Calibratory D-105 DC Precision Voltage Reference Standard

[timb]

And holy shit, those photos are scary as hell. You can buy a working Fluke standard (from the 1960's) for $200.

Those standards are +-10-20ppm. Why pay more for less? Why would anyone pay this goober $100 for a few components poorly hand soldered together?

How truly it has been said, "Do not throw pearls before swine." and again, "They will see but not understand. They will listen but not hear."

Clearly I possess that which you do not know. Otherwise your mind would be opened, and you would see the truth in front of your eyes . The old must pass away to make room for the new.  Everything is not as it appears. Else, my work could be stolen.
[/quote]

So which model Fluke calibrator was I referring to in my post? Does your unit go from 10uV to 40V and 1uA to 10A? Because the Fluke unit does. A lot more versatility for slightly more money.

Or you could get a DMMCheck for less than half what yours costs, and that does current, voltage and resistance! (Though at 5-10PPM.) Its also made on a professional PCB.

If you've sold as many of these as you claim, there's no reason you can't have PCBs made through OSHPark or somewhere.

Just the wires going to/from the banana jacks alone will create enough thermal EMF to invalidate your claims. Let alone the fact there's no support cutouts for the reference IC!

And guess what, smartass, I *have* done better. I've created a multitude of voltage reference circuits, including my own ovenized design with a discrete Zener.

The fact you're selling something for $100 that looks like a weekend project that was soldered by Michael J. Fox is pathetic. You should be ashamed.


Sent from my Tablet

[HighVoltage]

I was on the road for a few days and the unit I have ordered has not arrived. In the meantime I did ask the seller a few more questions and this is what he answered:

Quote:
That's just bad science to make a blanket proclamation that metal is better than plastic for accuracy. Either can be used. You just need to know what the effect of each is on the final product. Then, pick the one that will work the best. For our design plastic is better.

The 8-digit meter and the 732A belong to our calibration lab, which is owned by CalibratoryTM, LLC. I personally own a large part of CalibratoryTM, LLC, which also manufactures the voltage reference, D-105 10DC. But no man is an island! We all need other people to accomplish anything.

Ceramic packages allow greater internal heat to develop without melting the IC case. Ceramic or plastic really have no direct relation to accuracy. Case material is not reliable indicator of anything but heat resistance.

Our standard is ~2-1/2 times the accuracy of the Geller standard. But it looks like Geller stopped making the SVR-T. Whatever is the case, the SVR-T has fundamental design flaws that limit its use to certain restrictive environments. Tde bottom line is: there are many places in a design of this sort to shave off slivers of noise, jitter, instability and drift, if you know where to look.

But Geller did a pretty good job. He just lost faith that the device could be improved any further.
End-Quote

I am also shocked to see those pictures, especially ion comparison to the Geller Labs version and the Voltage-Standard PCB
Unreal

[Dr. Frank]

The photos of the construction of the reference and this aggressive self-manifestation of the seller here in the forum reinforces my scepticism, I made up from this exaggerated  description on ebay:

" Over a series of years the design has added filtering, noise cancellation, thermoelectric-effect suppression, high-granularity trimming, temperature compensation, dynamic settling, high-accuracy output, low drift, silicon aging, miniaturization, output jitter-damping, resistive balance, safety features, and excellent repeatability."


I don't see any circuit elements on the photos, and I doubt that any of his secret treatments of this reference, may really  deliver such a stability claimed.
In contrary, by using simple metal film resistors (obviously), which have big drift parameters, these will hardly give the stability needed, as hard as one tries to compensate that..

As a (real) scientist, I don't believe in magic either, and really being an experimental physicist, I only believe, what I can see and what I can measure. 

Therefore, I'm looking forward to get this D-105 into my lab to check the stability of this device!

(The German customs at Frankfurt airport will store that parcel for another 3 weeks, I assume).


OK, concerning the specification, the seller claims 2ppm accuracy, not uncertainty (!) against his 732A, not againt NIST (!).

That only means the reference is trimmed to better than 2ppm difference compared to his 732A.
There's no claim of the uncertainty of his 732A to SI!
Therefore, I truly believe this spec. parameter.

Therefore, the absolute uncertainty may be measured in any external lab, but that may differ from SI Volt apart of the claimed 2ppm.

But I think, that the claimed stability of 0.15ppm/°C and 1.5ppm/1kHrs in operation can be easily verified, my Veltins climate chamber is already waiting.
But I have to suppose from the foregoing arguments, that the reference will fail these claims.

Anyhow, let's see. I'm really curious!

Frank

[quarks]

I do not know what to say, but I know I do not like where this is going here and also in "my" LTZ1000 thread.

@Awesome14,
as you are the developer of the D-105DC, I must say for me your responses (so far) are not beneficial.
If your Refs are really "manufactured" as shown in the pictures, you better should not have sold a single unit, because this sure damaged confidence.
If you believe in your product you can easily improve it, as you badly missed to take care of proper production standards.

Finally I hope the dicussion comes back to objective exchange of opinions.
   

[EEVblog]

If I could explain it in a way that pleases you, I would do it. You guys like to know how everything works. In a practical sense, no one really knows exactly how the D-105 DC works. I have a good idea, but the design does not yield to clean mathematical analysis. If I ever get it figured out to your satisfaction, I'll let you know.

 

From the schematic it looks like the only magic happening is you have thermally coupled a thermistor that is part of the adjustment pin to the IC package.
Other than that it's a REF102 in a box just like any other such reference.

[EEVblog]

If you had understanding you would know it must be that way, or the device wouldn't perform to spec. Before you criticize, humble yourself. If the unit was pretty-soldered, or if it used anything but perf board, it would not perform to spec. It also would not power cycle well.

Sorry, but this is just nonsense.
There are four aspects to a reference stability like this.
Supply voltage, die temperature, thermal EMF's, and physical stress usually caused by thermal expansion of board material. And there are many way to avoid the latter.
Your poorly soldered and assembled perf board construction does nothing special to improve any of those.

[EEVblog]

I did not view the photos, but the units must be made a certain way, and the solder does more than make the connections secure. Perhaps someone here was in a rush. Or, perhaps the unit was vandalized. 

I'll make it easy for you, photos below (I hope the owner does not mind me re-posting them?). NOTE: Photos posted in Awesome's post above by accident, I'll leave then there instead of re-uploading.

You seriously expect us (electronics design professionals) to believe that this vero board construction and shoddy soldering is absolutely necessary to the performance of the unit? Really?

And lets assume for a minute it somehow is, how can you be so confident in that when by your own admission you admit you cannot explain or calculate how it works?

[TiN]

It's a trade secret, Dave, you miss a point.
Those trim pots are not even low-TC ones, few degrees aways and it's done.

P.S. I thought my references were look scary to post, seems have nothing to worry about if someone bought this for 100$.

[chickenHeadKnob]

I am interested in the binding posts, I know I have an incurable fetish. They appear to be the cheap plated brass kind one gets on ebay/deal extreme. I find it annoying that the chinese sellers list the metal composition of the parts based on the color and not any actual metallurgy. If they are brass then I would expect an additional contribution to thermal offset.

[mzzj]

chickenHeadKnob: Looks like brass. Probably not the biggest problem after seeing the internals and shady marketing bullshit 
Many (chinese) cheapo connectors work just fine with some common sense, ie let things to reach thermal equilibrium and avoid heating the connectors with your hands.
Banana binding post are  usually pretty good thermal conductors and there is very low thermal gradients over the lenght of the body.

Can't be worse than ass-whooping expensive Pomona tellurium-copper binding posts, out of 100 connectors something like 30 were mechanically crooked during manufacturing and totally useless 

[Zucca]

It´s maaagiccc...

and now stop joking... (I could not resist...)

Nothing against you Awesome14 but I suggest you to design a proper board for it. It would be a very good learning experience and on top of that you could see and touch how much you can improve your product by doing things in a proper way.

Normally board soldered like this are at prototype level, just to test if everything works as supposed (after breadboard or Spice simulation steps). It is time now to use a proper EDA/PCB CAD and try to do reach the level (after several iterations) of a final product ready for your customers. (Don´t you hear the voice inside you to go to a proper Cal Lab?Money really well invested if you believe in this product...)

OK, you probably thought the maturity level was ready for the market, but now if you want to save your project listen to those guys here. There is nothing more beautiful in learning how to do things well, and I thank God the Almighty for the EEVBlog where I can learn a lot of amazing stuff for free.

On the side here:

https://www.eevblog.com/forum/projects/project-kx-diy-calibrator-reference-sourcemeter/

you can find a lot of nice hints and suggestions, the brave TiN is trying to reach the best performance in voltage stability in a DIY enviroment.

A friendly regard,

[Jay_Diddy_B]

Hi group,

Time to step to back and start a little analysis of the circuit provided by Codeboy2K:




I can build a crude model of the circuit like this, do not worry that I used a simple, and hopelessly inadequate, BZX84C8V2 diode for the reference, it is at constant temperature in the model.


I have included a typical 10K NTC with a Beta of 3457 in the model. The temperature is being driven by the voltage source V3 representing temperature in degrees C. The thermistor model deals with absolute temperature.

I have assumed that the voltage pot is set mid-scale.



I have stepped the Tempco pot from 0 to 50K Ohms in 5K steps. The slope of the compensation changes from 0 ppm /^oC with the pot at 0, to -13.5 ppm with the pot set to maximum.




The compensation is always negative, and this is where I see the first challenge. The TI datasheet shows this:




The datasheet indicates that the tempco is defined by a box, the shape of the curve and the slope of the line is not defined.

A similar buried zener datasheet from Linear Technology, LT1021, shows that the temperature coefficient may be positive or negative:



I would expect the TI part to be similar (leap of faith required). The manufacture is trying to compensate the temperature coefficient internally, the error that remains is a measure of the inability to achieve a perfect compensation.

To compensate REF102 with a negative tempco, the NTC would have to be moved to the bottom part of the divider.

I can see that this compensation should work to improve the tempco, it does not address the more difficult issues of aging and humidity.


Regards,

Jay_Diddy_B

[codeboy2k]

Nice analysis Jay !   

The Gellar SVR-T and this design are very similar and based on your analysis I think even the Gellar design would have trouble compensating a reference IC that had a negative tempco.

[codeboy2k]

Those trim pots are not even low-TC ones, few degrees aways and it's done.

Actually, the TC of the trim pots and the resistor string doesn't matter.  It only affects the trim range of the reference, not the output value directly.

So, for example, if the TCR of the trim pots was 200ppm, and the values used were such that 1ppm of resistance change corresponded to 1ppm of voltage change, then it's only 1ppm of the trim range voltage, which is 25mV. So even 200ppm of 25mV is just 5uV change, or 0.5ppm of the reference output of 10V. 

EDIT: which is actually a good thing because it means one can use cheap trimmers and low tempco resisters here, and (2) it probably helps it survive transport without causing too much actual change in the output if the trimmer gets a small bump

[Dr. Frank]

Actually, the TC of the trim pots and the resistor string doesn't matter.  It only affects the trim range of the reference, not the output value.

So, for example, if the TCR of the trim pots was 200ppm, and the values used were such that 1ppm of resistance change corresponded to 1ppm of voltage change, then it's only 1ppm of the trim range voltage, which is 25mV. So even 200ppm of 25mV is just 5uV change, or 0.5ppm of the reference output of 10V.

Maybe not the absolute value is severely affected, but then the claimed T.C. of 0.15ppm/K cannot be met.
Following your calculation / estimation (?), it is 0.5ppm/K instead.

Theoretically, this T.C. influence can also be compensated by the NTC, although it is not in close contact to the trimming components.


Frank

[codeboy2k]

Maybe not the absolute value is severely affected, but then the claimed T.C. of 0.15ppm/K cannot be met.
Following your calculation / estimation (?), it is 0.5ppm/K instead.

Theoretically, this T.C. influence can also be compensated by the NTC, although it is not in close contact to the trimming components.


Frank

Yes, mine is an estimate (example) based on the trim pots having TCR of 200ppm and a 1:1 mapping of resistance change to voltage change in the trim divider. This is obviously not going to be true, it was just an oversimplified example to quickly illustrate how the TCR of the trimpot affects the actual output by a much smaller ratio. 

More reasonably is that there's a 5:1 ratio between the trimpot's TCR and the trim voltage.  Then a 200ppm change in the trimpot is 40ppm in the trim voltage which is 40ppm of 25uV or 1uV change due to 200ppm change in the trimpot . This is still 0.1ppm/K , so the claimed 0.15ppm/K might be achievable.  Your testing will confirm or deny it.  (where was this 0.15ppm/K figure claimed? I couldn't find it)

[Terabyte2007]

The claim that "it is set to +/-2ppm or *MY* 732A" is probably true.  Now, the calibration of the 732A is the next question.  Who calibrated the 732A, and what is the stated uncertainty [in the "scope of accreditations", from NVLAP or A2LA] for calibrations of voltage at a fixed value of 10V?  So, let's say the stated calibration uncertainty for the 732A is 10ppm, and the eBay reference is +/-2ppm of that, and now you have an absolute value of 10V +/-12ppm [to the SI volt].  How long will it stay at that setting?  How does it perform with temperature and humidity changes?  What is the time drift?  Does it have a drift with barometric pressure changes [most references do, including the 732A and 732B]?  Well, only frequent calibrations will be able to discern this.  The temperature and humidity changes can be tested in an environmental chamber.  The pressure changes can be determined in a pressure/vacuum chamber [not much pressure/vacuum is needed, so a simple hand pump can be used].  The device can be put into a box with a lot of desiccant packages, and the voltage can be monitored as the humidity approaches zero, then the desiccant can be removed and some hot damp paper towels can be added, and then monitor the output voltage for changes while the humidity goes up.  Usually, there is a delay in output change vs. humidity, ranging from 10's of minutes to weeks [depending on construction and water vapor infusion rates of the various materials].  It is a well known fact that epoxy [PC board material and IC packages] "swell" terribly with humidity, and this puts a "strain" on the die inside the package.  Pretty much any op-amp and/or voltage reference is going to respond to this strain unless the strain has been isolated [like in a hermetic package, and possibly with additional strain-relieving die mounting techniques].  The "zero drift" op-amps will self-compensate for this "die strain", but every other component type is likely susceptible to this effect [capacitors, resistors, and IC's used in the design].

Well said. I do have a pressure/vacuum/environmental chamber which I can get +/- pressure, temp, humidity, desiccant usage with temp/pressure/humidity logging. I am currently testing some long-term stability of this unit and will log all my data. Once this is complete, I can set up an experiment which can run up to 31 days and log all these parameters once a minute changing from high pressure/low humidity to low pressure/high humidity. It would be interesting to see what the logged values would be.

[codeboy2k]

...This design uses temperature compensation, which requires [to do it right] quite a bit of time in a chamber to set the TempCo.  So, this design is trading labor for better components.

I predict that these will require a *lot* of hand tuning to get them set right, then will drift like hell after that.

Wouldn't it always be better to use an oven and a good heater and feedback loop to keep the temperature at a fixed point, rather than try to compensate? If everything is at a fixed temperature, maybe only change a few tenths of a degree C, then it's temperature stable and this would eliminate all temperature effects.  You'd still have humidity and pressure to deal with, but the ceramic package can help with humidity. For pressure, maybe the the oven can be sealed at 1 ATM (or does the ceramic package take care of pressure issues too? Since it's hermetic, it might already be at sealed at 1 ATM on the die)

[dom0]

Wouldn't it always be better to use an oven and a good heater and feedback loop to keep the temperature at a fixed point, rather than try to compensate? If everything is at a fixed temperature, maybe only change a few tenths of a degree C, then it's temperature stable and this would eliminate all temperature effects.

Yes. Btw. with high end ovens we're talking about internal temperature changes measured in mK — Millikelvin.

Also probably all high end ovens are hermetic.

[VintageTekFan]

Yes. Btw. with high end ovens we're talking about internal temperature changes measured in mK — Millikelvin.

What's that in millidegrees Celsius? 

[idpromnut]

Yes. Btw. with high end ovens we're talking about internal temperature changes measured in mK — Millikelvin.

What's that in millidegrees Celsius? 

Half a bee's dick.

[quarks]

Totally off topic, but interesting
0°C is actually 273.15K
and 273.16K is a very special temperature, where water has all three aggregation states

[IanB]

So, I will email them and let them know they are wrong...

Yes they are wrong. 0°C is 273.15 K.

Also, "water boils at..." and "water freezes at..." are statements subject to all sorts of standardized conditions (absolute purity, standard atmospheric pressure, stable equilibrium). In any experiment you perform at home, water is unlikely to boil at 373.15 K.

[mzzj]

So, I will email them and let them know they are wrong...

Yes they are wrong. 0°C is 273.15 K.

Also, "water boils at..." and "water freezes at..." are statements subject to all sorts of standardized conditions (absolute purity, standard atmospheric pressure, stable equilibrium). In any experiment you perform at home, water is unlikely to boil at 273.15 K.

even if you have perfect standard conditions the boiling point of the water is not anymore 100,000 degrees like it was previously because of the latest temperature scale doesn't use it as a defining point like earlier ipts68 did. 

[IanB]

Also, "water boils at..." and "water freezes at..." are statements subject to all sorts of standardized conditions (absolute purity, standard atmospheric pressure, stable equilibrium). In any experiment you perform at home, water is unlikely to boil at 273.15 K.

even if you have perfect standard conditions the boiling point of the water is not anymore 100,000 degrees like it was previously because of the latest temperature scale doesn't use it as a defining point like earlier ipts68 did. 

Point noted. However, water is unlikely to boil at 273.15 K for the more fundamental reason that this is actually the freezing point of water...