Advantest R6581 8.5 digit DMM mini teardown/repair

[TiN]

What are the obvious parts? R008-R011? Not so obvious to me, TBH. 

And now input MUX ADC is cooked? Can you crack the hood open for us to see what's inside? You can heat it up (e.g. using small gas torch) and then cool the top with wet small towel.
Top ceramic likely will crack and fall off without resistance 

[TiN]

I did few quick checks on R6581T. This is definately not the "replace the fuse and she'll be alright" repair here. 

There is wierd issue with resistance ranges 100Kohm and higher, showing consistant error over 300 ppms, even after fresh ext cal.
I've used uncalibrated 5700A for this test and linearity test, and my FX protoref for 10V log.

EDIT: Added 3458As result vs same calibrator.

[Mickle T.]

Critical bug-fix for R6581T with firmware 004246-A02/007342-A01, 2006 yr.
Errata: an erroneously swapped bits in the relays control word are breaks the switching sequence of K006 relay and leads to large thermal EMF due to overheating.

[Kleinstein]

I did few quick checks on R6581T. This is definately not the "replace the fuse and she'll be alright" repair here. 

There is wierd issue with resistance ranges 100Kohm and higher, showing consistant error over 300 ppms, even after fresh ext cal.
I've used uncalibrated 5700A for this test and linearity test, and my FX protoref for 10V log.

A simple repair with just replace the fuse and maybe all the electrolytic caps is much less fun than such a hidden fault causing poor performance.

The problem wirh the higher resistors could be a problem with leakage currents, present with the external terminals. 300 ppm for the higher values could also be a kind of problem with waiting times.

The INL curve looks kind of interesting. It the ppm scale relative to the range or relative to the actual reading ?
I would consider the absolute error or relative to the range more suitable.

[TiN]

A simple repair with just replace the fuse and maybe all the electrolytic caps is much less fun than such a hidden fault causing poor performance.

Yea, indeed. However current schedule do not permit such time investment. Will do some more testing, but more in depth troubleshooting will be for later on.

I thought of leakage too, but 10K range is nearly spot on. That what puzzled me. I've used usual settings, OCOMP, DELAY 3, 4Wire, just like with 3458A. I'll post more detailed data, once I confirm there is no problem with the calibrator (that is repair project on it's own due to AC board failure). Calibrator is being tested now versus 3458A. So I can also use 3458A results as a reference for repeated INL test.

[szszjdb]

Hi,  Mr. Mickle.T and Mr.Kleinstein ,

Here comes the trying these day.

1. The INL issue are nearlly the same even with the new cap.  Please refer to the attached. There alway have a turn over point around 7-8V input. As I know , the 6581 only calibrate the positive 10V and negative 10/1/0.1V together with the zero point . It seems have no correction if the gain is not flat in the whole 10v range.  Which might lead to that?   The transfer error for 10v-1v is still in the 12-14uv range.

2. Futher more, I had replaced the U214/DG442 with reed relay, no obvious change viewed . I was worry about the leakage of the DG422 before.

3. I had chaneged the K006 to low EMF type reed relay.

Focusing on the ADC, but have no idea about the next step.

Hi, Tin,
Your INL issue is like mine. Hope to find solution!

Best Regards,
szszjdb

[Mickle T.]

There is wierd issue with resistance ranges 100Kohm and higher, showing consistant error over 300 ppms,

I have found a similar problem with 10 MOhm (and possibly 100 M and 1 GOhm) range: 100 k -> 1 M transfer error is about 1 ppm, but 1 M -> 10 M gives a repeatable +150...+180 ppm multiplicative error 

Have anyone tested a full version of R6581 (i.e. non-D and non-T)

[Kleinstein]

U214 (the CMOS switch in the ADC section should not be critical.

For the 10 V - 1 V transfer it is more like some problem with fast (during ACAL) / slow (manual) measurements. This would be the classical DA part from the integration cap. However there is not much more one can do about this. PP caps are about the best affordable caps at 20 nF. PS caps might be slightly better, but not really by much.  The effect of DA would be about the following: When measuring near full scale charge is stored in the cap and thus the reading a little smaller and the next reading (e.g. the zero) tend to read to high. To a first approximation and slow changing signals (e.g. 100 PLC mode) this would lead to reduced gain for all readings and thus not a problem. However when switching fast from a high reading to a low reading this would also bring the readings together, even with the zero readings in between. So in fast sequence the 1 V read in 10 V range would read a little higher and the 1 V in 1 V range a little lower. Thus the 10 V to 1 V transfer is expected to give a small error, towards too high a reading in the 1 V range. The factor is expected to be about the same for the 1 V to 0.1 V range transfer.  There might already be some software correction for this effect, but this is really hard to tell.

Another possible trouble would be, if for some reason the TC of the critical resistor array R200 (input part of the ADC) has increased. This could lead to thermal effects, like a time dependent response and additional INL from self heating. The performance could really suffer if R200 turns bad (e.g. higher TC). I am not sure if there is a possible independent test for these resistor array.  The drift in the non AZ mode ( 10 V range) might give some indications, but it is kind off difficult to get hard numbers this way. I would be very careful with R200 - this is a really critical part for the ADC and likely impossible to replace.

If it is just the 10 V to 1 V transfer one might have to live with a correction factor applied afterwards if the 1 V / 100m V are used for highest accuracy.



These time dependent effects would be best visible with the PC interface.


p.s.: The DA effect on the fast slow readings is more complicated than I thought: it depends on how the auto zero measurements are done. So it is hard to tell upfront how much error to expect and which direction.

[simon51]

Critical bug-fix for R6581T with firmware 004246-A02/007342-A01, 2006 yr.
Errata: an erroneously swapped bits in the relays control word are breaks the switching sequence of K006 relay and leads to large thermal EMF due to overheating.

Yes,This is a big bug!

[ap]

Critical bug-fix for R6581T with firmware 004246-A02/007342-A01, 2006 yr.
Errata: an erroneously swapped bits in the relays control word are breaks the switching sequence of K006 relay and leads to large thermal EMF due to overheating.

Sounds like you got some data for this meter. I have not been able to find any user manual in english (all  japanese) and no service manual at all. R&S who supports Advantest here has none either (meter not sold here). Any idea?

[Mickle T.]

Sounds like you got some data for this meter.

Unfortunately no. But I have found another bug

R6581T, firmware 004246-A02/007342-A01.
Errata: switching the measurement mode from 2W-OHM to DCI doesn't reset the corresponding bit in the control register U407, so the current source remains on until DCV mode is selected. This can lead to a spurious leakage when measuring low currents at 0.1-1 nA ranges.

[simon51]

Critical bug-fix for R6581T with firmware 004246-A02/007342-A01, 2006 yr.
Errata: an erroneously swapped bits in the relays control word are breaks the switching sequence of K006 relay and leads to large thermal EMF due to overheating.

Hi,  Mr. Mickle.T
 My 6581T firmware version is same with u, but the K006 dirve voltage is diffrent. I think it should be like K008. Look the picture please. How can I do. Thank you.

[Mickle T.]

I has the same relay drive voltage.

[simon51]

I has the same relay drive voltage.

Oh,I see.Thank U!

[simon51]

I has the same relay drive voltage.

After patch, It's OK!

[szszjdb]

Hi,  Mr. Mickle.T and Mr.Kleinstein ,

For the INL improvement , I had replaced the R200 with 4pcs 20k foil resistor(2.5ppm/C, 2 in series and 2 in parallel  ) , still found no improvement. The turn over  and transfer error are still 10-13uv at +1V input in 10v range and the transfer error is  2uv at -1V input in 10v range. The negative side performed much better than the positive as before.  All the test are steady and  repeatable.

I also checked the output waveform of the slope amplifier ,the U207 LT1220, and I don't know if it is correct. It is the red channel in the attached picture.

Any advice for the next step ?

Best Regards,
szszjdb

[Kleinstein]

The slope amplifier waveform looks about correct. I very much doubt there would be a problem here. 
A waveform that might be of some interest is the output of U205 (OPA177/AD707 in the integrator). This should show a signal similar to a square wave with reasonable fast (e.g. < 5 µs) settling. Due to the rather slow switching, I don't expect it to be critical anyway for the slow mode.

Replacing R200 is not such a good idea, as soldering can already effect the resistor. Chances are that the original C200 was tighter specs than the replacement. At least I would consider that resistor so critical to requite something like a TC in the 1 PPM/K range. Anyway R200 got the mistreatment, so it is to late. I thought my warning was enough to prevent this.

INL due to a higher TC of R200 would be a contribution proportional to U³ anyway. So it would not explain a different positive and negative sign.
A ratio that is slightly off could have an effect, but this would likely be more like an DNL problem at a few critical points. Chances are the replacement resistors are not that accurate, or at least not the same ratio as the old one.

For the low voltage turn over error there is still the possibility to have some kind of error in the zero measurement, like due to some leakage around the input MUX. Checking the turn over error for a few voltage could give an indication if it is an offset problem - it is still possible to have an error that looks like an offset, but is not.

A next step would be to get the PC interface running and check for step response. Another point to check with the PC would be for massive DNL errors - e.g. check a capacitor charging / discharging (e.g. from bias).

Is the transfer error for 10 V to 1 V range the same as for the 1 V to 100 mV range ?
A similar factor would kind of indicate a problem with INL, while a much larger error for the 100 mV range would indicate an offset problem.

While there are quite a few point't that can cause nonlinearity, I don't see many points that would change with age. One point might be aging caps in the decoupling and thus increasing supply ripple / ringing. However this is really hard to track down.

[szszjdb]

Hi,   Mr.Kleinstein ,

Thanks a lot!

I just replaced the R200 and do a fast check last night. The detail of the transfer and turn over error are as below. The transfer error are both 13ppm on 10V-1V and 1V-0.1V range, but just 1.3uv on 1V-0.1V range and 13uv on 10V-1V range . The turn over error is almost the same with the before.

I had replaced the cap on the power board and found some little improvement on the noise performance. None of the cap on ADC board had been replaced.

I will find the GPIB cable in next week.

Any other check should I make?

Best Regards,
szszjdb

[Kleinstein]

The meter does the 10-V - 1 V - 0.1 V transfers with negative voltages. The error in the 1-4 ppm range is not that large. Especially the 100 mV transfer could suffer from some offsets. 4 ppm of 100 mV is just 400 nV. This is not just during the internal transfer, but also during the external measurement.

The poor transfer for the positive sign corresponds to the turn over error and thus INL error. A 10 µV range error is too much to be a simple offset problem. So the main problem is like poor INL of the ADC or the amplifier.  For the amplifier in x1 mode I doubt it would cause so much error. So we are back to the INL error of the ADC.

It is kind of odd that the turn over error is so large with small voltages. This only covers linearity over a small range in the middle. Thus normally the turn over error is expected to get smaller at low voltages. It might help to get a few more points for the turn over error at smaller voltages - e.g. 0.25 V , 0.5 V , 0.75 V  - still in the 10 V range.

One possible source of INL around zero is capacitive coupling from the comparator towards the modulated current source around U202 / U203 and the reference inverter.  So one could test to add some shielding to these areas.
Similar there might be some similar effect of coupling via the supply again from the comparator / slope amplifier towards U202 / U203 / U210.
One could check some extra decoupling (e.g. an extra 100 nF) at U202 / U203 / U210.

Another possible test would be adding an offset to the ADC input. The idea would be to shift the problematic area a way from 0 V. The shift could be by adding a resistor (e.g. 1 M )  from the +19 V towards the neg. input of U200/U201.

[szszjdb]

Hi, Mr.Kleinstein ,

Thanks a lot!

The result of the turn over test for 0-1v input at 10v range is attached. The error is decreasing to around zero.

I had added 4.7uf cap on C214/C211,which is the decoupling cap for the comparator U208. No visible change found. I still need more time to change others decoupling cap around ADC and report.

I had also added the copper plate above the U200-U203,  no visible change found.

You means to add a bias to +17v or -19v ?

Any further advice?

Best Regards,
szszjdb

[Kleinstein]

Trying to add shielding and extra decoupling is kind of long shots. So chance are many tries won't change much.  However I am afraid that measuring supply noise won't help much either, as it would be small and thus likely not visible on the scope. And even if one sees something, it's not clear if real or just the wrong grounding point or an artifact of the probe. Anyway looking at the supplies again would not hurt.

The turn over tests for lower voltages look rather smooth. So it is more like a smooth type of INL error. Still quite a lot happens below 100 mV.
The 0.5 µV zero error can to a large part explain the 1.5 µV value for the 10 mV turn over.  So it does not look like the +-10 mV is that bad. It does not look like a very localized type of error that could happen due to capacitive coupling, but more like a continents bend over the +-0.8 V range.  The -8..-1 V and 1 to 8 V range might be Ok again - at least the turn over tests so far don't show clear trouble in the part. Above about +- 8 V there might be a problem again, but this might very well be a totally different one.

With this type of error I don't think it would help to add the offset - this would be something to help of there would be a very local problem at lets say +-50 µV, that by some coincidence just moved to zero. With the more continuous error it won't help.

For the ADC the range around 0 is not that special: the input current source has the added current from the -19 V and is thus just in the middle of the range. Also the Feedback during run-up is not that special at 0 V.

Most of the  error developing over the +- 0.8 V range makes me think about a diode drop.  There is a small chance the U108 in the input protection could cause trouble, not providing the U_in*1.07 +-1V. Another point worth checking might be the fast amplifier: are the very fast modes (<= 100 µs) working at least reasonably ?  A broken fast amplifier could load the precision channel.

Some more tests related to the input (like voltage dependent input leakage) could be done much easier with the PC interface.

Edit:
To dampen possible coupling to the reference one could also try adding caps (e.g. 10-100nF range) in the feedback of the reference amplifiers (7 to 17 V and 17 to 19 V step). Besides filtering possible switching effects it would also filter reference (and amplifier) noise and this way could give a slight reduction in noise.

[szszjdb]

Hi, Mr.Kleinstein ,

Thanks a lot!

I have checked the output of U108 to D003/004 and it is right the Uin+-1V. I also tried to desolder the D003/004 and found no improvement for turn over check.
I checked the reading when in 100US PLC and it is normal,  but it just have 6 digit and the number is jumping very fast in the last 2 or 3 digit. So hard to find out something. But I can find a replacement for U104 and try it.

I will add some decoupling cap around ADC and check the improvement for noise and INL.

I am waiting for the GPIB cable and report when install it.

Best Regards,
szszjdb

[Mickle T.]

Some R6581 A/D conversion waveforms.
P.S. I was surprised that PLC100 A/D timings <> 10x PLC10 ones in the RESET phase 

[Kleinstein]

If U108 is working correct there would be not need to desolder D3/D4. These diodes are important for ESD protection. Soldering in this area might also change the input bias.

I don' t think U104 will be a problem, if this is bad the amplifier would not work at all.

For the fast mode it is only important that it is working at all - no need to really check for precision. If would have been a really broken fast amplifier that might have a negative effect.

One could try to measure the turn over error in the 1 V range, to check if the error appears more like in the 0.1-0.8 V range or more like in the 0.01 to 0.1 V range, thus the same voltage at the ADC.  I know this could be difficult, as input offset and noise will be more important.

I don't think the trimmer R170 at the DC amplifier should be that critical - so one could check if it makes a difference. The trimmer would influence the coarse offset of the amplifier and no accurate adjustment should be needed here. There might be a slight difference if the fast amplifiers runs to the positive or negative rail if the slow amplifier is active. I don't know how the trimmer should be set, but a could imaging a setting to get a similar (but not too close) offset for the slow and fast mode could be about right - a small offset could be a another target as well.

[szszjdb]

Hi, Mr.Kleinstein ,

Thanks a lot!

I had checked the turn over error in 1V range before(on Feb.24) and it seems better than the 10V range. You reminded me that the ADC might be fine as it is also used in 1V range. So the issue is pointed to the DC amp..  But how to explain the test data from HI port to ADIN in 10V range? It is also fine in that measuring.
I will re-confirm the 2 test tonight.

I had trimmed the R170 before and it do change the offset of the precision channel, but have no effect on the turn over error.

Any further advice?

Hi,  Mr. Mickle.T ,
Could you tell me where is the test point of the red channel in the last picture you attached. Why have there a flat part in the beginning of the run down phase. Is there fix timing for S1024 to S1?

Best Regards,
szszjdb