Valhalla 2720GS

[maxwell3e10]

One can also do a calibration, that would change some string of numbers. From what I recall, it saves the last several calibrations together with dates. I am not sure if there is a way to perform a calibration on only a single reference and then do it several times to replace all numbers except presumably the serial number.

[branadic]

Branadic, do you still have  means to program the NOVRAMs?

No I don't, unless I get the TL866-II programmer running it.
https://www.mikrocontroller.net/topic/556243#7448143

One can also do a calibration, that would change some string of numbers. From what I recall, it saves the last several calibrations together with dates. I am not sure if there is a way to perform a calibration on only a single reference and then do it several times to replace all numbers except presumably the serial number.

I will try different things this weekend.

-branadic-

[PrecisionFreak]

Another possibility would be a permanently connected arduino (Mega256 with more i/o), that resets the processor 68A09 and takes over the bus. So you could change NVRAM contents very quickly  and then release the bus to give back control to the µP.  But take care of different Gnd-Levels, a battery-operated laptop would be best.

[dietert1]

There is a service manual online and it has a schematic "Reference Module" (page 40 out of 95) that shows the permutation of the X22C12 address bits. PrecisionFreak already mentioned unusual addresses. After simulating the permutation, branadics three memory dumps looked less random.

Apart from the fixed header there are subrecords (probably calibration numbers) tagged with 0014, 00E0, 00E8 and FFFF (tail). The subrecords are at the same position and have the same length in all three dumps. The serial number should be in the last subrecord tagged FFFF:
FFFF 0000 FFFF 0000 FFFF,
FFFF 0000 F7FF 0C00 FFFB and
FFFF 040C F7FF 0C00 F7F7.
An educated guess may give you a fourth valid example, e.g.:
FFFF 080C F7FF 0C00 F7FB

Regards, Dieter

[branadic]

Poor quality leads to pure misunderstanding. While it seems like the over temperature detection should be set to -0.8 V from the schematics in the service manual:
https://xdevs.com/doc/Valhalla/2720GS/Valhalla_Scientific_2720GS_Vol.%202_sm.pdf

the operation and maintenance manual says set it 0.1 V higher than the oven setting (page 9-9ff)
https://xdevs.com/doc/Valhalla/2720GS/Valhalla_Scientific_2720GS_Vol.%201_um.pdf

This adjustment is only required following component replacement
in the oven control circuitry on the REFERENCE MODULE PCB.
Note : 2720GS with serial numbers prior to 20-127 only have one
adjustment and no test point connector - contact Valhalla Scientific
for details if an early reference module requires adjustment.
a) Remove option "BAT" or "EBU" (if fitted) and the Ac power.
Allow 2720GS to fully cool for at least 8 hours with the top cover
and the REFERENCE SYSTEM COVER removed.
b) Remove the shorting link from all reference PCB test connectors
and the test connector at the rear top edge of the OUTPUT SYSTEM
PCB.
c) Apply AC power with the POWER switch in the STBY position.
d) Monitor the voltage between pins 5 (+) and 6 (-) on the respective
PCBs' test connector (front top edge) with a voltmeter with an
accuracy of better than 0.01V at 0.5V Dc.
e) Adjust RV1 for a reading of 0.5 +/- 0.05V DC on the voltmeter.
Note that if the ambient temperature is not 20C then this voltage
should be decreased by 0.02V per degree C above 20C. The user
should also note that the "HEATING" LED on the reference PCB should
be illuminated at this time.
f) Monitor the voltage between pins 3 (+) and 8 (~) on the test
connector with the above voltmeter.
g) Adjust RV2 for a reading of 0.1V more than that obtained in
step e) above.

That means for 23 °C ambient temperature the voltages should be set to 0.44V and 0.54V respectively.

-branadic-

[branadic]

There is a service manual online and it has a schematic "Reference Module" (page 40 out of 95) that shows the permutation of the X22C12 address bits.

I assume you are refering to that part of the schematic? Would you mind sharing the rearrangend Hex code?

As far as I understand the memory content was read in the address sequence 6,5,4,3,2,16,17,1 but 3,4,5,6,2,1,16,17 is needed to read the real content.

-branadic-

[dietert1]

With the correct unmapping of address permutation i got:

5A    C5    C5    C1    C1    00    00    00
00    00    00    08    00    EF    70    79
00    E8    30    43    EA    FB    00    E0
F2    3D    C7    BD    00    14    AD    8E
D0    00    E8    30    CF    46    36    00
E0    F2    70    55    E1    00    14    A3
8E    A3    00    E8    30    CD    30    7B
00    E0    F2    B6    DB    99    00    14
AB    8E    22    00    E8    30    C3    AF
C3    00    E0    F2    BB    68    B3    00
14    AB    8E    48    00    E8    30    C2
89    A2    00    E0    F2    36    12    34
00    14    AB    8E    BB    00    E8    30
41    67    A9    00    E0    F2    3A    3A
77    00    14    AB    8E    9E    FF    FF
04    0C    F7    FF    0C    00    F7    F7   

5A    C5    C5    C1    C1    00    00    00
00    00    00    08    00    EF    70    79
00    E8    9C    7D    BC    1F    00    E8
6B    D8    27    DC    00    14    AD    8E
E5    00    E8    9C    FD    27    FA    00
E8    6B    35    82    A1    00    14    A3
8E    F0    00    E8    9C    FE    B9    DA
00    E8    6B    3A    08    DA    00    14
AB    8E    EB    00    E8    9C    F9    F5
D5    00    E8    6B    3C    F5    04    00
14    AB    8E    79    00    E8    9C    FA
77    B5    00    E8    6B    D3    61    72
00    14    AB    8E    24    00    E8    9C
77    F4    EA    00    E8    6B    D6    F3
E9    00    14    AB    8E    FB    FF    FF
00    00    FF    FF    00    00    FF    FF   

5A    C5    C5    C1    C1    00    00    00
00    00    00    08    00    EF    70    79
00    E8    41    F4    DB    59    00    E8
66    3A    D7    5A    00    14    AD    8E
19    00    E8    49    09    B4    87    00
E8    66    B6    DC    CA    00    14    A3
8E    34    00    E8    49    0E    3E    0F
00    E8    66    33    7F    C7    00    14
AB    8E    74    00    E8    41    F7    B3
B2    00    E8    66    B9    F8    B3    00
14    AB    8E    AF    00    E8    41    F9
DA    B4    00    E8    66    3F    4F    4F
00    14    AB    8E    D1    00    E8    41
72    53    65    00    E8    66    39    34
87    00    14    AB    8E    57    FF    FF
00    00    F7    FF    0C    00    FF    FB 

The first example parses as follows:

Fixed Header                              
                                 
00e8   3   0   4   3   e   a   f   b
00e0   f   2   3   d   c   7   b   d
0014   a   d   8   e   d   0      
                                 
00e8   3   0   c   f   4   6   3   6
00e0   f   2   7   0   5   5   e   1
0014   a   3   8   e   a   3   
                                 
00e8   3   0   c   d   3   0   7   b
00e0   f   2   b   6   d   b   9   9
0014   a   b   8   e   2   2      
                                 
00e8   3   0   c   3   a   f   c   3
00e0   f   2   b   b   6   8   b   3
0014   a   b   8   e   4   8      
                                 
00e8   3   0   c   2   8   9   a   2
00e0   f   2   3   6   1   2   3   4
0014   a   b   8   e   b   b      
                                 
00e8   3   0   4   1   6   7   a   9
00e0   f   2   3   a   3   a   7   7
0014   a   b   8   e   9   e      
                                 
ffff      0   4   0   c   f   7   f   f   0   c   0   0   f   7   f   7

So there are 6 groups of 3 records plus the tail. The tail should have the different serial numbers.

Regards, Dieter

PS: I corrected a minor confusion between address bits 6 and 7.

[PrecisionFreak]

In the 156 page manual all GPIB commands are listed on pages 75ff, there are some to read the calibration values: GCAL0 .. GCAL87
get calibration constant
0 .. 7  Reference 1 through 8 voltage at the time of the latest calibration (in volts).
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.
16        Reference averaging coefficient (100% nominal)
33,35,37,39,41  Voltages of reference 1 at the times of previous calibrations (earliest to latest but one resp.)
34 .. 72  ..... for all 8 references
73 .. 77  Reference Divider rations at the time of previous calibrations
79 .. 86  presently measured reference voltages (as obtained by the OBRMS system) in volts

Also GSTA  (get overall status) gives some data of the 8 references in the first 8 characters of the reply.

I repeat my question: who has the possibilty to connect the device to GPIB?

Branadic: could you please post the OCR-version of the Vol. 1_UM.pdf?

[branadic]

Okay, here are some news. The unit is running with all 4 reference boards. It was "just" a matter of hitting the right buttons at the right time, no independant NOVRAM content required

1. Started the instrument with the fourth board fitted with an empty NOVRAM. The unit stops during boot sequence showing firmware version.
2. Pushed the SYS RST button, unit notice bad NOVRAM content
3. with the CAL key in CAL position and pushing the hidden key (below Last Cal) the instrument is able to repair the "broken" memory content, once the secret password was entered

IMPORTANT: Whenever the unit asks "CONNECTION ?" it expects a connention between SENSE terminals and NULLMETER, otherwhise an error new to me appears that tells you "CANNOT OUTPUT". Afterwards the voltage output doesn't work anymore. Ask me how I know 
It took quite some time to figure out how to get away with that error, as it is not clearly described in the manual (missing image too).

Meanwhile, I also adjusted all oven settings, which seems to have affected at least one of the references, but consequent Internal Cal should get away with it at some point once everything has stabilized again. That's at least what the manual tells you.

I repeat my question: who has the possibilty to connect the device to GPIB?

Not here, not at the moment, need to prepare something first, as all our original Agilent GPIB-USB adapters failed. I wonder why 

ranadic: could you please post the OCR-version of the Vol. 1_UM.pdf?

The file is too large to post it here

-branadic-

[maxwell3e10]

once the secret password was entered

Can you share the password?

IMPORTANT: Whenever the unit asks "CONNECTION ?" it expects a connention between SENSE terminals and NULLMETER, otherwhise a knew to me error appears that tells you "CANNOT OUTPUT". Afterwards the voltage output doesn't work anymore. Ask me how I figured that at out 

Thanks for figuring this out! I also had this problem and would go in circles doing self-tests until it concluded the Nullmeter is faulty and then work OK

[PrecisionFreak]

As the secret was found by me, I will not release it directly, but will give hints every few days how to find it.
One tip was given already: the cross sum (sum of all single digits) ist 31.
Today's tip: it is in the front-board EPROM

Dieter, your tipps are valuable!
But as I said, they often store 6-byte (48-bit)-values.
Wasn't there a 6 byte floating point representation? in Turbo Pascal?

My suggestion is: (my lines 5-8 are swapped with your lines 9-12, as A7 is swapped with A5, but Dieter's representation fits better)
Freak 8 byte wise
5A C5 C5 C1 C1 00 00 00
00 00 00 08 00 EF 70 79
00 E8 30 43 EA FB 00 E0
F2 3D C7 BD 00 14 AD 8E
AB 8E 22 00 E8 30 C3 AF
C3 00 E0 F2 BB 68 B3 00
14 AB 8E 48 00 E8 30 C2
89 A2 00 E0 F2 36 12 34
D0 00 E8 30 CF 46 36 00
E0 F2 70 55 E1 00 14 A3
8E A3 00 E8 30 CD 30 7B
00 E0 F2 B6 DB 99 00 14
00 14 AB 8E BB 00 E8 30
41 67 A9 00 E0 F2 3A 3A
77 00 14 AB 8E 9E FF FF
04 0C F7 FF 0C 00 F7 F7
END
5A C5 C5 C1 C1
00 00 00 00 00 00
08 00 EF 70 79 00
E8 30 43 EA FB 00
E0 F2 3D C7 BD 00
14 AD 8E AB 8E 22 00
E8 30 C3 AF C3 00
E0 F2 BB 68 B3 00
14 AB 8E 48 00
E8 30 C2 89 A2 00
E0 F2 36 12 34 D0 00
E8 30 CF 46 36 00
E0 F2 70 55 E1 00
14 A3 8E A3 00
E8 30 CD 30 7B 00
E0 F2 B6 DB 99 00
14 00 14 AB 8E BB 00
E8 30 41 67 A9 00
E0 F2 3A 3A 77 00
14 AB 8E 9E FF FF
04 0C F7 FF
0C 00 F7 F7

Dieter 8-byte-wise
5A C5 C5 C1 C1 00 00 00
00 00 00 08 00 EF 70 79
00 E8 30 43 EA FB 00 E0
F2 3D C7 BD 00 14 AD 8E
D0 00 E8 30 CF 46 36 00
E0 F2 70 55 E1 00 14 A3
8E A3 00 E8 30 CD 30 7B
00 E0 F2 B6 DB 99 00 14
AB 8E 22 00 E8 30 C3 AF
C3 00 E0 F2 BB 68 B3 00
14 AB 8E 48 00 E8 30 C2
89 A2 00 E0 F2 36 12 34
00 14 AB 8E BB 00 E8 30
41 67 A9 00 E0 F2 3A 3A
77 00 14 AB 8E 9E FF FF
04 0C F7 FF 0C 00 F7 F7

Dieter 5 or 6-byte-wise
5A C5 C5 C1 C1
00 00 00 00 00
00 08 00 EF 70 79
00 E8 30 43 EA FB
00 E0 F2 3D C7 BD
00 14 AD 8E D0
00 E8 30 CF 46 36
00 E0 F2 70 55 E1
00 14 A3 8E A3
00 E8 30 CD 30 7B
00 E0 F2 B6 DB 99
00 14 AB 8E 22
00 E8 30 C3 AF C3
00 E0 F2 BB 68 B3
00 14 AB 8E 48
00 E8 30 C2 89 A2
00 E0 F2 36 12 34
00 14 AB 8E BB
00 E8 30 41 67 A9
00 E0 F2 3A 3A 77
00 14 AB 8E 9E
FF FF 04 0C F7
FF 0C 00 F7 F7


Branadic could read the new NVRAM contents, but he cannot, missing a proper reader. Once again, I can give you my Elnec Jetprog for some months.  www.elnec.com/en/products/device-programmers/jetprog/
But I understand  that now, after his hard-gained success, the motivation for this may be < than small. ;-)

[branadic]

Branadic could read the new NVRAM contents, but he cannot, missing a proper reader. Once again, I can give you my Elnec Jetprog for some months.

He could, but lags a parallel interface on his modern computers

-branadic-

[branadic]

It is worth mentioning, that with the fourth reference board in place the averaged reference voltage has changed, leading to a shift of the output voltages and dropping out reference 2 and 7 (one of the references of my copy cat), which means the instrument now needs external calibration to account for that. Lucky enough I can use the ADR1399KHZ converted AN3200 and a 3458A as a first sanity check to calibrate the 10 V and the 1 V and a Vishay 100R resistor.

If the 2720GS should drop two references out of use, then this should also be used to indicate that an external calibration is necessary.

10V calibration - 2720GS "learns" the absolute volt
1V calibration - Used with the 10V step to "learn" voltage ratio

Ammeter Calibration - calibrates the internal ammeter (required every 3 years)

The user should note that the only “required” calibrations are that of 10V, 1V and the ammeter.

Ammeter Calibration ...required connections to a 100 ohm 0.1% resistor

Meanwhile, a second board with ADR1399KHZ was almost completed, it's just missing R15, which I currently try to figure out the required value.

-branadic-

[Messtechniker]

Branadic could read the new NVRAM contents, but he cannot, missing a proper reader. Once again, I can give you my Elnec Jetprog for some months.

He could, but lags a parallel interface on his modern computers

-branadic-

There are things like this: DELOCK 89445

[branadic]

More HV transformers arrived today. I can already spot some happy owners and more working 2720GS.

I tried calibrating (10 V, 1V, ammeter) my instrument today, which worked to some extend, but just a few minutes after adjustment with all reference being active the unit dropped reference 2 and 7 again. So I need to read the manual carefully again to spot any hints.

-branadic-

[PrecisionFreak]

New HV-transformers:
I hope they corrected the colors of the leads from the windings to the soldering posts, green and violet/blue were interchanged!
green is 115V, blue is 16V
If not, please inform them about it. They could use the correct colors of the primaries too, at least at the printing.
Appended you find my suggestion of the label (read it with monospaced (courier) font).

If you see the colors and connect the mating wires, the 16V circuitry will see 115V and not be happy about it.
Be careful and measure!
At the very first transformer the printed on description differed from the actual winding / connection.
Do you still have this transformer?  Maybe only the printing was false.
Or was the sequence 16 16 0 115 115?  Even then it could be rewired (separate the enamel wires at the soldering posts).

I suggest to use separate primary fuses (2 because of 115 parallel mode) for the HV-transformer.
And / or fusible 1k 10W resistors R201, R202 at the 1100V secondary.

Could you please measure the Ohms and Henries of the windings?
I am not sure if the 26 / 52 ohm is the 115v to 16V  or  115 to CT (center tap).
It could be 22 /44 ohms.

Where are the two shields E9 and E10 connected to?
I think they are connected to each other, but see a wire on the solder side of the mainboard.
Where does it lead to?

General hint:
Units that saw no voltage for longer periods of time should be powered up slowly, via a bulb or a variac with ammeter and sensitive fuse over several hours.

About the new references:  Probably they should burn in for some dozend hours before a new calibration.
Sorry for the lost time.

Tip of the day: digits are grouped by 2.

[branadic]

I today repeated calibration, but this time denied the drift analysis facility. First, all references were in use, but after some time ref 2 and some time later ref7 were dropped again. I will continuously retry external calibration, hopefully to be able to get all of them in use at some point in time.
Another solution would be to convert all of them to ADR1399 and start with empty memories.

-branadic-

8.1.3 OBRMS limits
At the time of an external calibration (the 10V calibration point) the current deviations of each reference are recorded and the absolute value of each reference calculated. These values are stored in memory in each reference module. From this time forwards the actual measured deviations from the average is compared with these calibrated deviations and any change above the limits in the table below will cause the action shown.

[branadic]

I've updated the schematic https://www.eevblog.com/forum/metrology/valhalla-2720gs/msg4991971/#msg4991971 with the correct signal path designators (+HS that is +18V unreg. with 1N4001 diode, +RS that is +6V, -ZS, -HS that is 16 VCT), that were barely readable in the scans. I still can't make full sense out of the voltage controlled current source, supplying both LM399. Hence, I'm still stuck to what is needed to convert the reference boards into ADR1399.

-branadic-

[Kleinstein]

The current source use D10 (1N827) as a voltage reference and than R16 and R18 to set the current to a relatively high value of some 6.2 V / 2.94 K , so a bit over 2 mA each.  The circuit uses the JFETs as drivers - a more conventional circuit would use PNPs or P-channel MOSFETs at the same place.

So the current should be high enough for the ADR1399 already and likely no change needed - just the added RC element for damping. C5 and C6 may already to that job though the ESR may not be that well defined.

[branadic]

I meanwhile measured the referene voltages, which to some extend could explain why the unit is dropping Ref2 and Ref7. Maybe Valhalla Scientific selected references being as close as possible together? Need to check, whether or not I have some LM399 left, that are way closer to the rest of the references.
However, if that is the reason I'm very surprised. I would have expected that during external calibration the unit calculates the mean for each reference combination being in use, so that the absolute value of each reference is not of interest and could deviate largely from the others.

Ref	Voltage [V]	Deviation from mean [µV/V]	Comment
1	6.9405	-1223.2
2	6.892	-8202.6
3	6.93	-2734.2
4	6.921	-4029.4
5	6.9045	-6403.8
6	6.919	-4317.2
7	7.058	15685.7	Copy Cat Reference
8	7.027	11224.6	Copy Cat Reference

mean 1-6: 6.918 V
mean 1-8: 6.949 V

-branadic-

[branadic]

Thanks for that explanation, Kleinstein. LM399 only requires 1 mA, while ADR1399 requires 3 mA, that's why I was concerned.

-branadic-

[branadic]

So I've checked all of my stock LM399, but couldn't find any that are close to the average of the existing ones inside the 2720GS. Hence, I decided to measured 12 ADR1399 that I have at hand:

Ref	Voltage [V]
1	6.932
2	6.939
3	6.9155
4	6.8915
5	6.905
6	6.909
7	6.936
8	6.943
9	6.9355
10	6.9355
11	6.933
12	6.943

replaced Ref7 and Ref8 with ADR1399-1 and ADR1399-11 and added the filter on RefBoard-4.
I left the instrument on for quite a while, but by now the instrument didn't complain. Seems, like it feels way more comfortable with that choice.

Next step is to replace the rest of the references with ADR1399 and add filter too. Since we know that it could deal with the reference values for Ref2 and Ref 5, which are on the lower side, I plan the following combinations:

RefBoard-1: ADR1399-8 and ADR1399-12
RefBoard-2: ADR1399-2 and ADR1399-7
RefBoard-3: ADR1399-9 and ADR1399-10

This way they are as close as possible together with the references I have at hand.

-branadic-

[PrecisionFreak]

>  but this time denied the drift analysis facility.
It's clear to me, all things must settle first (wait 90 days after calibration):  page 94 of 156:
8.1.5 Drift Analysis
After the 2720GS has been in use for some time (typically 3 to
6 months) the references and divider resistor networks will have
fully stabilized to the environment and any remaining drift rate
will be both small and also linear with time. Thus the calibrator
of the 2720GS can select for the change in calibration from the
previous data to be used for "drift analysis" (only if more than
90 days has elapsed since the previous calibration was performed).
With this enabled the drift rate is calculated and the value updated
every day to reflect the expected data at that time. If more
than one pair of data points are available for drift analysis
then a multi-order curve fit routine is used to forward extrapolate
the data in time. The resultant value of this is the value used
by OBRMS to compare each references' deviation from the average
thus even if a reference is consistently drifty then the 2720GS
will “learn” its' drift rate and automatically correct for it.

And page 103 of 156:
9.3.8 Request if to Use Drift Analysis
If the previous calibration was performed over 90 days previously
and the temperature was within 3C of the present temperature,
then the 2720GS can, if desired, perform drift analysis between
these data, and forward extrapolate this drift rate into the future.
The 2720GS will always request if the user wishes to perform this
activity. The user is not advised to utilize this facility in
any of the following circumstances :
a) If the 2720GS is under a year old.
b) If the previous calibration was obtained on different equipment
c) If either calibration is known to contain errors.
d) If the previous calibration was obtained by a different procedure.
e) If any component in the 2720GS has been changed or altered
since the previous calibration.

And I wondered how they would do the averaging, as all the active references are summed via the 10k00 0.1% resistors.
So the average voltage must be calculated from the active reference's actual voltages, I think.

What if you drive the ADR1399 with less current so they come in the LM399's range?
Or adjust their voltage via their temperature? ;-)

[Kleinstein]

For the references there is the analog average via the 10 K resistors. In addition there are individual paths via JFET swiching. So the instrument should be able to measure the difference to the average for all the references.

Changing the current to the reference would not change much - they are made to have a very low R_out. Adjusting the temperature is also no really an option with the very low TC due to the local fixed heater. There is only a tiny range by changing the heater supply and this way indirectly the temperature.
The reference voltage don't look that bad, so I would not expect the absolute value to be a problem. of cause adding different values to the average will also shift the difference for the old references. So one may have to disable the drift check for all the references, not just the new ones.  For the initial phase there is also a chance that the reference are actually drifting and this might even be enough to be detected - though there should at least be room for the +-4 µV typical popcorn steps to be inside the accepted range.

[branadic]

As I said, my assumption is that the two 7.xxV LM399 I'd fitted to reference board 4 were way too far off for the system (see image below). Once I've replaced them by selected ADR1399 the instrument was happy and put all of the references in use.

Meanwhile, all reference boards were upgraded with the aforementioned ADR1399 references and again the instrument appears to be happy and put them all into "in use" status. So this is the first ever Valhalla Scientific 2720GS/HSR-All-ADR1399 in the world

The instruments now needs to relax and the references to settle, before we can perform any reasonable calibration. But first noise measurements can be performed tomorrow, I guess. This way we can see were we are.

-branadic-