DAC precision question - requirement of applied DC voltage

[liletian]

DAC precision question?

HI All
I had a question, it is normal these days to have a DAC or ADC for 24 bits,
suppose the VDD is 1 V which means the resolution of ADC will be 1/2^24=5.96e-8, for such a small resolution, what will be the requirement of applied DC voltage? DOes it means the VDD must be 1 V instead of 1.000001, otherwise you can not
reach that resolution?
Thanks,
B

 

[saro_k_82]

Re: DAC precision question?

Sigma-delta Converters can resolve 24 bits., but their accuracy is much less. A low accuracy in the reference/supply does not prevent you from reaching the specified resolution

 

[liletian]

Re: DAC precision question?

Sigma-delta Converters can resolve 24 bits., but their accuracy is much less. A low accuracy in the reference/supply does not prevent you from reaching the specified resolution

can you explain more on this? ok, suppose in t1, V1=1.2 in t2 V1=1.2001, I will
think this will definitely affect your resolution, am I wrong?
tHANKS

 

[saro_k_82]

Re: DAC precision question?

Precision and accuracy are different.
https://en.wikipedia.org/wiki/Accuracy_and_precision

24-bit Sigma-delta converters are precise to 24 bits but are less accurate.
For the given case., it shifts the transfer curve by that much implies that it's range is changed. So you don't get accurate results., but it doesn't matter in some applications. In voice or xDSL applications the absolute accuracy of the ADC is even not a key feature.
Rarely you'll see some sigma delta ADC datasheets that can match accuracy with resolution but they wont be more than 16-bits.

//Copy pasted from the white paper **broken link removed** //

Resolution and Accuracy
One of the more confusing tasks when evaluating data sheets is to separate the accuracy
of an ADC from its resolution. Just because an ADC exhibits 16 bits of resolution does
not guarantee 16 bit accuracy. In fact, an ADC may not even maintain monotonicity (each
bit combination measuring a higher value than the next lower combination) over its full
operating range, especially over the temperature range. In some applications, resolution
is more important than accuracy; it may be more important to be able to resolve small
changes in a variable than to know the absolute value of the new signal level. Many 16-bit
resolution ADCs exhibit 14 bits of accuracy, and this is sufficient for a large number of
applications.
In some applications absolute accuracy is paramount; therefore overall accuracy becomes
more important than resolution. In these cases, it is important to look at the integral
nonlinearity specification. This represents an error which cannot, in general, be eliminated
with calibration.
Understanding the overall system requirements will aid in the selection of the appropriate
ADC without overspecifying its characteristics. Higher cost is generally associated
with ADCs that have more stringent characteristics. If dynamic accuracy is important, then
it may be better to choose a faster, lower-resolution ADC than one with high resolution
but longer conversion time.

 

[liletian]

Re: DAC precision question?

ok, it said for a 16 bits DAC, the accurate is about 14 bits, so it constrained the Vdc can not variable more than 1/2^14=6.1*10^5, right?
Thanks

Precision and accuracy are different.
https://en.wikipedia.org/wiki/Accuracy_and_precision

24-bit Sigma-delta converters are precise to 24 bits but are less accurate.
For the given case., it shifts the transfer curve by that much implies that it's range is changed. So you don't get accurate results., but it doesn't matter in some applications. In voice or xDSL applications the absolute accuracy of the ADC is even not a key feature.
Rarely you'll see some sigma delta ADC datasheets that can match accuracy with resolution but they wont be more than 16-bits.

//Copy pasted from the white paper **broken link removed** //

Resolution and Accuracy
One of the more confusing tasks when evaluating data sheets is to separate the accuracy
of an ADC from its resolution. Just because an ADC exhibits 16 bits of resolution does
not guarantee 16 bit accuracy. In fact, an ADC may not even maintain monotonicity (each
bit combination measuring a higher value than the next lower combination) over its full
operating range, especially over the temperature range. In some applications, resolution
is more important than accuracy; it may be more important to be able to resolve small
changes in a variable than to know the absolute value of the new signal level. Many 16-bit
resolution ADCs exhibit 14 bits of accuracy, and this is sufficient for a large number of
applications.
In some applications absolute accuracy is paramount; therefore overall accuracy becomes
more important than resolution. In these cases, it is important to look at the integral
nonlinearity specification. This represents an error which cannot, in general, be eliminated
with calibration.
Understanding the overall system requirements will aid in the selection of the appropriate
ADC without overspecifying its characteristics. Higher cost is generally associated
with ADCs that have more stringent characteristics. If dynamic accuracy is important, then
it may be better to choose a faster, lower-resolution ADC than one with high resolution
but longer conversion time.

 

[saro_k_82]

Re: DAC precision question?

If you are talking about the reference of the converter ..YES

 

[liletian]

Re: DAC precision question?

If you are talking about the reference of the converter ..YES

how could you get a such high DC precision in a single. what is the best DC
convert precision?
Thanks

 

[rfsystem]

Re: DAC precision question?

24Bit resolution is only useful for measuring equipment and relative measurements. The 24Bit mean 60ppb (part per billion). So if you have a very good reference with 10ppm/K temperature drift you have to stabilize the temperature to 1/160 of a Kelvin.

Also the noise requirement are very high. So how low should the bandwith if the noise resistance is e.g. 1Kohm. The resistor create 4nV/sqrt(Hz). So a noise bandwidth of 217Hz collect enough noise equal to a LSB.

So 24Bit is only for relative measurements, 10V low frequency designs and some esoteric audio applications.