Prema 6048 teardown

[branadic]

Hi everyone,

it just so happened that I bought a Prema 6048, which arrived today, so I now have a small collection of 8.5 digit meters (R6581D, S7081).
Since noone ever posted teardown pictures, this thread is addressing exactly that and also shows the hardware differences to Prema 6047, that Dave already torn down:

https://www.eevblog.com/forum/blog/eevblog-613-prema-6047-multimeter-teardown/

The meter arrived with a lot of Fluke stickers all over the place. Looks like it was calibrated in 08/2009?
Before powering the meter I visually inspected the meter and caps and took a few pictures, so this is just a first teaser, not a complete teardown yet.

The first difference is already known, Prema 6048 has a LTZ1000ACH reference, compared to LMx99 in Prema 6047. The reference board pretty much looks like how you're doin' and is a piggyback solution. Most likely from preaging and selection. Interestingly, they have choosen LT1078 instead of LT1013, while the circuit in the manual indicates LT1013 too. Maybe someone else can open his unit? Otherwise no fancy resistor magic going on here.
The reference is connected to where the LMx99 is populated on analog board in 6047. It pretty much looks like an afterthought. Some hermetically sealed RHK02 (now part of Vishay) 5k, 7k and 70k resistors can be found on the analog board.
But the most interesting part is, what is unpopulated in 6047. While we can find a PR BK7 on the analog board in both models, there is a second PR BK7 on the digital board. I was told that MickleT. revealed that this additional ADC is for temperature correction on the meter.

Edit: here is a link to MickleT.s findings: https://www.eevblog.com/forum/blog/eevblog-613-prema-6047-multimeter-teardown/msg441667/#msg441667

That's it for now, more pictures and measurement results on this meter coming soon.

-branadic-

[Dr. Frank]

Congrats.
The MM2020 took place a few days too early, obviously.
Would've liked to play with that famous instrument, and of course to check its calibration. (I'm still missing my 2ppm )
Please shoot some pictures of the instrument, while measuring something. Beautiful.
THX
Frank

[essele]

Interesting that yours looks like it's been hacked in ... mine is actually a standard part of the board...



More complete board photo...

[branadic]

Thanks for sharing. Looks like your analog board is a later version (6048-1-5), compared to the one that is in my unit (6048-1-2 and SN 1027).
Meter is still warming up with inputs shorted and offset is already down to -2.4µV starting at about 200µV on first power on. Maybe the unit was switched off for quite some time and need to stabilize first.

-branadic-

[z01z]

Do you know what those round traces are? Is there any component on the other side of the board, held by the screw?

[branadic]

They are just "board design", having equal distances to the screws. Somebody had fun making the layout for that board and signal traces are surrounded by guard traces almost everywhere to prevent leakage.

-branadic-

[branadic]

I made some progress on my unit.
First, I noticed some intermittent reset issue on the meter, indicating some power supply problem. So within the last days I fully recapped the meter by good quality Rubycon caps and one single axial Vishay cap on the digital board, renewed some bodge wires and gave the unit a careful cleaning and let it warm up overnight.
Seems like the reset issue is now gone, but I do have a slight zero offset of some -39µV now.
I also received some OTPs and a new zeropower SRAM module. Once my EPROM programmer arrives I will save the original EPROM and SRAM data and flash the latest firmware with the factory calibration block to the meter.
Finally the meter will need some adjustment I guess. More pictures of the unit in the next days.

-branadic-

[branadic]

So meter is running for some days now and I couldn't notice any further reset. So replacing electrolytic caps fixed that issue. 
Still waiting for the EPROM programmer to arrive.
Attached are some more pictures.
As my unit is missing the original feets I will need to design some replacements and get them printed. Thanks to essele I have an idea of their dimensions.

-branadic-

[branadic]

Feets printed and mounted (data attached in prior post), latest firmware updated with factory calibration block copied into an Atmel AT27C256-45PU OTP and M48Z12-70PC1 zeropower SRAM module replaced by a new one, thanks to TL866 II+ that arrived meanwhile.
Attached is the measurement with the input shorted. First 1.5h in 8.5digit mode (20s integration time per sample), second 1.5h in 7.5digit mode (4s integration time per sample). I would say, the meter is pretty quiet.
ToDo: Measure linearity and calibrate the meter.

-branadic-

[branadic]

And a measurement on one of my LTZ references in 8.5digit and 7.5 digit mode.
What I can conclude by now is, P6048 (20s) is faster than S7081 (51.2s) in 8.5digit mode, low noise, easy to repair and 7segment display will last much longer than any (custom) VFD or gas discharge display. It seems to be stable without any drift correction or ACAL functionality.

-branadic-

[branadic]

I measured INL of P6048. I'm not too happy with the result. Maybe someone else with such unit wants to share his result?
To me it looks like some buck of the result calculation?

-branadic-

[Kleinstein]

The error near zero looks like some kind of extra jump at or near zero.
Without that jump the INL curve looks actually quite good.

From the circuit it looks like there are relays used to switch the polarity. I have not found a detailed description, but it looks a little like they may use this between positive and negative side. A near 1 ppm jump would be some 10 µV (or 20µV ?), quite a lot !
There may be a calibration constant to correct the jump.

[branadic]

Well, calibration is done by offset calibration of the ranges first and by suppling either a positive or a negative voltage in the individual ranges next.
What I could do is recall the factory settings, to see if that changes something. If INL is different, then meter needs a full recalibration. Luckily I have already saved the SRAM, that stores the calibration constants, so nothing to loose here.

-branadic-

[branadic]

Performed a quick INL test with factory calibration data. Result have somewhat improved, but are still not amazing.

-branadic-

[TUMEMBER]

Put trend lines on the chart, it will be clearer.
You can now see a large number of colored dots.
It is not convenient for a human to evaluate a signal.
Describe the colors because you "forgot" - what is what.
Sorry, it's just a google translate.

[branadic]

Hi,

I currently have the reference board of the unit in my thermal chamber and t.c. looks rather large (positive and linear), so does noise (numbers and diagrams later this day), even though the reference is currently powered by 3x 6V leaded battery plus LT1763 to provide 12V.

The reference is powered by 12V inside the meter. It uses LT1078 in a low grade socket, BC547 and R80 (nominal 400k) is missing, temperature setpoint resistors R77/R78 (12k1 + 1k : 1k) are like R79 (100R + 20R), they don't appear of highest quality.

I was planing to replace the socket and LT1078 by LT1413. Furthermore I have a 13k:1k VHD divider and a 120R VHP202 resistor on its way. At least the footprint for R80 is present, so it would be possible to trimm the circuit for low t.c.

On the other hand Prema 6048 uses some temperature compensation. Is there someone with knowledge about it? Will I do anything bad to the meter, if I improve the reference, since temperature compensation is already accounting for this effects?

Edit: Attached a picture of the reference board. Seems like initially Prema wanted to use LMx99 and later decided it wasn't good enough for P6048. You can see, that the reference board is an afterthought and connected to LMx99 footprint. This is good, as I'm able to remove the reference board to characterize it. Newer board revision do have the LTZ reference circuit directly assembled to the mainboard.

Measurements of internal temperature showed ~13K above ambient temperature. The meter is fanless, though I think it would help to cover the reference board in some way.

-branadic-

[Kleinstein]

The chip is a LTZ1000A and thus normally does not need the 400 K resistor for compensation of residual TC. It may still help a little in some cases, but likely with a much higher value.

Using the simple LM199 footprint with just 4 wires may add some effect of wire resistance, as there is no proper Kelvin sensing close to the chip. This could explain some of the problems.
In this respect the low power LT1078 may not be so bad.

A cover for the reference is definitely a good idea and the lack of a cover may explain the higher than normal noise.
However there are rare case of LTZ1000 with extra popcorn noise - so it could be just bad luck.

The resistors in deed don't look great.  On the other side VHP202 is kind of overkill already for a LTZ1000 ref board with a proper connector. There is little need for a board with more mediocre interface. So the old resistors may be good enough in this context.

[dietert1]

Can't you wire in one of your other LTZ1000 references to find out about TC calibration? I mean before starting the revision.

Regards, Dieter

[branadic]

@ Kleinstein

the reference board only uses the power supply pins of the LMx99 footprint, while all other connections are done at proper points on the main board.
Refering to the datasheet the resistor divider (here 13k:1k) and the 120R resistor are the ones with tighter requirements. 120R is nothing you find in stock at lower grade. Since I was able to get both of them from someone without waiting weeks/months I took what I was able to pick up, even if it might be overkill.
I don't see popcorn noise and the reference board in the temperature chamber is connected in a Kelvin scheme fashion. So the interface is okayish and doesn't need modification. As I said, data later this day, as temperature profile is still running.

@ dieter1

good point, but unfortunately all of my references have the 10V boost, are trimmed and t.c. is compensated for the overall circuit including the 10V boost. So I currently have no zener output only boards at hand.

-branadic-

[branadic]

So here are the results, 0.2ppm/K, which is not too bad,  though I have seen better than that, but it's quite noisy.
Curious, the reference is missing the 1meg/100nF at the heater opamp.

-branadic-

[Kleinstein]

There are 3 capacitors - so may it is just the resistor missing.

Not having the 1 meg and 100 nF to limit the higher frequency gain at the temperature control coud in theory lead to a less stable regulation or even some oscillation. They are part of the frequency compensation to make the regulator a PI regulator as opposed to a nearly I regulator only. The exact values are likely not critical. The A version at relatively low temperature would have power and thus less loop gain could better have a slightly larger resistance and smaller cap to get similar overall loop parameters.

Even if not oscillating, a regulator close to oscillation can show more noise.

[branadic]

I managed to reverse engineer the board and found that the 1meg is populated with 100R. There is also a footprint for the unpopulated trimm resistor. So the board is pretty much standard, just some minor differences, such as LT1078.

-branadic-

[Kleinstein]

A 100 Ohms resistor in the compensation of the heater makes not much sense. I would change it to the more usual 1 M. With only 100 ohms the heater control could be very slow and chances are the higher noise is from temperature variations, especially without a cover.

[Andreas]

Mhm,

what is actually populated on the "original" board? (100R or 1 Meg?)

https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg448574/#msg448574

with best regards

Andreas

Edit: Ah found it:
https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg449750/#msg449750

so its more the 100R as the 1 Meg.

[branadic]

Andreas, someone populated 100R instead of 1meg, must have picked up the wrong resistor I guess???
Unfortunately I ran out of 1meg myself, so can't populate the proper value today. However, I already removed the socket for the opamp and since I had it laying around, populated LT1013 instead of LT1078.

-branadic-