Calibratory D-105 DC Precision Voltage Reference Standard

[Pasky]

Wow, this thread went on a wild roller coaster ride.  It was great for killing time at work reading it.  So basically he modified a circuit slightly and purchased an expensive calibration unit to tune it in?  I'm curious how much a Fluke 732A and 732B cost new and 2nd hand now.  Fluke's site asks for a quote.

[lowimpedance]

the datasheet is not even the actual working item

Perhaps there is someone out there, with an original Burr Brown datasheet?
In a book, on paper? 

And as it is a plastic IC, it has attracted some moisture over the years. What will that do with stability??

Burr Brown ref102 data sheet copy (first few pages only), FYI.

[IanB]

Appending a question to this thread, since it seems to have entered more enlightened territory:

If ceramic packages for voltage references are better than plastic packages for long term stability, why have manufacturers stopped selling them? Presumably there is a price premium that would justify the more expensive option, and if someone needs that stability then price would not be a big issue. So why did they disappear from the market? Were they obsoleted by something just as good?

[paulie]

Speaking of enlightenment...

We all realize [on this thread] that this is not going to be accomplished with cheap parts,

You can't build a voltage reference with cheap components, and expect stellar results.

You talk about reference diodes [or even cheap Zeners], and a resistor, but you are forgetting that this also needs an ultra stable voltage to generate the ultra stable current for the Zener.

So, having a discussion here about "cheap voltage references" is probably a little bit too far off-topic.

Perhaps you can start a "super cheap voltage reference" thread, as this is starting to pollute the quality of the LTZ thread.

Adding a thermistor and cheap voltage trim components [i.e. carbon-film resistors, and cheap pots] does not [magically or otherwise] improve the specifications of *ANY* monolithic voltage reference.

If you use cheap components to do this, then expect even a less pleasant outcome.  Also, not all voltage reference IC's are the same.

I was writing about some others that sell on the 'bay', that are using the cheapest junk they can buy,

Oh no.... Just noticed Dilligent seems to have picked up a new word of the day. I wonder where that came from? Certainly not me for whom cost is no object and only the best will do.  LOL

(lots more but didn't want my session to time out.)

[paulie]

So, I think it was the "suits" [lacking technical knowledge] that forced this change on us to save money.  That's right, greedy "bean counters" did it...

Cheap bastards!

[babysitter]

Does anyone remember the radioactive epoxy disaster? What would have this done to floating-gate references like FGA? "Precision reference with enclosure-controlled precision decay" ?

[JohnnyBerg]

but since tempco has so much influence on accuracy, drift etc .... why not just LM399 it all and "end of story"?

Price? Battery powered?

Building a "good" reference starts with a tradeoff between TC stability and price. Accuracy is less important, as it can be trimmed.

After that, you have to do everything right (board layout, decoupling etc) to get the most out of it. I think it is very hard to beat the specs in the datasheet. And if you think you can beat them, it is almost impossible to prove that. It takes a long time and a lot of data
And if you think you have beaten the datasheet, is it possible to reproduce those result in a production type? ;P

[JohnnyBerg]

**exploratory simulation into making a ref of some kind

And what resistors are you planning to use?

"Normal" 0.1% resistors have a TC of ~20ppm/°C (or worse) .. effectively killing the LM399 ..

Edit: and how are you going to trim your reference? Or are you satisfied with 2.3% accuracy? (3x 0.1% of the resistors + 2% of the LM399)

[JohnnyBerg]

panasonic 10ppm perhaps? this 1 sits @ 0.01% 

Still killing the LM399 
With 10ppm/°C you can use a $3,- reference like the LT1021 ..

actually after hearing so much about the "wildness" of variable trimmers ... wouldnt it be better to be without it and just have the output as it is ? or have a very small value trim (like blackdogs version with a 100R?)

It depends where you put the trimmer
If you put it inside a direct feedback loop, you have to put it in the oven. This is what Blackdog does.

I use a very large resistor in before the trimmer. I can only trim a few millivolts, but for me that is not a problem, as my reference is accurate enough. This trick won't work with a LM399, as it has a 2% accuracy, and you need to trim over a large range.

putting together the resistor framework is harder then it seems just by depending on what is available from E14 ... phew !

Welcome to the tough world of voltage standards 

Edit: serious: forget the LM399 when building your first refence. Why not try a REF5050A or even REF5050I? You can get 2ppm/°C, if you do your best 

[ebclr]

Take a look on this one

http://cs.utsource.net/goods_files/pdf/16/168867_ZARLINK_ZN458.pdf

Simple and good

[quarks]

It is disappointing that the SVR program attracted such strong reaction. 

Hello Joe,

it is very sad to read that you feel disappointed, because I think you earn the highest appreciation and should have payed tribute to you.
You can and you should be very proud of your SVR program.

As I stated already several times, I am a very happy SVR/SVR-T owner.
For me it is a great loss that you had to shut down. 

So many, many thanks for all of your effort to make this very useful and valuable product.

bye
quarks

[JohnnyBerg]

but if i dont really trim it, the 2% will not really matter am i right? say if i find a way to also use a smaller trimmer ? by buffering the trimmers "largeness" with fixed 5ppm stuff?

Whats the point in building a reference with 5ppm/°C stuff and a much more precise reference IC?

Why not use a REF5050I (or anything equal), that has 0.05% initial accuracy and 2.5ppm/°C typical (3ppm/°C max.) and uses no presision resistors and can be trimmed without fancy stuff?

[JohnnyBerg]

Take a look on this one

http://cs.utsource.net/goods_files/pdf/16/168867_ZARLINK_ZN458.pdf

Simple and good

Youre kidding?



ehh ..  0.003%/°C is how much ppm/°C ?

[blackdog]

Hi 3roomlab, :-)

My opinion is that you really can not make a cheap voltage reference, with reasonable specifications.
Everyone should stop with the word "cheap" and voltage reference in a sentence.
You are just kidding yourself...

I do a lot of testing on references but every time, the way you built the reference, is really very important.
Its a big part of the stability!!! and remember not to build the reference in a plastic box, METAL is wat you want.

I can create beautiful schematics but if the quality of the components and construction is not good, then it will never be a good reference.
It has already been mentioned many times on this forum, that testing of references is expensive.

If i send away my HP 3458A for Cal/Adjustment, its about 900$, for almost that price i can buy a KEYSIGHT 34461a :-)
And my 3458A is NOT 24/7 on, thats also expensive, i switch on the 3458A 24 hour before i start measuring...
Most of my daily testing is done, with TEK4050 en 34461 DMM's.

Last week i tested some LM399 schematics, the sensitivity for battery voltage change and the power that's needed for the LM399 heater.
My starting point was three 9V batteries in series, to keep the current for LM399 for the heater low.
The circuit does not start, because when you turn the circuit on, the heater current is really high.
If your batteries are getting older, the Ri is also higher, and the batteries cant deliver the peakpower that is necessary to start the heater.

Problem, problems, problems,

And now some pictures, poluting this topic 
These are the components i use, for the testing i do, with The LM399AH references.
The resitors are Rhopoint 8E16D, Copper band 0.1mm and Vishay 15PPM trimmer pots (1 piece 2x price of a LM399AH!)


7 Resistors, the 4K leg and 10K leg, maybe i put some 2 component glue in it to seal it a little.
The copper band keeps the temperatuur of the resistor nice close together, but i have to test it first, to see if its realy works!



Vishay 1280G trimmer


I'am going back to the LAB bench, for testing of the LT1236-5V LS8.
I do temp sweeps between 42,5C en 29C, most of them, are well within the specifications.

Kind regarts,
Blackdog

[JohnnyBerg]

i will need to explore what could i do to make something that can sit on the 4th or even 5th decimal point, as a project, as curiosity

The 5th decimal point, were talking 10ppm/count. Nothing fancy needed to realize this.

If you do your best, you'll manage even with a LM4040. And you will learn the most about stabilty when trying with a not so good device

[paulie]

for me, anything on or past the 3rd decimal point is exploratory for me.

Save your breath. Words like 5 digit or 3 decimal are not in the vocabulary for some of these guys. It should be obvious from some posts that these concepts and "cheap" are like garlic to a vampire (vampires are real you know...LOL). However we can learn a lot from guys like blackdog and dilligentminds because if you shoot for mars maybe you can reach the moon. For example note that virtually all the suggestions above are bandgap references so even ignoring the heater they are no competition for the likes of zener based LM399 for noise and long term stabilty. I heard it from them first then verified with my own tests so these guys do know what they are talking about.

As far as resistors I inherited several thousand 10ppm Vishay SMD and purchased some 25ppn to compare. I think they are actually the same parts because not only do they look the same (even if you grind them down which I did) but all tested better than 2ppm over 10deg range in my oven. So maybe not as good as wirewound or $40 foil but fine for the kind of references we are interested in. Specially considering the minimal effect on reference circuits as hinted by dilligent and proven by tests from janaf in the other thread.

ps. I have discovered that advice from guys even 10x smarter and more experienced  is not always correct for what I am trying to do. But with this in mind I do learn a lot from them. I'm also glad to see Awsome14 has come to similar conclusion and refuses to cave. Us  fraudulent schizophrenic nincompoops must stick together.

[Terabyte2007]

For what it's worth and for those who are interested, here are some results from my tests lately with the D105. My lab has been consistently at around 21.5C - 23C with humidity around 27-32% (Dry). I used my Agilent 34461A with an aperture of 100PLC and allowed a 1 hour warm up time before taking measurements. When taking measurements from the D105 I used my highest quality low EMF beryllium copper test leads with my Rigol DP800A PSU set to 14.500V. Warm up time on PSU was 1 hour also just to be consistent.

Each time the tests were conducted on my main test bench with the D105 connected between the two devices. The D105 was allowed to sit flat and undisturbed during these tests. Each of the tests were allowed to run for 1 hour each with periodic monitoring of the recorded output DC value.

Each test had roughly 1,060 samples taken over this 1 hour period. Each sampled time period was done at different times of the day such as morning, afternoon and evening. The reason for 1-hour sample sessions at multiple intervals was to simulate bench use as this device would most likely be used, plus I could not allow tie up that much time with equipment and personal time.

However, there are considerations to be taken into account with these tests. They are as follows:

1) Although the Agilent 34461A is within calibration specification, the unit is rated at 35ppm for a 1 year DCV accuracy. This must be taken into account. Also the desirable TAR rating is not quite adequate which should be at least 4 or higher so taking into consideration the claimed D105 specifications at +/-2ppm that would place the TAR somewhere around .0571428571 but allowing for a more realistic +/-10ppm it would be .2857142857. This is obviously far from the desired TAR rating which I cannot achieve with my current gear so there will be an error factor involved and this should be considered a general set of testing.

2) Humidity, I currently cannot test a wide range of humid environments which may alter the tests significantly. I will have to provide updated tests in the future when time is allowed.

3) Outside interference, As the current round of tests seem to indicate, the D105 is a bit susceptible to RF interference. I allowed my cell phone to sit too close to the D105 (About 3 feet) and a call came through which caused a slight spike in the measured value. It was only a few microvolts but enough to skew the test slightly. This may have also been introduced into the cables also.

4) Cables and connections, this may vary slightly depending on the type of cables used and there EMF rating. I know we are splitting ball hairs here, but all this does add up when we are talking about microvolts.

I am going to post 1 set of tests for now which is the latest test. All sets of tests have been very close to this one. If I have time, I will compile the rest and upload them.

Here is the data for the current test which is almost a month later after the purchase of the D105.

Seller claimed accuracy: 10V +/-2ppm

1hr warm up for all devices.

Latest Test Data: 02/25/2015
Start of time of test: 08:44:50
End of time of test: 09:44:53
Sample Count: 1,060
NPLC: 100
Input Impedance: 10M

Min Value: 9.9999030 VDC
Max Value: 9.9999159 VDC
Avg Value: 9.9999074 VDC
Delta Value: 00000220 nVDC
StdDev Value: 00000145 nVDC

Taking into account all the considerations this test looks to be in the range of 2-3ppm. Again all the considerations and some I probably forgot to include must be added to the equation. Based on what I can test and the given results, the unit is performing well. Even if we add a 20% error factor we are still in a comfort zone for most people using this on the bench for a low-cost DC reference. Again, I have not yet tested the unit under various climate conditions and changes. My lab has been pretty consistent at the listed Temp/Humidity.

Also, keep in mind this test is based on the average 9.9999074 VDC and not nulled to 10VDC. I am merely looking at stability at this point. Based on the seller's certificate of calibration which was 10.0000057 VDC with a 5.7 uVDC/.6ppm error, the accuracy figures would change a bit, but we would need to take into account the differences in calibrations and accuracy between his unit and mine.

Please Note: These tests have been completed with a TAR below the desired value. Even though the completed tests look promising, there are factors and deviation errors that may be present that I cannot take into account. Based on the current technology of a high quality 6.5 digit DMM, the results are not too bad. Aside from the internal construction of the unit being a bit dodgy, the unit is performing admirably, at least in the current series of tests I have conducted over the last month.

[JohnnyBerg]

when you say artificial aging, am i right to say meaning baking over higher than normal temp 100-150^oC? and cooling also? or no cooling? for say 1 week? then possibly this process is done on a ... bunch, and select a few good ones with least fluctuations.

No, artificial aging for references is done, purely by connecting them to a power source. No temperature involved.

[edavid]

Does anyone remember the radioactive epoxy disaster? What would have this done to floating-gate references like FGA? "Precision reference with enclosure-controlled precision decay" ?

No, because there was no such thing.  It was ceramic packages that were found to emit alphas.  I think the radiation level was too low
to affect floating gate references.  Anyway, once it was discovered, it was trivial to screen the clay for low radiation.

[blackdog]

Hi Group!


Just for the reference...
I have also done a one hour measurement on a 10V voltage reference.
To see the difference is the noise behavior of the D-105 and a Quad LT1021-5 and a lowpass 2x amplifier of my design.

Small picture, click on the link for the high res. picture


Link High Res Screenshot.
www.bramcam.nl/Diversen/Quad-LT1021-1H-10V-a.png

Al the setting where the same as Terabyte2007 used.

Kind regarts,
Blackdog

[codeboy2k]

...I have also done a one hour measurement on a 10V voltage reference.
To see the difference is the noise behavior of the D-105 and a Quad LT1021-5 and a lowpass 2x amplifier of my design.
...
Kind regarts,
Blackdog

HI Blackdog.... your reference is definitely quieter, more stable and seems to have far less excursions. I see about 5 uV (0.5 ppm) p-p noise, and the average over your 1H period is within 0.350 ppm of the min and max.

The graph of the D-105 over 1 hour shows 100 uV p-p noise, that's 10 ppm of noise right there.  The average voltage over the hour is within 1 ppm however (about 0.850 ppm)

This is the reference from this thread, right?
https://www.eevblog.com/forum/projects/10v-reference-(i-did-it-my-way)-no1/

Is this reference heated, as you said "housed in a 42C heated BimBox" ?

[blackdog]

Hi codeboy2k,

That is correct, but the schematic is not optimal, i wil change it, i will put a LTC2057 omamp in it.
And there is a stablility problem with the oven electronics.
No big problem, maybe 2 hours of work.

Kind regarts,
Blackdog

[Terabyte2007]

Hi codeboy2k,

That is correct, but the schematic is not optimal, i wil change it, i will put a LTC2057 omamp in it.
And there is a stablility problem with the oven electronics.
No big problem, maybe 2 hours of work.

Kind regarts,
Blackdog

Yes, definitely quieter. Do you sell these or is this just your own project.

[AlfBaz]

Sorry for the slightly off-topic question here but looking at terabyte and blackdog's trend pictures I'm a little confused.
It would appear you would need at least 8.5 digits to get this sort of resolution yet the 34461A's are touted as 6.5 digits
The other confusing aspect is (assuming at least 100NPLC) is the number of samples per second. I see in the specs it says at 100nplc it does 0.6/(0.5) readings/s

I'm currently looking through these meter's documentation in am attempt to understand how they're doing it but so far no good. Any links?

[LaurentR]

Sorry for the slightly off-topic question here but looking at terabyte and blackdog's trend pictures I'm a little confused.
It would appear you would need at least 8.5 digits to get this sort of resolution yet the 34461A's are touted as 6.5 digits
The other confusing aspect is (assuming at least 100NPLC) is the number of samples per second. I see in the specs it says at 100nplc it does 0.6/(0.5) readings/s

I'm currently looking through these meter's documentation in am attempt to understand how they're doing it but so far no good. Any links?

On the picture, 06:00 is 6 minutes.

As far as the number of digits, the 34461A returns the full floating point number for the measurement over SCPI (i.e. ADC + calibration function), so it has plenty of digits and there is quite a bit more effective resolution than the 6.5 digits on the display.
IIRC, the 34401A returns what is on the display, so no gravy.