24-bit ADC, what's the best these days for linearity (and low offset)

[SharpEars]

Looking for a good 24-bit ADC with the best linearity characteristics available today. Low offset/drift is preferable also, of course...

[Marco]

20 bit ADCs are the best for linearity, or rather Linear's LTC237X-20 20 bit ADCs are the best for linearity.

[Someone]

Audio DACs these days are usually delta-sigma type with inherently good DNL, but the INL is not that good.  They are also not that good for DC applications [instrumentation and industrial] because they have a large offset that drifts around-- not a problem for an AC coupled audio signal, but a big problem for instrumentation.  They *are* fast enough to get a 20Khz signal out of them though.  For a 24-bit sigma-delta DAC, you would be lucky to get 18 real bits of INL.  For a 32-bit delta-sigma DAC [TI sells one], you are *still* lucky to get 18-bits of real INL [called "ENOB" by the way-- look it up].  All of the extra bits are marketing hype for audiofools.

As you say, the DNL can be excellent which lends to a good SNR measurement and reproduction of a wide dynamic range. Paired with floating point processing the INL is down in the noise. Its not audiofool stuff.

For industrial and instrumentation, you are [currently] limited to about 20-bits in a single device [the Analog Devices AD7591].  The only alternative to this [for higher resolution] is to build your own DAC.  The literature is thick with different approaches to this, and my warning to you is that going below 1ppm INL is *REALLY* difficult.  Not talking about DACs, but ADCs, the HP/Agilent/Keysight 3458A DMM is well recognized as having the best linearity of any other DMM in the world [0.05ppm INL, 0.02ppm DNL], and it has been this way for many decades.  0.05ppm is [about] 25-bits of INL.  So, it can be done, but not without a lot of pain and misery, and a large pile of money.  You would have to own one of these 3458A's just to verify the linearity of your 24-bit DAC-- not much else out there [other than a Programmable Josephson Junction Array] would be up to the task.  Seriously, you are looking at many years of hard work to get to a 24-bit DAC with INL that is good enough to claim you actually have 24-bits of resolution-- it would be a remarkable feat!

It all depends on the sample rate needed, 23-24 bits are relatively easy at slow rates such as 1S/s. The OP really needs more detail to be able to provide any useful information.

[promacjoe]

Looking for a good 24-bit ADC with the best linearity characteristics available today. Low offset/drift is preferable also, of course...

unless I missed something, he is looking for a analog-to-digital converter "ADC". Not a digital to analog converter "DAC".

sorry I don't have an answer for you. Just wanted to clarify what you were looking for.

[splin]

For a 24-bit sigma-delta DAC, you would be lucky to get 18 real bits of INL.  For a 32-bit delta-sigma DAC [TI sells one], you are *still* lucky to get 18-bits of real INL [called "ENOB" by the way-- look it up].  All of the extra bits are marketing hype for audiofools.

Whilst I agree that most 24+ bit ADCs have INL specifications of 4ppm or more, equating to 18bit linearity at best, it has nothing to do with ENOB which is purely a signal to noise ratio (SNR) specification. You could conceivably have a very linear, but noisy, ADC with <0.1ppm INL error but < 16 effective number of bits. Or fabulous noise performance (i.e. resolution) but terrible linearity.

As Marco posted, the 20 bit SAR LTC237x ADCs have the lowest specified INL of +/- 2ppm max, 0.5ppm typical of any ADCs that I am aware of, but lower ENOBs than most 24bit delta sigma convertors. They have much higher conversion rates though so the ENOBs can be considerably improved by filtering/averaging.

Not talking about DACs, but ADCs, the HP/Agilent/Keysight 3458A DMM is well recognized as having the best linearity of any other DMM in the world [0.05ppm INL, 0.02ppm DNL], and it has been this way for many decades.

That is probably true but I suspect the reputation has a lot to do with HP publishing the results of their linearity testing in their in house magazine HP Journal. I don't know of any similar publication of linearity tests on other 8.5 digit meters (such as the Datron 1281/Solartron 7081/Keithley 2002 etc.) which, although they may have less impressive guaranteed specs, it may be that they aren't that much worse than the 3458A in reality. Anybody know how the Transmlle 8081 compares?

Linearity errors at these levels are hard to measure so it's hardly surprising that there isn't much freely available information - only meteorology labs will be likely to have much interest in such data. Testing linearity at these levels against Josephson junctions is expensive so perhaps it was a smart move by HP to put the results into the public domain. Perhaps HP's competitors didn't have the resources to undertake linearity testing at these levels or didn't have the business case to do so.

[deadlylover]

In regards to ADC linearity on 8.5 digit DMM's, I came across this graph on the brochure of Advantest's R6581 showing its linearity. I think they also had a picture showing the unit being tested at a JJA, I wouldn't be surprised if they actually operated one.

[Marco]

20 bit ADCs are the best for linearity, or rather Linear's LTC237X-20 20 bit ADCs are the best for linearity.

unless I missed something, he is looking for a analog-to-digital converter "ADC". Not a digital to analog converter "DAC".

sorry I don't have an answer for you. Just wanted to clarify what you were looking for.

Why are you quoting me in this response? I never mentioned a DAC, that was DiligentMinds.com. I have nothing to do with what he's talking about, at a guess I would say he was commenting on Blueskull's DAC but this is clearly the wrong thread for that.

OP wanted the most linear ADC and the  LTC237X-20 are the most linear COTS ADCs at the moment, not 24 bit ... but they are fast enough to do 24 bit resolution at audio frequencies using oversampling and adding either noise or a structured high frequency signal.

[branadic]

LTC2400 is still a good choice for an 24bit ADC with low INL. Its initial INL 4ppm can be calibrated to lower values (e.g. <1ppm) with simple resistor technics and a stable voltage source. See also:

https://www.eevblog.com/forum/projects/oshw-24bit-adc-measurement-system-for-voltage-references/

[promacjoe]

Why are you quoting me in this response? I never mentioned a DAC, that was DiligentMinds.com. I have nothing to do with what he's talking about, at a guess I would say he was commenting on Blueskull's DAC but this is clearly the wrong thread for that.

OP wanted the most linear ADC and the  LTC237X-20 are the most linear COTS ADCs at the moment, not 24 bit ... but they are fast enough to do 24 bit resolution at audio frequencies using oversampling and adding either noise or a structured high frequency signal.

because I pushed the wrong quote button. All the other groups have their quote button at the bottom of the text, This one is at the top. it is an easy mistake to make. it has been corrected.
I was not referring to you as being wrong, I was referring to the original author and what they wanted. This discussion was going in the opposite direction from what the author requested. This would not help him at all. I just wanted to get things back on track.