Ultra Precision Reference LTZ1000

[SeanB]

If you want a pressure equaliser bellows cheap go get an old mechanical fridge thermostat, there is a nice phosphor bronze capsule inside with a capillary tube attached. Vent the gas and solder the cap tube to the finished case and it will flex to accommodate the expansion of the fill gas or oil.

If I was building one I would fill it with an inert refrigerant gas like R134A,  all the housing has to do is be capable of being pumped down to a vacuum for a short time then filed and sealed with a standard method, either a schraeder valve or a soldered pinch tube.

[branadic]

My ultimate idea (not already a complete design) would be building some kind of "calibrator" from 0-10V with a LM399 as basis. Either with a PWM divider (EPN cirquit idea) (cheap but noisy) or 2 interleaved 16 Bit DACs giving about 28 Bits resulting (more expensive but less noise) resolution. When having 2 independent calibrators of this sort and a 24 bit low noise ADC with some relays or MUX switches you could build a "self calibrating" system which calibrates out the non-linearities below around 1ppm. Similar to Franks 10V/1V divider but with a binary adjusting scheme.

I was more thinking of using three references, several switches, an adc and "a dac."

1. Initial calibration for one of the three references. Therefor Ref1 is connected to the dac and output voltage of a specific digital word is measured, you just need to know its initial output voltage.
2. Ref1 is divided and act as the adc ref, Ref2 and Ref3 are divided to be within the range of the adc and are measured.
3. Ref2is divided and act as the adc ref, Ref1 and Ref3 are divided to be within the range of the adc and are measured.
4. Ref3 is divided and act as the adc ref, Ref1 and Ref2 are divided to be within the range of the adc and are measured.
5. Now the drift and the new digital word for Ref1 connected to the dac is calculated as well.
6. Start again at point 2.

After inital calibration the system needs to stay powered up.

[Andreas]

Also have a look on the Fluke 5440A self calibration scheme, it has a nice 0.1ppm linear DAC ( 2x 1bit) and automatic  range calibration, uncertain to < 0.5ppm, or better.

Frank

Hello Frank,

I had a look on the documentation (hope I picked the right one)
http://assets.fluke.com/manuals/5440B_AFsmeng0000.pdf

But with my experiences from a PWM-divider (see EDN cirquit)
http://www.edn.com/design/other/4326640/DC-accurate-32-bit-DAC-achieves-32-bit-resolution
I expect that besides the selection of the FETs (1 Ohm) in the 5440 unit a part of the linearity is hidden within the calibration constants.

Otherwise you have too much problems with "voltage dependancy of the switch (High/low)", "charge injection from PWM to output", "different rise times high/low" which makes it hard to get around below 2 ppm linearity even with tweaking all resistors.
See attached pictures. Linearity deviation is in mV referenced to 5000 mV of the reference. So 0.005 mV would correspond to 1 ppm linearity deviation.

There are different switches (CD4053, 74HCT4053 and MAX4053 without and with tweaking of the surrounding resistors) with different results ranging from about 100ppm to 4 ppm maximum deviation.

With best regards

Andreas

Edit: of course the genious trick of the 5440 with cutting out the non-linear range near zero (and full scale) by a offset would help at least for the tweaked cirquit.

[Andreas]

Part 3 LTZ1000A of Andreas (Todo list):

For the next 2 devices I plan the following.

- Using pre-aged Z201 resistors (100mW power cycling).
  the main intention is to have more room within the shielded area for additional improvement measures
  like cut outs on the PCB against the wirewound types.

- perhaps I can find also another suitable case with more room
  I would also like to have larger battery capacity.
  The 12 AA cells are only good for around 36-48 hours with heater on.
  I have already found the "Fischer TUG V 17" aluminium case which would be ideal in 160-220 mm length.
  But unfortunately its so new that there are no dealers who sell them in small quantities.
  The other solution would be the "ProMa EKG 3" aluminium case but the mounting possibilities
  for PCBs are very limited.
  Anyone better ideas for similar cases?

- use the additional room within the shield for some slots around the LTZ to keep thermal gradients
  away from the legs. The base emitter capacitors will remain as 0402/0603 capacitors directly at the LTZ.
  The LTZ will be mounted directly to the PcB (platet through PCB is necessary).
  My prefered solution up to now is using short legs and a slotted board.
  In this case I believe that the critical thermal junctions between PCB
  and LTZ have the lowest temperature differences.
  But if branadic gets other results with his thermal camera I will use the optimum solution.

- perhaps (depending on space used) additional isolation (slots) to the FET and the heater transistor.

- replacing the Hirschmann banana plug connectors with POMONA low emf connectors (equal height for both pins)

- buffered output with 10V probably with a LTC2057 low noise low drift amplifier (recently arrived from DigiKey)
  The resistors for the divider could be the trimmed version of the VSMP series which is offered by DigiKey.
  The fine trimming could be done by a DAC. And all on on a PCB cutout which can be thermally regulated.
  Of cause I will check the crystal oven heater which was recommended by branadic (good idea) as an option.

- I bought some LT1013A-devices in hermetically tight case (CERDIP) some time ago.
  One for the price of a half LTZ1000A. I wanted to use them to see if
  I could get more stability over time against the standard plastic case.
  And all pictures of references that I could see with an LT1013 had usually a TO-99 metal case
  which is now unobtanium.

  But now I have the LT2057 which is more stable over time and temperature,
  has more open loop amplification lesser problems with thermal emf and less noise than LT1013A.
  So at least I will give the LT2057 a try with the option to use the LT1013A as backup.

- The 14V voltage regulator and the battery pack should get at least a slot in the PCB for thermal isolation
  eventually further measures like additional (styrofoam-)wall between power supply and LTZ section.
  With a larger case I will put the batteries above the whole cirquit with a thermal isolation between
  the battery compartment and the rest of the cirquit. So the idea with the slot will only be for the
  local voltage regulator.

- Doing further thermal cycling (-18 degree celsius) after soldering (thanks Frank for the idea).
 
  With a AD586LQ-device I have made the experience that a gentle cyclical stress has a positive effect on
  the ageing rate. Of course every stress will give a new starting point
  (usually with a higher momentary ageing rate) for the device.
  But if you have luck the ageing will stabilize as can be seen on the graph.
  At the moment I repeat the same for a LT1236AILS8-5 device which has a too high ageing rate for my purposes.

  To the picture:
  Day 200 is start of the measurements with ADC13 and AD586LQ reference
  measuring a LTZ1000A by a capacitive LTC1043 2:1 divider (giving around 3600mV).
  The first firing of the reference is day 0.
  Initial ageing rate around 3.4ppm/sqrt(kHr) referenced to day 0.
  Around day 415: begin of cyclical stress during night with 15mA
  load at the output of the reference.
  Ageing slope increases new ageing rate around 2.4ppm/sqrt(kHr) referenced to day 415.
  Around day 460: Ageing stabilises to around 1-2 ppm/year cyclical stress continued.
  Around day 540: Finishing cyclical stress.

Any further Ideas for the LTZ?
How is the pre-ageing for the LTZ1000 done at the manufacturers of metrology equipment?

With best regards

Andreas

[branadic]

But if branadic gets other results with his thermal camera I will use the optimum solution.

It will take some time, but I'm working on that.

[Dr. Frank]

Part 3 LTZ1000A of Andreas (Todo list):

For the next 2 devices I plan the following.

- Using pre-aged Z201 resistors (100mW power cycling).
  the main intention is to have more room within the shielded area for additional improvement measures
  like cut outs on the PCB against the wirewound types.

Once again, I cannot recommend that at all.
This is necessary only for stabilization of the load life stability of the resistors, which is not the case for LTZ references (P< 10mW)
Also, you will get hysteretic resistance value shifts of several ppms, which you might want to remove by thermal cycling (similar to demagnetization)
If you use heremetically sealed VHP202Z, the drift is so low from the fab, that a "stabilization" as you want to do, causes more harm (i.e. several ppm of non reversible value shift) to the resistor than the ultra low natural drift (2ppm/6yrs.).

..
  The LTZ will be mounted directly to the PcB (platet through PCB is necessary).
  My prefered solution up to now is using short legs and a slotted board.
  In this case I believe that the critical thermal junctions between PCB
  and LTZ have the lowest temperature differences.
  But if branadic gets other results with his thermal camera I will use the optimum solution.

Try to use a single sided PCB, so all thermal junctions are on that side and can easily be shieleded thermally by a single layer of foam.
Again, the effects of slots in the board have not yet been explained, neithertheir physical mechanism, nor the quantity.

In contrast, a noisy power supply, or bad shielding cause shifts of up to several ppm, obviously by rectifaction of those AC disturbance signals on the temperature measuring Q1, therebyd shifting the temperature. To mitigate this effect, is much more important.
(I use an external PSU, so the transformer and its magnetic field is outside the case)

- The 14V voltage regulator and the battery pack should get at least a slot in the PCB for thermal isolation
  eventually further measures like additional (styrofoam-)wall between power supply and LTZ section.
  With a larger case I will put the batteries above the whole cirquit with a thermal isolation between
  the battery compartment and the rest of the cirquit. So the idea with the slot will only be for the
  local voltage regulator.

A magnetic shielding and multiple filtering of AC disturbance signals are much more important.

- Doing further thermal cycling (-18 degree celsius) after soldering (thanks Frank for the idea).

You have misunderstood me.
Thermal cycling means, that you apply temperature differences from the stabilization point (e.g. +50°C) cyclicly, with decreasing amplitude.
E.g. you have to apply -18°C (delta T ~ -68K), +100°C  (+50K), -10°C, +80°C, 0°C, +70°C, ... and so forth, until you are below ~ +/- 20K, where all hysteresis is gone. See also the patent of Pickering, realized in the Fluke 7001 reference (but which does not work properly in the 7001, I assume)

  To the picture:
  Day 200 is start of the measurements with ADC13 and AD586LQ reference
  measuring a LTZ1000A by a capacitive LTC1043 2:1 divider (giving around 3600mV).
  The first firing of the reference is day 0.
  Initial ageing rate around 3.4ppm/sqrt(kHr) referenced to day 0.
  Around day 415: begin of cyclical stress during night with 15mA
  load at the output of the reference.
  Ageing slope increases new ageing rate around 2.4ppm/sqrt(kHr) referenced to day 415.
  Around day 460: Ageing stabilises to around 1-2 ppm/year cyclical stress continued.
  Around day 540: Finishing cyclical stress.

I do not understand your diagram correctly, I fear, please help me:.

The drift shown comes from the AD586 only, the difference of both LTZ outputs is constant to ~ 1ppm in 1 year or so, therefore the drift rates of 3.4ppm/sqrt(khr) you have additionally drawn apply to the AD 586, but not to the LTZs.
Those are rock stable and need no further improvement. (I have measured similar behaviour over 4 years)

Any further Ideas for the LTZ?
How is the pre-ageing for the LTZ1000 done at the manufacturers of metrology equipment?

I doubt they make real pre-aging.
The LTZs and the peripheral components are so stable that the total drift mainly depends on the different LTZ samples, i.e. on fabrication variation.
The first might drift 1ppm/yr. (@60°C), the next 0.7ppm/yr. only, and so forth.
So they do a monitoring over 3-6 months only, and select the most stable references.

To my understanding of the physics, at such low drift levels, a pre-aging makes no sense, but causes more harm (higher drifts, hysteresis...) than it would really stabilize the reference even more.

I will publish my own results (measurements) soon, as an indicator for that thesis.

regards Frank

[babysitter]

Travelling Reference Info: My LTZ1000A cam back a few days ago, will visit Dr. Frank soon and after that it could travel south, to Quarks and Branadic !

[Andreas]

It will take some time, but I'm working on that.

Fine we are all curious on the results.

Once again, I cannot recommend that at all.

Try to use a single sided PCB, so all thermal junctions are on that side and can easily be shieleded thermally by a single layer of foam.
Again, the effects of slots in the board have not yet been explained, neithertheir physical mechanism, nor the quantity.

Sorry but my ageing experiment is already running some 1000 hours. So I should have known it last year in February. On the other side: when looking at the derating diagram the 100mW will heat about 17 degrees.
Thats nearly the temperature change between winter and summer of my "lab".  I will see the result in the run-in phase of the LTZ.

If using a single sided board I cannot use slots. So this will be only a option if the slots worse the temperature distribution between the pins. I have now a board without slots so I will most probably try one with slots the next time.

In contrast, a noisy power supply, or bad shielding cause shifts of up to several ppm, obviously by rectifaction of those AC disturbance signals on the temperature measuring Q1, therebyd shifting the temperature. To mitigate this effect, is much more important.
(I use an external PSU, so the transformer and its magnetic field is outside the case)

A magnetic shielding and multiple filtering of AC disturbance signals are much more important.

I will use a AC wall plug adapter only during charging. In operating mode I will use battery power.
The best shielding at line frequency is keeping far away from transformers.

You have misunderstood me.
Thermal cycling means, that you apply temperature differences from the stabilization point (e.g. +50°C) cyclicly, with decreasing amplitude.
E.g. you have to apply -18°C (delta T ~ -68K), +100°C  (+50K), -10°C, +80°C, 0°C, +70°C, ... and so forth, until you are below ~ +/- 20K, where all hysteresis is gone. See also the patent of Pickering, realized in the Fluke 7001 reference (but which does not work properly in the 7001, I assume)

Ok now I got it. But the first point at my side is the 270 degrees soldering temperature when using short legs.

I do not understand your diagram correctly, I fear, please help me:.

The drift shown comes from the AD586 only, the difference of both LTZ outputs is constant to ~ 1ppm in 1 year or so, therefore the drift rates of 3.4ppm/sqrt(khr) you have additionally drawn apply to the AD 586, but not to the LTZs.
Those are rock stable and need no further improvement. (I have measured similar behaviour over 4 years)

Yes you are right.
My test setup are the 3 references, fed trough a multiplexer, then divided by a capacitive 2:1 divider (LTC1043) and then measured by a 24 Bit ADC with a temperature compensated reference AD586LQ.
So if you assume the LTZ1000A as being constant you see the inverse ageing of the AD586. If the reference voltage of AD586 goes down you see the measurement going up in the diagram.
Unfortunately I do not have the absolute ageing rate of my LTZ1000 references up to now.

To my understanding of the physics, at such low drift levels, a pre-aging makes no sense, but causes more harm (higher drifts, hysteresis...) than it would really stabilize the reference even more.

I interpret the diagram above as follows: Every stress on the device will start a new ageing cycle with usual higher ageing rate as before.
It may be accidently in the test above or perhaps can be repeated: (I will know this in a few 1000 hours with a LT1236AILS8). After the stress (training) the reference is falling to its "sweet spot" remaining stable.

My theory:
I blame the hysteresis effects of hermetically tight packages on the "die attach" of the chip.
On AD586 change notes you can see that it is usually a silver filled epoxy compound.
I do not know how they dose the glue on the Kovar plate/lead frame of the housing but probably this
process is the one with the most errors in fabrication. And every manufacturer has its own mixture.
With temperature cycling the erratic connections of the die attach at the edges of the chip might be egalized somewhat giving better results.
The interesting thing with a LT1236AILS is that the hysteresis of the chip seems to have a time constant/delay of about 50 minutes (measured on one device). So the hysteresis curve on this device is drastically changed between a temperature slope of 0.1K/minute against 0.3K/minute. So also this effect could be some creeping effect of the die attach.

With best regards

Andreas

[Dr. Frank]

Sorry but my ageing experiment is already running some 1000 hours. So I should have known it last year in February. On the other side: when looking at the derating diagram the 100mW will heat about 17 degrees.
Thats nearly the temperature change between winter and summer of my "lab".  I will see the result in the run-in phase of the LTZ.

I think, that effect is so small (e.g. attenuation 1:100), that you won't be able to identify it, or distinguish it from other effects.

I will use a AC wall plug adapter only during charging. In operating mode I will use battery power.
The best shielding at line frequency is keeping far away from transformers.

The battery mode is the best you can do.
Although I use an improved DC - out wall plug adapter also, my LTZ refs still struggle from other AC irraditions.
There is even a small effect from the GPIB cable, if the measurement is running.
Therefore the next LTZ design will include multiple shielding and a double filtered DC supply.
 

Ok now I got it. But the first point at my side is the 270 degrees soldering temperature when using short legs.

I used heat removal pliers, and left the legs long.
If in doubt, whether the LTZ has "seen" this heating, you should do this thermal cycling afterwards.

For the first of my two references I was able to prove, that it did NOT see any hysteresis effect from soldering and only small other heating effects, because after a thermal cycling, its output value went back to exactly the initial value.
That means, that this LTZ reference  drifted less than 0.5ppm in about 3-4 years!!

So if you assume the LTZ1000A as being constant you see the inverse ageing of the AD586. If the reference voltage of AD586 goes down you see the measurement going up in the diagram.
Unfortunately I do not have the absolute ageing rate of my LTZ1000 references up to now.

Simply trust on the datasheet-stability of the LTZs..the AD586 drifts, but not the LTZs..

Well  , you indeed HAVE absolute ageing rates of your LTZs already!!
Simply create a diagram where you draw the change of the difference in output voltage over time, i.e. LTZ_1 minus LTZ_2 and LTZ_1 minus LM_2 (what's that kind of ref? not the LM1236 ?).
Those two graphs will show, that these 3 refs drift apart not more than 1-2ppm/2years!!

The individual drifts will not be more than those determined values, theoretically half of that for each one.
So, you already own two/three ultra-stable references, based on a design with "simple" wire wound resistors, like "babysitter" and me...

I interpret the diagram above as follows: Every stress on the device will start a new ageing cycle with usual higher ageing rate as before.

That's another argument for NOT doing a pre-ageing stress on the LTZs, as they might behave the same, but on a much smaller scale.


regards Frank

[Andreas]

Therefore the next LTZ design will include multiple shielding and a double filtered DC supply.

I have already tried to filter out the interference of a (switchmode) power supply X+Y capacitors + chokes. But with very little effect.
I guess one will have to use a transformer with a shield like on most metrology gear between primary + secondary side to get effective filtering.
So let me know if you have a solution.

Simply create a diagram where you draw the change of the difference in output voltage over time, i.e. LTZ_1 minus LTZ_2 and LTZ_1 minus LM_2 (what's that kind of ref? not the LM1236 ?).

LM #2 is one of my LM399 references now running 24/7 since 2008.

With best regards

Andreas

[Andreas]

Hello Ken,

thanks for sharing. This will be another point on my todo list: At least providing the space for ferrite beads on PCB at critical points.

Do you have any practical experiences with these parts mentioned in LT AN101?

When having a closer look to AN101 I see that J.W. has not used a real switching supply but has generated the waveforms by a "artificial circuit".
The main difference is that he supplies the spikes in differential mode to the input of the analog regulator (a LT1763 which I also use). The noise from a real switch mode supply usually clocks in as common mode so the noise will be equally on the +supply and the GND pin. At least this is true for a wall wart with 1.5 m cord length on the 24V side.
So instinctively I would use 2 ferrite beads on the input. Preventing that the input capacitor of the analog regulator will feed the noise on the GND line to the input of the regulator after the bead on the +supply again.
But why is he using especially this ferrite bead having only around 50 Ohms at 100 MHz. I would go for a higher Impedance e.g. 300-1000 Ohms at the critical frequency of 100 MHz if the DC voltage drop due to the DC resistance would permit this.
Further I would never use only a electrolytic capacitor alone on input and output of voltage regulator like it is shown in AN101. I always use additional 100nF ceramics to short the RF spikes to the GND pin of the regulator.

With best regards

Andreas
 

[Dr. Frank]

Monitoring Setup

If you want to find out how well your LTZ reference performs, you have to compare it to either one much more stable standard, or to several equally good ones.

This much better standard would be a JJ primary standard only.

Equally good ones would be supervised Zener standard banks as Fluke 734B, 7010 or Datron 4901, or an ensemble of at least 3 DYI LTZ1000 references, or similar.

Also, a precise transfer standard is needed, to be able to compare the raw 7.xxx volt precisely against the 10V references.

I use 2 different kind of references, the first one is the internal LTZ1000A of a 3458A, running on 65°C (instead of 95°C), and a Fluke 5442A which is based on the SZA263.

The HP3458A also serves as a very precise transfer standard, due to its linearity.

Both devices are well aged, 13 years for the HP3458A, 23 years for the Fluke 5442A, and both are running intermittently only. During power down times, their drift should be close to zero. Their combined drift over 3 ½ years is obviously 1ppm maximum, see figure 1.

Both of the DIY LTZ1000 references, including the 7,147V => 7,0000V => 10,0000V transfer derived from Ref_2, have been running continuously for 3 ½ years, and were compared also against the HP3458A. Therefore, my reference ensemble consists of 4 equivalent sources.

Long Term stability measurements

During the first 2 years, I have checked only, that all references were stable to within 1ppm of their initial values. Therefore, intermediate data points were missing.

In June 2011, I have moved my complete analogue equipment to the basement, where a stable room temperature of 20.0 … 22.5 °C is available during all seasons.
The room temperature may be constant to +/- 0.2K during several hours and to +/- 1.0K during successive measurement days. 
This measure improved the short term stability of all measurements significantly.
In other words, without a stable environment, sub-ppm stability measurements are not possible at all.

Ref_2 and its 10V output drifted about 1ppm in 3 ½ years. (See figure 2)
The 7,000V output has been calibrated initially only, but the 10/7 transfer was calibrated several times, about 2 times a year.

On Ref_1 I performed some experiments, so it was “mistreated” several times, i.e. the temperature control got out of regulation. Thereby it encountered several temperature trips to 100°C (estimated).
Afterwards, its output voltage then restarted at a lower value, and drifted much more than Ref_2, (see figure 3). As the output voltage of the LTZ1000 is lower at 100°C, the reference @ 45°C obviously memorized its short trip to 100°C, and drifted towards that direction.

Conditioning

The last such accident happened in June 2013, which left a hefty additional shift of -3ppm.

I remembered the patent of Pickering, to remove temperature induced hysteresis effects in the LTZ1000.

So I temperature-cycled the complete box to remove the hysteresis (see figure 4).

To do that cycling right, it is important to have big enough, but decreasing temperature differences, related to the stabilization temperature, in this case compared to +45°C.
This means, +100°C gives +55K, storage at -23°C gives a dT of -68K, +80°C equals dT = +35K and so forth.

I found out (see fig. 4 also), that below a difference of about +/- 20K, there is no big hysteresis effect.
So I really doubt, that the Fluke 7001 really did operate efficiently in removing hysteresis, because at a stabilization temperature of +45°C, the lowest available negative temperature difference at around 22°C room temperature (i.e. -23K) is too small to deliver a sufficient 'reset' effect. (The 7001 has been terminated by Fluke in the meantime.)

During that procedure, Ref_2 also showed a hysteresis effect, but in the end returned to its initial value of 3½ years before. This indicates that the cycling really removed the hysteresis in both references.

Ref_1 ended at about +1ppm, and its future drift behavior will show, if it’s now more stable. If it resides at about +1ppm, possibly the LTZ1000 had been heated too much during soldering, although I used a thermal transfer pincer and soldered very quickly.

Further effects

It is important to have a proper shielding (e.g. case connected to ground) and a quiet power supply.

An external wall plug power supply was used, to avoid magnetic disturbance from the transformer. In first instance, its output  of 18V AC induced disturbances on the LTZ temperature regulation.
The case had to be connected to ground of the LTZ output, and that induced a shift of 0.5ppm of each output. Exactly this same effect could also be demonstrated on another LTZ based reference, designed by 'babysitter'.
When I redesigned the wall plug in for an output of 25V DC, the LTZ references became more short term stable, and the 0.5ppm shift vanished.


Short Term stability measurements

Figures 5 and 6 show short term stabilities of Ref_1 versus HP3458A during 10 minutes and 35h, using an aperture time of 2 seconds, i.e. NPLC100.

The internal temperature of the HP3458A increased continuously from 33.7 to 34.8°C during those 35h. ACAL was performed only once, before the start of the measurement.
 
The 10min measurement shows  fluctuations of +/- 0.05pm, which is very well consistent with the transfer specification of the HP3458A,  +/-  0.1ppm, and especially with the noise specification of the LTZ1000, i.e. 2µVpp (equivalent to +/- 0.15ppm).

The average 35h drift of ca. 0.3ppm is well below the HP3458A 24h specification, i.e. +/- 0.55ppm, or its T.C. of 0.5ppm/K without ACAL.



Conclusion:

The basic LTZ1000 circuitry, output of around 7,2V, running on 45°C and built with wire wound resistors is capable of drifts well below 1ppm/year. Noise and short term stability are below +/- 0.1ppm.

Sophisticated shielding and a low noise DC PSU are very important for obtaining that degree of stability.

The influence of additional ‘gimmicks’ as summarized in part #1, have not been demonstrated yet, but they have to compete in value with the effects demonstrated here.

<end>

[Andreas]

To do that cycling right, it is important to have big enough, but decreasing temperature differences, related to the stabilization temperature, in this case compared to +45°C.
This means, +100°C gives +55K, storage at -23°C gives a dT of -68K, +80°C equals dT = +35K and so forth.

After such a "mistreatment" with a output voltage shift (shorting the 7V output by a unpowered ADC) I simply switched off the power of my LTZ#1 for some minutes to cool down and switched on again. The deviation was much smaller so I repeated power cycling on the LTZ until the deviation was small.

With best regards

Andreas

[Dr. Frank]

After such a "mistreatment" with a output voltage shift (shorting the 7V output by a unpowered ADC) I simply switched off the power of my LTZ#1 for some minutes to cool down and switched on again. The deviation was much smaller so I repeated power cycling on the LTZ until the deviation was small.

With best regards

Andreas

I did not mention that: Although I unplugged the box for several days after such an event, the Ref_2 output recovered a little bit, but afterwards quickly drifted towards negative values.
Therefore, "resetting" at 21°C did not really help.

Another indicator: Under normal conditions, the outputs recover stably to their initial values, i.e. to within a few tenths of a ppm, if the box is switched off for a longer period of time and then switched on again..

My wife tested that for me when she saw the illumination of  the multiple socket outlet, and switched it off to 'save energy'    (the box consumes around 2,5W).
3 weeks later, I was very relieved to determine, that absolutely no drift occurred afterwards.

Frank

[Andreas]

Hello Frank,

Would be interesting if the LTZ1000A are easier to get to the initial condition than the LTZ1000 or if my accident was simply not so hard as yours.

With best regards

Andreas

[Dr. Frank]

Hello Frank,

Would be interesting if the LTZ1000A are easier to get to the initial condition than the LTZ1000 or if my accident was simply not so hard as yours.

With best regards

Andreas

Hello Andreas,

currently I don't want to test that again. No experiments on the revised box any more..
Perhaps I could do that when I assemble those five new LTZ1000..

I assume that the hysteresis depends on sample variations.
And the LTZ1000A should heat up to a higher temperature than the LTZ1000, due to its higher thermal isolation.

Frank

[Andreas]

currently I don't want to test that again. No experiments on the revised box any more..
Perhaps I could do that when I assemble those five new LTZ1000..

I can understand this ;-) And I do not know if I wanted to repeat the experiment with only having 7 devices.
Except when having one candidate with a very large drift over time.

I assume that the hysteresis depends on sample variations.
And the LTZ1000A should heat up to a higher temperature than the LTZ1000, due to its higher thermal isolation.

Of course so we would have to test many samples to get a statistical prove.
And dont forget: I have only a voltage regulator in SO-8 which will go easier to current limiting than other voltage regulators.

With best regards

Andreas

[branadic]

So let me know if you have a solution.

http://www.haufe-uebertrager.de/

delivers shielded (toroidal core) transformers. If the transformer is not shielded it should be shielded within your case. An important point is the way you arrange the transformer compared to the pcbs. Place it so that coupling by its field into your pcbs is as small as possible. You don't need to use expensive mu metall but galvanically tinned metal sheet / rf sheet metal is a perfect choice.
I would avoid using switching supplies in such an application at all and use linear components instead. But if not avoidable my tip is: There is a good book by Würth Elektronik you might want to have a look to, you will find all the info about ferrite beads and filter strategies using them.

[Andreas]

At 100Mhz, a few pF looks like a low impedance.  This is not an easy problem to fix, and there is no guaranteed "recipe" that will always work in any situation-- theory is not very helpful here-- only experience.

Good Luck!

-Ken

True words.
I worked on EMC optimized PCB layouts for some years. The 3rd truth is that you have to fight against every mil line length producing high impedances for filter capacitors shifting the maximum usable frequencies towards dramatic lower values.

So let me know if you have a solution.

http://www.haufe-uebertrager.de/

delivers shielded (toroidal core) transformers. If the transformer is not shielded it should be shielded within your case. An important point is the way you arrange the transformer compared to the pcbs. Place it so that coupling by its field into your pcbs is as small as possible. You don't need to use expensive mu metall but galvanically tinned metal sheet / rf sheet metal is a perfect choice.
I would avoid using switching supplies in such an application at all and use linear components instead. But if not avoidable my tip is: There is a good book by Würth Elektronik you might want to have a look to, you will find all the info about ferrite beads and filter strategies using them.

Hello branadic,

perhaps I have not explained it correctly: I do not want to use a transformer within my device. I am looking for a solution of keeping external mains disturbances like the "green" switchmode supplies away from my cirquit so that I could (optional) charge while measuring. My shielding and guarding concept does not allow a transformer within the device.
For this I would need a outer shield connected to earth ground additional to my existing housing which is a floating guard.

And there is a difference between the magnetic shield (a outer ferromagnetic housing around the transformer) which they offer as option for standard parts and a electrostatic shield (a single ended isolated metal foil connected to earth ground) reducing the coupling capacity between primary and secondary side of the transformer.

So I will most probably try to get some enhancement by some ferrites.
Which of the books did you mean on the Würth homepage?
The "Trilogy of Magnetics" or one of the other books?

With best regards

Andreas

[branadic]

Which of the books did you mean on the Würth homepage?
The "Trilogy of Magnetics" or one of the other books?

Yes, this is the title I think to remember to. I have bought this book for work and think it's pretty helpful.

[babysitter]

Of course... Pk4ts. There is the Same Part in normal thermal emf available for comparisons, too.

Sent from my mobile

[babysitter]

IT is in their t&m catalog! I dont know a Distributor, price around 12 Euro each. Minimum Order around 60 euro. Br Babysitter

[quarks]

so far I have not found a source for the SLS410-TS and the SL425-A/TS, if you find them please share
(what I do not like is, both have solder connection only)

[Dr. Frank]

so far I have not found a source for the SLS410-TS and the SL425-A/TS if you find them please share
(what I do not like is, both have solder connection only)

I've ordered the PK-4TS directly @ MC Germany, therefore I assume that you may also get the other parts there.
Perhaps you ask them first, also for pricing.

Frank

[babysitter]

Price was about 7-8 eur for the plugs iirc.
Minimum order of ~60 euro in germany !

Greetings from the Babysitter