Building a 7 decade voltage calibrator

[sync]

Yes - ADC feedback. As I now have the LTC2400 ADC running - and in addition have built the Linduino clone (expanded Arduino with a C++ library called LTSketchbook) to control / read all LT devices, all I need is another DAC (16 bit should be OK - I have one). I do not have to copy the Code Comparator PIC running ASM.

The AN86 code comparator is part of the control loop. With it and the ADC the precision is defined by the ADC not the DACs. That's the trick. You can use relatively imprecise DACs, resistors, amps.

[What will I divide?] Alternatively I can make a 2 Bit DAC from a LM399 board ouputting 10 Volt. Another possibility for reference will be 2 or 4 stacked LT6655 5V or 2.5V on separate batteries. This device will with quality relays be a programmable 2 Bit DAC with resolution 2.500.... Volt. I can then sum the output of this 2 Bit DAC to the output of the LTC2756 configured as a 0-2.500.... DAC.

A clever idea. Maybe there are linearity and stability problems on the switching points (2.5V, 5V, ...). I would use the same reference for both DACs. Then the reference drift affects both in the same way.

[quantumvolt]

"I do not have to copy the Code Comparator PIC running ASM."

What I am saying is not that I can drop the Feedback Loop, I say that I can use C++ on Arduino in stead of the looooooooooong (for me cryptical) PIC assembler code in AN86.


"Maybe there are linearity and stability problems on the switching points (2.5V, 5V, ...)."

Imo the idea is so simple it cannot fail. Imagine 4 batteries in series - each nominal 2.5 Volt. By tapping at the "end points and joints" - either directly - or through 3 capacitive dividers of the (4x2.5=)10 V, you will get something like 0, 2.5, 5.0, 7.5, 10 V.

Now change one battery with a 0-2.5 V 18 Bit DAC (LSB for 2.5 V and 18 Bit is just under 10uV which the LTC2756 gives stable), and add relays so that you can tap the full span 0-10V (i.e. 0, (0-2.499...),  2.5+(0-2.499...),  5.0+(0-2.499...),  7.5+(0-2.499... ), 10) with 18 Bit resolution. You now have the topology for a 1PPM 10V Source (10uV resolution for 10V).

The implementation is only technical details. You can use a rotary switch as in a KV divider - or you can use jumpers on a matrix/breadboard. But then it is only manually controllable ...

If you use digital control by relays and ADC feedback, you can tap the physical joints of the 2.5 v references, and then correct the output via the 18 Bit DAC. You will not need LTC1043.

If no feedback - open loop - is chosen, you must tap 2.5, 5.0, 7.5 V from resistor or capacitor precision dividers.

I think one has to use different references for the 2 Bit and 18 Bit DACs because of galvanic isolation (different ground is needed for series summing). This is not a problem. The more LTC6655 you put in series, the more stable the end result (or fractions of it).

But maybe it is a better idea to buy a 20 Bit DAC. Then you can add 4 Bit and get a (nominal) 0.1 PPM 7 Decade Voltage Source    This idea came up because I already have an 18 Bit DAC.

Maybe I am wrong in details - and personally I think it is almost impossible to make a stable 7 Decade 0.1 PPM Source - but all I am saying is that you can get "finer resolution" voltages between 0 and 6 Volt by series 4 pcs. AA batteries and put a multi turn potentiometer over ONE battery (divide 1.5 V) instead of putting the pot over ALL of them (the full 6 V).

[edavid]

The AN86 approach looks interesting. An similar but slightly different approach would be to PWM between 2 dacs. So same dual LTC1599 setup. You set one DAC to word N, and the other to word N+1. Then you use an analog switch to switch between the two DAC outputs, followed by an integrator/LPF. The duty cycle then determines where in the N...N+1 range you end up. And same as in AN86 you use an ADC for feedback.

Since the two input voltages for the analog switch are very close this will (hopefully) minimize the amount of switching noise.

Why not switch the same DAC between the 2 codes?  Then you don't need an analog switch.

[sync]

What I am saying is not that I can drop the Feedback Loop, I say that I can use C++ on Arduino in stead of the looooooooooong (for me cryptical) PIC assembler code in AN86.

Ok, i misunderstood you. Yes, that assembler in AN86 is not easy to read. They better used C.

Now change one battery with a 0-2.5 V 18 Bit DAC (LSB for 2.5 V and 18 Bit is just under 10uV which the LTC2756 gives stable), and add relays so that you can tap the full span 0-10V (i.e. 0, (0-2.499...),  2.5+(0-2.499...),  5.0+(0-2.499...),  7.5+(0-2.499... ), 10) with 18 Bit resolution. You now have the topology for a 1PPM 10V Source (10uV resolution for 10V).

The problem is when the DAC has a span of 0.001-2.498V and you add 2.501V. Then you can't get 2.499-2.502V (simplified to mV differences).

I think one has to use different references for the 2 Bit and 18 Bit DACs because of galvanic isolation (different ground is needed for series summing). This is not a problem. The more LTC6655 you put in series, the more stable the end result (or fractions of it).

A LTC1043 can provide galvanic isolation (within the supply voltages). The problem with different references is, that they drift differently. Today the 18 bit DAC reference has 2.501V and the 2 bit one has 2.498V Tomorrow they have 2.499V and 2.502V. How do you compensate this?

But maybe it is a better idea to buy a 20 Bit DAC. Then you can add 4 Bit and get a (nominal) 0.1 PPM 7 Decade Voltage Source    This idea came up because I already have an 18 Bit DAC.

Good 20 bit DACs are expensive. But you can use a 18 bit DAC and PWM/ dither between two values. That gives you more resolution. But i think the INL won't get better.

[Andreas]

Hmm,

just some thoughts:
- when I use a DAC with 16 or more bits, I will do this with the only aim to get a low noise output signal.
  So I would never do some kind of switching noise or dithering at the output of the DAC.
  every DAC switching creates unwanted noise or glitches.
- If I want to dither then I would use some kind of the EDN cirquit with 2 cascaded 16 Bit PWMs.

- 7 decade resolution = 10V output with 1uV (RMS) noise can be done with some experience
  when all is battery supplied and interfaces good galvanically isolated.

- stability or even accuracy is more difficult.
  With the best (pre-aged) references, a good design something
   in the range of 5-10ppm/year as stability can be made with standard parts of catalog distributors.
  But of course it will not have the fast settling times of a professional calibrator.
  Accuracy will be even more limited if you dont have a josephson array at home.

On the other side: even the HP 3458A with 8,5 digits is specced for 4-8ppm accuracy/year in the best range.
And finally: with a LTC6655 you will not get near to 10ppm/year stability because of humidity and hysteresis effects.

With best regards

Andreas

[quantumvolt]

Now change one battery with a 0-2.5 V 18 Bit DAC (LSB for 2.5 V and 18 Bit is just under 10uV which the LTC2756 gives stable), and add relays so that you can tap the full span 0-10V (i.e. 0, (0-2.499...),  2.5+(0-2.499...),  5.0+(0-2.499...),  7.5+(0-2.499... ), 10) with 18 Bit resolution. You now have the topology for a 1PPM 10V Source (10uV resolution for 10V).

The problem is when the DAC has a span of 0.001-2.498V and you add 2.501V. Then you can't get 2.499-2.502V (simplified to mV differences).

I understand this. But if one accepts that this is a DC gadget and allow settling time in the order of 100 milliseconds or a second, then voltages and relays probably will have stabilized and the LTC 2400 will tell the software about the glitch and you can maybe use the offset to output the missing values.

"A LTC1043 can provide galvanic isolation (within the supply voltages)." Good point. I didn't think about that. I guess you are right about using common reference.

But there is another possibility with the LTC2756 (at around USD 35 it is not too expensive). Let's say you have a reference 10 V, and you add LTC1043 chips to make a "ladder" 0, 2.5, 5, 7.5 and 10V. The LTC2756 is softspan so you can configure it as +- 2.5 V (it is powered +- 15 V clean external DC which is stepped down two more times in linear regulators). You now connect the +- 2.5 V DAC span successively to points 2.5 V, 5 V, 7.5 V on the ladder.

Depending on your configuration you now have the glitch at 0V and 5V, at 2.5 V and 7.5 V, or at 5 V and 10 V. In all cases you get a span of 5 V w/o glitches (but maybe the LSB will be 19.1 uV - 2 ppm). Also the span is firmware programmable / changeable while the ADC is powered up (it is programmed by the first byte of the 4 bytes SPI data transfer). Furthermore it has 2 pins that allows trim of up to 2048 LSB for offset and also adjust gain by applying 5 V. By experimenting with it, learning it's behaviour and maybe also do a quick calibration against an external reference  and meter before each working session, I think most of these problems can be solved in software / firmware.

Anyway - I do not build usable lab equipment - I just experiment. So it is a very nice and flexible device for me 

Here is how I found it (the LTC2757 is parallel output - but the same guaranteed 4 ppm DAC with +- 1 LSB INL).

[quantumvolt]

- stability or even accuracy is more difficult.
...
And finally: with a LTC6655 you will not get near to 10ppm/year stability because of humidity and hysteresis effects.

I have given up all dreams of a fix-point voltage reference with high accuracy and stability in time.

Whatever I do doesn't matter - I am doomed by practical and financial externals to a 10 yo uncalibrated Agilent 34401A and a (very good bang-for-buck) Geller ceramic reference 3458A transfer-calibrated board. I have enough frustrations as it is when I cannot reproduce a measurement.

On the volt-nut email forum (which has many smart and knowledgeable people - but also a lot of morons with too much dollars and no brain), a clever guy recited the saying about a man with one clock who "knows" the time versus the man with two clocks showing different time (however small the difference between the clocks - you will never again know what time it is ...)

[mrflibble]

Why not switch the same DAC between the 2 codes?  Then you don't need an analog switch.

Because with just one DAC you'll have to deal with your DAC settling all the time. Not the best for a low noise output. So you use 2 DACs, let them settle, and then PWM between those 2 DAC outputs.

[Andreas]

Why not switch the same DAC between the 2 codes?  Then you don't need an analog switch.

Because with just one DAC you'll have to deal with your DAC settling all the time. Not the best for a low noise output. So you use 2 DACs, let them settle, and then PWM between those 2 DAC outputs.

And then you will learn something about digital feedthrough and additionally charge injection of a switch. So this will also be no low noise solution.

With best regards

Andreas

[mrflibble]

And then you will learn something about digital feedthrough and additionally charge injection of a switch.

And there I was thinking that by keeping the two analog inputs at almost the same potential + carefully chosing the switching patterns you could keep the errors due to charge injection effects to a minimum.

Feedthrough will be an issue, agreed. But then again, the regular AN86 approach isn't static either...

[RikV]

Is this topic still alive or was it moved somewhere else?
I'd like to build one for myself and the existing software would of course be of great help?

[Andreas]

Hello,

I am aware of this thread:
https://www.eevblog.com/forum/metrology/diy-acvdcv-calibrator-with-ad5791/msg3087976/#msg3087976

But do you really want to build one by yourself? (and how do you adjust and calibrate it?)
https://www.ianjohnston.com/index.php/onlineshop/handheld-precision-digital-voltage-source-2-mini-detail

with best regards

Andreas