Calibratory D-105 DC Precision Voltage Reference Standard

[idpromnut]

@Awesome14: Isn't the image you posted showing +/- 1ppm, so ~2ppm over the measured period?

EDIT: ah, wait, drift is an absolute measurement right? So it's 1ppm drift (+ or -) from 10.000000.  My bad!

[HighVoltage]

That was a terrible review, because the device wasn't used properly.

I think it would help, if you had clear markings of "input"and "output" on the binding posts

[codeboy2k]

All units with temperature compensation use the TIREF102C.

Thanks for the update awesome14! That's good to know, I like the REF102 specs better.

I was really disappointed to see it hand wired, and would have preferred to see a manufactured PC board in there. Please don't say it's not meant to be opened.  It should be safe to open any reference, as long as I don't mess with the settings. People open up 8 digit DMM's and Fluke 732's.  Opening it should not be a problem.

Despite that, it DOES WORK as built and calibrated by Awesome14, and it holds it's value to the best of my ability to measure it. The naysayers need to keep that in mind.   The temperature compensation works, I've not seen any changes in output and my home lab fluctuates about 10 deg C from morning to night (it quickly cools down over night!). So if i'm getting 2 ppm/C then I can't see the 20 ppm change on my meter (which is expected, since I only have a 5-1/2 digit meter).  When I get a better meter I'll be able to measure it better.

The value in your product is in the calibration you offer.  I can build it just as easily as you did, but I can't get it trimmed near 10V like you did, so that's valuable to me.

Good to see that you decided to make a PCB (I think you said so in one of your previous posts).  There's a lot of good information here in the forum, and lots of smart people who have good advice to offer you. You can make a really nice, clean and small PCB to fit in there, and it will significantly reduce your construction time and effort, and the product will still function as it should, and will ultimately be better in the end.  So consider that advice when you do, and you'll make a good product. With Gellar Labs no longer making temperature compensated references, you are in a unique position and have a head start on anyone else jumping in to fill that void. Go for it!

No one likes smoke and mirrors, so they might poke you on that, so remember to stick with the measured facts, only science and engineering here.

Cheers!

[codeboy2k]

That was a terrible review, because the device wasn't used properly.

Oh.. don't say that and just end there, because everyone , myself included, wants to know how it should be used then.
What did he do wrong?

I know he didn't use it at exactly the same voltage and temperature that was marked on it, when it was calibrated.  Is that what you mean?

[TimFox]

"Despite that, it DOES WORK as built and calibrated by Awesome14, and it holds it's value to the best of my ability to measure it. The naysayers need to keep that in mind.   The temperature compensation works, I've not seen any changes in output and my home lab fluctuates about 10 deg C from morning to night (it quickly cools down over night!). So if i'm getting 2 ppm/C then I can't see the 20 ppm change on my meter (which is expected, since I only have a 5-1/2 digit meter).  When I get a better meter I'll be able to measure it better.

The value in your product is in the calibration you offer.  I can build it just as easily as you did, but I can't get it trimmed near 10V like you did, so that's valuable to me."


I agree with this statement excerpted from codeboy2k's response above.  My measurements also are consistent with the spec, and I thought the price was reasonable for the calibration included.

[macboy]

You want the relay modules. The general purpose relay module will work for you. It does 10 DPDT connections I think. The solid state modules are basically an array of mosfets designed for signaling applications.


Sent from my Tablet

OK, I ordered the wrong multiplexer. So, I ordered the general-purpose keithley 7011-S.  It's supposed to be here next week Wednesday. I wonder, does anyone know if every relay in the multiplexer requires custom compensation. I could also read the manual.

In the 7011-S specifications, they state <5 uV (microvolt) contact potential per relay contact, but <2 uV per contact pair. You should always use a pair for switching the hi and lo side inputs at the same time. The thermoelectric potential is lower because it is assumed that the pair of contacts will be around the same temperature and therefore around the same potential, effectively cancelling each other out, as voltage is measured differentially across the pair.

There are some other cards (such as the 7067, of which I have a few) that have special low contact potential relays for <1 uV per pair. With 10 V, that's a 0.1 ppm error term.

[TimFox]

Any updates on long-term stability measurements of the device?

[gridleak]

If this transfer standard could actually maintain even 10ppm/year stability I would have to consider that Calibratory has some very specific empirical knowledge regarding those commodity 1% tolerance metal film resistors. Let's call them CMFRs. Anyway, that's what they look like. The specs. for commodity metal films is +/-1000ppm/year. Thus, the D105 should become useless after 12 months. That said, I believe that the stability specs published are for CMFRs at rated current, and 80C temp. I haven't found data on CMFR stability at 23C over a year. I might conjecture that Calibratory has a secret temperature cycling procedure to improve stability of CMFRs. Perhaps CMFRs become more stable with time, just as the reference does. Merely having an ample inventory of 20 year old CMFRs may permit Calibratory to select the ones that have remained stable for that long.  In order to maintain records on hundreds, or thousands of CMFRs for years on end would take a particular type of individual.  Indeed. Has anyone on this forum ever built two identical reference circuits, one with CMFRs, and the same one with foil or ww to actually compare stability at,  a small temperature range (+/-5C) above and below 23C  for over a year or more?  Then, of course, there's that 1K trimmer. That seems too much range, to me. As a disclaimer, I've purchased a D105, but I don't have a metrology lab setup that can track the progress or demise of my D105 over the course of a year. According to Calibratory, the D105 should remain unpowered, when not in use, for greatest stablility. That sentiment is not without support from some respected metrologists. In one scholarly paper LTZ1000 references were only powered on when needed. Aging was significantly less than other units left on continuously.  IMHO, this is truly a Wait and See thread at this point.

[timb]

The LTZ1000 and LM399's should only be powered on when needed due to the internal heater. The hotter you run a semiconductor the faster it degrades.


Sent from my Smartphone

[SeanB]

If you are using it as a standard leave it on, it will degrade slightly with time but the temperature cycling will be worse for drift.

[Awesome14]

If this transfer standard could actually maintain even 10ppm/year stability I would have to consider that Calibratory has some very specific empirical knowledge regarding those commodity 1% tolerance metal film resistors. Let's call them CMFRs. Anyway, that's what they look like. The specs. for commodity metal films is +/-1000ppm/year. Thus, the D105 should become useless after 12 months. That said, I believe that the stability specs published are for CMFRs at rated current, and 80C temp. I haven't found data on CMFR stability at 23C over a year. I might conjecture that Calibratory has a secret temperature cycling procedure to improve stability of CMFRs. Perhaps CMFRs become more stable with time, just as the reference does. Merely having an ample inventory of 20 year old CMFRs may permit Calibratory to select the ones that have remained stable for that long.  In order to maintain records on hundreds, or thousands of CMFRs for years on end would take a particular type of individual.  Indeed. Has anyone on this forum ever built two identical reference circuits, one with CMFRs, and the same one with foil or ww to actually compare stability at,  a small temperature range (+/-5C) above and below 23C  for over a year or more?  Then, of course, there's that 1K trimmer. That seems too much range, to me. As a disclaimer, I've purchased a D105, but I don't have a metrology lab setup that can track the progress or demise of my D105 over the course of a year. According to Calibratory, the D105 should remain unpowered, when not in use, for greatest stablility. That sentiment is not without support from some respected metrologists. In one scholarly paper LTZ1000 references were only powered on when needed. Aging was significantly less than other units left on continuously.  IMHO, this is truly a Wait and See thread at this point.

OK, so where to start. With a 45-day burn-in the device drifted < 1ppm average after 90 days. So, we upped the burn-in to 70 days, and the units are still within +-1ppm after 120 days, and counting. We've also changed our calibration procedure, which produces a device that will hold +-1ppm over at least 20 days. It's too soon to tell, but I think the units will hold to +-1ppm for at least 90 days.

We didn't change the spec from +-2ppm, because I'd rather give buyers greater precision than what is advertised than try to milk every penny. Our lab acquired a Fluke 732B that arrived back from the calibration lab a few days ago. We also purchased another 732A, three Fluke 8846As, (2) multiplexer rigs by Keithley (7001 ), and standard resistors. We tested and calculated the internal resistance (0.17 Ohm) of the units and the current draw at 10MOhm (3.0mA).

We've sold units to buyers in 35 nations, and had only one minor issue. The temperature drift is still 0.2ppm/degree C. About the drift in the resistors over time: worst case scenario figures are not typical. The components on our boards operate in a mild environment.

The device is a hit, worldwide. In our latest eBay listings there are about 100 units sold. And we have a 100% feedback score. 

[timb]

Oh lord, here we go again!

Let's see the empirical data to back up these statements. How often were measurements taken? Was the device connected to the same meter(s) and left undisturbed the entire time? Post a dump of the raw data for the entire 120 day period, please.


Sent from my Tablet

[acbern]

It has been made obvious throughout this blog that this design is, at best, a hobbyist approch to supply a somewhat precise (but in the end unknown) 10V reference voltage. And to be fair, more is not expectable at that price.

From a principle design perspective, using a reference (with an unknown tempco, could be up to plus or minus 10ppm/C for the LT1021C)  and compensating it with a thermistor may result in some temp compensation by coincidence, but this is by no means guaranteed, it may actually become worse. So in real life use, it may or may not compensate environmental temp drifts. That allone makes the 2ppm statement useless (ignoring anyway the fact that the 2ppm number misses any tolerance information with reference to national standards).
Overall, when you do a error propagation calc of all the possible contributors of error, lets face it, this item probably is rather in the 50ppm (K=2) level uncertainty worst case or more once on a users desk (initial drift, initial standard tolerance, worst case temp drift...).
One could go on, like no low EMF binding posts used (not expectable at that price). And so on and so on.
All this makes it an item with undefined specification, maybe usable to check 3.5 or 4.5 meters in a hobbyist environment, but thats it.
Any more statistics will not change this, given its basic design and build standards.

[eas]

The device is a hit, worldwide. In our latest eBay listings there are about 100 units sold. And we have a 100% feedback score.

There are multiple hypothesis that could be consistent with these results. You have a hypothesis that you have a natural bias towards, the one that proves you right, and makes you money.

An alternate hypothesis is that your customers don't know what they don't know, and aren't able to adequately validate your claims. And further, that you have, through cunning or dumb luck, chosen a combination of price point and claimed precision such that anyone with the knowledge and equipment to validate your claims isn't willing to invest the time, money and energy necessary to prove or disprove them.

I love the idea of an outsider proving the experts wrong. I know though that that can cloud my judgement. So, I look for independent sources of evidence. If you are both genuine and truly clever, you'll find ways to turn some of your critics into allies, rather than continuing to use them as foils to promote an underdog narrative.

[lars]

In the end of August I couldn’t resist buying a D105 just out of curiosity. It arrived in a week to Sweden but at the same day it arrived I went on a two week trip abroad so the D105 was left in my home for two more weeks before power-up. It showed +8ppm against my other voltage references but that may not say so much as I am not fortunate to have a calibrated Fluke 732. But what caught my attention was that it seemed to drift with temperature. So I did a temperature scan between 16 and 26°C. It showed a 13ppm difference. I contacted Calibratory LLC and they were very helpful. I got an instruction how much to turn the temperature compensation pot. Adjusted and did another temperature scan and the difference between 16 and 26 was just 1ppm. As the second-order temperature coefficient is quite high on the REF102 (and most all other buried zener based 10V refs) around room temperature I didn´t bother about the last ppm over 10°C range. I set the output to what I believed is 10V.

On October 6 I started to continuously check the D105 (which I call D105-1 in the chart below as it was my first D105). In the first two weeks the output dropped 7ppm and when was relative stable until I started a humidity test a week ago. On Oct 13 I got my second D105 (I call it D105-2). This time the D105 was spot on!! But in the next one and half weeks it moved 4ppm upwards before it stabilized.

Now afterwards I have checked my weather station and found the indoor humidity was almost 60% in September that is quite normal here in autumn. In the beginning of October it fell very quickly to about 40% and has stayed at 35-50% since when (We have had very nice weather in October and November)

On Nov 11 I placed both D105´s together with one of my Geller labs SVR boards in a two liter ice cream box together with 150grams of silica gel I have prepared earlier to have almost 90% relative humidity. Now after a week the SVR hasn´t changed but the D105´s has went up about 15-20ppm.

This test is very similar to tests I have done on six SVR boards with AD587LN´s with different datecodes from 1994 to 2006 (Note: the AD587LN were not ones supplied by Geller Labs). That test also revealed a change of up to 20ppm for 40-50%RH change. The interesting with the AD587LN was that one with datecode from 2004 had a very fast negative change with time constant of about a day or two and that counseled the long term positive change. So for slow humidity changes this looked very good. The other has just a small or almost zero fast negative sensitivity to humidity and the positive long term humidity sensitivity were shown as seasonal variations.

After having said this I still hope we can keep the discussion on a technical level and that nobody says wasn´t this what I already thought. I think we really should encourage Calibratory LLC to develop this standard more and be open minded. I really liked the SVR-T boards as simple voltage transfer standards from a Fluke 732. I also liked that Geller Labs was so open with design, schematic and BOM´s (including my NTC compensation with both first and second order compensation that Joe could have kept for himself). What I missed on the SVR was a box, banana jacks, reverse power polarity and EMC filtering. At least three of these requirements are satisfied with the D105. For EMC protection I can live without that by just turning off my mobile. In my lab I have no high EMI sources and no large transmitters nearby.

So my conclusions are: My two D105 are humidity sensitive but I like the design and concept of a voltage transfer standard in small package. And I really hope for a good open discussion.

Lars

[bingo600]

Hi Lars

So volt-nuts get to eat a lot of "Glass"   

I have a and a few SVR's , and a single SVR-T (thnx for the design) , Joe mentioned you did it 

/Bingo

[eas]

Thanks for providing some independent data, Lars!

[ltz2000]

The construction of these units is superb. There is nothing left to chance.

Here comes an offer. Let me do a proper printed circuit board and mechanical redesign for your product. Free of charge.

It costs you absolutely nothing. PCB and mechanical design was my day job for many many years, so I am sure that you will get more than you pay for.

Electrical, mechanical and thermal things taken into account. Binding posts mounted directly on the PCB, no wiring required. Probably saves your assembly time by 90%.

You will receive a set of ready processed manufacturing files. All you need to do is send them to a board manufacturing house of your choice.

The new design probably replaces some of the "magic" with improved reliability and stability, but I believe your customers can live with that.

For some reason I have a feeling what the answer is, but you never know...

[Macbeth]

lars, very interesting results. It's good to see the Calibratory D105 performs as expected. I think in your own case the ultimate 10V standard could be made by making an array of paralleled standards using both the D105's and the SVR's thus:

Code: [Select]

                         Geller SVR stage

+18V o----------------+----- SVR#1 10V -----+-----o
                      |                     |
                      +----- SVR#2 10V -----+

                      +----- SVR#1 GND -----+-----o
                      |
GND  o--+-------------+----- SVR#2 GND -----+
        |                                   |
        +--- D105#1 10V ---- D105#2 GND ----+
        |                                   |
        +--- D105#1 GND ---- D105#2 10V ----+

           Calibratory "null" stage

There would appear to be some redundancy in this array, and I would not recommend the "null" stage in other designs, but if you have it you may as well work with it. In this case anyway.

Those Geller SVR's seem pretty amazing. Shame he had to retire. Are the voltagestandard.com VREF10-003 up to any snuff? No magic claimed and proper board, and free recalibration after 12 months. Cheaper than D105 too.

[orin]

<snip>
I really liked the SVR-T boards as simple voltage transfer standards from a Fluke 732. I also liked that Geller Labs was so open with design, schematic and BOM´s (including my NTC compensation with both first and second order compensation that Joe could have kept for himself). What I missed on the SVR was a box, banana jacks, reverse power polarity and EMC filtering. At least three of these requirements are satisfied with the D105. For EMC protection I can live without that by just turning off my mobile. In my lab I have no high EMI sources and no large transmitters nearby.

Hi Lars,

I have an SVR-T (thanks for sharing the design with Geller Labs!) as well as a regular SVR.  I put my SVR-T into a nice metal box with feedthrough caps for the power supply, and added an LM317 based regulator inside the box to get the 15V for the SVR.  So, what EMC effects are you seeing with the SVR-T and what would you do about it?

I found that reversing the leads from an SVR to a 3455A would cause a drop in magnitude of the output by about 10 ppm!  A 0.047uF capacitor across the output helped.  I never really worked out what caused that.  I think it was OK with the 3456A.

I was thinking of adapting the output protection circuits from the Fluke 732 for both my SVR-T and Fluke 731B - any comments on that?

Orin.

[lars]

Hello Orin

Here comes some general comments to both you and others.

I have don´t know how the Fluke 732 is protected but my general thoughts of  EMI protection is to use the normal bypass capacitors on the PCB, use a ground plane or as close as possible, and use ferrite beads in series. The banana jacks makes the “Faraday case” difficult for high frequencies as it is difficult to bypass without inductances. Therefore the EMI protection should be already on the board.

Four examples of problems I have experienced in my home lab:

1. Mobile phones, gives GHz signal but can of course be avoided (turned off). As it isn´t a CW signal it also gives varying offsets on the voltage reference output so is easily seen.

2. 10MHZ from OCXO. Two years ago I often had a 10MHz OCXO connected with unshielded cables to an Arduino during a GPSDO controller development. This shifted some of my volt refs several ppm and as this was a CW signal it was a constant offset that might be difficult to recognize.

3. Florescent lamp. I added a florescent lamp over my lab bench and that gave a broadband noise all over both the frequency band and the room and disturbed almost everything.

4. 50Hz into pin 8 of AD 587. If I remember correct I had no bypass capacitor on the noise reduction pin 8 on the AD587 and had an NTC very close connected to a mains powered DMM. The AD587 also had +15V from a wall wart. This gave if I remember correct an offset of 60ppm! Your 3455 DMM problem might have been 50HZ hum into the output that really benefit from bypassing.

So my conclusion is that all frequencies from Hz to GHz can affect the voltage reference if not properly taken care of. My greatest concern is also the signals that give systematic errors that you might not recognize. If you check your voltage reference somewhere with different EM fields around and when go back to your lab you can have very different results. This is why I do EMC tests in my professional work but it is not so easy at home.

If you mean over voltage protection that is a completely different story that also is needed in a professional product. On my DIY voltage references I haven´t cared of that (yet).

Lars

[lars]

I think in your own case the ultimate 10V standard could be made by making an array of paralleled standards using both the D105's and the SVR's thus:

Many thanks Macbeth for the perfect idea.

So volt-nuts get to eat a lot of "Glass"   

/Bingo

Yes as a true volt-nut I eat a lot of "Glass" (icecream) just to get the boxes. Not only for humidity tests but also to store all my junk DIY voltage references.

Lars

[gilbenl]

We also purchased another 732A, three Fluke 8846As, (2) multiplexer rigs by Keithley (7001 ), and standard resistors. We tested and calculated the internal resistance (0.017 Ohm) of the units and the current draw at 10MOhm (3.0mA).

I'm currently working on an automated cal/measure rig, which uses a Keithley 7001. Which cards are you using? What cabling? How are the standards connected to the DUT? I'm particularly interested in the resistance side of things. How did you wire the 4-wire ohms through the 7001 (+I, +S, Ohm G and -I, -S, Ohm G)?

[robert_]

I just read trough this entire thread. You should really charge money for that kind of entertainment, we really laughed out behinds off instead of getting any work done today

Somehow this guy reminds me of Chris-Chan or Drachenlord, with more religious WTF added.

Now for the productive part, did anybody try to build the same circuit without the special god-grade magic and special-needs soldering? I would, but i dont have anything to verify the performance to see if it works just as well or better.

[Jay_Diddy_B]

Robert_

I was Laughing Out Load when I read 'Special Needs Soldering'

You can see a demonstration of 'SNS' in this video:



Enjoy!!

 'SNS' starts at about 1.00

Jay_Diddy_B