5V-ADC-Reference-Board-attempt based on LTZ1000 and LTC1043

[Echo88]

For my experiments with the 32-Bit-ADC-chips LTC2508-32 and AD7177 i want the best possible reference one can build at home.
Since the chips only accept maximum 5Vref one can use a LTZ1000 but has to divide the 7.2V down via high cost resistors or via capacitor-dividers (or via transformers...).
Since the LTC1043 capacitor-building-block-ic is said to be very long term stable and inexpensive i used this road.
An attempt on blueprint-board with the following circuit: LTC2057-buffer -> LTC1043 / 3 -> LTC1043 * 2 -> LTC2057-buffer delivered sufficient ~4.8V at ~1µVpp, while the 7.2V from the LTZ1000 came from a separate KX-Board.
See also this thread for more info: https://www.eevblog.com/forum/metrology/ltc2508-32-bit-adc/msg1727468/#msg1727468

Since it works i now want to design the LTZ1000 and the whole buffer/divider-stuff on one board, based on the LTZ-reference-board suggested by Dr. Frank.
See this thread were he shows his board: https://www.eevblog.com/forum/metrology/mx-reference/msg1297126/#msg1297126
I chose this board as a basis since it follows all suggestions for a nice and quite reference-design:
-all components THT, therefore one does not need to fear microphonic effects, humidity/temp-board stress
-star-ground
-good thermal design
-foil/PWW-resistors usable
Dr. Frank was so nice to send me his board-files so i can modify his board directly. Thanks again for that. 

In the attachment you find my early attempt to layout the board, please critize it and tell me were i can improve it. After successful tests the board will go Open Source.

A few words and thoughts about the board and changes to Dr. Franks-Layout:

-the following components arent in the original LTZ1000-schematic from the manufacturer: C6/C5/C4/R6/R11, Dr. Frank and Andreas designed them in due to measured spikes, i left them in the design so one can still populate them later if needed
-LTC1052-buffers with lowpass at the input to dampen chopper-spikes, should be able do safely drive the 1µF-capacitive loads, but i didnt find a manufacturer-suggestion for a capacitive-driver-design
-CD4060N and 4MHz-crystal produce the 500Hz-clock to synchronize the two LTC1043 so one gets the low 1µVpp-noise, but its also possible to populate C9 and C10 and depopulate R16/R14 and let them run independently   
-im thinking of stuffing the 500Hz-generator (CD4060N and crystal) into a small separate shielded case and connect the 500Hz-clock via SMA/SMB-connector, this would shield the board from the 4MHz-crystal, which might add noise?
-i tried to design a good star-ground, but maybe my solution is not sufficient?

Thanks,

Echo

[branadic]

Just an idea... you want it low noise, so why not using 2DW232 based reference to feed ADC and compensate for drift by measuring with the ADC against LTZ1000?

-branadic-

[Echo88]

The LTZ1000 has about 1.2µVpp and the used ADC is not much better, so the 2DW wouldnt make much difference in the noise calculation. Also the LTZ1000 is fully characterized per datasheet, the 2DW isnt.

Your suggestion might be more useful with an even lower noise integrating ADC like the 3458A i think. Also the 3458A would have the guaranteed linearity to compensate for the 2DW-drift compared to a LTZ1000.

[MiDi]

Nice project, I am  looking forward for this as I do not know anyone who has done a 2/3 divider with LTC1043 and posted the results 

My thoughts after a quick look at the schematics:
- Pin 14 of IC4 is missing connection to GND 
- V+ of IC4 has to be above output voltage at Pin 7, V+ is marked as 12V 
- no buffer between IC4 out and IC5 in: not shure if this matters, but it looks like it leads to an imbalance of the switched capacitor values
- LTC1052 is not the lowest noise chopper, perhaps ADA4522 or LTC2057 could be an alternative

Driving capacitances above a few pF with op-amps is worth a whole thread itself, there are a couple of principles to handle that:
- decoupling with resistor
- RC-compensation/snubber
- transistor buffer

More literature: Microchip, ElectronicDesign, Stack Exchange

Andreas has proposed a single LTC1043 design of a 3/2 divider, wondering if a single 2/3 divider is possible too

Building the divider with LT5400 could worth a look at... anyone with experience with this part?

[Echo88]

Thanks for the countercheck.  Indeed ground at Pin14 of IC4 will be corrected.
Vout of IC4 is 2.4V (7.2Vref/3) and therefore 12V supply is sufficient or am i wrong in my thinking?
I hoped that no buffer between IC4 and IC5 is necessary, but that might be the case.  I just build this circuit on a blueprint and measured it with a 3458A, which is not really a good way to test for noise. I really need to get my hands on a low noise amplifier to check this. Maybe someone can suggest a good LNA which can be bought or easily build?
I hoped to build the whole circuit in THT (less drift through humidity and therefore board-stress), but there are no modern Autozero-OP-Amps. Since AZ-OP-Amps shouldnt be influences by boardstress (or am i wrong?) the LTC1052 can indeed be substituted by better OP-Amps in SMD.
The capacitor-driver-design was not calculated by me and it sure would be better to see if other AZ-OP-Amps would include such a capacitive-driver-application-example in the datasheet, so that could be trusted.
Andreas single-LTC1043-design is interesting and worth a try to see if it behaves stable and without overshoot, i just went the save way by using two LTC1043 to get my needed voltage.

By using the switching-capacitor-solution i wanted to avoid resistor-drift, which would plague the LT5400: influences by humidity through boardstress and more noise than a PWW/foil-resistor.

Im a little short on time at the moment and need to get a LNA to test this circuit noise-wise. Thats why it didnt continue till now.

[MiDi]

Vout of IC4 is 2.4V (7.2Vref/3) and therefore 12V supply is sufficient or am i wrong in my thinking?

If my eyes do not fool me, in the schematics of first post I recognize IC4 (left side) as x2 and IC5 (right side) as /3 
You could flip them, but that would lead to higher noise.

I really need to get my hands on a low noise amplifier to check this. Maybe someone can suggest a good LNA which can be bought or easily build?

I had the exact same thought when I built a LTC1043 circuit
To get an idea for DIY LNAs look at the thread DIY 0,1-10Hz LNAs.
The upper bandwidth of 10Hz might be a bit too low for this purpose

The capacitor-driver-design was not calculated by me and it sure would be better to see if other AZ-OP-Amps would include such a capacitive-driver-application-example in the datasheet, so that could be trusted.

I guess there is no ootb solution for this, you have to evaluate the circuits to best fit your application 
A decoupling resistor suffers on changing currents (bias, leakage e.g. of cap), a transistor buffer suffers on recovery of load changes, ...
The In-the-loop compensation circuit could be a good compromise, but there are no values given for Rout for choppers...

By using the switching-capacitor-solution i wanted to avoid resistor-drift, which would plague the LT5400: influences by humidity through boardstress and more noise than a PWW/foil-resistor.

The datasheet gives quite good values and as this is made of one piece, the tracking should be quite good, but without experience I would not trust this too much...
Not shure if the higher noise would matter in this application as the choppers introduce some noise at their inputs.

[Echo88]

Now i get what you meant. I interchanged the /3 and *2 blocks in the schematic.   
Really should have crosschecked everything before the layout-attempt. I wanted to use the already existing 12V-power-supply for the LTZ, thats why i wanted to stay under 12V and did the /3 and then the *2-blocks. Can you elaborate why the noise would be lower if *2 /3-stages would be used? If that might lower the noise one could of course integrate a low noise 18V-supply for a LTZ *2 = 14.4V-stage.

Yes, since the LTC1043 and choppers might produce broadband noise its really necessary to also have a LNA for the kHz-Region. Sigh. Need to read the thread and start to look at Vref-noise. At least one can think about circuits while waiting for the LNA-input-cap to stabilize. 

[David Hess]

In-the-loop compensation circuit[/url] could be a good compromise, but there are no values given for Rout for choppers...

The output resistance can be calculated or at least estimated from the saturation voltage versus output current graph.

Not shure if the higher noise would matter in this application as the choppers introduce some noise at their inputs.

Chopper stabilization not only removes drift but also low frequency 1/f noise so at low bandwidths (very low) chopper stabilized amplifier noise is lower.  This is why chopper stabilized amplifier noise is usually also given from 0.01 to 1 Hz as well as 0.1 to 10 Hz.

If lower noise is necessary, then a chopper stabilized amplifier can be used to correct the offset and remove 1/f noise from a standard precision operational amplifier but I have never seen this done in a reference circuit.  Usually the reference has a lot more noise than the operational amplifiers.

[Andreas]

wondering if a single 2/3 divider is possible too

Hello,

at least as subtraction 7.2V - 1/3 (7.2V) should be possible with a single LTC1043.
But you have to arrange the capacitor values in a way so that no overshoot is done.
So 3*1uF for the output in series (tapped at 4.8V) could be a option + 2 flying 1uF (or 1.5uF) capacitors.

with best regards

Andreas

[MiDi]

Can you elaborate why the noise would be lower if *2 /3-stages would be used? If that might lower the noise one could of course integrate a low noise 18V-supply for a LTZ *2 = 14.4V-stage.

If first stage does /3, the noise introduced in this stage is amplified by 2 and adds to noise. The point is: does it matter? Would be hard to guess...

[Andreas]

The point is: does it matter?

Hello,

if the 2 devices have a synchronous clock: you will probably not be able to measure it.

with best regards

Andreas

[Andreas]

wondering if a single 2/3 divider is possible too

Hello,

totally forgot: there is already a suggestion from me:

https://www.eevblog.com/forum/metrology/ltc2508-32-bit-adc/msg1729499/#msg1729499

with best regards

Andreas

[TexasRanger]

Hey Echo, any updates on your project?

[alex-sh]

Hey Echo, any updates on your project?

I was wondering about the same. Any updates?

[Echo88]

Ah, sorry. No updates, i didnt continue the project like intended. After destroying my ADC, its peripherals and seeing that the reference is not the biggest noise/drift-contributor, i kind of got sidetracked by other projects. 
As far as i recall the 2/3-divider worked nicely built on perfboard, but had a bit offset due to missing input caps at the LTC1043. Didnt measure the resulting noise i think.

[TexasRanger]

Oh thats sound bads, I also destroyed one LTC2500. Luckily Analog Devices sended me 2 LTC2500 free of charge to my Workplace (Metrology Lab). You can try your luck at the bottom of each product page.

Yeah, the shown Input Circuits in the Datasheet don't feature Low 1/f Noise and Drift OR Low THD and Input Range.
They didn't even bothered to mention, that you need a Input Buffer for the Single Ended to Differential Conversion.


I'm currently trying to use other OPAMPs to drive the LTC2500.