K2001 ADC

[GigaJoe]

looks like 2 version,

approx 1993 , no composite


711 should be where R811,  just a single 602 as integrator .....

like ...  probably possible to add mod ada4077 + 2 resistors ....  question:  would it screw up something ( at most) ,  does it help with noise ( goal )

[LooseJunkHater]

Can you provide some background on what this is, as well as clarify what you're wanting to do?

[Kleinstein]

There are 2 different versions. One with and one without a 2nd OP-amp. The OPA177 has slightly better low frequency (e.g. 30 Hz to handle 1 PLC or 3 Hz for 10 PLC) noise than the OPA602.  Still the integrator noise is only one noise part. One may not even see the difference.
The divider at the compund integrator is used to slow down the precision OP-amp. So the relatively fast ADA4077 or similar OPA205 may want a different divider than the OPA177.  The 1 K / 100 ohm in the schematics are a bit odd choice: rather low resistance and a bit much for the relatively slow OPA177 already. Beside a little less noise the 2nd OP-amp can give a slight DNL improvement: the constant impedance at the integrator is no longer important and less error from not having it. In theory there may be corrections in the software, so that a better integrator could result in a worse result from the corrections for an not ideal integrator.

The ADC in the K2001 is quite noisy for different reasons, with no easy way out. In addition there is a gain of 0.5 for the 20 V range and the input buffer is also not super low noise.

[GigaJoe]

thanks,
so " do not rock the boat" ....
if i understand it correct: if i mimic schematic , 1.1K basically loaded 177 , skewed DNL, but as it constant , it may hardcoded in a software ,with part i dont have ...
and if i replace 602 by let say composite , it still negligible due to overall design ...
so in overall fight with noise the way by averaging only,  means ... ideally to make this noise as close as possible to uniform ,  but ... electronics usually produse gaussan noise ..

[Kleinstein]

There are several reasons why the ADC is relatively noisy. One point is switching only one reference side and add the other side as on offset. This is an integral design point an nearly impossisbly to change. As a result there is quite some noise gain ( ~ 5.3)  for the integrator (and similar also some other noise sources). The OPA602 is relatively low noise for an old JFET op-amp aready. This is not so much white noise, but the 1/f noise as much is about relatively low frequencies like 3 or 30 Hz for 10 or 1 PLC operation. With the somewhat complicated AZ cycle of the K2001 even lower frequencies may matter too.
I have not looked at the noise in detail - maybe if I find time I could enter the numbers (needs still a few thoughts for design detail) to my spreadsheet.


For the linearity the problem would be some hard coded (or after special factory calibration) corrections for the relatively slow OPA602. The simple 1 OP-integrator input acts like like some resistance ( 1/ (C_in*2Pi*GBW)) to the virtual ground. And this may need a correction for the fast run-down step, as it looks like this happens with a different impedance at the input. With the rather large integration capacitor they may still get away without a correction.

For a compound integrator with ADA4077 + OPA602 I would more choose some 10 K and 1 K for the divider. One only needs an effective speed ratio of about 10 (can be as low as 6 with suitable RC to ground a fast amplifier that is good driving capacitive loads). No idea why they choose so much devider with the slow OPA177 in the other version.

[Kleinstein]

The point for possibly upgrading the integrator would not so much be noise, but the linearity. One could test for the DNL errors that are expected from the simple integrator without keeping the impedance constant. The expected error would be a saw tooth like contribution to the INL with jumps when the RU steps change. In between the differential gain would be a bit lower than ideal. One may be able to see the issue with 1 PLC conversions.

[GigaJoe]

it can be really difficult for me to test linearity even if improved
Guess that idea of change getttin discarded ...

[Kleinstein]

The relatively high noise level makes INL testing a bit challanging. One could give it a simply try by looking a slow drifting voltage (e.g. RC discharge). If I found it right the Run-up modulation is 4.5 or 5 kHz. With 20 V range and 1 PLC this would make some 500 mV per RU-step.
So a signal would be something like a slow RC discharge from some 1.5 to take some 15 or 30 minutes and record data at 1 PLC non AZ at the full rate.
After subtracting a fitted exponential and maybe some averaging, one may be able to spot jumps that reat about after 500 mV. I my crude estimate is right I would expect something like 5 to 10 µV.

[GigaJoe]

second look at schematics :

added polymer 100uF for zener 8V and 6V
input short for max resolution no filter , 100 counts measured - 0.23 - 0.27 uV  P-P  ( 8V seems more important then 6 )
10V with filter ave-100, has around 6uV P-P ; but this most likely contributed by V source .... pattern of moving back and forth aperiodic,   guess some random spikes change the picture ...
there is no circular moving of measured result as was before ,   it aperiodic now ,   most likely caused by source ....

I'm kinda happy for such result ...  maybe next round to change something  ... who knows ...  I need a better source and logging for now ..

[Kleinstein]

The digital offset compensation (AZ mode) should be relatively fast. It's purpose is to reduce the very low frequency noise. Ideally there should be not big jumps for the offset, but more thermal like fluctuations. My suspicion is that the offset correction does not work all that ideal in the K2001 (and also quite some other meters).
In the Allan variance curves there is usually some extra noise bump visible already with a shorted input. The somewhat complex AZ cycle with addtional steps for gain correction complicates things and there can be additional jumps in the gain (some are seen with the DMM7510 during warm up). One point that could go wrong is the wrong internal zero point (e.g. at one end of scale, or anominal zero that is a bit off) for the ADC scale, so that gain correction could carry over to the offset.
Tests with an external voltage add noise from the external reference, the meter internal references (ADC zener and long term LM399) and the gain corrections. The transition between the 2 separate references in the K2001 is a tricky part with not many things known. It looks like things are different for 1 PLC and 10 PLC (or longer). The longer than 10 PLC cases are likely just averaged from 10 PLC readings and not that interesting for the understanding.
For the start and for the ADC noise the shorted input case is more suitable to look at ADC noise. I is the easier case to understand first.

A good way to looks at the noise is the allan deviation with relatively long data sets, something like 10-50 x the longest time of interest. So this would be a few hours unter stable thermal conditions and well warmed up.

p.s.
Some of the visble jumps could well be from the LM399 ref. These commonly show popcorn noise jumps of the some 4 µV which would translate to some 6 µV with 10 V voltage.

[Kleinstein]

The statistics is about noise. With enough averaging one gets the noise (or at least what the simple statistics thinks) down. Accuracy is a different thing and inludes effects like temperature drift, aging and the adjustment / calibration.

The test with the external 7 V reference tests the amplifier, ADC and the refences. To get only ADC and amplifier one does a noise test with a shorted input.

Especially with the K2001 it can make sense to look at the noise not just with the simple statistics, but also plot the Allan deviation. This because the K2001 is known to not have white noise as the simple statistics assumes. So the simple noise estimate is to be taken with a grain of salt.

[Kleinstein]

The Allan deviation curve looks like there is something wrong. The slope it is going down too steep. It should initially be like the square root (or a little less) and will normally only get less steep and at some point turn up from drift.

[miro123]

I think al;so that there is something or many thinks wrong.
My guess is that this is measured with input short - but even then i expect to flatten and go up
Other option is that is used wrong function in alan python module

[Kleinstein]

The new curve looks much more realistic.  The y-scale it quite zoomed in and x and y scales thus quite different.
The maximum in the Allan dev. curve is quite common with Keithley meters, but the exact source is not known. Chances are it is linked to some background averaging in the auto zero mode that also includes gain correction.

[GigaJoe]

some draft ....

coefficient a bit need adjustment ...   but how it looks like on my non filtered data , again .. 10NPLC , no any filter
spikes cutting really well ,    not sure what would be next ... averaging , then stats ...

[Kleinstein]

The noise level indeed looks high. Even the 9 µV_pp are not really good. The K2001 is known to be a bit noisy in the 20 V range. Part is from the extra divider (gain -0.5 at the amplifier) that doubles the ADC noise and also from the the buffer, that can have quite some 1/f noise. This 1/f noise could be quite different betwen units.

The extra Lf noise from the AZ mode oddity can also add to the noise.
 In theory the protection with MOSFETs can also contribute a bit more than usual thermal EMF, but it should not be that much unless the temperature changes a lot.

Leakage at some switches is another possibity. For a test one could check the input current, if it is higher than normal.

[Kleinstein]

If they have a mix of DG211 and DG411 they probably have a reason for this. If possible one would reduce the BOM and use only eine type. The DG411 is lower resistance, while the DG211 is lower leakage, capacitance and likely charge injection. So both have there pros and cons.

The 20 V range is actually not using the CMOS switches very much (only a few in the gain setting).
The positive input side gets switched to ground with relatively low resistance. I would not so much expect leakage there to be an issue. If the CMOS switches fail they tend to fail completely and not so much start leaking.

Before doing anything to the meter, it would be a good idea to first do more tests and get more data.
One point would be the noise for the 2 V and 200 mV range (with a shorted input). Here it would likely be data for 10 PLC with no extra digital filtering / averaging and than look at the Allan variance curves.
Another point is the input leakage (e.g. looks at the voltage dirft with some 10 nF of low leakage capacitor (PS or PP type) at the input. Leaking switches (specially JFETs would often also effect the input current.

Another point is looking at the supply ripple - extra ripple there could cause higher than normal noise.

AFAIK the strong points is having a 20 V high impedance range, good gain stability (low TC) and relatively high BW AC ranges. The 200 mV range may be reasonable low noise. AFAIK the K2001 is a bit noisy by design and not really made for highest precision or low noise.
It is a rather old design, that is likely only kept alive because of some (military, avionics ?) customers.

[Andreas]

Hello,

in comparison to my K2000 in 10V range the std dev over 100 samples seems much too high.
on a 7V reference I typically have below 1 uV std dev.

So it should be maximum the double for a 7.5 digit instrument in the 20V range. (or even better).
I would also look if there is excessive ripple on the power supply.

with best regards

Andreas

[Kleinstein]

From the old DMM noise thread it looks like the K2001 is quite noisy in the 20 V range also for there units. In comparison the 2 V range looks quite good.
https://www.eevblog.com/forum/metrology/dmm-adc-noise-comparison-testing-project/msg726789/#msg726789
So the 20 V noise measurements could be real and normal for the meter.

There may be a principle issue with the 20 V range. One possible issue could be the buffer amplifier (U341 = AD548) in combination with an AZ cycle that is not good in suppressing 1/f noise. Another weak point could be settling after switching between 0 , 20 V and 7 V reference.

It may be worth checking what AZ cycle is actually run for a 10 PLC setting in the 20 V range. A first step is looking at the amplifier output with an DSO.

[Kleinstein]

An OPA140 (or cheaper OPA141) on an adapter board would be an option for a lower noise buffer. It still depends on the AZ cycle if this would really make a difference, but it could.

[GigaJoe]

Continue:

NE5534 ( U802 ) on ADC that basically kinda null detector or something ...
at first did try to replace by ADA4077 ,   but my gut tellin me it wrong ..   mostly not enough speed ... and i was correct ... gettin random number generator. wow - a bit wrong timing and everything skewed ... a good news that the point ...

2nd shot:  OPA140  (correction caps intended for NE - ignored) .... and it looks much better then NE5534, it really much better as after about one hour of warmup only last digit jumping without any filter .... my guess to idealize, it need to replace LM311 by something much faster and precise.

I'm thinking ...  that NE5534 has X20 amplification, in consideration of offset and noise amplification..... What happens if I bump it to let say X40
Will it increase precision of "null detection" therefore decrease the noise ?

the next thing: OPA602 , im not sure how much the noise i'm able to reduce by replace it.  TI official  telling OPA145 as a replacement ...  so def OPA140 would be ok .... Im' thinking ..... ( composite on small add-on board with ADA4077 + OPA140, same as an original schematic, but diff opamp  looks much more attractive)

for reference, schematic https://i.ibb.co/840yf0s1/ADC-K2001.png

damn ...   it gettin better and better

[Kleinstein]

The slope amplifier has 2 purposes:
1) it amplifies the signal around zero crossing and this way makes the noise and offset drift of the comparator less relevant. The comparator also sees more voltage and thus gets faster. The amplifier noise is usually better then the comparator noise. There are still other noise sources, especially the integrator (OPA602) and the 5.1 K resistor at the slope amplifier input.
2) it sets the bandwidth for the detection of the zero crossing. It is a balance between enough speed and low bandwidth to reduce the noise. For the speed there is GBW / gain and C882*R800, which should be the more relevant part. Still the ADA4077 may well be too slow and add too much delay.

I would not expect much of an improvement from more gain. The LM311 is not that slow and a gain of 20 is already quite a bit when it comes to noise. Precision of the LM311 should not matter that much, with the extra gain in front.

The OP602 (U809) is the fast part of the integrator. That op-amps noise has a limited effect for the final charge (like the NE5534) and thus mainly for short integration time (maybe 1 PLC, but mainly less). A good substitute could be OPA1641 or OPA1677 (with an adapter PCB). Higher speed could help a little with linearity. No need for a low offest (OPA140).
The more relevant OP-amp for longer integration (e.g. 10 PLC) is the slow OP-amp (U813 OPA177) it sets low frequency noise with a noise gain of some 5.3 . The current noise is hurting even more.  The ADA4077 (or similar OPA205) could be a substitute here, but may need a change (larger) to R864 to compensate for a higher GBW. One would want R864/R865 >= 6 * GBW(U813) /GBW(U809) and maybe larger to get a similar value for R865/R864 * GBW (U813), even if the rest of the integrator is faster.

[GigaJoe]

Kleinstein, thanks a lot for input ...

My current step, try to replace an old ADC with single OPA602.  So i removing OPA602 and substitute it by  composite OPAMP1 + OPAM2
What would be ideal types for both opamps to substitute it ?

just wonder ..  i thought about 5K resistor, but more concern about 100K resistor . so calculation for 100K give me around 1 microvolt - is it right?
still why  not like 1K 20K .  or less load with 100K on opamp output so better recovering time ... it seems timing are crucial , as with 4077 , only 2 digits after comma was solid ... rest random flipping

[Kleinstein]

A single OPA602 would give a poor integrator with rather high low frequency noise. I think the 1 OP-amp version used the expensive OPA627.

A single OP-amp solution could use an OPA140/OPA141/OP1641.

For a 2 OP-amp solution one could use something like ADA4077 or very similar OPA205 for the precision part (replace OPA177) and OPA1677 for the fast part. The divider between the OP-amps has 2 options:
a) aim for fast settling and a ratio like  2:1 to 3:1  (for the amplifiers above).
b) aim for settling slower than the modulation so a ratio more like 200:1 to only correct the slow part, similar to the K2001 2 OP-amp version.

Smaller resistors at the slope amplifier would cause more loading to the integrator. This would definitely be an issue with the 1 Op-amp version and even the 2 amplifier version may not like 1 K, but maybe 3.3 K.
The 100 K resistor is less of an issue as it is relative to the output and thus after the gain of 20.
The noise of the slope amplifier should be mainly an issue with short integration, less at 10 PLC.

[GigaJoe]

actually my first post with the image ... single OPA602A
okay ...

so options

1 - throw out OPA602A  in place OPA140 - see result ...
2 - add on board with ADA4077 ( like it)  + hmm ..   I have OPA1678 (a lot ... ) technically  simply shorted one to ground to test an idea ...  I probably go to this way if opt-1 give unsignificant result

original K2001 2 OP-amp version resistors 1K and 100 so 10:1