Ultra Precision Reference LTZ1000

[TiN]

But experiment results and execution often vary. Hence the discussion.

I still have difficulties measuring TC in sub-ppm zone. I had shown TC test with ramp +24C to +44C with stock HP 3458A A9 PCBA, which did not allow to calculate any TC. Perhaps method should be changed to test two different temperatures for very long time, instead of ramping temperature (speed was +1K each 2 hours) ?

There are no extra heat sources on LTZ module (except 5mA LED perhaps?). Also I did not see any measurable supply voltage dependency on output when tested before. Output was same either with 10V input or 16V. I usually use 15V or 12V from battery.

[Andreas]

I still have difficulties measuring TC in sub-ppm zone. I had shown TC test with ramp +24C to +44C with stock HP 3458A A9 PCBA, which did not allow to calculate any TC. Perhaps method should be changed to test two different temperatures for very long time, instead of ramping temperature (speed was +1K each 2 hours) ?

Hello Illya,

you will have to find out where your measurement problem is.
I have a typical repeatability/noise of around 1uV for a 7V reference.
So over 20 deg C you should be able to detect at least 0.05ppm/K.

I measure either with (good warmed up, my K2000 needs longer than the 34401A) 6.5 digit instruments in 100mV range
as difference measurement against my "best" 7V reference,
or with one of my 24Bit temperature compensated ADCs (if the room temperature gradient is not too high).
https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg907048/#msg907048

I average all measurement values within 1 minute. (This is possible since I have checked for a gaussian distribution).
And for the ADC I sometimes have to build a sliding average over 20 minutes.
I use ramp speeds typical of 0.1 deg C / minute. (not below 0.05 deg C/minute and not above 0.3 deg C/minute).
I measure ramp up and ramp down to average out temperature measurement delays.
Without a closed loop I would not be able to detect drift/hysteresis of my measurement setup.
And without continous ramping I would not be able to detect non linear response of the LTZ.
(e.g. large drift when the temperature regulation loop is out of control).

So I think its not a good idea to do only two temperature points.
Of course a constant temperature might be the starting point to look how to improve the stability of the setup.

I use also battery supply for my references and for my ADCs to avoid ground loops and mains noise.
For critical measurements I try to use a grounded metal plate where I put all my devices and wiring on.
(Where is the side wall of my desktop?).

With best regards

Andreas

[TiN]

Thank you for settings, that's helpful. I think part of problem that I was measuring direct reading, not the delta like you. Even though I use 3458, it's own TC much larger than source, and my environment is very far form desired +-0.1K and still air conditions... I'll have to build thermal chamber for 3458A to solve this...

Last night results with LTZ chip 8 (dark-blue) vs HP A9 PCBA (cyan, TEC cooled +23.40°C) and my old LTZ module (orange) : https://xdevs.com/dcv_kltz8/
That's with 1 minute average (10 samples). Raw samples are displayed by green dots.

[TiN]

There is 3458, but following this idea Keithley 182 or Keithley 2002 + EM A10 preamp on 20uV range will work much better.
Currently I have least 2 pairs of LTZ modules which are within <10uV from each other. Thanks for idea, I'll try this.
I have suitable relays for this job.

[TiN]

And many many boxes of thermal and air insulation around everything.
Ok, added one more item into lloooong ToDo list. I have few more urgent projects to finish next weeks, but after that I'll be fully equipped to start on this challenge.

[Dr. Frank]

Ken and TiN,

sorry to disturb, but I think, nV- measurements are over-engineered, and not really solving the problem.

I agree, to measure the difference between an environmental- temperature - stabilized LTZ1000 and a environmental -temperature-changing DUT, if the 3458A itself cannot be used in a constant environment.

Anyhow, if you want to determine a 0.01ppm/K dependency, as worst case, that would be about 71.5nV/K  for a typical 7.15V output.
Measuring, let's say over a 10°C range, that would be about 715nV.
Both orders of magnitude can be well measured with a 3458A, and using statistics would give a reasonable T.C. = dU/dT calculation.

The problem at that point is of course the extreme sensitivity of such a measurement, which a nV meter would probably not improve in principle.

I mean, what do you really measure, the T.C. of the LTZ1000 circuitry, or any thermocouple (asymmetry) inside the measurement loop?

Even if you reverse the output of the LTZ1000, you will still have thermocouples left between the relays and the output directly at the LTZ1000 pins, which you cannot distinguish from the LTZ-circuitry -T.C.

Therefore, this LTZ-T.C. measurement, and all this trimming to lowest possible T.C.s (below 0.1ppm/K level) is mostly esoteric... or do I miss something ?

Frank

[TiN]

Dr.Frank

For me it's not so much for nV-measurements themselves, but rather to change method of measuring TC to see if results correlate.
As with just direct measurement LTZ's output over environmentally unstable 3458A does not provide any clear result, leading to erroneous results or inability to see relation between LTZREF temperature shifts and output voltage variations. I don't feel comfortable just applying average(256) filter and analyzing likely meaningless data.

Here just today's example with raw data.
* HP 3458A A9 PCBA in TEC-thermostat controlled at +23.4°C with +/-0.04°C, from 10pm to 11pm.
* Cyan line, moving average 10 samples at NPLC100
* Variation Vout = 1.4uV(pk-pk) which is double of your theoretical 715nV over 10K.

Or I interpret result incorrectly, and we dealing here with furry carpet stairs?
Also old initial results on all three modules show much higher TC (0.5 to 1.5ppm/K), not talking about <0.1ppm/K TC's.
So still need clear answer for myself if it was design problem in module/components or measurement error.

Alternative to that is building thermal chamber for DMM to stabilize its conditions, which technically is easier, but would occupy lot of space in the homelab.

[Andreas]

Therefore, this LTZ-T.C. measurement, and all this trimming to lowest possible T.C.s (below 0.1ppm/K level) is mostly esoteric... or do I miss something ?

Hello Frank,

Perhaps its not so essential if one has a temperature stable home lab in the basement.
But with my lab under the roof I have typical 6 deg C (24-30 deg C) variation over the day in summer.
Extreme values may be up to 10 deg C during one measurement.

So my references have to be a factor 10 better than someone elses with only 1 deg variation.

Besides that I had more problems with EMI-noise (from PC + USB-cables) than with thermocouple voltages.
If I can keep air drafts away from the setup by cloth and use thermally coupled connectors (D-SUB) the later should be avoided.
The EMI can be detected by putting the hand on/over the power supply batteries or the wiring.
I had up to 38uV output voltage difference (depending on sample) between unbuffered and buffered LTZ output before soldering the 100nF capacitor across the output.
(capacitive load isolation cirquit is necessary to avoid oscillation of the buffer).
After the change the difference between unbuffered and buffered output (LTC2057 offset) were near my noise limit (1-2uV).

https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg846835/#msg846835

With best regards

Andreas

[Dr. Frank]

That *is* why they all us "Volt Nuts", no?

What, ME telling?  O0 O0 O0

I really think, at this point, there is a natural, physical limit, where it's not useful anymore, trying to measure something, which is buried too deeply in other effects, like e.m.f.s, zener and circuit noise, etc. If you were able to have a "chopper" principle available, (like Lock-In), to cancel all e.m.f.s, and other drifts, then you would be done... but that's not the case here, as  these effects are inside the circuitry.

Also, a nV meter, i.e.  < 10nV noise, would dig into the 0.001ppm/K region, which makes absolutely no sense anymore.. simply look at this order of magnitude..

That has also nothing to do, that we as "amateurs" would be limited in means, compared to standards labs. I think that we are pretty close to them.
And they also are limited by the rules of physics. I really doubt, that they would give much better StD or T.C. figures for that circuitry.



I like to take the chance to present my latest measurements on a newly assembled LTZ1000 circuit, PCB and schematic by Andreas (many thanks again!!).
It's running on 45°C, the PWW resistors are selected, to give a theoretical T.C. of about +0.045ppm/K, 430k resistor for "T.C. compensation" assembled.
Measurement is done in constant environment, max. 0.8°C RT change, absolute voltage measured at NPLC 50 with the 3458A, which also was stable to 0.2 / 0.5°C, internally.

The first diagram shows the datasheet assembly, i.e. only these 3 cap's, and the LTZ in a socket. No electrical or thermal shielding, besides a styrofoam cup on top of the LTZ.
1h-noise is about 750nV, or about 0.1ppm. Not the best performer of my five new LTZs. 



Second diagram shows the same LTZ, soldered fix into the PCB w/o cutting its legs. I avoided to heat the chip during soldering.
The circuit is equipped with all capacitors, as defined by Andreas, except C14, C15 (fear capacitors), and C13, which I think would spoil the ovens feedback constant of C1, R7. The metal shield is also in place, but no thermal enclosure on PCB bottom yet.




1h noise has gone down to about 270nV, or 0.04ppm. These 0.5ppm dips, and glitches were gone, at least these are much, much smaller.
So, all these capacitors, which Andreas has introduced into his circuit really improve the performance.

What's left is a random medium term noise, or drift, of about 0.05 .. 0.1 ppm, see trend line (200 points average).
That may further be improved by completing the thermal shield, and putting the whole assembly into the aluminium box.

The 16h stability, that is the combination of the 3458A and the LTZ1000 circuit, is on the order of about +/- 0.1ppm.

First conclusion, this is a quite stable measurement, mainly due to stable temperature.. it really can be done that way..

Second conclusion, despite the stable output voltage, you still see a lot of medium term noise, or variation, on the order of several 100nV.
Maybe this is caused by thermal draught, but probably also by zener instability, i.e. random walk, popcorn noise, or whatever you would like to classify this.
And I really cannot imagine, how to  principally improve this, so to be able to extract changes on the order of 0.01ppm, caused by temperature variation.

Third conclusion, although there is RT change, and 3458A temperature variations, there is no correlation visible between T and U. In other measurements also, I never have seen a correlation.

Illya wonders, why he also does not see a correlation in his measurements.

Maybe, that's because there is no effect, or the effect is buried under other effects.

Well, of course, I will also try to measure the T.C., when my box will be finished.
But I will definitely stop at this 0.05 ppm/K (~100nV) perception frontier..


Andreas, I really think that 0.05ppm/K is realistic, and that all your LTZ circuits are already having that T.C. or better, by design. No need to worry further. 
At a 10°C change, that would be below a 1ppm output change, which would be extraordinarily fine.

I agree, that shielding and EMI is a superior problem over the T.C. .. I also see changes on the order of ppm, depending on the configuration of the shield (outer case to ground, or using the inner shield as a guard).

Frank

[TiN]

Andreas,

Is the schematics posted on page 58 latest? I'd like to try few capacitors you used on my module to see how it affects results.
If you have updated one, please post in PDF format for reference. Thanks.

[Andreas]

Hello Illya,

If you mean this one:
https://www.eevblog.com/forum/projects/ultra-precision-reference-ltz1000/?action=dlattach;attach=189655

yes it is the latest schematic of my current PCB. Only the 100 nF capacitor at the output of the buffer is missing.
And I do not populate C16-C19. (no filtering against the inner shield).
Edit: C14,C15 cannot be populated together with fast OP-Amps like LTC2057. C13 is only necessary if C14 is populated (see Frank).

If you have the datasheet cirquit not all capacitors are possible without cirquit modification.
E.g. C9 will require R19 and C8 to prevent oscillation.
R21, R22 and C22 prevent the output buffer from oscillation with capacitive loads.

So from datasheet cirquit I would start with C11 and C12.
And if possible add C8,C9 and R19.
See also revision history.

With best regards

Andreas

[TiN]

I've added C8,C9,C12 (film SMD caps) and R19 (10K Fluke wirewound). Had also to R13 (50K Fluke WW) or circuit would not start.
Output shifted +0.96ppm. Set to datalog, will see how it goes in some hours. Wiring/settings not changed.



Live data

[Andreas]

Hello Illya,

I am missing C11.
For R19 and R13 simple SMD or metal film resistors are sufficient.
A PWW resistor for R19 might pick up more magnetic noise.

C11 and C12 are directly at the LTZ pins.
See also my very first PCB LTZ1000A where I added some components manually.
C11 + C12 at the LTZ circle.
R19 as metal film below the LTZ.

By the way.
I have difficulties to interpret your diagrams.
For me there is too much information on them.
I would prefer a diagram with not more than 3 lines on it.
(1 LTZ unfiltered+filtered + 1 temperature sensor).

With best regards

Andreas

[zlymex]

In my opinion, the measurement of T.C. of a voltage reference should be done by comparison instead of direct measurement. This is mainly because the T.C. and hourly stability of an 8.5 digit meter is usually not as good as the reference to be measured. Some multimeters such as 3458A, they have very good noise and short term(<=10 min) specifications but not very good hourly or daily, the period often required when measure T.C. of voltage references.

The reference should be better(ideally much better) than the voltage to be compared, should be noiseless, good short-term stable and most of all low T.C. If the low T.C.  cannot be guaranteed, use constant temperature chamber, temperature lag, or temperature correction/compensation, or some/all.

By means of comparison, back-to-back(as mentioned above by DiligentMinds.com) is a very simple and effective. However, it only works for two references of equal nominal voltage, we cannot back-to-back an 7V with an 10V for instance. A switch(preferably an automatic one) can be used here together with a low-noise, high linearity meter such as 3458A to compare references of different voltages such as 7V and 10V. The switch can be made of an latching relay with 2-Z contacts(or even 1-Z). Automatic switches have the advantage of multi-readings and averaged later to give more confident result.

I don't think the thermal EMF matters very much, because the reference now days are very large in value(>=7V). Plus, constant thermal EMF has no effect on T.C. measurement result.

Below is a comparison measurement of mine(with my 4910-AV) for 16 voltage references for 30 hours, purple line is the ambient temperature in degree C.


All the curves are very thin indicating a low noise system.
Many curves are straight(not vary with temperature much) indicating low T.C.
Some curves show clear variation with temperature, and can be further processed to obtain the actual T.C.
The curve marked with grn4 is the worst, indicating there are problems with it( https://www.eevblog.com/forum/metrology/repair-datronwavetek-4910-voltage-standard/msg885426/#msg885426 ) 
Andreas probably excuse me for violating his saying of 'not more than 3 lines' because my lines are much more straight.

[Dr. Frank]

In my opinion, the measurement of T.C. of a voltage reference should be done by comparison instead of direct measurement. This is mainly because the T.C. and hourly stability of an 8.5 digit meter is usually not as good as the reference to be measured. Some multimeters such as 3458A, they have very good noise and short term(<=10 min) specifications but not very good hourly or daily, the period often required when measure T.C. of voltage references.

The reference should be better(ideally much better) than the voltage to be compared, should be noiseless, good short-term stable and most of all low T.C. If the low T.C.  cannot be guaranteed, use constant temperature chamber, temperature lag, or temperature correction/compensation, or some/all.

By means of comparison, back-to-back(as mentioned above by DiligentMinds.com) is a very simple and effective. However, it only works for two references of equal nominal voltage, we cannot back-to-back an 7V with an 10V for instance. A switch(preferably an automatic one) can be used here together with a low-noise, high linearity meter such as 3458A to compare references of different voltages such as 7V and 10V. The switch can be made of an latching relay with 2-Z contacts(or even 1-Z). Automatic switches have the advantage of multi-readings and averaged later to give more confident result.

I don't think the thermal EMF matters very much, because the reference now days are very large in value(>=7V). Plus, constant thermal EMF has no effect on T.C. measurement result.


All the curves are very thin indicating a low noise system.
Many curves are straight(not vary with temperature much) indicating low T.C.
Some curves show clear variation with temperature, and can be further processed to obtain the actual T.C.
The curve marked with grn4 is the worst, indicating there are problems with it( https://www.eevblog.com/forum/metrology/repair-datronwavetek-4910-voltage-standard/msg885426/#msg885426 ) 
Andreas probably excuse me for violating his saying of 'not more than 3 lines' because my lines are much more straight.

Hello, Zlymex,

principally you are right about the 3458A stability, on the other side, its LTZ reference is as good as the external LTZs, we want to measure.

The back-to-back method is better only, if the backing reference is better or equally stable as the DUT, otherwise, the absolute method is fully equivalent, as far as the A/D and amplifier stage are stable, too..
This is the case also for the 3458A in the 10V range.


In the volt-nuts thread, somebody measured several solder junctions; all had about 3µV/K, which is 0.5ppm/K related to 7 V.
Therefore, this is relevant if you resolve fluctuations to 0.1ppm or smaller.


Your diagram is 15ppm wide, in contrast to mine, which in comparison is zoomed 15  times, ie. 1ppm wide, resolving 0.01ppm or better.
So yours only looks better at the moment..

Would you mind providing a similar resolution, so to better see, how stable these 732B references are, in comparison to LTZs, and in comparison between differential vs. absolute measurement?

Thank you

Frank

[zlymex]

......Would you mind providing a similar resolution, so to better see, how stable these 732B references are, in comparison to LTZs, and in comparison between differential vs.

Sure. But now it's getting very late here, I'll do it tomorrow.

..soldered junctions; all had about 3µV/K, .....

This per K is for temperature difference between two thermal junctions.
When we change the temperature of a voltage reference for 10 degree K(say) for testing T.C. we don't get 10 degree K difference for thermal EMF junctions. Usually very much less if properly handled.

[Dr. Frank]

This per K is for temperature difference between two thermal junctions.
When we change the temperature of a voltage reference for 10 degree K(say) for testing T.C. we don't get 10 degree K difference for thermal EMF junctions. Usually very much less if properly handled.

.. provided if you balance all junctions in pairs.. if you have an unpaired junction, that might be different, isn't it?

But you're right, practically, such effects mostly cancel out.

Frank

[TiN]

Andreas, Dr.Frank
Sorry, I just got used to overdo my logs with all data possible and often include multiple instruments logs, as I have only one GPIB dongle to run experiments.
I did excel graph with three datasets over same sample time.

This is RAW samples, no average at all. NPLC100, AZERO ON. Only math is in excel to remove offset, so graphs match same point.

.

Based on this, I see no difference in result, and all three measurements resemble Dr.Frank's second graph closely, both in span <0.2ppm and spike behavior.

Baseline - is my original circuit with PWW's, which I shown on previous page with plenty photos. TEMP? was lower, room was airconed.
Trimmed - same stuff, but cut legs so LTZ chip sits within 1mm of socket. TEMP? around +44, no aircon
Andreas - same as trimmed, but with caps/resistors mod. TEMP? around same, no aircon

[Kleinstein]

Solder junctions are only a problem if there are temperature differences inside the solder junction itself.  If you want to compensate, it is important to have the same thermal conditions on both junctions. If the thermal conditions can't be matched well it is likely better not to add an extra junction, but better make sure the temperature gradient at the junction is small. 

With the joints at the LTZ itself there is essentially no way to compensate. With the typical LTZ1000 circuit there are only the connectors, where you have extra connections. Here the junctions are usually balanced / symmetric anyway. But it is still important to keep temperature gradients small in these areas and get the gradient of the thermal chamber in the pure, preferably unbend wire.

[Dr. Frank]

Andreas, Dr.Frank
Sorry, I just got used to overdo my logs with all data possible and often include multiple instruments logs, as I have only one GPIB dongle to run experiments.
I did excel graph with three datasets over same sample time.

This is RAW samples, no average at all. NPLC100, AZERO ON. Only math is in excel to remove offset, so graphs match same point.

.

Based on this, I see no difference in result, and all three measurements resemble Dr.Frank's second graph closely, both in span <0.2ppm and spike behavior.

Baseline - is my original circuit with PWW's, which I shown on previous page with plenty photos. TEMP? was lower, room was airconed.
Trimmed - same stuff, but cut legs so LTZ chip sits within 1mm of socket. TEMP? around +44, no aircon
Andreas - same as trimmed, but with caps/resistors mod. TEMP? around same, no aircon

Yes, seems to be similar. In the end, they have to be..
Though, the indicators are incorrect, the lines are 0.1 ppm apart, so the vertical arrows are 0.2ppm and 2ppm.

Frank

[zlymex]

.......
Your diagram is 15ppm wide, in contrast to mine, which in comparison is zoomed 15  times, ie. 1ppm wide, resolving 0.01ppm or better.
So yours only looks better at the moment..

Would you mind providing a similar resolution, so to better see, how stable these 732B references are, in comparison to LTZs, and in comparison between differential vs. absolute measurement?

Thank you

Frank

Here is the same chart but 1ppm wide, 0.1ppm per line for voltage, for 18 hours. I measure these references by 3458A, NPLC=50, AZ=on like always.
The green line is raw 732B data except offset.
The orange line is 732B compared to 4910-AV, eliminated most of the influence from 3458A, but of course added influence from 4910AV.
The other two lines are channel 1&2 of 4910 compared to 4910-AV.
As can be seen from these 4 lines that comparison is better than measurement.

[Dr. Frank]

Here is the same chart but 1ppm wide, 0.1ppm per line for voltage, for 18 hours. I measure these references by 3458A, NPLC=50, AZ=on like always.
...
As can be seen from these 4 lines that comparison is better than measurement.

Thank you very much, zlymex!

Your statement depends on, what you are after.

If you compare both methods concerning pure noise, see left frame, 4h - 6h, in your diagram, then I would guess, that both methods give equivalent results.
Would be nice, if you could do an StD analysis on this time frame for both methods, and I estimate both give about 100..200nVrms of noise.



If you take the right frame, 10h - 12h, there is an additional thermal drift visible, caused by the T.C. of the 3458A, as its internal temperature rises proportionally to the RT.
This is the parameter w/o ACAL, i.e. <=0.5ppm/K, which compares to <=0.15ppm/K for backing of each of two references, as per their individual specifications.

From your measurement, I would estimate the 3458As T.C. as having about 0.2ppm/2K = 0.1ppm/K, and the backing method about 0.05ppm/2K = 0.025ppm/K.

Therefore, the references "beat" the 3458A in the category of their T.C.s only, but not necessarily concerning noise.

The absolute method may be a bit more noisy, due to the different noise figures (at NPLC100) of the 3458A, for the 10V ( => 100nVrms), and the 100mV (=> 20nVrms) range.  Your StD analysis may tell.

Here's again my latest measurement, absolute method, Andreas optimized circuit, 2.5x zoom, with stability noise figures for 1h. Let's see, how this compares to the 732B.


Frank

[zlymex]

Hi Dr. Frank.

The noise or drift is time frame dependent.
For normal low frequency noise, we are talking about 0.1Hz to 10Hz, the noise of 3458A in this time frame is superb.
By specification, 3458A will drift less than 0.1ppm in 10 minutes, which is still very good.
What about 24 hours? 0.5ppm, which is not very good, considering the drift of 732B is only 0.3ppm for 30 days.

If we assume linear behavior in between, the two lines intersect at about 1800 second. In another word, 3458A performs worse than a 10V standard for hourly measurement or above. That is why I prefer comparison rather than direct reading for T.C. measurement of voltage reference which usually takes several hours.

It may argue that the actual or selected 3458A performs better than specified, the actual or selected voltage references may perform even better than specified.
The drift of 3458A is not only thermal, it consist of other factors that drift in longer period even if the temperature is stable, that can be seen from my previous chart.


Although this 732B is not LTZ based, one channel of 4910 (LTZ based) performs very similar to 732B.

[Dr. Frank]

Hi Dr. Frank.

The noise or drift is time frame dependent.
For normal low frequency noise, we are talking about 0.1Hz to 10Hz, the noise of 3458A in this time frame is superb.
By specification, 3458A will drift less than 0.1ppm in 10 minutes, which is still very good.
What about 24 hours? 0.5ppm, which is not very good, considering the drift of 732B is only 0.3ppm for 30 days.


If we assume linear behavior in between, the two lines intersect at about 1800 second. In another word, 3458A performs worse than a 10V standard for hourly measurement or above. That is why I prefer comparison rather than direct reading for T.C. measurement of voltage reference which usually takes several hours.

It may argue that the actual or selected 3458A performs better than specified, the actual or selected voltage references may perform even better than specified.
The drift of 3458A is not only thermal, it consist of other factors that drift in longer period even if the temperature is stable, that can be seen from my previous chart.

Although this 732B is not LTZ based, one channel of 4910 (LTZ based) performs very similar to 732B.

We both agree on that, as you are talking about the timely drift, over a 24h time frame.
Per specification, the 732B is 4 times more stable than the LTZ inside the 3458A.

That would be 0.1ppm/24h drift for the 732B, and 0.4ppm/24h drift for the 3458A.
Both values are worst case, and are valid for virgin references only.
This will go down for mature references, so probably it won't play a role here.
The drift we both see in our ~ 24h measurements, is caused mainly by temperature changes, not by timely drift.

Also, my 3458A is pimped to 65°C, so it will also have a drift of < 2ppm/year, if I would run it continuously.
In my case, I really compare the external reference on par with the 3458A, regarding this parameter.

The 732B will probably drift more, due to the resistors drift, which does not apply to the 3458A (due to ACAL).


The 3458A '10min Transfer Accuracy' is not identical with the timely drift, as this includes the linearity of 0.05ppm max.
The timely drift would be theoretically about 0.03ppm.


Anyhow, when discussing your measurements, I focused on the 1h or 2h 'pure' noise, where the timely drift has practically no effect under all circumstances... and where both of our 3458As are performing much better than these worst case parameters from the specification..

So, I'm still interested on your 1..2h Standard Deviation / Noise-RMS numbers of the 732B vs. 4901, to get an idea, how well Andreas modification performs.

Btw.: For 24h and longer, you may calculate the difference of both methods from the 3458A datasheet:

Absolute method:
10V range => 100nVrms noise, 5µV/K T.C.

Differential method:
100mV range => 20nVrms noise, 0.2µV/K T.C.

Therefore, concerning noise, the differential method is only about 5 times better, but will be superior, if the references are on the order of 100nVrms noise also.

Concerning T.C., the differential method is even 25 times better, therefore constant room temperature is mandatory for such absolute measurements.

Therefore, having no constant RT, requires the differential method.

Frank

[zlymex]

Hi Dr. Frank.

It's true that there are some 732Bs with not very good stepping up resistors making them drift over time. But there are many good 732Bs that drift very little, being used as secondary standards in many labs through out the world. Plus, there are other references such as 4910 that employ PWM method in 7V to 10V step up which inherently don't sufer from resistor drift. I always use my 4910-AV as the reference, which not only with low noise, but drift very little over time(<0.2ppm per year by average).

On the other hand, there was a design/implementation flaw in 3458A by using that big customize chip which contains many paring resistor with very strict requirements that often fail to meet. Your 3458A is good today, my 3458A is good today, doesn't mean they will be good always, doesn't mean other 3458As are all good. My 3458A was bought 10 years ago 1st hand from an authorized Agilent dealer, but about two months after, I found out it was drifting all the time, they provided free repair of course - they replaced the big customize chip. I myself once try to fixed a drifting 3458A that ended up with suspecting that chip and can do nothing about. I Think TiN has the similar feeling about it.
Even with 'good' 3458As, they require frequent ACAL to cancel the drift of resistors within the big chip(better hourly), which is not possible for T.C. measurement that often automatic, continuous for several hours.

HP/Agilent knows the flaw and prepares many big chips for replacement(not publicly available I'm afraid).


Attached also some bad big chips that one of my friends replaced when repairing 3458A.

To summarize, 3458A is a legend multimeter with unbeatable low noise and linearity for 10V range, but probably apply to short period only. In order to make the best use of it, I use frequent comparison(to a good reference, by a scanner/switch) rather than direct measurement.