. I had imagined that this is designed to shield it from the effect of external thermal variations. However, I had wondered why the board had those rather large cut-outs around the base of the LM399, so thanks your comment. This would seem to explain that they are there to reduce the physical/mechanical stress on the legs of the component. I hadn't appreciated that such stresses could also affect the reference.

I notice that the board has a number of precision fixed resistors, but also a pot to trim the value. I can see how this is being used to set the reference to exactly 10.0000v, but surely even a precision 10 turn pot is, by comparison, significantly more prone to drift over time? Doesn't that negate the point of having a stable reference? Would it not be better to have a reference of a known fixed value, even if it is not exactly 10.0000v, but that is stable over the long term?

[MrBungle]

The cutouts on the LM399 aren't actually required. There was an experiment someone did which was flawed and everyone copied it (caution: there's religion here!). This was to stop thermal variations in the board from creeping up the legs of the devices. When you get into 7.5 digit and above then there are problems with dissimilar contact materials as well on the banana plugs (contact potential) and you end up with all sorts of expensive and exotic materials. Still cheap than audiophile crap though!

The precision pot only makes up a tiny proportion of the dividers so any thermal drift is divided by a few orders of magnitude. The divider is pretty insignificant if you use proper low tempco resistors. You don't get perfect ones on those reference boards. The really really stable ones are in HC49 type packages and tend to cost £20-30 a resistor (!)

With respect to the output value, this is a bit of a pain this one. The LM399 uses a buried reference zener and the optimium value for these for drift is 6.95 volts. Ultimately if you have a reference to calibrate something with you want to do it near full scale on the range so the ADC slope count is perfect so if you have a 3V scale (like my 3478A) then you really want a 3.000000V reference really! However the 34401A I have is 10V range so you want a 10.000000V reference. The latter is pretty standard across the board these days which is where I suspect the 10V reference comes from.

Again it's all a trade off. Everything is. And it's really annoying

Edit: another diversion. If you look at the HP3458A reference that's kind of ultimate. It uses the LTZ1000 reference which is expensive but 0.05ppm/oC to deliver a stable 8.5 digit reference (!):

1. https://xdevs.com/article/volt_xfer/
2. https://xdevs.com/fix/hp3458a/#a9pcba

That's pretty much the reference example other than weird stuff at NPL

[Craig Sawyers]

I have two Cropico zener voltage references. One is a 10V standard. I had that recalibrated about ten years ago, but it had not changed since it was built in the late 80's at a 20uV level.

The other is what they used to call the ESC (the Electronic Standard Cell) and gives outputs of 1.01861V and 1V.

I attach a paper about the design (not sure how I got that), and it shows the level of detail to get a piece of kit like that that will hold it calibration to 0.2ppm over long periods.

I also attach some images from someone who took a Cropico ESC apart. Three nested die-cast boxes, and critical parts sealed in copper blocks. The final couple will be in another post.

Craig

[Craig Sawyers]

Final two images

Craig

[Craig Sawyers]

Just found the cal data. Initial cal was by Cropico in 1990 and was 10.00000V with an uncertainty of +/-8ppm.

When I had it recalibrated in 2008 (so 18 years after initial cal) it was 9.999947V with an uncertainty of +/-0.2ppm. So it was lower by 53uV in 18 years, or 3uV per year.

However, since the initial cal had an uncertainty of +/-80uV, the new calibration is actually within the uncertainty limits of the initial cal.

Craig

[chriswood1900]

Craig
Thanks for posting the pictures and the Cropico paper it was fascinating to read. Do you have any other information or circuit diagrams of your unit.
Regards
Chris

[Craig Sawyers]

Just unhelpful ones that are in the ESC instruction. Attached. I don't have anything for the VS10, but they say in the single sheet marketing bumph that it is like the ESC, but has got a 5 milliohm output impedance. I think they left off the resistive dividers of the ESC, and put at least one op-amp in to boost the zener voltage to 10V. Of course the gain setting resistors have to be precise, aged and undrifting of course.

Craig