Prema 6048 teardown

[Rax]

After performing the DCV cal on the Prema on the .2V, 2V and 20V ranges - using my 5440A, which seems to be vetted by a KS-calibrated 3458A - I am seeing what to me like pretty reassuring agreement. For instance, between the 2V "non-DIV" input and the 2V "DIV" input, there's sub-unitary (ppm) divergence between the two hardware paths (preliminary conclusion). Of course, not very easy to say if what I'm seeing confirms the Prema, the Fluke, of both of them in "joint venture." I'd need to do more work to clear my data and experimental conditions.

But to me this provides some level of comfort with the linearity of the Prema ADC away from zero crossing, as it compares with the pretty good linearity the 5440A in known to have.

[Rax]

The other provisional conclusion I feel I can draw is that I'm seeing much more consistency from the Prema than before the upgrade.

I feel that the zeroing is now pretty reliable and and the drifts I used to see (and not know whether they are caused by the Prema or the 5440A) are no longer occurring. If anything, the zeroing of the Prema seems just about as reliable and resulting in a level of relative accuracy (between the two) as the AUTOCAL the 3458A seemed to provide (and direly need all the time). In fact, for my particular samples, the Prema needs zeroing less than the 3458A needed the AUTOCAL.

[Kleinstein]

The upgrade via better OP-amps can be really woth it.  The change OP77 to OPA140 seems to largely fix the weak point in the INL.  The OPA140 and LT1007 for the reference are major contributors to the zero point drift. Here there can still be variations in the actual parts used. Even in the original configuration the stability of the zero can vary between units. If the zero is stable there is no need to change things. If it is drifting it can make sense to try an alternative chip (could be just another LT1007 or similar or an AZ OP).

How often the zero for the 6048 or ACAL for the 3458 is needed  can also vary. Some units can be more stable than others.

For the prema meter the gain of the ADC and input is set by 1 resistor ratio. The 2 V range may be slightly more stable in this respect, using 2 x 5 K and thus equal value resistors. The other ranges still also still have only 1 pair of resistors, just exchanging one of the resistors to a different value. So there is no clear prime range with like with the 3458. The 6048 does not look that it could easily measure it's own reference directly. So a kind of ACAL would not work so easy for the Prema meter. The test in 2 ranges is more like a self test, but no way to correct drift.

The 3458 has additional resistors in the ADC gain and amplifier gain. So less stable to start with, but able to correct. On the other side it has a stable zero.

[Rax]

A bit of a follow up post (medium term observations, at least in my book) to the issue of the op amps upgrades and zeroing. It partly reconnects with some offline conversations with dietert1.

I'm very happy for how it behaves from that perspective. I had it off for a few days and just turned it on. From a cold turn on it started around maybe .00000250V with a short on the input, and after a good one hour warmup it's at about .00000035-45. I don't think that's a bad zero by saved values from the prior run by any stretch of the imagination.

I think this is further proof the op amps upgrades are conducive to bettering the performance of this great meter.

[branadic]

We learned something new at MM2024 about Prema 6048. Some participant, let's call him Manfred L., gave us some interesting info. In case you were wondering about the bodge wires in your unit (soldering side), in particular at the 5 V logic rail, this is not someones repair attempt of a broken trace but a factory bodge made by Wilfried Stricker himself to decrease trace resistance. Don't attemp to remove them, but leave them as they are and where they are!
Hopefully we can soon report what removing them will lead to

-branadic-

[Rax]

We learned something new at MM2024 about Prema 6048.
-branadic-

OK. Bring on more novel finds on the 6048.

[meandeev]

We learned something new at MM2024 about Prema 6048. Some participant, let's call him Manfred L., gave us some interesting info. In case you were wondering about the bodge wires in your unit (soldering side), in particular at the 5 V logic rail, this is not someones repair attempt of a broken trace but a factory bodge made by Wilfried Stricker himself to decrease trace resistance. Don't attemp to remove them, but leave them as they are and where they are!
Hopefully we can soon report what removing them will lead to

-branadic-

Hi Branadic,
got the broad hint  ;-)
Background for everyone else: I bougth a prema 6048 3 years ago with a strange error, that wasn´t solved by us until now.

In the picture that I made when the unit arrived one can see the bodge wires at the digital board.



I removed them, because I couldn´t imagine that they are an "official repair". I didn´t made any measurements before that, but later the meter shows a strange behavior, shown in the second picture:



strange jumps of the measured values.  The red line are the values from the Prema 6048, blue are the values from a HP34401 that measures PIN6 of  U10 (~ the voltage of the Prema LTZ1000). Please ignore the jump at 1:05:00, at that time I switched off/on the prema for a second.

If I have some spare time I will start to investigate that again, also with installed bodge wires.

[dietert1]

When i started working on that board - after receiving our Prema 6048 from former contributor essele - i found those bodge wires, too. As far as i remember it is an improvement of the +5 V supply network.
I also remember adding several buffer caps on that board. The original design is poor in comparison to a typical HP device of that time. Those have almost one buffer cap per logic chip.
I also bought a new 2 MHz crystal, as the oscillator wasn't in good shape. Don't remember whether that solved the issue or other mods were necessary, like grounding the crystal housing.

Regards, Dieter

Some more information: The processor board does have one buffer cap per IC, my memory error. Appended two new images of our Prema 6048 processor board:
Top:
- 2 MHz crystal replaced, its housing connected to Gnd
- Three 22 uF elyt buffer caps renewed
- New 5V power supply cap is 10 000uF/16V/85°C
- Rectifier bridge got 5x5 cm copper sheet for cooling. Still gets pretty hot.
- FRAM was installed by previous owner
Bottom:
- 2x 470 nF across rectifier
- Supplementary wiring fixed with glue

[meandeev]

I found some spare time and so I investigated the error of my Prema a little bit deeper (with removed bodge wires): with a logic analyzer and decoded the SPI value from the BK7 chip. At 10V one can see nearly only 2 values alternating. I don´t have a clue, what the SPI values are or if I decoded it right.



A new timeseries with a stable 10V shows spikes again. Because the lid of the prema was open, the measured values are not that stable. But you can can see clearly the spike of ~120µV. The x-axis is the sample number of my logic analyzer, that  I tried to synchronize (around sample 205 000 000)



I recorded the SPI-values in parallel, you can see the values start to oscillate at around 205 000 000:




The 5V-rail at capacitor C2 was also measured with a 8-bit AD converter. It oscillates all the time, but normally the amplitude is stable:



But the amplitude start to oscillate at the same time, when the SPI-values start to oscillate:



And also stop, when the SPI values stop to oscillate:



So the next step will be: install the bodge wires again and the enhancements that dietert1 described

[dietert1]

The LM309 datasheet claims that no output capacitor is needed for stability. Then C2 isn't that important.
Your "scope" traces suggest that C1 requires replacement. Probably its capacitance is less than 20 % of nominal value.

[meandeev]

I reinstalled the bodge wires as well as a new C1 with 10000µF and a new CR1. For CR1 I used a B500C7000, because the datasheet of the  former installed B40C3700 doesn´t allow such a big C1. Also the 2MHz quarz was replaced.
Unfortunately this was not the solution for my problem.



The 5V rail looks clean(er) (maybe also because of better ground) and the amplitude doesn´t oscillate anymore, when the SPI values start to oscillate.




So now I´m clueless again where to search for the error  :'(

[branadic]

That's really unfortune
Are the ground wires thick enough, similar to the original ones? Is there anything else you have "cleaned up"?

This weekend I've tested another Prema 6048 white, that was dropped here after MM2024, for INL with the original OpAmp arrangement. Similar results to what we've already seen. New OpAmps are on their way and should arrive in the next days (OPA189 but also OPA140), will give them both a try to see, if I can spot a difference.

-branadic-

[meandeev]

The wires are 1.5mm^2 copper. On the digital side nothing else was changed.
Because you have a 6048 on the bench: please can you measure the PLL voltage PIN9 from U7?
And make an oscilloscope picture from PIN9 of U8 (the Prema IC)?
Thanks!

[branadic]

Pin9 of U7 measures 0.678 V on that unit. Attached are images of U8 pin9 with display set to infinity over a course of 20 s, as well as U10 pin 6 @ 10 V.

-branadic-

[meandeev]

Thanks for the measurements!
I´m more and more convinced, that I have a problem with the PLL locking…
It seems that the output pin9 from the Prema-IC (U8) has an extreme high output resistance. It makes a big difference if I use a 10Meg 1:10 probe or a 1:1 1Meg.  As you told me, you used a 10:1 probe.

For the records here my measurements from my Prema 6048:
With 10:1 10Meg the upper voltage of the peak is ~11.4V, the lower „floor“ ~5.92V, the peak width is ~220µs, and the Prema shows 10.243489V for a 10V input, Pin 9 of U7 is 2.13V


With 1:1 1Meg the upper voltage of the peak is ~11.3V, the lower „floor“ drops to ~5.5V, the peak width is now 1.85ms !!!, and the Prema shows 10.261289V for a 10V input, Pin 9 of U7 stays at 2.13V
Without any connection the Prema shows 10.242167V for a 10V input.


A rough calculation from the "floor" drop gives 90k - can this be?

[Kleinstein]

The voltage levels for the PLL signal look about right. The following OP-amp has some 6 V at the non inverting input, as a reference level.
For a PLL phase-frequency dedector (e.g. as in the 4046) it is normal to have a signal with 3 states: positive , negative and high impedance, usually with much of the time at high impedance. The prema meter has an integrator in the loop filter and should thus have only short pulses. However the scope probe already disturbes the DC balance, even with 10 M and more with 1 M. The 5.9 or 5.5 V case are likely 6 V from the OP-amp side and the the 100 K and probe causing the voltage drop. The 1.8 ms of pulse length sound about right so that 18.2 ms with 1.1 M to GND is compensated by 1.8 ms via 100 K.

From the observed waveform it more looks like the PLL is working. The better point to probe may be the output of U6 or the PLL test point. This should be a relatively stable voltage with little to no detectable ripple. The exact voltage will depend on the current mains frequency and R56 setting.

Normally probing the PLL should not effect the reading. Some effect is possible via additional capacitive coupling to the PLL and the larger ripple at the VCO tuning signal. With normal operation coupling to the PLL and the residual ripple can cause part of the INL error.

[branadic]

Precision pinheaders arrived, so I was able to populate U9 and U10 with OPA140. Results attached. The improvement is crystal clear. Next is OPA189 for comparison.

-branadic-

[branadic]

For some odd reason OPA189 performs worse than OPA140 on this unit. 
So OPA140 is the way to go. 

-branadic-

[branadic]

Good evening,

this weekend I had a visitor in the lab with a Prema 6047 and Prema 6048. As usual INL in the 20 V range was measured against 3458A and something interesting was found.
The Prema 6048, populated with U9 = OP77 and U10 = LT1007, showed the typical INL issue (Prema6048-1.png) before the OpAmps were replaced with OPA140 (Prema6048-2.png).
The Prema 6047, also populated with U9 = OP77 and U10 = LT1007 didn't, but showed a slightly different behavior (Prema6047-1.png and Prema6047-2.png).

While both meters are absolutely identical except for the fact that Prema 6047 is missing the BK7 for temperature measurement on the digital board and uses LM399 instead of the LTZ1000 reference circuit, I wonder if the INL issue is really due to the OpAmps used for U9 and U10. Instead I think the real issue is something like capacitive loading of the LTZ1000 reference circuit, with U10 being part of the loading.
The Prema 6048 I had (see the first posts in this thread) was one of the very first ones, using a Prema 6047 ADC board and a seperate LTZ reference board connected to the ADC board. I remember me trying to measure the t.c. of this reference board and having huge trouble with oscillation, without any difference between using LT1078 and LT1007.
We now see a totally different behavior of the ADC with an LM399 connected to it. We also know that LM399 has pretty low dynamic impedance — 0.5Ω.
Unfortunately, we didn't have enough time to measure Prema 6047 with OPA140 populated, but this would be something interesting to test. I also think equipping the Prema 6047 with ADR1399 could be an interesting investigation.

-branadic-

[dietert1]

As far as i remember the 6048 has a buffer meant to correct for multiplexer resistance in its switched reference current source. That buffer opamp had input protection diodes which would cause transient load/current spikes from the reference source (LTZ1000 circuit), causing the typical nonlinearity near zero input voltage. No surprise that the magnitude of the effect depends on details of the reference circuit.
I also remember that the ADC board i was working on at the time had a place for an optional buffer cap for the LTZ000 reference. It was missing in our instrument, probably because capacitive load can cause instability. I remember adding an RC combination similar to the snubber recommended for the ADR1399.

Regards, Dieter

[Kleinstein]

Loading the reference via switching spikes and the diodes inside the LT1007 (U10) is possible. However this would not be limited to the range just around zero, but a possible issue with any ref. switching. AFAIR the point for the INL was that the OP77 (U9) is a bit slow, so that the integrator input voltage is not fully settled for very short reference pulses that happen near zero. So an INL cause be a low speed of U9, not from U10. The way the reference is switched leaves the ref. side of R31 floating when off. This makes the integrator sensitive to incomplete settling with short pulses.

Loading the reference with a capacitor (C49 = 100 nF) is a possible issue. The LM399 ref has a better chance to tolerate the capacitor. For the LTZ version one would ideally not have C49 or a resistor of some 100 ohm in series to C49.

Using a ADR1399 could be a sensible upgrade for the 6047. The effort is moderate and so far they seem to behave well: not just half the noise, but also low drift seen so far (still a bit limited statistics).

[branadic]

@ dieter
As far as I can see, there is no difference between 6047 and 6048 integrator board schematic except the reference (see attachement).

@ Kleinstein
How do you explain that 6047 doesn't show the INL issue, if the schematics are otherwise identical? All that we can see for the 6047 is a t.c. sensitivity, visible around zero.

-branadic-

[dietert1]

Kleinsteins little theories sound interesting but don't lead anywhere. Discussing at that level of half knowledge doesn't resolve anything. You have to spend some serious time to find out what is going on. I did that work for the 6048 and the facts can be found above in this thread.
If the 6047 doesn't have that linearity problem, i don't understand the point of this discussion.

Regards, Dieter

Edit: Yesterday i happened to watch this repair video:

A good example of what it may take to fix the problems of a legacy device. Probably the 6047 has enough other problems to work on..

[branadic]

If the 6047 doesn't have that linearity problem, i don't understand the point of this discussion.

The answers to that is:
1. Both meters have identical schematics, except for the reference.
2. One of them shows the INL issue in the 20 V range (6048), the other doesn't (6047).

While the OpAmp (OPA140) replacement you suggested solved the issue for the 6048, I somewhat doubt the root cause (OpAmp speed, ...) and then the question is, what is the difference between both meters? That is what I'm interested in to learn something from that excercise for the future. For me this points towards the reference and its output impedance. But I might be wrong. 

-branadic-

[Kleinstein]

Chances are the INL problem in the 6048/6047 may not effect all units but could be a borderline case. If as large as measured in this thread, they should have caught it at Prema in the design / initial test phase. So it could depend on individual parts (e.g. the OP77 or LT1007 or prema chip) within the normal variations.
Another point could be a slight difference between the schematics and actual circuit. Dieterts writeup of the mods show an extra capacitor at the integrator (around U9), that could make a difference, though hard to tell without knowing the internals of U8.

The impedance of the reference is a clear difference between LM399 and LTZ1000 based. The LT1007 may cause a current spike on switching the feedback / load though not for sure. This would be a part one could check.