my noob journey to lower DMM noise (keithley mods)

[Kleinstein]

I would not change the supply regulation. There is not much (if anything at all) to gain from changing it, but much to loose - a not so well designed voltage regulator can easily turn into a power oscillator and damage quite a lot. So for the internal voltage regualtion the integrated regulators like 7815, LM317 are perfectly OK, they are quite reliable.

The still not so very high input impedance is more a question of keeping the board clean. Also repeated soldering could have changed to board material itself an added to leakage there - nothing you can do about it. Also the LTC2057 has a higher bias current, but this is essentially the price you have to pay for lower voltage noise.

[3roomlab]

back during post #71, i have already left the "building" with the 3 analog section VREGs all removed along with the main smoothing caps, heatsinks etc. they are no longer on the original PCB but on a sub-board above the GPIB section. (partially was i suspect, to have stressed them during a short circuit accident)

the main reason for the replacement is that, certain critical sections of the circuit (like the LTC2057) are powered by the unregulated part of the power.

happy new year everyone and happy modding!

[Kleinstein]

The LTC2057 allready has it's own volateg regulation (the two zener diodes), and current sources give a rather good isolation from the rest. So any ripple / noise from that supply will not show up in the output.

With the updated OP (LTC2057 instead of LTC1050) there is not much more room for improvements: There is also no large unacouted noise source - so no chance to find a lucky magic cap that will give a significant improvement. For a significant lower noise  in the low volagte (0.1 and 1 V) ranges one would need a radical different input amplifier design (no more separate buffer stage, JFET amplifier, AZ at front end)  but this also requires different control (e.g. AZ mode) - so it's more like building an DMM from scratch.

It looks like the noise in the 10 V range is significat higher, when measured in volts. So the main noise source here seems to be ADC part of the circuit, not the input amplifier.

[3roomlab]

yea i think so too. while scoping the A/D MUX, i see the noise, but then again this is the only bench DMM i have taken apart, maybe this unit is way way out of spec since it is never calibrated from since 2007, maybe they forgot about this unit since maybe it could be badly out of spec / poor performance.

but so far it is all good, i am just tying up some loose ends (regulator/zener/JFET/optos), and i think it will be the end of DCV mods. then maybe i can start ... to calibrate, or start to make something to calibrate

[Kleinstein]

Even with such an old DMM and with the modifications, there is not much you can do abput the calibration. It's verry likely still quite good. It's rather expensive to get a reference or meter good enough to make a true calibration - thats why such meters are usually send in to special calibration labs.

With the modifications it may not be mechanical stabe enough to do that. Also for normal private use, there is no real need for a fresh calibration. Even if 10 years old it's likely still good enough.

The only thing one can do is a plausibility check, to see if it is way out of specs.

[3roomlab]

i wonder, have anyone on this forum tried to self calibrate? and know what to expect? esp of a K2000, TiN?
it will be a waste to let such a "senior" LM399 lay idle, now that the front end is a diff animal from the original

**update
after reading some measurement articles again. i have come to understand uncertainty
= reading error + range error + tempco error

which means if a K2015 is to measure 10v on 10v range, 1 is to expect 250ppm/2.5mV uncertainty (for 90day, 20+5ppm). i cant find any 24hr accuracy numbers, so if i plug K2000 numbers in (15+4ppm), a 24hr uncertainty for K2015 could be 190ppm/1.9mV (all excluding tempco error which is 3ppm).

did i get that right? but now, how do you re-certify or certify a DMM still have such a range of uncertainty?

[Macbeth]

I've (part) self calibrated my 2015 DC which was consistently +30uV out to my other insruments and volt ref and my 2000 AC which was faulty when I bought it and repaired it.

Hint: Use SCPI for cal and you can go through the steps up to your abilities and then skip to the end and save. If you manually cal from the front panel then you have to go through the whole shebang, HV and resistance too.

I will emphasise I only "calibrated" my instruments to agree with each other and my cheap ass Hao Qi Xin volt ref. They are all singing from the same hymn sheet but are likely a couple hundred uV out from reality 

Also, it is easy enough to make a backup of the calibration EEPROM using something like a TL866 and I highly advise you do this just in case!

I'm looking at making a better volt ref (though my Hao Qi Xin is remarkably consistent at 10.00275V!) and also making a HV supply tracking a ref using a esi Dekavider I recently obtained, or else going the DAC route.

[3roomlab]

i see, looking at the service PDF. i think i have zero gear to try anything serious atm, unless only shorting.
it is good to know i could selectively 1) SCPI init cal 2) choose cal/steps 3) save/lock. that sounds simple enough.
the AD584 you mentioned, i think there is no more L version. i guess i might try to grab a K version, just for sanity checking.

i just managed to finished NPLC1000, 900 samples @ 0.1v. just to see its plot.
 @ 200 samples, this NPLC1000 have STDEV of 26nV and pp of 163nV (which is = NPLC 10 of dr.diesels 7510, wow haha, old machine 100x effort to reach same STDEV)
**updated NPLC summary (10v +NPLC 500)

[Kleinstein]

For an absolute voltage reference, one can buy a reference circuit that is calibrated. This is usually much better than the bare ref. chip.  So it's something like a small board with a AD58x , max6350 or similar that someone with supposibly a good DMM has measured. So you the board and a note about the measured voltage.

[3roomlab]

further experiment

substituting JFETS Q109 113 120 with MMBFJ113 (40v NXP)
and changing new 11v zener DDZ11ASF-7

input impedance takes a dive, but it no longer shows the large "NPLC" timing artifacts
plot is V vs Gohm
tried a 2nd plot NPLC 2 for sanity check
and the 3rd NPLC 6

it is as if the J113 switches better?

**update after 2nd wash, the +ve impedance is still not "happy"

[deadlylover]

after making some inquiries to some china shops selling AD584 type references, i think this time of the year is a bad time to order anything serious. as new year just flipped and chinese new year will be here soon. it doesnt seem like serious gear can be produced seriously ever since from last month. ... time for long long long wait me thinks

Where are you located? (your country tag is broken for me =P)

If you're local I can measure something for you to about 8ppm uncertainty (95%) on 10VDC, otherwise, the shipping charges from Australia is more than enough to build up some nice LTZ1000 references. 

I think there's a good chance your meter is still within 1 year spec. I have an old 34401A from ~1997 that has never been adjusted, and it reads 19ppm high which is basically the 24 hour spec. 

[3roomlab]

thanks for the offer @deadly lover
i am inside a tiny "village" with 2 humongous gambling resorts  ... singapore

i did think about LTZ1000, but atm to keep things simple, the cheap AD584 looks like a step i can take more easily. i dont think i will be sending anything "active" out to fellow forumners to ascertain any kind of standard. but i might take up your offer to measure some passive 0.01% resistors? (they are cheaper and lesser hassle to move around?), or maybe i would ask if the local tek centre can validate resistors  , but i dont think they will, judging from the 14-16 days they need to reply a calibration email    , they then re-routed me to a 3rd party lab instead of a proper follow up.  i have no idea how things work in tektronix. so i might just try to calibrate this gear my self

[Kleinstein]

The 4393 and J113 fets seem to be more or less identical - depending on the manufacturer. Sometimes the 2N4393 seems to be specified for lower leakage and would thus be better. Also the voltage rating depends - sometimes 30 V,35V or 40 V. Anyway the garantied specifications are something like < 100 pA, sometimes even only < 1 nA.

What you want in this application is more like <10 pA at 40 C - so not every unit will work, thoug h many will be likely acceptable.  So one might want to check the fets before soldering them and than hope soldering does not change to much.

The three changes fets are more or less used in ohms mode only, so they Q113 and Q120 might contribute to leakage in DC mode - Q109 should be isolated anyway. The strange jagged curves are more like an artefact from rounding, calculation the slope, so nothing to do with the hardware.

[deadlylover]

thanks for the offer @deadly lover
i am inside a tiny "village" with 2 humongous gambling resorts  ... singapore

Hmm actually, I sometimes find myself at Changi Airport for a short 3-hour layover once or twice a year, I remember seeing a postal service there that I could use right? I might be swinging by within the next 2-3 months.

[3roomlab]

100v VREF ... in theory for now (updated using AN19fc , fig 30 method)
i think i have got some parts of the PWM wrong
(gladly appreciate any tips on this 1)

the VREF side approx draws 10mA (in simulation, R6=1k, lower R6 = lower draw)

exploring possibility of
":CAL:PROT:DC:STEP5 100"

**update
i think i found some new problem again ... ohmmmmmmsss
plot of ohms vs seconds. PTF56 10k verified (voltage-std) ... but whats with the rainbows?
awaiting next batch of JFET 40v 4393/J202

and in the course of discovering keithley 2700 PDF, i discovered my ohms section is indeed faulty from a long time ago
https://www.eevblog.com/forum/projects/keithley-2015-repair-and-the-input-buffer-replacement/msg825815/#msg825815

[Kleinstein]

For generating the raw 110 V, I would consider a royer converter. Its rather simple and low noise because its resonant - just no regulation but a fixed voltage ratio like a transformer. Winding the transformer is not that difficult.

The linear amplifier part look resonable, except for the way the supply for the OP is generated - I would stick to the more classical resistor/ zener and emitterfollower.
The 100 V regulation has no provision for capacitive loading - so a few additions may help. It would also be a good idea to limit the current to something like 1 mA - just for the safety.

[3roomlab]

as with this very "noisy" thread about my adventures, (after many weeks of toying around with my used K2015) i am walking around a large unknown building called "calibration and adjustment" of DMM trying to go inside and play. and while looking at some ebay stuff, i stumbled on something familar, this VREF unit called the D105 that claim to have 2ppm stability (some of you know it has a famous thread  in this forum). and i came back to re-read some of the stuff in there. edit** i have parts i do not understand, hopefully somebody who knows more can enlghten

having experienced the kind of noise i dislike, i saw the last few posts where blackdog showed his quad REF based on linear VREF vs the D105 comparing stability and noise. and then on closer look i started to notice something about the D105 plot, which now on 2nd pass, i started to understand what all the accuracy and low noise could possibly mean.

i (hijacked 2 plots from that thread) superimposed 3 plots, blackdogs' quad VREF 10v, 1 of my best 10v NPLC100 (plot 15-Dec30-0101_LOG.ods) shorted test run and D105's 10v (in correct order as it appears in the pic, top middle and bottom). but as i didnt want to squeeze more of the K2015 plot into a smaller volt scale, i have to offset it and show it as double the original Y-axis size (the left side still showing the 2000nV Y-axis size). for explaination, i have sized both yellow plots to have similar Y scale size, which is the original 2uV per div, my shorted plot is magnified 1uV per div. the X scale is nearly the same for all 3 plots at 6min per div or 1 hour total.

if the VREF have very high noise, it will add this noise/uncertainty by its high rms noise value? or by its pk-pk ? is my understanding here correct?
so if i imagine (if i want to calibrate or adjust this DMM in this 10v scale) my DMM shorted STDEV uncertainty = 0.3uV, D105 STDEV noise = assume @ 2uV, the instance of measuring this D105 uncertainty = 2.3uV ? and if i use a better VREF source like blackdog's, if the uncertainty is assume @ 0.4uV, then this instance of measuring the uncertainty is 0.7uV ? do i sound right in theory?

[3roomlab]

from this article
https://www.wmo.int/pages/prog/gcos/documents/gruanmanuals/UK_NPL/mgpg11.pdf

page 5
As a ‘rule of thumb’, roughly two thirds of all readings will fall between plus and minus (±) one
standard deviation of the average. Roughly 95% of all readings will fall within two standard
deviations. This ‘rule’ applies widely although it is by no means universal.


which means if i measure 10v, with my STDEV being 0.3uV, my reading uncertainty should not just be +/- 0.3uV but should be +/- 0.6uV ?

[Kleinstein]

Noise from uncorrelated sources adds as power not as voltage. So if you have 2 µV and 1 µV from two noise soures, the resulting noise is not 3 µV but the square root of (2²+1²) µV or about 2.2 µV.

The uncertainty given as a +- value due to noise is usually 3 times the RMS value, though this can vary from context to context.

When doing a calibration / adjustment, there is usually enough time to use a long integration / averaging time. So noise is usually not a problem in this context.  A problem might be high 1/f noise - here longer averaging does not really help.

[Macbeth]

i had a peek at what is going on in my local national met lab, they are using k=2. they are based on this french lab http://www.bipm.org/en/bipm-services/calibrations/cms_em.html. i assume that they are not the only entity doing this service to every country, they look like contractors which anybody with alot of money (and scientists / gear) could setup and run.

Here is the fixed link for you!

(You left a full stop '.' at the end of the URL so it failed  )

Both links look very interesting 

[3roomlab]

i had a peek at what is going on in my local national met lab, they are using k=2. they are based on this french lab http://www.bipm.org/en/bipm-services/calibrations/cms_em.html. i assume that they are not the only entity doing this service to every country, they look like contractors which anybody with alot of money (and scientists / gear) could setup and run.

Here is the fixed link for you!

(You left a full stop '.' at the end of the URL so it failed  )

Both links look very interesting 

ah thanks
ok 10 months of me in a time machine away from the bench ... i had to re-connect the last question (10months ago) which i didnt quite understand

in the multiplying of the STDEV/RMS by 3, is it related to what is called coverage k value in this document (https://www.wmo.int/pages/prog/gcos/documents/gruanmanuals/UK_NPL/mgpg11.pdf) page 23 item 7-4. which then makes confidence level of the uncertainty to 99.7% ... i think.

i had a peek at what is going on in my local national met lab, they are using k=2. they are based on this french lab http://www.bipm.org/en/bipm-services/calibrations/cms_em.html i assume that they are not the only entity doing this service to every country, they look like contractors which anybody with alot of money (and scientists / gear) could setup and run.

if i understand the 28nV/V correctly, am i right to say it is saying their 10v reference is 0.28ppm in uncertainty?

[Kleinstein]

28nV/V is 0.028 ppm.
The web page is from the french metrology lab, it's the JJA they use and the limits of there comparison tools. So no surprise to find a really good value - but likely rather expensive. This is more like a lab where normal calibration labs may send there high end calibrators.