Power supply for voltage references

[dietert1]

Yes, a similar setup was also used for data transmission in the isolation interface of the HP 3456A DVM. Don't know whether they use that interface during measurement or autozero. I think if the core is kept off by plastic insulator or spacer this will have more capacitive coupling than with (partial) air spacing as in the Cleverscope image. But that's something to decide by measurements.
Up to now i decided to make something old school that is cheap and hard to beat. Anyway, i can't imagine running that mains transformer close to a voltage reference. Neither can i imagine running a high frequency converter near a voltage reference. Integrating and optimizing such a setup will be a major engineering effort.

Regards, Dieter

[doktor pyta]

AFAIK one of the Fluke/ Hart Scientific thermometers used two akkus or super caps. One was charged and the other was connected to the load at the time.
This was patented solution.

[Echo88]

Links to the circuit in the F1595a mentioned by doktor pyta:

https://download.flukecal.com/pub/literature/Recent%20Advances%20in%20Resistance%20Thermometry%20Readouts.pdf
https://patentimages.storage.googleapis.com/94/8a/8b/fdbc915129e6b9/US8120327.pdf

[fcb]

AFAIK one of the Fluke/ Hart Scientific thermometers used two akkus or super caps. One was charged and the other was connected to the load at the time.
This was patented solution.

I quite like that approach.  Could probably work if you can get a high-enough isolation/low enough capacitance between across a relay.

I thought about using a motor, long plastic shaft and generator (in reality two stepper motors), if driven by low frequency could be low noise.  Could even mount motors onto a PCB and use a belt.  Risk of creating a van-der-graff generator?  Although you could run the belt past a numer of generators if you needed to power a few channels.

[The Soulman]

AFAIK one of the Fluke/ Hart Scientific thermometers used two akkus or super caps. One was charged and the other was connected to the load at the time.
This was patented solution.

I quite like that approach.  Could probably work if you can get a high-enough isolation/low enough capacitance between across a relay.

I thought about using a motor, long plastic shaft and generator (in reality two stepper motors), if driven by low frequency could be low noise.  Could even mount motors onto a PCB and use a belt.  Risk of creating a van-der-graff generator?  Although you could run the belt past a numer of generators if you needed to power a few channels.

If you want to go mechanical, why not use two (brushless) electric motors with a belt or a plastic axle between them, one driving the other?

The switched battery/supercap idea is kinda clever tho, something like a beefed up ltc1043.

[fcb]

AFAIK one of the Fluke/ Hart Scientific thermometers used two akkus or super caps. One was charged and the other was connected to the load at the time.
This was patented solution.

I quite like that approach.  Could probably work if you can get a high-enough isolation/low enough capacitance between across a relay.

I thought about using a motor, long plastic shaft and generator (in reality two stepper motors), if driven by low frequency could be low noise.  Could even mount motors onto a PCB and use a belt.  Risk of creating a van-der-graff generator?  Although you could run the belt past a numer of generators if you needed to power a few channels.

If you want to go mechanical, why not use two (brushless) electric motors with a belt or a plastic axle between them, one driving the other?

The switched battery/supercap idea is kinda clever tho, something like a beefed up ltc1043.

I literally said that and you've quoted it.

[The Soulman]

AFAIK one of the Fluke/ Hart Scientific thermometers used two akkus or super caps. One was charged and the other was connected to the load at the time.
This was patented solution.

I quite like that approach.  Could probably work if you can get a high-enough isolation/low enough capacitance between across a relay.

I thought about using a motor, long plastic shaft and generator (in reality two stepper motors), if driven by low frequency could be low noise.  Could even mount motors onto a PCB and use a belt.  Risk of creating a van-der-graff generator?  Although you could run the belt past a numer of generators if you needed to power a few channels.

If you want to go mechanical, why not use two (brushless) electric motors with a belt or a plastic axle between them, one driving the other?

The switched battery/supercap idea is kinda clever tho, something like a beefed up ltc1043.

I literally said that and you've quoted it.

   Sorry about that, first time I've read that to quickly and thought you were thinking of a motor driven commutator to
use instead of a relay.
Reading again, I don't know what I was thinking.

[guenthert]

     I don't quite see the need for extreme measures for the power supply of a DC voltage reference as typically they just feed long scale DMMs measuring DC, which take their sweet time integrating the signal (Fluke 8508A takes a whopping 25s for the conversion at best resolution -- twice as fast as the old Solartron 7081) or being compared to other sources using a null meter (Fluke 845A specifies a patience-testing settling time of 5s at the most sensitive range).  Substituting a nano voltmeter like Keysight's 34420A for the null meter speeds up the measurement considerably, still for best results it'll integrate over multiple power line cycles.  That of course assumes that there won't be some rectifying of the signal in the meter.  For null meters AC rejection is specified and can be tested. 

     I'd think other instruments, e.g. spectrum analyzers, have much more stringent requirements for the noise-lessness of their power supplies.  How are they doing it?

[SilverSolder]

[...]
     I'd think other instruments, e.g. spectrum analyzers, have much more stringent requirements for the noise-lessness of their power supplies.  How are they doing it?

One thing I've seen done is the use of a shielded transformer for the mains supply, to avoid common mode noise capacitively leaking across from primary to secondary, I guess.

[exe]

     I'd think other instruments, e.g. spectrum analyzers, have much more stringent requirements for the noise-lessness of their power supplies.  How are they doing it?

Looking at tear-downs, there are many-many individual power rails in modern instruments, and most if not all of them are smps. I think in R&S instruments I've seen completely integrated modules that don't require any external components. Probably, this way parasitic inductances are small, thus reducing the noise.

I've also seen decoupling capacitor networks, where multiple caps are in parallel. But you can't just wack in several caps and hope they they will have a combined effect. Usually it only makes things worse. So it takes some engineering to do it right.

[Dr. Frank]

     I don't quite see the need for extreme measures for the power supply of a DC voltage reference as typically they just feed long scale DMMs measuring DC, which take their sweet time integrating the signal (Fluke 8508A takes a whopping 25s for the conversion at best resolution -- twice as fast as the old Solartron 7081) or being compared to other sources using a null meter (Fluke 845A specifies a patience-testing settling time of 5s at the most sensitive range).  Substituting a nano voltmeter like Keysight's 34420A for the null meter speeds up the measurement considerably, still for best results it'll integrate over multiple power line cycles.  That of course assumes that there won't be some rectifying of the signal in the meter.  For null meters AC rejection is specified and can be tested. 

Hi guenthert,

I thought just like you, until I found out in practice, that ordinary linear power supplies, w/o shield and isolation, obviously  create ground loops which may shift the reference voltage by several tenths of ppm. These may also manifest in disturbance from mains into the circuit, which may create additional shifts.

The worst effect was observed during the difference measurement between two mains supplied LTZ circuits, using a DMM like 3458A or 34465A, I don't remember. That measurement was complete toast, only huge spikes.

I did not repeat that measurement any more by using my KEI 182A nV meter, but maybe I'll do that now that I have a Fluke 7000, against a mains or battery supplied LTZ. 
Frank

[branadic]

At least a yesterdays measurement with K3458A in 100mV range worked, measuring one of the F7000 (cell2) powered from mains against a F731B, also powered from mains. The readings were stable and consistant. So it is possible, at least on commercial products.

-branadic-

[niner_007]

I quite like that approach.  Could probably work if you can get a high-enough isolation/low enough capacitance between across a relay.

There are a lot of relay options out there, series connected relays might also be an option

[Kleinstein]

Ideally the reference should not be very sensitive to mains hum an similar. In most cases the circuit where such a reference is used / needed is relatively low impedance and thus should not be effected very much. It is more like that most meters would also inject some signal. In some cases good isolation at the reference could compensate for not so ideal isolation at the meter. The main critical cases I see would be using a KVD or Hammon type divider, as these have relatively high impedance.

So chances are one could get away with a relatively simple mains supply (like classical mains transformer with separated windings and linear regulator to avoid possible surprises from EMI). For the few really critical measurements one could use battery power - this is kind of the standard to compare too.

[Dr. Frank]

A few years ago, I made measurements with 3 different setups, always using a 3458A, NPLC 50 in difference or absolute configuration. I used 3 references, Ref_1 and Ref_2 in one case on same supply, LTZ #5 in separate case and with separate supply:

1. difference between two separate references, with 2 separate mains supplies, about 9mV apart
2. difference between two references in one case, on same mains supply, about 47mV apart
3. absolute measurement on one of the references

For all measurements I then calculated the relative deviation in ppm.
First case made really ugly spikes, up to 100 times bigger than the absolute measurement.
Even the 'quiet' areas between 7h-8h were much noisier compared to the absolute measurement.

LTZ #5 meanwhile runs on batteries, so I might repeat this test.

REF_1 and REF_2 were my 16 years old prototypes, standard LT circuit, LTZ #5 is Andreas design, practically immune to external disturbances.
Absolute measurements on latter references, as well as on the FLUKE 7000 usually show noise of max 0.2ppm_pp only, in 'quiet' areas even half of that

Frank

[dietert1]

Here is todays 24h log of two LTFLU references, difference measurement with HP 3456A. Vertical scale is +/- 0.5 ppm again. Those two references are running continuously on mains, using the old school power supplies as shown above. That setup includes a second HP 3456A to measure temperatures, two unmodified Arroyo temperature controllers (with SMPS) and a HP 59401A isolated GPIB DAC for temperature control. This represents development status of May 2020.

Regards, Dieter

[MK]

The construction details of the GR shielded bridge transformers are interesting- https://www.ietlabs.com/pdf/GR_Experimenters/1935/GenRad_Experimenter_Oct_1935.pdf
You can also shield a toroid with copper tape.
Me, I'd probably use a couple solar cells, lit up with some big incandescent lamp! Can't get much better isolation than that.

I have a couple of double shielded tranformers made by Singer/Gertsch and they are also 1:4.

What is the reason for these double-shielded transformers often being made with a 1:4 turns ratio?

Perhaps that guarantees that the output volts count as "SELV" even with 230V going in?

[MiDi]

Perhaps that guarantees that the output volts count as "SELV" even with 230V going in?

SELV has to be <= 50 VAC or 120 VDC in air and w/o protective earth connection.

[branadic]

Xmas time, reverse engineering time.
LCR meter powered up to measure some inductance, multimeters to measure coil resistance, caliper to get some dimensional measurements and handheld multimeter to measure frequency on the F7000 with LT1533.
Based on that I will try to replicate this mystical Pickering xformer circuitry for some DIY voltage references I have. A spice circuit was already build to prove the reverse engineering was correct.

-branadic-

[dietert1]

What is the coupling capacitance from primary to secondary of the transformer?

Regards, Dieter

[branadic]

Forgot to measure it and when I thought of it, the unit was already put back together. But I can measure it once my replica is done. Boards already designed, need to order them in the next step.

-branadic-

[Andreas]

Hm,

in the paper I read from 2 "shield screens" (primary + secondary) consisting out ouf a polymer.
Where are the shields connected to? Case or Guard?

With best regards

Andreas

[Echo88]

Measured the coupling capacitance of two DIY-Pickering Transformer without shield:

13pF and 21pF

Other normal COTS transformers suitable for the LT1533 have the following coupling capacitance for comparison:

CTX02-13834 (suggested in the LT1533 datasheet): 9pF
WE 314619 (substitute for some other CTX02): 25pF
WE 30804 (substitute for some other CTX02): 86pF

Measured with HP4192A at 1MHz.

In the end the ultimate test is to see wether someone can spot the difference when the powersupply is switched over from battery to DCDC, while measuring the 10V reference in opposition-mode.

[dietert1]

I'd guess the one shown above by fcb will be like 1 pF.
There is nothing mystical about that pickering transformer. The two core idea is the same, except the position of the two cores is different, since it is meant for power transfer. So you don't want one but more than one turn for the coupling coil.
From the images i can see that the DIY Pickering transformers have more distance between the cores than the original one. So the original one will have more coupling capacitance, maybe 30 or 40 pF. And the shields don't really help, since the guard can't absorb high frequency spikes very well.

Regards, Dieter

[branadic]

Andreas, primary shield is tied to primary GND and earth / case, secondary shield is tied to secondary GND.

-branadic-