Project: pimp a Keithley 2000

[Andreas]

Hello,

although the Keithley 2000 is a fine 6.5 digit instrument,
which outputs resolution in excess of 6.5 digits over the
communication buses there are 2 points which are annoying:

First there is a temperature dependency (T.C.) of the reading
in the order of around 1 ppm/K which makes the higher
resolution nearly useless.

Second there is a large dependency on orientation of the
instrument when tilting on the left or right side.
Peak values  are +/-70..80 uV (temporary) measured on a stable 7.2V reference.
with +18/-42uV in steady state giving a span of 8.4 ppm.

On a HP34401A I could reduce the tilting effect from 3.5 to around 1.4 ppm
simply by better thermal isolation of the LM399 reference on both sides of PCB.

For the tempco the idea is to put a temperature sensor within
the Keithley. Evaluation can be done by a software on PC which
samples the readings via RS232 or HPIB-Interface together with the temperature.
There are some unused spare pins on the RS232 connector and the trigger connector
which can be used for connection to PC.

A possible solution would be a RS232 dongle which appends a temperature reading
to every measurement. Another idea is a high resolution secondary display
connected to RS232 showing the corrected readings.

The trick is to find a place which represents best
the T.C. behaviour of the device.
So critical locations might be the resistor dividers,
the LM399 and the surrounding scaling resistors.

One challenge is that a sensor shall not compromise the
isolation barrier of the floating part of the instrument.
So a location on the earth referenced part
near the critical location is preferred.

Since T.C. is probably different in every range of the DMM I want to restrict the
compensation to the 10V and 100mV range.

with best regards

Andreas

[Andreas]

Pre-Tests:

As pre-test I did a warm up test with 4 NTC sensors and a stable 7V reference (LTZ#5).
First NTC on top of the K2000 in the middle of the instrument.
2 further sensors at the front sense connectors
Last NTC on top of the metal shell of the RS232 connector.

If I use the RS232 NTC as correction value (green dotted line) it follows the actual
instrument reading within 1-2uV when starting the measurement 30 minutes
after switching on the cold device.
So a stability improvement of a factor 10 over my 18-32 deg C "lab temperature"
range seems to be realistic.

with best regards

Andreas

[Andreas]

Analysis preparation:

For further analysis I placed the 4 NTCs within the instrument.

First NTC near the trigger connector (there are several unused SMD-Pads which are connected to spare pins).
2nd near the LM399 but on the earth referenced side of the PCB.
3rd on top of the SMD resistor array.
4th near the 100:1 input divider

The LM399 gets a additional thermal isolation.
Unfortunately there seems no easy way to isolate the LM399 also from the bottom side
of the PCB without completely removing the main board.

And I also think that placing the very hot LM399 directly at the
critical resistor array is a design fault.
So my hope to reduce the tilting effect is nearly gone.

with best regards

Andreas

[Andreas]

Analysis:

first picture in 10V range with LTZ#5:
warm up cycle from 30 minutes after switch on with temperature variations of around 4-5 deg C.
Correlation calculation shows that the NTC near the trigger connector has the best fit for T.C.
T.C. to this location is around +0.55 ppm/K. (green curve is interpolated value).

2nd picture with difference voltage of LTZ#5-LTZ#1 (around 60mV) in 100 mV range.
no clear correlation visible within 2 deg C temperature variation.
-> will have to test if a correction in 100mV range makes sense.

3rd picture tilting effect.
with the LM399 isolation the tilting effect is now +16/-33uV in the critical directions in steady state or a span of 6.8ppm.
So not much improvement up to now.

with best regards

Andreas

[Andreas]

Modifications:

I finally decided to put the NTC between Pin 5 (GND) and Pin 9 (RI = unused on Keithley 2000)
of the D-Sub connector which is very close to the trigger connector.
So temperature evaluation can be either done by a dongle in the RS232 line or by a one
of my 24 Bit ADCs connected over a 2nd RS232/USB connection.

Further I use a larger "hat" on the LM399 including slots for the vertical resistor arrays so that also
the SMD array gets a little better shielded against air drafts.

With best regards

Andreas

[Andreas]

Results:

Tilting is now +11/-35uV or a span of 6.4 ppm so against the first measurement 8.4 ppm a improvement of 2 ppm.

T.C. in 10V range against NTC position on RS232 connector is around +0.5 ppm/K
measured over a 8 deg C environment temperature range.
All measurement values are 10 NPLC values averaged over one whole minute.

A linear correction shows a large improvement of stability.
(around +/-2uV instead of 25 uV).

when comparing the Allan diagrams then I get:
without correction: starting at 0.3 uV std deviation for a 1 minute average its going up to 6 uV for a 4 hour (250 minutes) interval
with linear correction: also starting at 0.3 uV and staying there but for very long times (10 hrs) the value goes up to 0.55 uV std deviation.
with 3rd order polynominal correction: the standard deviation stays below 0.3uV (decreasing also for long averaging intervals).

So the conclusion is that I will have to use 3rd order polynominal correction for best performance.
Since the difference between internal and room temperature is around 8-8.5 deg C I will do a normalisation to 31.5 deg C internal temperature to meet the calibration temperature of 23 deg C.

One side effect:
The readings have gone up by about 1 digit (around 7uV or 1 ppm) due to the additional isolation.
Unfortunately in my case this is the wrong direction to the "true volt".
But this is still well in the accuracy specs of the instrument.

todo: 100uV range

[Andreas]

Further steps:

[Andreas]

Spare:

[Andreas]

Spare2 :

[CalMachine]

First (Reserved) (Will sell this space for advertisement)

[3roomlab]

o goody goody !
these are my noob mods, pls use them freely
https://www.eevblog.com/forum/beginners/o-qn-on-funny-nplc-multislopenoise/?all

some schematic update (in kicad)
https://www.eevblog.com/forum/repair/repaired-keithley-2000/msg814435/#msg814435

[Smith]

I guess the LM399 references differ a lot in temperature stability. My 2000 is quite stable when heating up, my work 2000 absolutely not. It drifts for over 8 hours for about 600-700uV on my 10V reference. One other 2000 at my job has max 1 or 2 digits drift when heating up (10-20uV). The most stable one was also the newest one. The older the unit is, the more drift it seems to have. Quite strange, as I would expect the opposite.

[Kleinstein]

There are different parts than can drift with temperature:
The first is the LM399. Here the the initial drift is more mike heating of the chip itself and that is very fast. After that, there should not be very much temperature drift. There is no good reason to expect less temperature drift on older units - this is more or less random, on how close the LM399 is to the sweet spot. It is aging where old units might be better at.

A second important part is the resistor array. Drift here can effects the gain of the ADC and reference scaling. How much an effect a change in temperature has, depends on the individual resistor array. They can be rather different - some can be much better than others and also the quality of the arrays could change over age of the units. So there could be a tendency for newer units to have better quality as the manufacturers might have improved quality control of process stability.

Tilting also seems to have an influence on the NTCs. So well possible this is mainly a temperature effect. Tilting could also couple through the mechanical effect, though this is less likely.

[Andreas]

these are my noob mods, pls use them freely

Thanks for that,
will have a closer look at this after my project.
But I fear the modifications are too extreme to me at least for a good working unit.

With best regards

Andreas

[Kleinstein]

It is interesting to see that temperature correction is already visible in the Allan deviation plots at surprisingly short times. So the thermal contact of the NTC seems to be good enough. Without so much up front averaging one could extend the curves a little to shorter times.

The changes the user 3roomlab did were quite a lot. However the main improvement was due to changing the AZ OP in the input buffer. In the original K2000 this is the main noise source in the 100 mV and 1 V range, when measuring close to zero. For the 10 V range this should be only a minor improvement.

AFAIK and learned from the data of the K2000, the noise of the K2000 in the 10 V range is mainly from the ADC circuit and the voltage reference. The LM399 reference is relatively noisy and thus could very well be the main noise source in the 10 V range.

There is not much one can do for the noise when measuring a significant voltage, but there is a chance to reduce the noise for measurements near zero. Due to the switching between a positive and negative reference to get zero, reference noise in the kHz range (the frequency used to switch between reference and negative reference in the ADC) will be effective and is quite large (about 2 times the noise of the input buffer, when in the 10 V range). This noise contribution could be reduced with filtering of the reference. A first test could an additional capacitor to filter the negative supply.

[VintageNut]

My K2000 from eBay died after a week or so of usage. I offered to send the unit back for a refund but the seller told me to keep the dead unit and gave me a full refund.

I sent the dead unit for repair. The entire main board was replaced for USD $560 which included a full calibration, adjustment and second calibration after the adjustment. This was a one-time bargain price because I asked nice. 

The K2000 report has 10V and 1V ranges within 1ppm. I have used the K2000 to adjust my 731B units to be fairly close to 10V.

I find that the K2000 is fast to warm up and appears to not be drifting. After the next calibration, I will know more about the long term drift of the K2000.

I agree that the newest K2000 main board is very stable and usable for those of us volt nuts with a limited budget. 

[saturnin]

I would like to share my experience with K2000 modification too. Some time ago, I decided to upgrade 7K resistor (R194) in the constant current source circuitry. It is used for generation of 1mA current  for 100Ohm and 1kOhm ranges.

For some reason, Keithley decided to use cheap 7.06K PTF56 5 ppm/°C resistor with long-term stability ± 0.04 % (1000 h rated power (0.125W) at +85 °C). It might be the reason, why measurements on 100Ohm and 1kOhm ranges were out of specs after some years of service. Ranges 10kOhm and above were just fine. So I decided to replace PTF56 resistor by RNC90Z type (6.98K, 2 ppm/°C, stability 0.05 % for 2000 h, 0.3 W at + 125 °C). After the update, day-to-day variations of measurements on 1kOhm range are much smaller.

The disadvantage of the update is you have to re-calibrate your unit since the value of 1mA constant current is changed. But it was not an issue in my case since I wanted to re-calibrate it anyway. (It did a partial calibration of zero, 10V, 100V and 1kOhm ranges only.)

[Andreas]

I guess the LM399 references differ a lot in temperature stability.

Yes the sweet spot (lowest warm up drift) seems to be at a output voltage of 6875mV.
https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg422367/#msg422367

It drifts for over 8 hours for about 600-700uV on my 10V reference.

I hope this is mostly during the first 2 hours.
after that it should be only environment themperature changes.
(or small changes in the 1-2 ppm range until final heatup).
otherwise I suspect that there is something wrong.

There are different parts than can drift with temperature:

thats true.
so for the first 15 minutes warmup the LM399 should give the mayor contribution to drift.
after that the scaling resistors dominate.

I find that the K2000 is fast to warm up and appears to not be drifting. After the next calibration, I will know more about the long term drift of the K2000.

The 2 Keithley 2000 of my former employer
(which are calibrated regularly)
drift about 1-2 ppm/year in the 10V range.

I would like to share my experience with K2000 modification too.

Thanks for that. I will keep that in mind. Perhaps with a taylor made resistor of Edwin.
On the other side: our calibration guy already told me that the stability
of resistance measurement on Keithley 2000 is not the best. (compared to HP34401A)

With best regards

Andreas

[mimmus78]

My k2000 was ~14 ppm up on the 10V range in 15 years that was not calibrated. K2001 is just 4ppm up from my 3458a in more than 10 years that was not calibrated. Yes they are very stable and since I have them (3 years) they are always 9 to 11 ppm apart on the 10V range.

Inviato dal mio Nexus 6P utilizzando Tapatalk