Valhalla 2720GS

[dietert1]

.. enough to be detected - though there should at least be room for the +-4 µV typical popcorn steps to be inside the accepted range.

In my experience the ADR1399 does not exhibit popcorn noise when used properly.

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.

Regards, Dieter

[maat]

In my experience the ADR1399 does not exhibit popcorn noise when used properly.

I wouldn't go that far in saying, but they are considerably better than the LM399. So far out of the 10 diodes I am testing two show popcorn noise (see attachment, its channel 6 and 9). This is way better than the 2/3 quota of the LM399 though. I will post more details in the ADR1399 thread when I get to 1000 h (500 h so far).

[dietert1]

Apparently there is something wrong with your test. I would not wait another 500 hours before trying to find out what may be wrong.

When i got the  ADR1399KEZ (SMD package) evaluation kit, there were several things wrong. But at least i did not have to think about soldering temperature profiles and the like. And they had the board cutout that everybody thinks is for thermal reasons. Meanwhile i think it is important to avoid mechanical stress. With the KHZ device (TO-46-4 package) it's easy to avoid mechanical stress by bending the four wires in S shape. A reference soldered into a board with straight wires cut to 5 mm will deteriorate. Another one is fixing the thermal cap against movement. Then the RC filter.

Regards, Dieter

[ch_scr]

(...) Once I've replaced them by selected ADR1399 (...)

To automate such selection tasks in general, I've made up a python script. With the example at hand, it looks like this:

Code: [Select]

grouping_optimizer.py 8 6.932 6.939 6.9155 6.8915 6.905 6.909 6.936 6.943 6.9355 6.9355 6.933 6.943
Average value 6.93712
Error sum 3928.14
Average error 491.02 ppm
With these elements and corresponding errors:
(6.932, 6.939, 6.936, 6.943, 6.9355, 6.9355, 6.933, 6.943)
['-738.78', '270.28', '-162.17', '846.89', '-234.25', '-234.25', '-594.63', '846.89']It can also be asked for a specific target value as well:

Code: [Select]

grouping_optimizer.py 8 6.932 6.939 6.9155 6.8915 6.905 6.909 6.936 6.943 6.9355 6.9355 6.933 6.943 --target 6.918
Average value 6.92519
Error sum 15394.62
Average error 1924.33 ppm
With these elements and corresponding errors:
(6.932, 6.9155, 6.905, 6.909, 6.936, 6.9355, 6.9355, 6.933)
['2023.71', '-361.38', '-1879.16', '-1300.95', '2601.91', '2529.63', '2529.63', '2168.26']As you can see, this one works happily with multiple elements of the same value. It can also show the iterations and handle priority elements much like the resistor divider ratio optimizer.
Hope someone finds it useful.

[branadic]

I performed some noise measurements today. Strange, I can see some low frequency hum at 2.3 Hz and 12 Hz, the source of it is yet unknown. Maybe the ovens? The service manual indicates, that in some revision C7 was added across R10 = 10 M. None of my boards has this capacitor installed. Looking at page 27 of the service manual we find 1 µF across the 10 M of the output system oven controller. So maybe this is something I should add to my boards too. Also C8, most likely 100 nF, was added between pin 4 and 8 of IC2 later, which is missing on my revision C? boards.

We can compare the noise to ADR1399 converted AN3200, which is way quieter.

I found that the clock is running a bit too fast. The clock, displayed on the lower display, is ahead by a few minutes every day.

Real Time Clock
The real time clock functions are generated within IC102 (2720-084 or 2720-071). This is a CMOS IC supplied from the supply labelled "+B" on the schematic, this is generated either from the battery or from the +5V supply whichever is higher. This selection is performed by the diodes D2 and 3 with C3 providing a smooth transition during changes in the power supply.
The real time clock is provided with a 32.768KHz reference clock by the crystal Y1 and is adjusted by means of C8 (the 32.768KHz should be monitored between TP1 and TP2 (GND)).

Seems like this is some old explanation, as Y1 a 4 MHz crystal in the schematic, while Y101 is a 32.768 kHz crystal. The test points are nevertheless there, so I will check them next and adjust C105 (not C8) accordingly.
I also did my first GPIB communication with the instrument for the "GCAL" - Get CALibration constant

Code: [Select]

0 through 7 - Reference 1 through 8 voltage at the time of the latest calibration (in volts).
+6.94570030
+6.94534459
+6.93843745
+6.94167089
+6.93890978
+6.93802343
+6.93563261
+6.93402463
8 - Reference Divider ratio (divider used for 0.65V, 1.3V, 650mV and 1300mV ranges). The data in nominally 100% and is the data at the time of the latest 1V external calibration.
+099.987381
9 - Nullmeter scaling coefficient (1.073% nominal)
+000.003951
10 - Nullmeter high range zero offset (1.19% nominal)
+000.004493
11 - External reference scaling coefficient (14.39/100/143.9% nominal for the EXR1/7/10 respectively).

12, 13 - Internal DVM. scaling and zero coefficients respectively (0.119% and 0.085% nominal resp.)
+000.000221
+000.000154
14, 15 - Internal ammeter zero and scaling resp. (0.119% and 0.098% nominal resp.)
+000.000226
+000.000183
16 - Reference averaging coefficient (100% nominal)
+100.000320
17 - 2:1 sense buffer scaling (for 1300mV,1.3V,13V,26V,130V and 1200V ranges) (100% nominal)
+099.998918
18 - 2:1 sense attenuator scaling (for 26V range) (100% nominal)
+100.001274
19, 20 - 10:1 and 100:1 sense attenuator scaling resp. (100% nominal)
+099.993863
+100.003231
21 - Negative polarity DtoA convertor offset (0% nominal)
+000.000000
22 through 32 - 650mV, 1300mV, 0.65V, 1.3V, 6.5V, 13V, 26V, 65V, 130V, 600V and 1200V range zero offsets in voltage units
+0.00000000
+0.00000000
+0.00000000
+0.00000000
+0.00000000
+0.00000000
+0.00000000
+0.00000000
+0.00000000
+0.00000000
+0.00000000
33, 35, 37, 39, 41 - Voltages of reference 1 at the times of previous calibrations (earliest to latest but one resp.)
+0.00000000
+0.00000000
+0.00000000
+0.00000000
+6.95000000
34, 36, 38, 40, 42 - As above for reference 2
+0.00000000
+0.00000000
+0.00000000
+0.00000000
+6.95000000
43, 45, 47 ,49, 51 - As above for reference 3
+0.00000000
+0.00000000
+0.00000000
+0.00000000
+6.95000000
44, 46, 48, 50, 52 - As above for reference 4
+0.00000000
+0.00000000
+0.00000000
+0.00000000
+6.95000000
53, 55, 57, 59, 61 - As above for reference 5
+0.00000000
+0.00000000
+0.00000000
+0.00000000
+6.95000000
54, 56, 58, 60, 62 - As above for reference 6
+0.00000000
+0.00000000
+0.00000000
+0.00000000
+6.95000000
63, 65, 67, 69, 71 - As above for reference 7
+0.00000000
+0.00000000
+0.00000000
+0.00000000
+6.95000000
64, 66, 68, 70, 72 - As above for reference 8
+0.00000000
+0.00000000
+0.00000000
+0.00000000
+6.95000000
73 through 77 - Reference Divider ratios at the time of previous calibrations (earliest through latest but one resp.)
+000.000000
+000.000000
+000.000000
+000.000000
+100.000000
78 - User set Nullmeter zero offset (by SET ZERO command) in volts.
+0.00000000
79 through 86 - Presently measured reference voltages (as obtained by the OBRMS system) in volts
+6.94570657
+6.94533921
+6.93843947
+6.94166848
+6.93891909
+6.93802837
+6.93563030
+6.93401944
87 - Presently measured reference averaging system error (as obtained by the OBRMS system) in %
+100.000307


-branadic-

[branadic]

By the way, does anyone have one of the options, such as EXR1/7/10? Would be nice to get some images and the NOVRAM content... yes, another one

-branadic-

[branadic]

This morning I've added 1 µF X7R 0805 caps between pin 1 and 2 of the corresponding opamp on all of the boards and yes, the oscillation vanished. So definitively something to add, if not already populated.

-branadic-

[PrecisionFreak]

Very very good, branadic!  To adjust the RTC real time clock: variable capacitor C105 on page 9 of 99.
Thank you for the GCAL values.  BTW, I found three missing GPIB-commands not mentioned in the manual:
string $9672 4 ; "SRNG"   set range?  set random generator?
string $96FC 4 ; "GSEP"   get separator?
string $9703 4 ; "GTRM"   get terminator?

About storing voltages in NOVRAM:  results from https://www.omnicalculator.com/other/floating-point
(the results are not that exact !?!  64bit should be 401BCCCCCCCCCCCD, not 401BCCCCCCCCCC00)
the hex representation of e.g. 6.95 is:
Single-precision (32-bit) floating point
6.95 converted to 32-bit floating point is
01000000110111100110011001100110 (40DE6666H) with:
Sign: 0
Exponent: 10000001
Fraction: 10111100110011001100110
The real value of this representation is 6.94999980926513672.
The machine error resulting from this conversion is 1.90734863281250010587911840679e-7.

Double-precision (64-bit) floating point
6.95 converted to 64-bit floating point is
0100000000011011110011001100110011001100110011001100110011001101 (401BCCCCCCCCCC00H) with:
Sign: 0
Exponent: 10000000001
Fraction: 1011110011001100110011001100110011001100110011001101
The real value of this representation is 6.95000000000000018.
The machine error resulting from this conversion is 1.77635683940025046467781066895e-16.

Still missing is the 6byte (48bit) representation as used in "real"-type in turbo pascal.
https://moddingwiki.shikadi.net/wiki/Turbo_Pascal_Real  (they count the 8 bits from 0 (MSB) to 7 (LSB), not from 7 to 0 as usual).

[branadic]

I today adjusted the 32.768 kHz of the RTC using our 12digit, GPSDO driven frequency counter, but the clock still runs too fast.
Even if I adjust the frequency to the lower adjustment range the clock runs too fast, which makes me belive that during the instrument being turned on, the clock is driven from the 4 MHz not from the 32.768 kHz. Could that be?

-branadic-

[PrecisionFreak]

You mean, after turning on, the µC reads the RTC and then the clock runs software-driven?
The same happens in every? PC.  You could be right.  There are timer-driven interrupts.
What happens if you read the clock via GPIB (GCLK  get clock), which one is asked, hardware or software clock?
Will the RTC be re-read after a Reset?   37   RESE   7.6.1.1   Perform Device Clear

> The real time clock is provided with a 32.768KHz reference clock
> by the crystal Y1 and is adjusted by means of C8 (the 32.768KHz
> should be monitored between TP1 and TP2 (GND)).

> The microprocessor obtains its clock from an internal oscillator
> controlled by the 4MHz crystal Y1 and the capacitors C2 and C3.
Change the values of C2 / C3?

> 11.5.1.3 Interrupt Routines
> There are two sources of interrupt to the I/O uP in the 2720Gs.
> These are :
> i) Real Time Clock Interrupt
> This interrupt is generated by the real time clock and occurs
> every 500uS (approx.). The uP uses this interrupt to multiplex
> the displays and also for timing functions. The front panel
> RESET key is actioned within this routine also.

Ah, IC102/19 (MC146818) generates the IRQ of IC10/5??  (MPU MC6803G)
This means  the 32768 Hz Xtal is the source of RT
Please measure this frequency (2048 Hz?)
Datasheet of MC146818 can be found at page 3-697 (741 of 876):
http://www.bitsavers.org/components/motorola/_dataBooks/1983_Motorola_8-Bit_Microprocessor_and_Peripheral_Data.pdf
Generated interrupts can have various sources.  See page 3-712 (756 of 876).

[branadic]

From what I can observe the output noise is mostly dominated by varying low frequencies of the oven controls, not by the noise of the references, the DAC or the output stage, which is limiting the performany. Every now and then one of the controls kick in and ruin the output voltage with low frequency oscillations. Not sure how to improve on that. Thermal insulation?

-branadic-

[alm]

From what I can observe the output noise is mostly dominated by varying low frequencies of the oven controls, not by the noise of the references, the DAC or the output stage, which is limiting the performany. Every now and then one of the controls kick in and ruin the output voltage with low frequency oscillations. Not sure how to improve on that. Thermal insulation?

Do you know if the interference is thermal (temperature fluctuations) or electric (e.g. the oven controller modulating the voltage rails)?

[dietert1]

As far as i understand the original circuit would work in on-off mode like a common refrigerator. Adding a capacitor into it makes it an I-controller. In general it would still not be stable until you also add a P-term for damping. Only then you get the desired PI-controller. I posted a schematic here that contains a simple recipe.
https://www.eevblog.com/forum/metrology/thermal-fluctuations-on-pcb/?action=dlattach;attach=1719803

To arrive at a good solution one can test and tune the dynamics injecting some test signal (external error signal). E.g. a rectangular wave with a period of one or two  minutes. Then you can see whether the oven controller has critical damping. As an example the image shows such test results for a 50 Hz digital PLL that i implemented for an integrating multimeter. Here the lock phase changed every 20 seconds.

Regards, Dieter

[branadic]

Do you know if the interference is thermal (temperature fluctuations) or electric (e.g. the oven controller modulating the voltage rails)?

I simply can't answer the question at the moment, but have attached measurements at different times, showing all sorts of situations. The instrument is sitting in a temperature controlled lab (23 ± 1°C).
I always wondered, why the instrument is specified with 30 µVRMS wideband noise (10Hz to 10 kHz) in 13 V range, no matter what version and how many reference boards you have (2720GS/54-4T, 2720GS, 2720GS/HSR). First I thought the noise limit is coming from the amp, but the schematics doesn't lead to the assumption that the noise is that high. If on the other hand the limits are given by the temperature variation, causing way larger output voltage variation that could explain the specs.
One thing I could do is adding some non-conductive foam to the ADR1399s inside the oven and see what happens. That is not too much of an effort, as I'm quite busy at the moment.

As far as i understand the original circuit would work in on-off mode like a common refrigerator. Adding a capacitor into it makes it an I-controller. In general it would still not be stable until you also add a P-term for damping. Only then you get the desired PI-controller.

As you can see from the images below, we have all sorts of situations, so in case you are right about that I wouldn't expect situations with "low" noise in between, or would I? 
I only added, what was given as a modification on board revision B in the service manual, so the missing 1 µF as well as 100 nF at the supply rails.

-branadic-

[branadic]

First INL measurement wasn't necessarily successful as instrument switched between ranges
Need to figure out how I can force the instrument to stay within the very same range. Someone already suggested to test setting the output voltage and use INCR command instead. I will give that a try.

-branadic-

[branadic]

Seems like even ADR1399 can exhibit popcorn noise. During noise measurements I found, that one of the references installed to one of the reference boards, is jumpy.

-branadic-

[branadic]

For a few days I was now monitoring the repaired, upgraded and converted 2720GS with 3458A and want to share first results. The 3458A is continuously powered and performs ACAL each morning auto-magically.
We can clearly see, that it took the unit quite some time to stabilize after I performed some noise measurements with the lid of the instrument removed and the LNA directly connected to the reference boards (2x5 pin header at the edge of each reference board).

-branadic-

[alm]

Thanks for showing the data. Looks like it's been stable within roughly 0.5 uV/V over the past days?

Could you plot voltage against temperature using whichever temperature sensor works better, excluding the part where you did the noise measurements? That might give a decent measure of temperature coefficient (of the entire setup) if other factors like drift aren't too large. If drift is a factor, then do it over only the first days when there were more temperature fluctuations.

[branadic]

Not at the moment, I would like to observe the reference a bit longer to separate drift before.

-branadic-

[branadic]

Here is a first update with 670 h of stability data measuring the 2720GS/HSR-ADR outputting 10 V with a 3458A.
At some point I was brave enough to hit the internal cal button of the 2720, which lead to the jump of the output after about 260 h. Before that a slight increase in output voltage was already visable. That's also about the time when we had some experiements running in our lab, which lead to increased lab temperature during the day, making it hard to barely see anything as the t.c. of the whole setup was hiding additional details. However, while these experiments are now over and the lab is getting quieter we can again see some slight increase of the output voltage.

-branadic-

[dietert1]

As far as i remember you put ADR1399 references into that calibrator. Then a drift of 1 or 2 ppm over the first 600 hours isn't that unusual. There seems to be some correlation with the temperature peaks though, more apparent after the calibration. Might be a problem of the meter. I mean one would better compare to another calibrator.

Regards, Dieter

[branadic]

Yes I did put ADR1399 into the unit. I'm just reporting results since these were LT marked ones. To be honest, I expected way more drift in the first 600 h, but also opposite direction.

-branadic-

[Kosmic]

I got a package from Germany 

The transformer crossed the Atlantic and survived. Still smelling fresh varnish!



Thanks branadic for all your efforts, much appreciated!

[Kosmic]

As for my 2720GS, before I install the transformer, I want to backup all the X2212 on the reference boards. I'm currently working on a programmer based around a cheap STM32F103. More info later.

[PrecisionFreak]

Hi Kosmic, please be careful with the colors of the wires, the isolation hoses of 16V of the replacement are green (gr for german grün), but the 16V wires of the original trafo are blue and vicy versa.  See https://www.eevblog.com/forum/metrology/valhalla-2720gs/msg5034961/#msg5034961
After lot of dicussion with Langer, they agreed to change this to the correct colors in future orders.
Also the three shield windings have the same color, but the one which shields the primary winding should be connected to chassis (PE), but the other two go to board Pins. How to distinguish?  Langer did not understand this.

Also I promised to give tips for the secret password: The numbers are in eeprom adress range 07xx of the front board (TMS27C256-FrontPanel-2720_V61.1_8_18_88.BIN). Remember the other tips! The first person to post the secret will be the champ of the day.