Voltage reference for 24bit ADC?

[asrock70]

Can you advise how to solve the voltage reference for 24bit ADC so as to have the least impact on the accuracy (les LSB) of the ADC?
For example 24bit ADC AD7763
in datasheet say suitable reference is ADR434. I thnig NO!
24bit ADC have LSB 0.059ppm and ADR434 have TC 1ppm/C ==> error 17digit/C
They ADR434 supplemented by two pole LPF with -3dB on ~15Hz and -35dB on 1kHz. ADR 434 have noise 0.1Hz - 10Hz 6.25uVp-p
I admit I can not quite tell what the impact of noise on the precision reference ADC.
Can you advise?

 

[FvM]

I suggest a somewhat more serious analysis. AD7763 itself is already specified with 0.0002 %FS/°C typical gain drift, which calculates to 2 ppm/"C. Drift of other circuit elements can easily reach a multiple of this order of magnitude.

On the other hand, if you want better voltage reference TC than 1 ppm, you most likely need a thermostat or digital drift correction.

The "impact of noise" completely depends on your application. In a SD converter Vin/Vref is continuously evaluated as measurement signal, so you can calculate impact of noise by integrating the known reference noise spectrum over the ADC spectral window which has the wellknown (sin(x)/x)^n shape.

 

[asrock70]

Yes, AD7763 also has a thermal drift, amplifiers around him, surely, but all this can be well corrected automatic calibration to accurate reference.
Thermostated reference I want to avoid.
I'm thinking about MAX6341 + 10bit digital potenciometer + temperature sensor , but maxim in the datasheet skimp information about input trim so I do not know how to achieve control +-150uV.

Noise
I use AD7763 on 20MHz with all digital filter on Pass Band Bandwidth 31,25kHz,
MAX6341 have outut noise density 25nV/Sqrt(Hz) on 1kHz
LPF behind Voltage Referebnce have on 1kHz attrnuation -35dB
25nV x -35bB ~0.45nV * Sqrt (31.25kHz) = ~80nV
It can thus calculate?

 

[FvM]

25nV x -35bB ~0.45nV * Sqrt (31.25kHz) = ~80nV
It can thus calculate?

No, the noise power spectrum has to be integrated over the 31 kHz window to determine total noise power.

 

[asrock70]

I'm sorry, we got this, but even there, there am I missing
http://videos.analog.com/video/products/amplifiers-and-linear/756529027001/Spectral-Noise-Density-to-RMS-Noise/

 

[Avla 0652]

First of all, 24 bit = it is not an accuracy, 24 bit is a resolution of the ADC.
the accuracy of a high precision ADC usually is much worse than resolution.

And moreover, the accuracy of Vref usually is much worse than accuracy of ADC.

Many high-precision ADCs are supposed to be used in "ratiometric" circuits
without Vref source
for instance like this
https://www.maximintegrated.com/images/appnotes/748/DI45Fig08.gif

 

[crutschow]

What is you application that you need near 24-bit accuracy (.06 ppm)? You would likely need NBS voltage standard references to meet such a requirement.

 

[asrock70]

At the outset I know what is the difference between resolution and accuracy.
What's going on? Actually the most accurate voltmeter which can be built from commercially available components. A little bet :|
Number of errors ADC can be removed using different calibration tricks, but as with any measurement wants it done a solid reference standard, a voltage reference.

 

[FvM]

What's going on? Actually the most accurate voltmeter which can be built from commercially available components.

So you'll want refer to the proven methods used in high performance (e.g. 6 1/2 Digit) DMM, like thermostated or digitally compensated reference. Obviously the AD7763 data sheet isn't intended as instrument construction manual.

 

[asrock70]

Use any as LTZ1000ACH was the first choice,. but
- LTZ1000ACH have TC 0,05ppm/C
- Amplifer LT1006 have 0,2ppm/C
- min 3pcs Resistor make 1,5ppm/C
I thnig my solution with temperature sensor and digital potenciometer + MAX6341 is it easier, cheaper, and potentially as an exact solution
What I use ADC truth is not yet certain, but the AD7763 seems like a good choice, or not very accurate ADC with at least 1kSPS.
In Data sheet AD7763 is aplication recomandation among others, Instrumentation
But it is important that these recommendations are more fantasy writers datasheets.
Accuracy quoted in DS is decent and I especially did some measurements, and minimum bit on my development board seems to be a very solid piece of silicon. But if anyone know of a more accurate ADC, recommend

 

[schmitt trigger]

Analog Devices has many, many tutorials on ultra high precision analog apps. But as other posters have already explained, it will take both tempco compensation and digital drift compensation to fully reach a 24 bit resolution.

 

[GUMY]

You can not make temperature compensation with temperature sensor, because reference TC is not exact and fixed value. It is different between individual components.

 

[FvM]

Use any as LTZ1000ACH was the first choice

Yes, it's internally thermostated.

Are you actually considering to design a precision instrument with AD7763? Unless you need it's high speed, this won't be a good idea. The differential amplifier brings up several issues regarding drift and linearity that aren't obvious from the datasheet specifications.

But I don't want to blame this particular part. You should be aware of similar "surprizes" with every component that's stretched up to the performance limits. It's a long way to make an instrument meeting all intended specifications. And part of the explanation for the rather high price of the retail instrument.

 

[carpenter]

LTZ1000ACH is internally thermostated reference with 0.05 ppm / C, but as he writes astrid, all other component as output amplifier, and few resistor will swim with the temperature. I'm not sure whether it will be a significant difference between the MAX6341 and a solution with LTZ1000ACH.

It also struck me that Analog has no datasheet slightest hint parameter internal amplifiers in the AD7763.
On the other hand, maintains that the AD7763 using the internal amplifier has Integral Nonlinearity 0.00076% , truth 0.00076% is approximately 17bit.

 

[FvM]

It also struck me that Analog has no datasheet slightest hint parameter internal amplifiers in the AD7763.

I recently asked ADI support about additional data, e.g. amplifier common mode range. They said, there is no information available.

On the other hand, maintains that the AD7763 using the internal amplifier has Integral Nonlinearity 0.00076% , truth 0.00076% is approximately 17bit.

0.00076% (8 ppm) would be O.K. for most applications, although probably not for a 6 1/2 digit DVM. But according to my observations, you get easily much worse nonlinearity due to parasitic effects like sampling clock "backlash". I have no reason to assume that the datasheet numbers and curves are fake, but apparently you need some effort (and possibly unsaid means) to reproduce it.

 

[carpenter]

You are right.
It strikes me that two things
1. Level datasheets falls. For example, the MAX6341 data sheet is not written as input TRIM works. The amplifiers in the AD7763 has already been mentioned.
2. Analog Device is a leader in the ADC, but even they can not find one converter that should be the real accuracy of at least 21 bits.

 

[carpenter]

A good choice for such a precise instruments seems to be a 31bit ADC with INL 0.5ppm ADS1282

 

[asrock70]

ADS1282 seems to be a very interesting circuit, but I do not understand why it has such a brutal 31-bit resolution?
You do not have to experience these circuits someone?