Cheap eBay AD584 Voltage References - My experiences

[intabits]

A few months ago, I bought three cheap (AU$3ish) AD584-based voltage reference modules from a couple of Chinese eBay suppliers.
When the first couple arrived, I was a little disappointed to see that they came without the "Calibration Certificate" that was shown in the eBay photo. However,  the third one did have the certificate, so I ordered 2 more from the same supplier. (And have a further 2 more on the way...)

(pictures below - I don't know how get them to go here)

Of course given the price, I was suspicious that they might be useless fakes, and I was almost certain that the "Calibration Certificates" would prove to be total bullshit. However, these units did appear to provide output voltages as per  the specifications, but at the time I only had rubbish test instruments, and was unable to verify the measurements shown on the certificates with even the vaguest degree of certaintly.
But seeing that the 3 certificates I had were not all just reprints of the same figures, did instill a little confidence. Furthermore, they are hand-written with visible pen imprints.

Things changed recently when I acquired some decent voltmeters, a Fluke 8800A (5.5 digits)  and a HP 3456A (6.5 digits). So I was now in a position to properly test these modules.

All the modules use AD584JH devices, with various date codes as shown below.
Also shown is the reference temperature from the calibration certificates (modules #1 & #2 had no certificate)

   Date code   Cal. Temp (C)   
#1   0923      -
#2   1236      -   
#3   1793      11      
#4   d144      24
#5   9223      14

BTW, the modules that did come with certificates had other distinguishing features:-
- They were supplied in very thick ziplock bags (vs the usual thin bags)
- The AD584s on them have a duller matte finish (vs quite shiny) Especially #3 & #4 
- They have smaller reverse polarity protection diodes.

Below are the claimed measurements from the certificates for each range (said to be taken from an Agilent 34401), and the
results I measured using the HP3456A and Fluke 8800A. These two instruments, although not calibrated, were in very close agreement, the Fluke sometimes reading 1 or 2 counts different in the last digit, usually on the high side.

BTW: I saw the HP reading 10.zeroes on a lab calibrator/reference when I bought it, and together with the close corroboration from the Fluke, I am very confident in the accuracy of its readings. The HP is assumed to be the more accurate of the two, and the figures shown below are from it. (in the default 5.5 digit mode, set to NPLC=10, with autozero and filter enabled)

The references were powered by a 16v bank of LiFePO4 batteries to eliminate any chance of power supply fluctuations or noise.

All equipment was powered on and acclimatized for several hours before the measurements were taken, at an ambient temperature was 19.5C.

Range         2.5v      5.0v      7.5v      10v
#1            
Claimed         ---      ---      ---      ---
Measured      2.4983      5.0014      7.5003      10.0029
#2            
Claimed         ---      ---      ---      ---
Measured      2.4967      5.0005      7.4973      10.0007
#3            
Claimed         2.4990      5.0013      7.5002      10.0020
Measured      2.4990      5.0013      7.5000      10.0017
#4            
Claimed         2.5001      5.0011      7.5026      10.0032
Measured      2.5001      5.0011      7.5026      10.0033
#5            
Claimed         2.4992      5.0025      7.5018      10.0041
Measured      2.4993      5.0025      7.5019      10.0048

Averages       2.4987      5.0014      7.5004      10.0027


Conclusions
These modules do appear to be very stable and quite accurate, and when supplied with a measurement certificate, there is a good chance that the figures can be relied upon. (though I've also read that these can sometimes be faked).
It would seem they can provide a very cheap and reliable reference source for calibration of 4.5 digit (and possibly even 5.5 digit) multi-meters.
Also, the very close agreement between the certificates and my measurements gives cause for even greater confidence in the accuracy of my new test equipment.

[SeanB]

 I would have more confidence in 3 and 4, as they have the oldest dies, with the most ageing in them to something approximating being stable.

[Rbastler]

Good to see that those chips are relatively stable. Thank you for the comparison!
I have a few of them and I'm planning to build a reference with them for the time, when I'm away from my MV116.

[cellularmitosis]

I would have more confidence in 3 and 4, as they have the oldest dies, with the most ageing in them to something approximating being stable.

I know that's true for zener-based references, but is that also true for bandgap-based references?  (The AD584 is a bandgap)

[Andreas]

   Date code   Cal. Temp (C)   
#3   1793      11      
#4   d144      24
#5   9223      14

- The AD584s on them have a duller matte finish (vs quite shiny) Especially #3 & #4 

Datecode 1793 really?? (would be far past christmas this year)

I also doubt that someone does calibration of any gear between  11 or 14 deg C.
(you would need a warm jacket).

You could also suspect that the matte finish has been done to sand the original marking.
(perhaps they have repackaged a DIP device into a metal can?)

I would have more confidence in 3 and 4, as they have the oldest dies, with the most ageing in them to something approximating being stable.

I know that's true for zener-based references, but is that also true for bandgap-based references?  (The AD584 is a bandgap)

After my experience: after every stress to the die (e.g. soldering) a new ageing cycle may begin.

Usually on bandgaps the ageing drift is somewhat higher than on buried zeners.

with best regards

Andreas

[cellularmitosis]

Cross links:

Mark Hennessy wrote up a nice blog post about several of the different AD584 board designs available:

https://www.eevblog.com/forum/metrology/are-cheap-ad584-units-worth-it/msg921732/#msg921732

http://www.markhennessy.co.uk/ad584_references/

TiN took a look at one of the "KKMOON" units using a few 8.5-digit meters:

https://www.eevblog.com/forum/testgear/any-experience-with-cheap-ad548-referance/msg987298/#msg987298

[cellularmitosis]

I also ordered a few of the same units you have.  intabits, thanks for giving me a prod to get off my butt and post about these!

I ordered mine from ebay seller "alice1101983", which were $2.73 each.  http://www.ebay.com/itm/311500253890

My strategy was to look for units which looked rough and scratched up, hoping that was an indication that they were not counterfeit parts.

All of mine came with the little hand-written cal note against a 34401A.  I just measured all of these against my own 34401A, who's calibration history is unknown.  I've attached the results.


----

Measurement setup details:

I used two 9V batteries into an 7815 to power the boards.  Each board was allowed to settle for about a minute upon power up, and was allowed to settle for 10 to 20 seconds each time the jumper was moved to select a different output voltage.

My 34401A was allowed to warm up for about 45 minutes, was set to "slow 6 digit", 10G Ohm, and 3Hz filter.  A dish towel was used to cover the DUT and leads.  My AC has been set to 73 (and hasn't been changed in 24 hours+).

Connections were made using about a foot of solid core twisted pair from an old ethernet cable, with "dupont wire" jumper sockets on one and and screwed to banana adapters on the other end (Cinch 108-0753-102 https://www.digikey.com/product-detail/en/cinch-connectivity-solutions-johnson/108-0753-102/J10108-ND/565813 )

[cellularmitosis]

Here are some shots of the PCB layout.

[cellularmitosis]

"ML005-V1.1" appears to be the model number of this design.

While on taobao recently, I came across what appears to be a different product from the same designer, the "ML006-V1.2", which has a similar style but is based on the LM399.

"ML" appears to refer to the taobao seller, "Miller Electronics", or miller16235

https://world.taobao.com/item/538328618275.htm

It looks like his general approach is to build boards around salvaged components, most of which appear to be available individually on taobao.

I've recently gotten a few of these boards in, and I'll start another thread about them after I log some measurements with my 34401A.


edit: attaching screenshot of the taobao listing of the LM399 board

[cellularmitosis]

A few weeks ago I also logged some data from one of the AD584's (in the 10V position), using my 34401A.  Here's one of the runs, which is about 24 hours.  The vertical scale is in microvolts.

This really opened my eyes to how sensitive these measurements are.  Anytime I'm in the same room, the signal is relatively unstable.  When I'm asleep, or at work, or in another room, the signal is much more stable.

I also captured temp and humidity data, but I'm still working on getting multiple csv file to show up on the same plot...

The first attachment is the overall plot (minus some data at the beginning and end when the AD584 was switched off), followed by a marked up version of the plot.

The third plot is a zoomed in section when I was asleep, followed by the same plot with a 10-point rolling average.

I've also attached the CSV files ('data' is the 10V signal, 'env' is temperature and humidity).

[intabits]

Datecode 1793 really?? (would be far past christmas this year)

hehe, I didn't twig to that. I just copied from notes I took when I first got them.
Looking at the actual device, the date code is d017 (d=delta symbol), much more credible.

I also doubt that someone does calibration of any gear between  11 or 14 deg C.
(you would need a warm jacket).

I found that surprising also.
But maybe he does have a warm jacket...

You could also suspect that the matte finish has been done to sand the original marking.
(perhaps they have repackaged a DIP device into a metal can?)

They don't look sanded to me, just old.

[intabits]

...  thanks for giving me a prod to get off my butt and post about these!

Wow, you really did get off your butt and start posting!

My strategy was to look for units which looked rough and scratched up, hoping that was an indication that they were not counterfeit parts.

All of mine came with the little hand-written cal note against a 34401A.  I just measured all of these against my own 34401A, who's calibration history is unknown.  I've attached the results.

I don't think you can trust the ebay pictures though. I'm pretty sure that one of the shiny ones i have without the cal note had a photo of the matte type, with  a note, in a thick bag.

Your notes are  in the same handwriting as mine, and your measurements and cal note figures also match very well.
Your cal notes look to be done in the springtime, all the temps were in the 20's.
All in all, these things seem to be the real deal, at a good price.

Those LM399 modules look interesting. I'd like to get a couple, but TaoBao!?!? WTF?!
I look forward to your results on those.

I also plan to do some long-term data logging of the AD584's and a couple of LM399s, including temp and humidity.  It will be interesting to watch as we come out of winter here, and then track it well into summer.  I was thinking of just using some cheap ADS1115 16bit modules, but now I'm starting to think they may be a bit too crappy. Maybe I'll include some 24 bit ADCs in the mix, and while I'm at it, I can also add in an old LH0070-OH 10v ref. that I have here. (which shows as 10.00050v on the HP)

[cellularmitosis]

A quick update which I had forgotten to post earlier: I think I found the original Miller Electronics listing for these boards on taobao: https://world.taobao.com/item/520786269812.htm

He includes this note in the listing description:

"This is designed for flashlight forums designed for the voltage base! Offer to the brigade of the flashlight forum.

The purpose of this benchmark is to verify the new ML102-V9.0 charge cut-off accuracy.

Because most people use the multimeter watch, are tens of dollars three and a half, compared to what is absolutely necessary!"

ML102 refers to another one of his projects, a charger for 18650 lithium cells.  So apparently a bunch of flashlight-nuts needed to be able to calibrate their handheld DMM's (in order to verify the charger cut-off voltage), so he started offering what is effectively a budget calibration service by selling these voltage reference boards.  A pretty industrious guy!

Here is his list of "Miller originals": https://shop57989019.world.taobao.com/category-1051401568.htm?spm=a312a.7700824.w5003-16825319266.10.6f1cbbceoxMwVA&scene=taobao_shop

[WhichEnt2]

That $2.73 ref from alice1101983 currently not supplied with the cal notes.
Received one today.

[cellularmitosis]

That's a bummmer.  I appreciate you letting us know!

[WhichEnt2]

Another parcel from alice1101983 received today. Now with the calibration notes.
Difference is in the package and numbers on sticker. Module in ESD shielded bag comes in the first attempt.

Their writing of "9" is flipped. Weird.

[rhb]

I got 3 units from "andwind"  via eBay.  All the AD584JH devices are shiny and bear an 0923 date code.  Ambient was 81 F.  Meter was a 34401A which recently checked to within 1 digit of a new DMMCheck Plus calibrated to 1 ppm before shipment and had been on for an hour.  These did not include a calibration sheet and were an impulse buy at $2.83 while looking for something else.

First column is initial and second after 5 minutes powered on

#1
10.00171   10.00156
 7.50066       7.50059
  5.00171      5.00041
  2.49890      2.49889

#2
10.00079    10.00085
  7.49881       7.49884
  5.00171       5.00164
  2.49905       2.49921

#3
  9.99237       9.99238
  7.49532       7.49533
  4.99685       4.99689
  2.49919       2.49920

I bought them to evaluate aging behavior, but clearly they will calibrate a 3.5 digit DMM out of the box without the calibration sheet.

[WhichEnt2]

Another incoming from alice1101983, no cal papers, module in esd shielding bag.
And seller tells that he's not have any control about it. He can just guarantee that package contents match to lot description and pics.

[intabits]

I gave up trying to get more with the cal paper, I probably have enough anyway.
I did ask several sellers if they had the paper, including the ones that had supplied them previously.
But they all said they only had without papers...

[GregDunn]

Lots of useful and interesting info in these threads (yeah, read pretty much all of them including Mark's articles).   

Is there a current (2018) consensus as to the best (eBay or other) source for one of these modules that is reasonably likely to come with a cal sheet - or is it pretty random and best served by buying a handful and crossing your fingers for luck?   

[intabits]

Seems to be luck nowadays.
I've had with and without from the same seller...

[GregDunn]

Seems par for the course...     With any luck, maybe Malone will produce some more voltage reference boards when he comes back from vacation.

@intabits: Have any of the reference boards you've bought deviated significantly from each other?  I'd just like to be able to calibrate my 4.5 digit DMMs well enough that I'm within 1% for general lab use.  Maybe the typical reference is good enough for that without a cal certificate?

[Kleinstein]

Unless the AD584 is a fake, it should be well better than 1 %. With 1% or even 0.1% accuracy one can not really calibrate a 4.5 digit meter to a useful level.  It is only a crude, but maybe sufficient plausibility test at a few points.

If the existing meter and the reference agree reasonable well, chances are good everything is OK. If there is a significant difference, I would not discard the meter, as chances are it's the reference board. The one thing you don't get with these cheap references is a guarantee you can really trust.

An important point missing using such a unit for calibration is a reliable traceability back to a primary standard / national metrological  institute.

[GregDunn]

I understand that there is (for all practical purposes) no traceability on these inexpensive references.  My observation is that in tests against instruments which actually do have a traceable calibration, I've seen them consistently exceed 0.1% accuracy.  That's good enough for what I'm doing - IF the variability of these references is small enough.  That's what I'm trying to get an assessment of.  If the kind of results rhb got (earlier in this thread) are typical, they would be good enough for my purposes.

I actually hooked up all my DMMs last night to the same 5V power supply - a cheap one, but analog and extremely stable after warm-up.  I gave them all a considerable time to stabilize (around 1/2 hour).  All the Flukes measured within 0.02% of one another, and none of the others deviated from the Flukes by more than 0.16%.  Which was most accurate?  No way of knowing without a calibrated reference, of course, though I have an opinion.    

[GregDunn]

First of all, thanks to the posters for feedback about these units.  I ended up buying a likely-looking one off the Bay which promised a cal certificate.  We'll see.   

Also I picked up one of the nicely-packaged KKMOON units from Amazon, which happily has a hand-written cert from an Agilent 34401A reference.  It's clearly not a duplicate of the others I've seen, so at least we'll give them points for creativity.     This one is now cooking in the basement lab, hooked up to my bench meters.  I'm going to check the rig periodically throughout the day; the basement is quite stable in temperature year-round so it will be interesting to note what, if anything, self-heating does to the measurements after a while.

Interestingly, immediately after turn-on, my Keithley 197A read exactly the values on the cert.  The Fluke 8600A was a few mV higher; they usually track with that amount of offset, so that's normal.  The Dana 4200 reads a few mV lower than the Keithley, again normal.  Stay tuned.

Edit: Very impressive.  I monitored temp (24C) and voltages on the bench for about 12 hours and the Keithley sat right on the listed voltages the entire time, to the limits of its resolution; very stable.  Not bad for a nearly 30 year old DMM.  The Fluke read about 2-3mV higher, again rock solid.  The Dana started about 5-6mV low and drifted up to about 3mV low at the end of the run.  Of course, there's no way of knowing the "real" values of the voltage reference, but the fact that the Keithley read the same values as on the cal sheet - all 4 voltages - for the whole time is either great good fortune or perhaps corroboration of the transfer calibration used on the reference.  I won't pronounce judgement.  But I'm happy with this very inexpensive reference.

[intabits]

@intabits: Have any of the reference boards you've bought deviated significantly from each other?  I'd just like to be able to calibrate my 4.5 digit DMMs well enough that I'm within 1% for general lab use.  Maybe the typical reference is good enough for that without a cal certificate?

Sorry, didn't notice that this was directed at me.

At least for the units I have, I'd say they should be easily good enough to get a 4.5d any DMM to within 1%.

As to deviations, I haven't done much with mine except for this:-


Which is a mess that I've been running for year now.
It's made almost entirely out of ebay crap - I just wanted to get something logging away...
It consists of:-
1/  Two large hot-swappable LiFePO4 battery packs, powering:-
2/  An Arduino UNO with a data logging shield (RTC & SD card).
3/  A DS18B20 temperature sensor, plus a DHT11 (temp and humidity sensors)
4/  A REF5040 providing 4.096V to:-
5/  A cheap ADS1115 module (4Ch 16 bit ADC), which converts the output of:-
6/  2 of the AD584 modules
7/  2 eBay "LM399"s (outputs are divided by 2 for the ADC)

It logs timestamp, 2 temp, 1 humidity, and 4 VRef readings every 10 seconds (way too often, but storage is cheap).
It's sitting in my back junk room, where it collects a lot of dust, and is well exposed to the elements (near an open window, and a doorway that is almost open to the outside)

Below is the full year output (DS18B20 temp and the VRefs). There are a few glitches, and a month of missing data that I'm still kicking myself for losing.
Obviously, my software also needs a *lot* of work.
The strong upward trend (and a lot of other variation) I put down to the crappy ADC. But it can be seen that the 4 VRefs track each other quite well, and the two AD584s are nearly always with an LSB of each other.


This can be seen better here, which is 10 days in the middle of the above. 


At some stage, I'm hoping to do some analysis of all this, trying to see what correlates with what. Next version will use a better ADC (and maybe two, with one kept further away from the environmental changes)

[GregDunn]

That's quite impressive - in spite of large ambient temperature swings, they all seem to be pretty stable.  And, as you say, tracking is really good.  I appreciate you sharing that data!

I hope to have another test run when my other references arrive (assuming they all have cal sheets).  One of them is an LM399, and they're all supposedly aged.  I won't be able to log data as completely as you did, but given that my temperature is essentially constant I think I can get away with fewer samples on a long term test.  It will be interesting to see the meter variances, if any, on a single reference and the differences between references with all the meters hooked to it.  Depending on what I see from the other references, I may be able to tell which meter is most accurate.

[intabits]

Made some minor updates to my previous. Most importantly, I'd left out of the BOM, the REF5040 providing 4.096V to the ADC. Thus each LSB is 125uV.

Since the the two AD584's match each other to with a couple of LSBs on the 2.5V output, over a wide time, temperature and humidity range, I was wondering if similar applied to the 5.0V, 7.5V, and 10V outputs.
If so, using two with a floating power supply, they might provide a very stable 2.5V reference between the 5V of one and the 7.5V of the other.
All the common mode environmental issues would be cancelled out...

[GregDunn]

Now I have two AD584 references: one the KKMOON branded item off Amazon (claimed checked with a 34401A), and a generic AD584 off the Bay (the one with SMA connectors for the outputs as well as edge connector pins; Fluke 8846A reference meter).

I ran the KKMOON for a solid 24 hours after warmup (temp was about 23.7-24.0C) and all I ever saw was the typical ±1 LSD display error from any of my meters.  It was listed as "10.00325" on the hand-written sheet.  My Keithley 197a registered... 10.0037 and the Fluke 8600 and Dana 4200 were pretty close.  Here are the deviations I read vs. the supplied data:

Keithley: +0.0045%
8600A: +0.027%
8800A: -0.0015%
Dana: -0.009%

Now, the generic AD584.  Only about 3 hours of warmup so far, but I strongly suspect this won't change any further.  It's been solid since about 5 minutes after power-up (meters, of course, warmed up overnight first).  It had the printed cal data, but all the listings of this seller showed different values so if not actually measured, they tried to be creative.  However, after measuring the 10.00260 pin with the Keithley, I got... 10.0027.  Deviations from supplied data:

Keithley: -0.0010%
8600A: +0.024%
8800A: -0.0015%
Dana: -0.0060%

I have no idea what the "real" values of the references are, but after seeing that the meters all cluster around the written values with very small std dev and variance (yes, I actually calculated them.  Nerd.) , I'm impressed.  I would not have been surprised to see them all read higher or lower than the reference, given the low variance.  Oh well, here's a few data points for the forum at least.   

Edit: added the new 8800A/AF measurements and corrected a calculation.

[myf]

Hello,

I hope I'm posting in the right place.

I am looking for a cheap voltage reference and I do not have access to a 7.1 / 2 digit multimeter.

I have an inexpensive 4.1 / 2 digits multimeter uni-t 61E. Maybe I'll buy another 4.1/ 2 or 5.1 / 2 digits multimeter later.
I want to control my multimeters. Are all the numbers correct?

Are these AD586 with their own calibrated sheet sufficient for this purpose?
Is one device enough, or are there a lot of funny experiences that require 2 or 3 devices?

Which vendors deliver everytime the device with this genuine calibrated sheet on ebay.com or aliexpress.com?

On eBay, I find this article at around 10 €. Is it fair and serious, or do you know other cheaper ones?

http://www.ebay.co.uk/itm/AD584-Precision-Voltage-Reference-2-5-7-5-5-10v-Calibrated-By-Agilent-34401/261972330425

For search, the article number is 261972330425.

What are your advices?

Have a good day !

François, from France.

[GregDunn]

I have only tried one eBay vendor, and fortunately they shipped the unit with a cal sheet.

Unfortunately, I don't think anyone has found a vendor which is certain to provide cal data.  I bought the KKMOON branded Amazon unit

https://www.amazon.com/gp/product/B00Z033Y6Y/

and a board from the following vendor:

https://www.ebay.com/usr/vrg5612

But while I'm reasonably certain the Amazon unit will come with a cal sheet, there's no guarantee any eBay vendor will provide one.  I just took a chance and it worked out.

[Kosmic]

Which vendors deliver everytime the device with this genuine calibrated sheet on ebay.com or aliexpress.com?

Even if they give you a calibration sheet, you really think you can trust them ? Even if they took the trouble of taking some measurements you don't even know if their meter is calibrated.

I don't think those board are more accurate than the chip sitting on them. So you should look at the specsheet of the AD584 et AD586 and decide if the accuracy you get out-of-the-box is enough for you. Ex the AD584J is 10V +- 30mV.


If you want more accuracy you should look for other more expensive products (ex DMMCheck Plus voltagestandard.com or the 10V ref from FisherInstruments.com)

[Harfner]

I bought this unit https://www.ebay.de/itm/AD584-4-Channel-2-5v-7-5v-5v-10v-High-Precision-Voltage-Reference-Module-AD584JH/153094942682?ssPageName=STRK%3AMEBIDX%3AIT&_trksid=p2060353.m2749.l2649

It does contain an AD584JH with a datecode from 93 and 4 resistors. On the back are values written by hand supposedly measured from this board. The voltages are measured to 4 1/2 digits, the resistances to 5 1/2 digits.

I checked with a HP3478A and a Prema 5000, both several years out of calibration. If I assume both meters to be within 200ppm (their 1-Year limits are 120-150 ppm), than the values are genuine.

[GregDunn]

I should probably add that I acquired a third reference, though this one is an LM399-based board with a noticeably different "cal" value.  The values read off my meters all cluster closely around this one too, though with a slightly higher std. dev. value.

In all three cases, the mean of the measurements is within 0.01% of the written value given for the reference board.  Is it possible that randomly picked digits give a number which is the mean of a tight cluster of values given by my meters in all three cases?  Of course.  But I believe they were actually read off a high precision meter, of unknown calibration heritage.  The fact that my meters (all of wildly different ages, design and use history) agree within a small fraction of a percent to the written values gives me some confidence that the values on the stickers were obtained in good faith, no matter their absolute accuracy.  The whole point of calibrating them (if done with a traceable reference) is to get an idea of what the actual voltage is - you already know it's not going to be 10.000000V.

And, if they're aged, the accuracy is less important than the stability.  Even the worst AD584 is spec'ed at 25ppm long term drift - on the same order as the output noise.  The LM399 is much better than that.  Good enough to at least ensure that the value, whatever it is, doesn't drift far enough to cause a problem on any of my meters.  They're clearly all inadequate to use as a transfer standard for something like a 3458A.