Ultra Precision Reference LTZ1000

[martinr33]

What I am missing here, then, is why did the autocal cause an obvious shift in the reading if the meters are stable over time and temperature?

[TiN]

What I am missing here, then, is why did the autocal cause an obvious shift in the reading if the meters are stable over time and temperature?

I think reply is obvious if you consider own meter's noise from it's own LTZ and front end/ADC which is about observed 0.2ppm minimum.

[martinr33]

What I am missing here, then, is why did the autocal cause an obvious shift in the reading if the meters are stable over time and temperature?

I think reply is obvious if you consider own meter's noise from it's own LTZ and front end/ADC which is about observed 0.2ppm minimum.

On second look - that dip cannot be an autocal, as it takes way too long to get to the new level. . Autocal should show up as a discontinuity in the readings, not a slope covering multiple datapoints.

So something else made that dip happen.

[TiN]

snip.

Yup, AVG100 moving filter . Sorry about that. Together with the way D3.is plots graphs averaged line is rendered as continuous slope. You can pull CSV from my site and check yourself in RAW points.

[martinr33]

So something else made that dip happen.
[/quote]Yup, AVG100 moving filter . Sorry about that. Together with the way D3.is plots graphs averaged line is rendered as continuous slope. You can pull CSV from my site and check yourself in RAW points.
[/quote]

OK, now I found it - about a 1.3uV drop (13 counts, .13ppm) at 23/12/2017-17:52:45

Why would you do an autocal in response to temperature changes if the meter is not affected by temperature? Seems that that add unnecessary uncertainty.

[branadic]

Looking in a different way on the data the autocal effect comes more visible. However, CH6 seems to be quite uneffected by changes in temperature.

-branadic-

[TiN]

Why would you do an autocal in response to temperature changes if the meter is not affected by temperature? Seems that that add unnecessary uncertainty.

Because of two reasons:
* I had implemented ACAL code in python app before actual tempco test for DCV 10V range was done.
* ACAL is used also for resistance tests, for which I did not done full characterisation yet. I will get back to this once my Fluke SL935 standard come back.

[Andreas]

PS: Two references are just on their way to a round trip via dhl, so I'm very curious, how their measured values can be reproduced afterwards.

Hello Frank,

I never had issues with my LTZ1000A references when transporting them (hot) to calibration and back.
But I would perhaps do a "drop test" several times before I give it into mail.

The only issue I had was with a LM399 reference after transport for first calibration.
After that I mentioned that the readings with my ADCs on LM399#2 have shifted about 4 PPM (30uV).
The only explanation I could find is that I had not put any Top/Bottom marking onto the card box which I have used to keep air drafts away.
So the orientation of the LM399 within the box was not clear before and after the calibration.
And I could reproduce the shift by tilting the reference in different orientations.
So I learned it is essential to check the references before/after transportation.

LM399 also seem to be sensitive to tilting for several days when being heated. Reported here:
https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg1014749/#msg1014749

With best regards

Andreas

[Echo88]

I couldnt find a more fitting topic, so i hijack this thread to ask: Which setup do you use to measure your references on the long run, to calculate drift and compare them?

I think a good technique is to sequentially connect each zener-reference against each other in opposition, this will leave a voltage <200mV so one can use a simple DMM in 200mV-Range, compared to just measuring each zener-reference after another while using the expensive 3458A in 10V-Range. And when switching the polarity of the references during measurement one could average thermal-emf-errors which might occur.
Heres a good article about it: https://www.researchgate.net/publication/228900801_Complete_Characterization_of_a_Low_Thermal_Scanner_for_Automatic_Voltage_Measurement

But i still dont understand if a cheaper DMM in the 200mV-range would affect the drift-measurement or if the selfdrift of the DMM-selfdrift would cancel out during the long-time measurement. Maybe someone can point out how its done?

Also: I still would need a good reference with a known low drift to calculate the absolute drift of each of my LTZ1000A-references, is that correct?

Also it surprised me that in the Data Proof scanners 160/320 the wires are internally soldered to the Pomona-jacks, instead of copper spade lugs/copper-wires being used throughout the whole device like i would have expected: https://doc.xdevs.com/doc/Data_proof/img/bot_open.jpg

Since i have a Keithley 705-scanner mainframe and there are no <1µV cards available with the necessary type-c-contact-relays to copy the circuit in Figure 1c of the Article, i think about building myself a fitting card based on OptoFets (no contact life limitation). Of course i would characterize the resulting thermal emf with my Keithley 181 Nanovoltmeter.   

These questions emerged because i now have my 7x LTZ1000A-reference which i want to measure.

[Kleinstein]

If you doesn't have a high quality meter, using one reference to subtract from the other is a real option. It still relies on that one reference to be good.

I would not switch the reference polarities, as the extra switches tend to cause more trouble than good. At least the modern meters tend to be about as good as reversing the polarity, when it comes to offsets. The low voltage range will only con tribute a little to the overall voltage, so it's (gain-) drift would usually not be that important.

Soldered connections are generally not a problem - thermal EMFs only develops if there is a temperature difference across the different materials.

Optofets can be a little tricky for low volts DC, as there is some self heating. So they are not per se DC accurate. 

[Andreas]

Which setup do you use to measure your references on the long run, to calculate drift and compare them?

Hello,

for the first: I try to get as many calibrations as possible from calibrated equipment. In best case I have the calibration report so that I can narrow down the uncertainity of the calibration points after some years.

2nd: I do dayly intercomparisons of my ADCs (ADC13 is one of the most stable) and known good references.
So I can detect unusual drifts. E.g. the jump on LTZ#1 by -1.9 ppm on day ~820 (green dots) due to a short on the unbuffered output.
I use a relay scanner with latching signal relays for this. (low heat generation). After each switching I wait some time to let the thermals settle.

But i still dont understand if a cheaper DMM in the 200mV-range would affect the drift-measurement or if the selfdrift of the DMM-selfdrift would cancel out during the long-time measurement. Maybe someone can point out how its done?

Also: I still would need a good reference with a known low drift to calculate the absolute drift of each of my LTZ1000A-references, is that correct?

lets do the math: 
- 30 ppm error in 10V range are 300 uV.   
- 50 ppm error in 100mV range only 5 uV.  (assuming that the error is larger in the 100mV range against the native 10V range).
And yes you will need a well aged known good reference so that you can do differential measurements.
LTZ1000 references tend all to drift down over time.

So I use my well aged LTZ#2 as reference for the new references LTZ#3-#6 to do weekly ageing measurements.
Several ADC´s measure the absolute value and 2 DMMs in 100mV range the difference to LTZ#2.

These questions emerged because i now have my 7x LTZ1000A-reference which i want to measure.

Welcome to the club.
Before starting with ageing experiments I would fully characterize the references as good as I can.
(Tempco, variation of power supply, EMI sensitivity (changes when using battery supply, or laying the hand over cables, mobile phone ...), tilting, ....).

With best regards

Andreas

[branadic]

I do have the first data (6.5 hours) of my LTZ1000#1 (one of three references based on Andreas LTZ1048B battery powered boards with battery monitor, LTZ1000 none-A and Rhopoint 8G16A resistors with 12k:1k for temperature setpoint). Measurement will be running until tomorrow.

MEAN: 7.1490899458 V
STD: 267.55 nV

-branadic-

[Echo88]

Thanks for all the suggestions. I still need to get the Rhopoint-resistors and have smd-resistors installed at the moment to i could test the references at least after building. I still need to evaluate my scanner-setup and maybe might buy a cheap Keithley 2000, which will then monitor the references on the long run, which are compared in opposition to each other and to a known reference.
Each KX-reference will be placed in a Hammond-case, together with a LT3042-regulator, input-filter and an output-buffer. This should lead to robust references which can age undisturbed.
The tempco-test will of course also be done once i have all references assembled.
I dont want to use my 3458A for long-run-measurements since i need it for many other measurements.

So much to do...

Need to find the post about output buffers, i think it was mentioned that the internal protection diodes of the OP-amp could be activated during a short on the buffered output?

[hwj-d]

Each KX-reference will be placed in a Hammond-case, together with a LT3042-regulator, input-filter and an output-buffer.

Have a look at LT1763 too.

[Echo88]

I already have the LT3042 and fitting boards, but thanks for the suggestion. Of course the LTZ1000 doesnt need such a good regulator, but hey.

[cellularmitosis]

Need to find the post about output buffers, i think it was mentioned that the internal protection diodes of the OP-amp could be activated during a short on the buffered output?

I used Andreas’ output buffer:

https://www.eevblog.com/forum/metrology/px-reference/msg1386669/#msg1386669

https://github.com/pepaslabs/px-ref/blob/master/kicad/releases/v2.2/basic-ltz1000.pdf

[hwj-d]

Hi cellular,
osh park will send me your boards v2.2 in a couple of days. 

[MisterDiodes]

Cellular -
Just be aware of the (very low) output drive current of a '2057, these are usually never meant to be used directly as an output buffer... If you need any drive current at all you probably want a simple, cheap, low noise emitter follower / current limit circuit in your output buffer feedback loop.  See the Fluke 732b schematic for inspiration.

Otherwise that chopper noise can feed thru to your Vref output once you start drawing real current, and you'll start seeing offset errors also as the output current draw goes up past a mA or two.  This effect is -not- listed on the datasheet directly. 

It just depends on what your application, and just something to keep an eye on.  It will probably work OK as-is if you're only driving DMM inputs, but be aware of what happens if you try to drive (for instance) a higher speed ADC / DAC Vref input that needs real current flow.

[Andreas]

Just be aware of the (very low) output drive current of a '2057,

You mix that up with a LTC1050.
Short cirquit current on LTC2057 is above 30 mA (sourcing).

With best regards

Andreas

[MisterDiodes]

Andreas:
"Short circuit" current means you have zero voltage output signal - which doesn't apply here.  A signal of "Zero" is useless.  That spec is more to calculate overload protection for the amp.

You will begin to see increased chopper noise and Voffset loading errors at around 2 or 3mA out from a '2057 - LT apps recommends to keep the output current draw on those choppers as low as possible for best performance -  < 1~2mA preferred, and certainly <<<< 5mA.  The less current the better in terms of noise performance.

Look at the topology of how the demodulator section works on the output side of any chopper amp, and you'll realize why you can't pull much current and keep output noise low.

That's why chopper amps will normally have some sort of follower amp or transistor drive stage if you want to supply any real current flow.

Like I said - if you're just driving a DMM input section then a bare-naked chopper amp output can work - but if you need some current flow you normally have some sort of boost follower.

Cheapest, most reliable and lowest noise solution is usually a transistor follower - and a larger area transistor can add a very negligible amount of noise.

You'll see LT1010's in some of the app-note literature but those are pretty noisy (especially at frequencies you weren't planning on) - even within the chopper amp feedback loop you see much more noise added to the output than you probably want - depending on what you're doing, of course.

[Echo88]

So basically like the output buffer of the Datron 4910 (Fluke 732B doesnt show full schematics): https://doc.xdevs.com/doc/Datron/4910_4911/4910%20c20090120%20%5B8%5D.pdf Page2
But instead of the OP27 i would take a zero-drift-op-amp to not compromise the drift of my reference, is that correct? I just dont understand why they inserted C123 and D102.

[BU508A]

I already have the LT3042 and fitting boards, but thanks for the suggestion. Of course the LTZ1000 doesnt need such a good regulator, but hey.

This is a nice article about a low noise voltage regulator made by Peter Märki, ETH Zurich.
I've attached it to this post, but it can be found online here:
https://people.phys.ethz.ch/~pmaerki/voltage_regulator_2013_0.02/

[Kleinstein]

Even if the OP could deliver enough current it could be a good idea to limit the power consumption of a precision amplifier and thus use a separate driver stage. Heat loss from the chip can lead to chip internal temperature gradients and thus extra offsets, even with an AZ OP.

The noise of the current driver would not be such a big issue, as it would be inside the loop. It would be only at higher frequencies that the noise will coming from the driver. If higher frequency (e.g. kHz range) is an issue one might even consider a kind of compound amplifier instead of just a current driver, since there is quite some higher frequency noise from the LTC2057.

The Datron output amplifier includes some filtering. So it is not just a buffer. Here there is no need to replace the OP27 with another AZ OP. It would be more like changing to an JFET based OP to get less input bias. D102 is just for level shifting. So the output can go to higher voltage.
C123 is setting the cross over between the LTC1052 and the OP27 to a rather low frequency (around 0.5 Hz).

I would more like look at the compound amplifier in the LTC2057 data-sheet, or just use the two transistors for the current limited emitter follower.

[MisterDiodes]

So basically like the output buffer of the Datron 4910 (Fluke 732B doesnt show full schematics): https://doc.xdevs.com/doc/Datron/4910_4911/4910%20c20090120%20%5B8%5D.pdf Page2
But instead of the OP27 i would take a zero-drift-op-amp to not compromise the drift of my reference, is that correct? I just dont understand why they inserted C123 and D102.

Echo - Just for a concept example, if you get the Fluke 732b manual at Xdevs site, go to page 113 - and everything you want to know is mounted right at the business end of the output terminals. Current boost and limiter functions are provided by 2 X 2N3904's...  These transistors are buffering the output of the final low-drift op-amp, and offering current limit as well.  On the Fluke 732b current is limited to around 12mA max for safety (short circuit at the output terminals causes no harm) - which is set by the 22 Ohm R1.  As you try to draw more current, the voltage developed on R1 turns on Q1, and as Q1 starts to conduct - that in turn begins to shut down Q2.  The signal 10V High Output Stage comes from an op-amp stage and it's output resistor-isolated from the Q1 & Q2 follower / limiter.

The output terminals will see around 1ppm voltage error as you approach 12mA according to specs, but in actual use the error is typically (much) less than that.

This transistor buffer current boost + limiter design is very common for DC driver designs, and when you use large area transistors you're looking at only a couple nV/rtHz noise added (if that).  Which is generally less than most op-amps and a whole lot cheaper and more reliable.  Sometimes you'll see additional diode drops or zener are in the current limiter driver line (from op-amp to follower) to raise the base-driver operating point headroom on the output driver / limiter - this also has the effect of making the current limit cut-off point more defined with slightly less voltage error, and sometimes puts the op-amp output level in a more quiet place relative to power supply rails.  It does add a little more noise though.

There are lots of variations to this type of output buffer circuit.

If you're driving a fast ADC  or DAC typically you'll need an op-amp behind the '2057 to get some push-pull current flow at higher freqs.

[Echo88]

Thanks for the info and description.
The buffer should only act as a short circuit/overload-protector so the LTZ1000 behind it wont be harmed, since ive read that some people here had the misfortune of shorting the LTZ-output and thereby ruining the established smooth drift.