Valhalla 2720GS

[maxwell3e10]

Well, this makes it easy: 25 60 02 12 58

[Kleinstein]

A capacitance measurement should tell which shield is close to which winding: e.g. the shield to PE should have the largest capacitance to the primaries.

[branadic]

The primary winding is the inner one, so you can tell which of the screens is the inner one too (the closest to the core), which has to be connected to earth. The other two screens are physically tied together on the mainboard, so it doesn't make a difference to which solder point you are connecting them.

-branadic-

[branadic]

Here are results of my Valhalla2720GS-HSR-ADR1399 running for almost 2200 h.
One might be suprised by the bad behavior, but that can be explained. The measurements suffered from EMI of what I called a scream cube (a 1 m³ fine particle filter that was running near one of our test benches) nearby. This scream cube was switched off 11 days ago and we can see how temperature in the lab stabilized and output noise went down to a reasonable level since then. We can also see some jumps that resulted after performing internal cals of the Valhalla.
Since the scream cube will not be turned on again, the related project has ended, I guess it makes sense to repeat the noise measurements, but also to further observe how stability turns out from here.

-branadic-

[branadic]

An update of my Valhalla 2720GS/HSR-ADR1399. I'm approaching 5000 h of operation after my repair, modifications and upgrade. Since removing the "scream cube" from the lab the extreme spikes disappeared and never showed up again by now.
At some point I had to readjust the oven settings of my copycat reference board 4, as the reference board was continiously turned off by the unit due to an "oven high" failure. The components and thus the voltages had drifted quite a bit. Since I made the adjustment the board was never turned off again.
What I observe are some large intermittent deviations, which I think are due to the units internal intercomparison of the references, as they appear at least at around 12:00 pm. I can tell, as ACAL of 3458A happens every day at 8:00 am, hence the gap in the data.
The location of the equipment near the AC in the middle of the room is far from ideal and the south side of the lab with its mirrored glass window front isn't perfect either, but it is how it is. There is clearly some temperature depence that needs to be ruled out.

I also received data of an INL measurement of another 2720GS against a Datron 1271 from some fellow voltnut that I attach. Again, results are disappointing given the claims Valhalla made: "The 2720GS delivers the performance required to calibrate the world's most accurate 7.5 and 8.5 digit DVM's yet does even more."

-branadic-

[branadic]

As already stated, I replaced all 6x LM399 references in my unit and also equiped my copycat reference board with them too, so a total of 8x ADR1399. Unfortunately, noise of the instrument didn't improve as I have hoped for. So I looked into the schematics https://xdevs.com/doc/Valhalla/2720GS/Valhalla_Scientific_2720GS_Vol.%202_sm.pdf and noticed IC202, IC315 and IC4 in the chain, all three of them being ICL7652CPD. So I today performed the exercise of removing them one after the other, replaced them with a socket and fitted a single AD4523-1 in each position, with the following outcome:

IC202: noise didn't change at all
IC315: noise went up by a factor of ~2.5
IC4: noise went up by a factor of ~10

The instrument is now back on ICL7652 and noise is back where I started (8-9 µVpp at 10 V output, 0.1 - 10 Hz). The only thing that was successful today, while I was at it I tightened the Cal keyswitch.

-branadic-

[Kleinstein]

The ADA4523 is a rather extreme case: low voltage noise, but really high current noise and thus not suitable for more than some 10 K of source impedance. In addition the ADA4523 may be more picky about supply decoupling and a bootstrapped supply and has a slight chance to oscillate. The higher speed could also be an issue. A more moderate replacement would be a LTC2057 or AD8638.

The ICL7652 are not very low noise by todays standard, but should still be way better than 8 µV_pp for a 0.1-10 Hz band. It is still not clear how the noise was measured and especially the upper BW limit.


A point to check could be the noise with different voltage settings. E.g. with the output set to 0 V still is the upper range should not include ref. noise. One could also look at the DAC output directly, without the output amplifier/buffer.

[dietert1]

You may have a look here once more: https://www.eevblog.com/forum/metrology/valhalla-2720gs/msg5040739/?topicseen#msg5040739.

Regards, Dieter

[branadic]

It is still not clear how the noise was measured and especially the upper BW limit.

Output set to 10 V, measured with a 0.1 - 10 Hz LNA, 80 dB amplification by pipelie.

You may have a look here once more: https://www.eevblog.com/forum/metrology/valhalla-2720gs/msg5040739/?topicseen#msg5040739.

Thanks, that is something I could look at.

-branadic-

[branadic]

When looking at the Valhalla schematic i found the 10 KOhm averaging resistors bad. In my LM399 arrays i have been using 274R resistors and i think one should replace the 10K resistors by 1K. I think they used 10K since it is a 2-wire circuit, but one doesn't need full precision for the averaging coefficients.

Did that, but without a change in terms of noise at the 10 V output (13 V range that is). Haven't yet checked the 6.5 V range, but this sort of proofs that the noise is generated somewhere in the chain after the references have been averaged.

-branadic-

[chuckb]

The data sheet for the ICL7652 lists the current noise at 0.004 pa / rt Hz. This number is most likely not a measurement. It may be a shot noise calculation based on Bias current.
The LTC2057 looks like a good replacement with 0.2 uVpp noise. I have not calulated the output noise from the opamp current noise of 0.12pa / rt Hz. But the LTC2057 has the lowest current noise, of the low voltage noise chopper opamps.
The ouput noise could be from residual PWM current spike noise or another grounding issue. My latest project took several revisions to get the grounding correct. Just going from memory I believe the Datum 4910 PWM filters had more sections.
Good luck!

[Kleinstein]

There are OP-amps with less current noise than the LTC2057. It still has a good combination of current and voltage noise, especially among the higher voltage ones.
Chances are the OPA186 should be lower current noise (at least that is what the data-sheet says), but is also higher voltage noise.
The ICL7652 should be low current noise, maybe not all the way to 4 fA/sqrt(Hz), but it may not be that much worse.
Current noise specs with AZ OP-amps are anyway not very reliable: quite some false ones, calculated from the bias current and even the real value can depend on the conditions of use. There is a good chance to have correlation in the current noise for the 2 inputs - so a single number is not enough to describe it fully and there are different ways (resistor at one side of both) to have the resistance to test.

The reported noise still looks higher than expected for the ICL7652 - so there are likely other sources.
There are different stages that could be followed to see if the higher than expected noise is alread there at the reference, the PWM DAC or only at the output stage.  Ripple in addition to noise a possibilty.

[branadic]

The 0.1 - 10 Hz at the 6.5 V output appears to be similar to the noise of the 10 V output (13 V range). So one of the things to look at is the noise of the averaged reference voltage.
Someone suggested to stop the DAC by pulling the jumper PL201 connected to the 16.777216 MHz clock at the reference generator.

-branadic-

[branadic]

Here is an update:
At some point I stopped my long-term observation (LongtermStability.png) and adjusted all oven temperatures down to 45 °C. From there, I started a new observation (LongtermStability2.png). The huge peaks are from me entering the unit and adjusting the "oven hot" alarm levels.

-branadic-

[branadic]

Attached an update of the Valhalla 2720GS/HSR-ADR1399. After adjusting all the 5 internal ovens to 45 °C and over temperature alarms the unit is now rather stable and almost all of the intermittent peaks have now vanished too.

-branadic-

[faraday]

There is 5x this units in France for sale. I would really like to have one of this

https://www.ebay.co.uk/itm/115950380650?_skw=valhalla+scientific&itmmeta=01JKV8NA368CDP26CZHM4B0DD0&hash=item1aff2e666a:g:h9wAAOSwYUJlNXJ1&itmprp=enc%3AAQAJAAAA8HoV3kP08IDx%2BKZ9MfhVJKmhe1gBMu%2FHftxww9HDX14R4qJ3gJK1UazCwIXR3T%2BQYJiByCrL8Y1WVg9couy0tFVDxPbkRRDMZociZaS4jLRA2BsgJ03pb1w%2BN5X5deX0N%2Bye6FnryY%2Fszv8i4H1rSFWazpM%2F9roSrAyvCUV%2Faw%2Fc4DwfSicPrZgu16FIL7%2BbBleDBWyPZjGv9ISNBhxueGcn153N%2Fa0Zp2LVDDeyDJZtbnSlVY03tttUhRYbvlJAhFwU0xzbNnK4zbcMZYNSCcBFk%2Ba8mg4yZP6hoHTDiI6JPxT0EFzapFcW%2BMqcGx1scw%3D%3D%7Ctkp%3ABk9SR9yh1eieZQ

[picburner]

They have been for sale for quite some time and the seller has not yet been able to sell them.
The problem is only two of them seem complete (from what I can see in the photos), the other three are missing pieces.
Before the seller asked 1500 eur for shipping, now he ships for free...

[branadic]

The offer is totally overpriced and the seller does not accept lowball offers.
In best case, you can make 2 working units with what is present. One HV power supply board is missing to increase the number to 3, but it contains a special toroidal transformer, so nothing you can replicate easily.

It's sad, but unless the seller asks for a real market value I think they will stay with him.

-branadic-

[dietert1]

It's hard to tell from branadics last post, but the stability appears to be marginally sub-ppm (observed sigma maybe 3 uV in 10 V or so). Considering the amount of work he put into the device, that's somewhat disappointing. But i neither have that Valhalla calibrator nor a Fluke 57xx to get a serious comparison.

Regards, Dieter

[ZhuraYuk]

Is it possible to determine the internal resistor structure of the RN301 proprietary resistor divider? It is located at main microprocessor board.

[branadic]

Is it possible to determine the internal resistor structure of the RN301 proprietary resistor divider? It is located at main microprocessor board.

At least not for me, I don't have plans to disassemble the unit just for that at any time.

I was triggered by someone and thus looked into the filter section of the unit, as noise still is unsatisfyingly large. While I haven't measured to spectral content just yet, I've simulated the 3rd order filter as given by the manual in LTspice and compared it to the 5th order filter used in 5700A [1] and the 3rd order filter that Macaba designed [2]

[1] https://youtu.be/SPz6u9fxJgc?t=69
[2] https://github.com/macaba/TheManhattanToolkit/tree/main/source/application%20specific/LPF-A-v1

So it seems to me that there is potential to improve the filter section in the unit as filter response isn't that great at all.

-branadic-

[branadic]

I managed to measure the spectral content with the output set to 10 V using a MSO46 in high resolution mode (16 bit). Here is what I got.

-branadic-

[branadic]

My 2720GS stopped working. 
When I recently entered the lab the output was still on and outputting 10 V, but the display showed no actual time anymore "00:00:00". A reset of the unit stopped with the message "Setting Defaults", after showing the firmware version 61.61, but other than that, nothing. 
However, this already points towards the possible culprit.
First I've checked the voltages of the low power supply, but they all appeared to be fine and within spec. Next thing to do was to turn on switch 2 on the back of the unit, to read back factory configurations, but that didn't work either and the unit still was stuck at "Setting Defaults"
Okay, let's check what the manual has to say about that:

"SETTING DEFAULTS"
Indicates that non-volatile battery maintained data has been lost. If occurs often, then replace FRONT PANEL ASSEMBLY after checking rear-panel configuration switch (see 5.1.2.2)

5.1.2.2 Section 2
When this section is in the on (down) position, the 2720GS will completely initialize the battery backed up data (except for the real time clock) whenever the 2720GS is reset. The user should note that it is not recommended that the user leave this switch in the on position as the 2720GS will “forget" any faults detected if this switch is on when the power is re-applied or the 2720GS is reset.

Time to take a look at the front panel assembly.
First thing to note is, the schematics in the manuals that are around show a 128k and a 64k EEPROM, while the unit I have has a single 256k EEPROM, so they must have changed that at some point.
A check of the content of all EEPROMs showed no difference to the backups I have taken some years ago. I then checked the 32.768 kHz real-time clock with the unit turned off, an almost perfect fit. But when I switched on the unit, the frequency shifts by about +200 Hz and shows huge variation. Okay, that is weird. 
The RTC chip in this unit is a MC146818AP, a real-time clock plus RAM and that is where I'm at right now, most likely that chip is the bad. Unfortunately, I don't have a replacement for that at hand, but already ordered some NOS.
Fingers crossed they arrive soon, because while I was in the unit I've replaced C308 and C309, 100 nF/50V foil caps in the 3. order PWM low pass filter, by 220 nF/50V C0G/NP0 ones in order to improve noise. I want to measure if that modification made any difference and hence why I have taken the spectrum of the 10 V output in my former post.

-branadic-