7.xV to 10.00000V AutoCAL circuit

[Kleinstein]

If you want to use the LTC2400 to compare the 7 and 10 V directly towards zero, you need to use just one divider and a buffer amplifier to switch between the 7 and 10 V.  Separate dividers just don't make sense as you need 2 stable dividers to replace one.

The alternative to using a divider would be subtracting a certain voltage. If this voltage is well proprotional to the 10 V this is OK and can give that 1 extra bit of resolution. Still drift from the divider enters, and the ADCs gain error enters, unless extra switching is used.

[e61_phil]

If you want to use the LTC2400 to compare the 7 and 10 V directly towards zero, you need to use just one divider and a buffer amplifier to switch between the 7 and 10 V.  Separate dividers just don't make sense as you need 2 stable dividers to replace one.

The alternative to using a divider would be subtracting a certain voltage. If this voltage is well proprotional to the 10 V this is OK and can give that 1 extra bit of resolution. Still drift from the divider enters, and the ADCs gain error enters, unless extra switching is used.

Isn't that exactly what I've described in my opening posting?

I use the LTC1043 to subtract 5V from the 7V reference element. And I will only use one stable divider. The drift will by given by the 1:2 divider and the ADC only.

[Alex Nikitin]

I think I will need a stable divider by 2 and the stable LTC2400 for long term stability. Otherwise you will need some switching between different divider, I think.

The idea is to switch the same divider between two input voltages 10V and 7V, in that case the divider accuracy and long-term drift does not matter, and even the long term stability and accuracy of the reference for the LTC2400 does not matter. What matters is the stability of the LTC2400 ratio measurements (should be very good) and the noise on the measurements (so there will be a compromise between the speed and accuracy - if you reduce the noise by averaging you increase the drift influence from the reference and the divider). For the best result a by-stable relay can be used for switching however a solid state switching should also be possible with a little bit of trickery to get it's impedance out of the loop.

Cheers

Alex

[e61_phil]

I think I will need a stable divider by 2 and the stable LTC2400 for long term stability. Otherwise you will need some switching between different divider, I think.

The idea is to switch the same divider between two input voltages 10V and 7V, in that case the divider accuracy and long-term drift does not matter, and even the long term stability and accuracy of the reference for the LTC2400 does not matter. What matters is the stability of the LTC2400 ratio measurements (should be very good) and the noise on the measurements (so there will be a compromise between the speed and accuracy - if you reduce the noise by averaging you increase the drift influence from the reference and the divider). For the best result a by-stable relay can be used for switching however a solid state switching should also be possible with a little bit of trickery to get it's impedance out of the loop.

Cheers

Alex

In this approach an additional divider for the reference of the LTC2400 is needed or a second short term stable reference. But that shouldn't be a problem. Also a nice idea You eleminate the long term drift of the LTC2400 and the divider, but the LTC2400 should be very linear for this approach. Jim Williams and Andreas showed that is possible.

A buffer (LTC2057 for example) at the output of the CMOS switch should be enough to get rid of the switch impedance, I think.

I've drawn my idea on a whiteboad. Perhaps, it was not clear to everyone due to my crude english (the left part is only the LM399 1mA current source the right part behind the dashed line is the AutoCAL circuit).
Another adavantage of the higher digital supply is the easy application of a 5V DAC on the 10V output.

[Andreas]

In your configuration the divider quality (tempco and long-term drift) is still important. If you only measure the voltage ratio using the same divider for both voltages, only a very short term divider stability is required, so it is one uncertainty less. Even LT5400 has some drift

The LTC1043 has virtually no drift over temperature and time.
Practically you have something like 16nV/K or 0.006 ppm/K which is mostly due to the offset drift of the buffer amplifier.

https://www.eevblog.com/forum/metrology/t-c-measurements-on-precision-resistors/msg528192/#msg528192

I've drawn my idea on a whiteboad.

I would make the 7->10V transfer with one single OP-Amp. (saving at least one Op-Amp if you need to buffer the 7V for the ADC).

with best regards

Andreas

[e61_phil]

The two OP-Amps on the left side are only used for the current source. It should be no problem, to connect the "main" OP-Amp directly to the zener and use a "normal" precision dual opamp for the current source.

[Kleinstein]

The 2 critical OPs are the ones to buffer the 7 V and the 5 V output of the LTC1046 divider. For the 7 V the load from the LTC2400 input might be to much to connect it directly to the reference.
The other OPs are not that critical. The one in the center to output the 10 V is corrected by the ADC/DAC, so like the resistive divider it only needs to be reasonably stable to keep the DAC range small.

[e61_phil]

Has anyone experiences with a good buffer OP-Amp for the LTC2400?

My first guess would be an auto-zero opamp combined with a buffer stage.

[Andreas]

Hello,

I never tried a buffer stage.
just a LTC2057 or a LTC1050 together with a R-C low pass of 820R + 470-820pF like in Schematics "Widerstandsteiler"
https://www.eevblog.com/forum/projects/oshw-24bit-adc-measurement-system-for-voltage-references/msg434154/#msg434154

the largest problem is that the LTC2400 generates  transients at the input during conversion.
without the low pass these transients are rectified at the output stage of the OP-Amp.
This generates a negative offset near zero and a positive offset at full scale. (ADC seems to have too much gain).
If the capacitor is too large you get non-linearities. There is a sweet spot where the gain is exactly 1.000000

For the "current source" it is sufficient to put a 3K resistor from 10V output to the zener.

with best regards

Andreas

[e61_phil]

For the "current source" it is sufficient to put a 3K resistor from 10V output to the zener.

I read in the datasheet the long term specifications with 1mA +/- 0,1% and I thought it could be nice to be within this 1mA +/-0,1% even with different LM399 (voltages) for comparision.

[doktor pyta]

I came up with another idea of using multiple LTC1043 however LT Spice seems to have huge problems in simulating the circuit.
I'm sending You zipped .asc file so You can simulate the circuit (or make the simulation working).
It seems that it would be not so easy to make the circuit stable.
Finally, if someone has 2 or 3 pcs of LTC1043 he could check the concept in the real world.
PS. The print screen is to small to fit all the schematic, but it explains the idea.
Comments invited.

[Andreas]

Comments invited.

Hello,

the trick that you have to do with a LTC1043 as divider is in phase 1:
put all flying capacitors in series to the output capacitor.
in Phase 2: connect all flying capacitors in parallel to the output capacitor.
Only in this way the tolerances of the capacitors play nearly no role.

I cannot see this principle in your cirquit to generate the 1V part.

With best regards

Andreas

[doktor pyta]

Only in this way the tolerances of the capacitors play nearly no role.

Well, not exactly.
Speaking of LTC1043, the 'inverting' topology introduce the same error as 'differential to single ended' topology so 2ppm.
Both these topologies are unaffected by capacitors tolerances.
Proposed circuit uses 'single ended to differential' topology stacked to obtain 10V from 1V input. The main problem I see is the pulsed current taken from the output on an opamp.

[d-smes]

Does this accomplish objective:   Use divide-by-two from 10V to form reference for LTC2400 (buffered, of course).  Then use differential-to-single-ended circuit on cover of LTC1043 datasheet to translate the 10V - 7V differential to ground potential.  Buffer this and use as Vin of LTC2400.  Now everything is at ground potential and you're using only one LTC1043 plus two buffers.   LTC2400 output ratio of 3V / 5V minus the reference (calibrated) ratio is your ACAL error term to the correction DAC.

[Kleinstein]

For the 7 V to 10 V step you have to distinguish 2 cases:
1) you have some 6.9-7.1 V source and want to get exactly 10 V out   (e.g. reference source). Here you need the fine adjustment. The LTC1043 alone can not provide this solution.

2) you have about 7 V and need a stable about 10 V, e.g. as a reference for an ADC. In this case you might get away with a factor close to 10/7 , maybe even  3/2. Here the LTC1043 circuit could do the trick as not fine adjustment is needed.


Using the LTC1043 to transfer the 10- 7V difference to ground would work, but the advantage is limited. There is no big deal having the LTC2400 positive tied to the 10 V output. The whole circuit of ADC, µC and DAC does not have to be ground referenced.
 

[d-smes]

I don't understand Kleinstein response.  For point #1, I agree the 7V has a wide initial tolerance but the main objective is to compensate for 7V to 10V amplifier drift.  All the LT1043 does with this voltage is divide it by two so the resulting ~3.5V +/- 1ppm (from data sheet divide-by-two circuit) is within the range of the LTC2400 as a input signal.
I don't understand point #2.  By taking the differential 10V - ~7V = ~3V and translating the result to ground, the ~3V signal is within the LTC2400's 5V reference range, so a accurate ratio can be established.  The LTC1043 isn't clear what accuracy of differential-to-single-ended conversion is, but the CMRR and Charge Injection discussions in the data sheet implies that's where this chip excels.  While I agree the white-board concept posted by the OP would work, I submit having everything ground referenced is easier and has less hardware.  The math of the ratio is different, but a correction term from the ratio to drive the DAC can still be derived.

[Galaxyrise]

Only in this way the tolerances of the capacitors play nearly no role.

Well, not exactly.
Speaking of LTC1043, the 'inverting' topology introduce the same error as 'differential to single ended' topology so 2ppm.
Both these topologies are unaffected by capacitors tolerances.
Proposed circuit uses 'single ended to differential' topology stacked to obtain 10V from 1V input. The main problem I see is the pulsed current taken from the output on an opamp.

I agree that you should be effectively free from capacitor tolerance issues, even tapping off 7V from the middle like that.  Like you, I'm also suspicious of driving the 1u caps directly from LTC1051s (I get about 50mVpp oscillation in LTSpice. Easily fixed in spice with some resistance.)  However, I'm also suspicious of your outer feedback loop; the 1043s take a long time to stabilize and don't do it smoothly--I would guess U12 oscillates.  And it does exactly that in my stripped down LTSpice run.  (Just one 1043+1051.)

Simulating the 1043 in ltspice is super fiddly.  I'd found some advice about it somewhere, involving changing the integration type to Gear and messing with some of the tolerance values, but I still would have to mess with the switching frequency and couldn't always get it to work.

I've also played around a bit with doing LTZ1000 -> 10V with 1043s (I used 3, and wrote a program to solve for their arrangement) in LTSpice, but I don't have the switching knowledge to practically implement it without spoiling the precision DC output.