Advantest R6581 8.5 digit DMM mini teardown/repair

[Pipelie]

yesterday I had measured the Input of 6581 to TP200 via Keithley 155 at 30uV range and logged the output of K155 by 34401A, Signal source is from F5700.  The voltage of input to TP200 is calculated by this equation: (reading of 34401 - Average of all reading) * 30/1. 

[Kleinstein]

The difference from input to TP200 would be mainly the offset of the input amplifier in non AZ mode. For this, only some 0.5 µV_pp over quite some time is remarkably stable for a JFET amplifier. This kind of confirms that the INL error is likely due to the ADC itself, not the input amplifier.

Is the 2nd curve just a second run ?

[branadic]

Checked all the power rails and ESR of the caps of the analog power supply. The caps read fine, though I found at least something weird. The -18V rail reads -19.38V. This rail uses a discrete voltage regulator based on A1015 PNP transistor plus a diode and is derived from the rail that also feeds 7924 regulator for -23V rail. So there is at least something wrong here. Need to find out what.

-branadic-

[ramon]

Maybe Toshiba? Legendary low noise, excellent hfe linearity 2SA1015

[branadic]

Well, I don't expect the transistor to be broken, as the zener used is marked 20, so -19.38V would make sense. I wonder if this is normal or if they were running out of 18V zeners?

-branadic-

[Pipelie]

yes, the 2nd curve just a second run, each run takes about 2 hours and 20 minutes.

[Kleinstein]

The higher voltage is odd. However there may be different versions.  Depending on the FETs used, -18 V may be on the low side to turn off the FETs in the input MUX, if there is a -13 V signal.

A problem with ripple can also happen if the caps have too low an ESR or just variations in the regulators. Usually the output impedance of the voltage regulator is pretty (actually not so nice) close to a low loss inductor. So add a capacitor of the right size and low ESR and one will see a peak in the power supply impedance. With some bad luck this may just be in the 1-10 kHz range.
It can be pretty hard to filter the ripple current from the integrator + slope amplifier (some 2 mA) at the low impedance supply. It may be simpler to filter at the sensitive part (U202) as there is already some 10 Ohms in the supply, but only 100 nF for filtering, which does not help much for 5 kHz. The critical supply should be the -15 V, as usually the negative side PSRR is worse.

The reference amplification around U209 and U211 already used a different supply, so no problem there. However the LTZ1000 uses the +15 V. At least the OPs have GND as negative supply - so should be less sensitive.

If one has a sensitive scope with good averaging mode (box car averaging type) one may be able to see if there is 5 kHz ripple (e.g. 1 PLC non AZ mode and trigger from the start of integration). At 5 kHz the PSRR of the AD707 /OP177 is pretty poor (some 45 dB), so ripple in the 0.5 mV range could already be a problem. The low PSRR for the negative supply seems to be normal for slow OPs, as the internal compensation is usually relative to the negative supply - so it is not a poor quality OP, but just a slow OP.

[branadic]

TiN already confirmed, that the -18V rail measures about -19,4V on his unit too and the zener is marked 20 in his unit as well. So no issue here.

The slow opamp theory is somewhat doubtful, as TiNs unit is almost identical, though different date of fabrication with small differences, no extra local filtering, but better INL performance compared to my unit. So the question is, what is actually different, what has maybe aged and is now running out of limits or is it simply component variation and he is lucky to have "this" component, what ever it is, that results in overall better INL?

-branadic-

[Kleinstein]

There could be a difference in the voltage regulators and capacitors. The speed of the OP177 can also vary - at least the data-sheet allows for quite some range. A low PSRR in slow OPs is a general tendency and compensation towards the negative side explains this. So the PSRR value is not just a test limit with a large margin, but more like a typical value with not much variance.

The different performance between units is indeed puzzling. Bad luck with impedance peaking could be an issue.

[branadic]

Some small correction, the unit I have and that is also with TiN, have AD707 installed. It was replaced by OP177F in later versions.
Next step is, even if they still read fine, replacing the caps on the analog power supply board as a maintenance service for longer reliability, since they are still original after all this years. Not sure if that changes something.

-branadic-

[branadic]

Replaced the caps by good quality Rubycons and measured again, INL hasn't changed, though I haven't expected anything else. Time to swap some opamps.

-branadic-

[branadic]

So here's a small update.
I have replaced all AD707 by OP177F and also changed all LT1056CN8, this slightly improved the overall INL, still it's not as linear as S7081 or K3458A.

-branadic-

[Kleinstein]

That is quite some change from just changing the OPs. The difference from AD707 to OP177 should not be very large.

The curve looks different, but only a small improvement.

[branadic]

Thanks to pipelie I received a brand new VFD for my unit yesterday. Manufactured in 2017 
In case someone needs one, please contact pipelie.

-branadic-

[branadic]

Replaced the VFD last weekend. It's a pita, but pretty straight forward though. Remove the front panel (6 screws), release the front panel board (5 screws), cut all pins of the old worn out VFD, clean all vias, rescue the rubber from the back of the old VFD that is fixed with double sided adhesive tape, put it on the new VFD, assembly VFD to the front board, get all the flimsy pins in place, solder all the pins, give everything a good clean and install everything back in reverse direction.

It took quite some time and I forgot to make pictures during the process, but the result is nice. Everything is bright and even.
Also attached a picture of the back, beautiful structures (chrome) and an interesting construction. Maybe I will take some detailed pictures of the old one, showing the grid and plate wires.

-branadic-

[leighcorrigall]

Hello. Excuse my random hijacking of this thread, but can anyone tell me what a typical "INTERNAL-TEMP" value would be? I am reading between 32 and 36 °C since I turned ON this 2005 variation R6581T. Seems to be off by quite a bit. I would say my apartment is well below 30 degrees.

Anyone know about the temperature monitor? Maybe I will attempt to calibrate it or at the very least, replace it. 

Other than the odd temperature reading, the instrument seems to be working fine with a calibration sticker from 2005/08/01 to 2015/07/01. I inspected the unit and it is clean. The instrument measures a Caddock USF370-10.0M-0.01%-5ppm precision 1 MΩ resistor (4-wire) within specifications and measures voltage from a calibrated Keithley 2200-32-3 reasonably well too.

[Mickle T.]

The LM35 temperature sensor is located near the input muxю
During the last 3.5 years I was reading between 36 and 43°C.

[leighcorrigall]

The LM35 temperature sensor is located near the input muxю
During the last 3.5 years I was reading between 36 and 43°C.

Thank you, Mickle, for this quick and informative response. I am surprised how hot the shielded enclosure gets, but I suppose the airflow is minor. Out of curiosity, what do you do with your temperature measurements?

[Mickle T.]

Out of curiosity, what do you do with your temperature measurements?

These temperature measurements simply accompany long-term drift measurements 

[leighcorrigall]

Master Reference Module Capacitors

I also change the R2 and R3 to metal foil type and add the C4 47UF for the power rail. It might be useful to the long-term stability.
I also tried the C5 with 22nf but found some oscillation occurred, so removed it.

Hi SZSZJDB,

I noticed that you have populated C4 and C5 on the Master Reference Module of the R6581.

So did HLDIY: http://bbs.38hot.net/forum.php?mod=viewthread&tid=28698&highlight=6581

Your capacitors are switched compared with HLDIY. Why is that? I am also curious about what capacitor models you tried. I may consider upgrading too.

Thanks.

---

Another feature I want to understand is why the trace between C1 and the OpAmp has been severed on my R6581T. Someone else (EDIT: TiN) on this thread also has this manufacturing edit, so I know it was intentional.

Few more photos on article WIP..

[TiN]

Most likely they mistakingly connected opamp ground (V-) to the zener signal ground instead of the power ground.

[Kleinstein]

+1 to TiN: it looks like a change to the OPs ground.

I don't think the capacitor models are very critical on the reference model. If at all leakage of C1 may be an issue. With only some 22 nF this could be a low leakage type like PP or PS or in moden times maybe COG MLCC.

Changing the capacitors at the LTZ1000 circuit can be tricky - it a slightly delicate loop with additional gain. C4 at the supply should be not as critical.

The reference noise in the 5 kHz range (modulation frequency for the ADC) could contribute quite a bit to the ADC noise - a point that they seem to have ignored in the design. So some additional filtering may make sense. However this should be more done after the normal ref. circuit. A first test could be for example at U209 (a cap at pins 2 and 6). If this gives a significant noise reduction (e.g. like 10%) one may consider more filtering to get some 1.7 times the effect.

[leighcorrigall]

While examining Mickle's simplified diagram of the R6581, I came across the R201. Has anyone replaced this resistor network before? I understand that Kleinstein has mentioned it a few times. 

Unlike most of the other resistors, these do not have printed values.
Anyone have details about these epoxy-dipped creatures?
R100
R201
R230 45k/45k/500

I am guessing that the resistance values of R201 are 15k/2k/2k/17k EDIT: 13k7/1k83/1k93/16k4. Can anyone confirm this? 

[Kleinstein]

R201 is a custom resistor network.  If the gain stability is OK I see no reason to change or touch it.
So far in most cases it looked like it performed quite well - so no real need to change it. This network is not super critical for performance.
Mickles schematics has slightly lower values, but the ration seems to be close to  15/2/2/17. Only the ratio should be relevant.

R200 is the critical resistor network I would better leave it untouched, unless there is a real problem with it. It can effect the gain drift and linearity.

For custom resistor networks with different resistors there usually is not much info - maybe a cryptic part number and date code.

R230 is only for the auxiliary voltages for ACAL - so nothing really critical. It maybe a more standard divider  - though still not common.

[serg-el]

Shematics R6581 v.5