Scanner/Multiplexers for voltage references

[branadic]

Hi folks,

multiplexing between different voltage references is neat for automated test setups.
So while some of you have already build the DIY Low Thermal EMF Switches, others owned an expensive Dataproof scanner such as Model 160, 164 or 320, I tested my Prema 2080, which can be configured for 80x common ground, 40x 2W measurement or 20x 4W measurement, for thermal emf.

Therefore, I shorted two channels on one of the D-sub50 connectors on the back, connected K2182A to its output and switched between both channels. The connector on the back was -  metrology-like - covered with a towel and a Lemo - PTFE cable - Stäubli SLS4-B connector 22.2642-* connected scanner and nanovoltmeter.

Turns out the emf measured is <<100 nV and thus good enough for measurements on voltage references in the typical 10V range, even with a 8.5-digit meter. The scanner uses G6AK-234P-ST-US 12V relay. I received the full schematics of the scanner on request at Prema, which is basically used the same way in Prema 5017 SC.

So here you have it, even though it is specified to have 1 µV typ. and 2 µV max after 1.5h warmup only, it performs much better than expected.

What is your solution?

-branadic-

[TiN]

Not bad, thanks for data.

How was TEMF measured? Was is just 2182A measure scanner output with direct cable, and select CH1 or CH2 shorted inputs? How short was made?

Perhaps more details on method, to allow reproduction by other scanners?

[Mickle T.]

Prema Type 2000 10x4W scanner have about the same 100-200 nV TEMF offset.

[branadic]

How was TEMF measured? Was is just 2182A measure scanner output with direct cable, and select CH1 or CH2 shorted inputs? How short was made?

As I said, I made a D-sub50 connecter with a short on channel 1 and 2 using copper wire and standard leaded solder, so nothing special or low thermal. The setup was warmin up overnight and the connector covered with a towel.
K2182A was read via serial interface and GNU Ocatve, while P2080 was switched manually. Prior to measurement ACAL on K2182A was performed and measurement done using Rel button, while Ch2 was active on P2080. Hope that answers your question?

Prema Type 2000 10x4W scanner have about the same 100-200 nV TEMF offset.

Ah, thanks MickleT., is that the scanner with the fancy unobtanium relay you showed some years ago in another thread?

-branadic-

[TiN]

Abovementioned Data Proof 160A results. It is version without silly connectors on the back, just pure copper single-strand cables come out to the DUT.



Data points collected with Keysight 34420A nVM on 1mV range CH1, NPLC100. nVM measure A-B HI difference, while LO is shorted with copper bar (just like photo shows)



Pretty happy with results. I might replace relay board that house CH5,CH6,CH7,CH8 relays with spare board, to rectify higher TEMF for these channels.

But even bad CH8 is still within maximum vendor spec, just a tiny bit under 35nV. 35nV error in case of 10V DC voltage measurement error would translate into 0.0035 ppm, which can be ignored for all practical reasons.

Looks inside, with massive metal lid removed from isothermal box for clarity:



Latching relay card:



Rest of teardown photos are in ancient ManateeMafia's article.

[dietert1]

Your scanner has six cables in one photo and eight in the other one. What happened?
I was wondering how the cables enter into the temperature protected inner cabinet. There seem to be some centimeters of cable between outside and inside plus some kind of thermal clamp.
Could you indicate which one was the "bad" relay you want to improve, i mean the position in the inner cabinet?
As far as i understand there is something like an analog bus to collect the output signal, with card edge connectors in between. If that is correct, your < 10 nV results are wonderful.

Regards, Dieter

[Echo88]

The card edge connector is only used for the control/protection signals.
The cables used to connect the references are going directly from the outside through the insulation foam into the internal housing, soldered to the visible big vias on top of the cards.

https://doc.xdevs.com/doc/Data_proof/img/rly_top.jpg
https://www.eevblog.com/forum/metrology/teardown-data-proof-160a/

[niner_007]

I looked through and bought some research papers on the subject, and seems a lot of people had great success with motor controlled rotary switches, resulting in ultra low EMF, I don't have the papers in front of me, but it was 40nV or much less in their measurements, IIRC.

The setup was basically an Electroswitch series C4 rotary switch in a dual pole, multiple throw configuration, being rotated by a stepper motor. The setup is fairly simple, and the switch is fairly inexpensive. Something like C4D0206N-A, gives you a straight two 1:6 muxes, just perfect for reference measurements.

[branadic]

Even though the Prema2080 contributes <600 nV absolute offset from input to output it is still good enough to be used for voltage references, as the difference between two channels seems to be <100 nV. So the 40x 2W channels available on the unit configured as two channels for one reference with polarity reversal still results in 20 channels available in total and offset of the setup of <50 nV. I call that good enough, even though I have only measured channel 1 and 2 by now.

-branadic-

[e61_phil]

Here is my DIY multiplexer. It uses Panasonic relays. The performance seems to be in between a real Dataproof and the Prema.

Keithley 182 was shorted in front of the multiplexer and after that connected to the multiplexer.

[TiN]

Which relays? Maybe post some more details

Replaced Data Proof relays in channel 5,6,7,8 and restarted zero test. This time we catch warm-up time.
It took about 8 hours to fully settle under typical +/-20nV limit.

[dietert1]

A more contemporary method would be using a little ADS1256A ADC with integrated multiplexer. Board with reference is € 20 at amazon.
If you look up the datasheet page 8, offset drift with PGA 64 (measuring small voltage differences) is within +/- 15 nV/K. Input referred noise of the PGA is 27 nV (chopper).
Today i got two such ADC modules to monitor a 5x LM399 reference setup at 7 V and to monitor some other references at 10 V. Will report later.

Regards, Dieter

[branadic]

A more contemporary method would be using a little ADS1256A ADC with integrated multiplexer. Board with reference is € 20 at amazon.

Not sure how that is a contemporary method, unless you mean buying stuff off Amazon, instead of using what you already have at hand. 

I've spent some cotton-wool underneath and above the relay card in Prema 2080 and repeated the measurement.

First 500s: both Stäubli SLS4-B connectors shorted at one of the P2080 posts
500s - 1000s: both connectors connected to the output of P2080 and switched to shorted channel 1, which is an imbalance in thermal mass
1000s - 1500s: channel 2
1500s - 2000s: channel 1
2000s - 2500s: channel 2
2500s - 3500s: MULTI-CONTACT 4mm short at the ouput of P2080, which added a large thermal mass

Lower diagram is zoomed in, the difference between the channels decreased, I'm happy with that.

-branadic-

[dietert1]

If you have a museum, that's fine with me. And if your pieces are working (more or less), even better.
Yet somebody should mention how those things would be done nowadays. For me a solution that is available as a regular product is more interesting than using 30 year old surplus.

Regards, Dieter

[e61_phil]

Here is some more 30+ years old equipment

I played around with my HP 3488A switch unit. No low thermal cables, just Hirschmann stuff not even gold plated..

The Keithley 182 was shorted before connecting it to the multiplexer. Therefore, the measurements show the absolute voltages not just channel differences.

Channel was switched to the next every 120s. One measurement point is the average of 10 measurements (100ms integration time)

[branadic]

If you have a museum, that's fine with me. And if your pieces are working (more or less), even better.
Yet somebody should mention how those things would be done nowadays. For me a solution that is available as a regular product is more interesting than using 30 year old surplus.

Regards, Dieter

I wouldn't call a Prema 2080 a museum device and on the other hand you could have just kindly asked for the schematics, if you want to rebuild it. It uses components that are easy to buy at any distributor and a lower voltage relay version with some "contemporary" low volt controller like Arduino (in the past people simply programmed their microcontrollers themself instead of clicking stuff together, not sure if I would call that an improvement but obviously that is contemporary), Nucleo board or whatnot can be used.

-branadic-

[Kleinstein]

A more contemporary method would be using a little ADS1256A ADC with integrated multiplexer. Board with reference is € 20 at amazon.
If you look up the datasheet page 8, offset drift with PGA 64 (measuring small voltage differences) is within +/- 15 nV/K. Input referred noise of the PGA is 27 nV (chopper).
Today i got two such ADC modules to monitor a 5x LM399 reference setup at 7 V and to monitor some other references at 10 V. Will report later.

Regards, Dieter

The ADS1256 is limited to voltage inside the supply (5 V max). So the multiple inputs on a ADC chip are not a full replecement for a multiplexer for 7 or 10 V sources. It can work for things like the differenes, but keep in mind the limited voltage range.

[dietert1]

Built several some multiplexers with bistable relays, that's boring. The results are in my contributions to the precision resistor TC thread. In my opinion precision measurements of voltage references are always measuring small difference voltages between voltage references. Measuring a reference with a high resolution DMM fully includes the ADC errors (bad!). If you think measuring voltage differences is difficult, i can tell you it is easy, once you understand the isolation and guard status of your equipment. Even Mr. Frank recently got nice results using the difference method.

The ADS1256 was made for measuring thermocouples. It's something like a nanovoltmeter with a low thermal EMF input multiplexer. Depending on ambient setup it reaches about 10 ppm accuracy and is good enough for monitoring the five LM399s with an accuracy of about 100 to 200 nV. Resolution will be ten times better. I ordered some MAXIM SPI isolators for it, so with a guarded isolated supply it can be used like a DMM. There is tested firmware since i used the part before to implement a linear TEC controller.

By the way:
The NASA Perseverance robot that recently arrived on Mars has an electric energy generator that produces up to 110 W by thermal EMF (Seebeck). The temperature difference stems from 4.8 kg of Plutonium 238, half life about 8 years.

Regards, Dieter

PS: In order to measure larger voltages one can use a differential voltmeter setup. I recently made a PWM divider for our 5x LM399 reference. That is already good for a fraction of a ppm and still under development.

[branadic]

In the last days I've set up an automated measurement measuring Prema2080 with K2182A, switching through all channels (1 to 40) and acquired quite some data.

Figure P2080-TEMF3: The relay card was fully covered in cotton wool, initial zero was done with a short at the scanner output, afterwards channel 1 (very left) to 20 were monitored (1000 samples per channel / 500s), at around minute 180 I connected the short again at the scanner output, then measured channel 21 to 40 (very right) and finally connected short at the scanner output again.

I then removed all of the cotton wool at the component side of the board.

Figure P2080-TEMF4: The initial zeroing was done with scanner switched to channel 1, each channel was monitored over 1000 samples (500s) before switching to the next channel and this for ten complete runs.
As can be seen there is quite some overall distribution and difference between the first 20 channels (connector X2) and the second 20 channel (connector X3). It can also be extracted that some channels are close to each other, so that they could potentially be used for reversal measurements of voltage references without contribution of large errors.
We can further extract that there is obviously some thermal gradient, so that improving thermals could improve overall behaviour. At least for channel 21 - 40 this can be directly correlated with the position of the relays inside the scanner, while on channel 1 to 20 it's less obvious.

Is someone experienced? Would a copper plate covering all relays improve thermal gradients significantly and thus TEMF?
Just for clarification, the scanner is fully within spec (typ. 1 µV, max 2 µV after 1.5 h warmup time), but I just wonder if it can be made behaving a bit better.

-branadic-

[Echo88]

You could design a pcb(70µ copperplane would be good) with relay cutouts, that is laid on the bottomside of the existing relay-pcb and then thermally couple it to the relays with Vishay THJP for example...but i doubt it will be yielding the same results as a design where theres no heat producing stuff next to the relays.

[niner_007]

I wonder what if you heat the relays with a heater substrate, or keep them relays in an oil bath.

[Andreas]

Is someone experienced? Would a copper plate covering all relays improve thermal gradients significantly and thus TEMF?

I fear no. In my experiments with 1.5 mm thick Aluminium sheets I have seen quite large temperature differences within few cm.

The first thing that I would do is looking with a thermal camera where the gradients are. (If I had one).
I fear that the processor, the voltage regulator and the bridge rectifier produce a lot of heat.
There also seems to be an old 74LS device on the PCB which probably could be replaced by a CMOS device.
In my scanner the average current consumption of the whole cirquit is well below 1 mA.
So I would try to replace the heat generating parts. (or at least remove the voltage regulator from the PCB).

with best regards

Andreas

[branadic]

There is at least one possible solution, that I will give a try. Between board and metal sheet is a distance of 8 mm. Filling this gap with cotton wool didn't much. So I will install a 6mm aluminum plate with the shape of the board and put some 2mm thermally conductive, but electrically insulating silicone rubber pad (4 kV/mm) in between. I hope to decrease thermal gradients this way a bit better.

-branadic-

[branadic]

Meanwhile, I made an aluminum plate, installed it to the scanner and the silicone pads arrived, which are sandwiched between aluminum plate and relay card. I also repeated the measurement relative to channel 1. The picture slightly changed, which proofs that the scanner has a thermal issue, but the result is still not satisfying.

-branadic-

[dietert1]

The datasheet of those bistable OMRON relays https://www.mouser.de/ProductDetail/Omron-Electronics/G6AK-234P-ST-US-DC12/?qs=CX134%252BdLMDH9uKRlS2Zpog%3D%3D does not contain the terms "thermal" nor "EMF" nor "Seebeck".
It does contain though a specification for "Min. permissible load". Somewhat surprising for a gold plated contact, but it may be unreliable for dry switching.
Can you check that?

Regards, Dieter