I see the datasheet numbers are given assuming the internal reference. Fig 49
shows that the reference thermal drift is not ideal, and the text suggests that
an external refrence should be used to improve the gain accuracy/drift. Is the
gain drift only dependant on the reference or are there other ADC internaml
factors to be accounted for?
Then assuming I go for an external reference (ADR130), what would be ADI
recommendation with respect to noise in order to keep the specified input
referred noise as specified (~0.9 LSBrms) or better?
I’m assuming a sample rate of 20Msps. Sample rate affects the gain temperature
drift because there is more “droop” of an internal boosted supply at low sample
VREF drift is the primary factor in ADC gain drift. Other temperature dependent
elements in the VREF signal chain can cause a small ADC gain drift over
temperature, but this will be a small portion of the temperature overall drift.
For example, the designer did some analysis and estimated that the temperature
drift due to elements in the VREF signal chain (other than VREF drift itself)
would contribute about a 0.13% gain drift from -40C to 85C at 20Msps. Over this
same temperature range we see from Figure 49 in the datasheet that the drift of
VREF itself is typically from -5mV at -40C to 2mV at 85C for a total VREF
temperature drift of 0.7% from -40C to 85C. So the temperature drift
contribution of VREF is over 5X the temperature drift contribution from other
It would seem that from a noise perspective the ADR130 would not be a major
contributor. We also established that the input referred noise using the AD822
as a reference buffer yielded identical results to using the ADC’s internal
We looked at AD9257 input referred noise with internal VREF (integrated
reference within the ADC) vs. external VREF (using the AD822 setup shown below
in the AD9257 Design and Integration Files:
cts/EVAL-AD9257/eb.html). The results were identical between the two.
When properly filtered, the relative noise contribution from VREF will be
small, i.e. it is other noise sources that are the main contributors.
Looking at the ADR130 noise specifications at 1V from the datasheet, it says
Voltage Noise (0.1Hz to 10Hz) is 6uVpp. One AD9257 LSB is 2Vpp/2^14 = 122uV.
There is much more to it than this, but its still consistent with other noise
sources being dominant. This would especially be so because the noise from the
ADR130 would not be correlated to the other noise sources and they would
combine as root-sum-of-squares.