TMS320F28035: Understanding ADC gain error mentioned in F28035's datasheet

Sheldon,

Answers inline:

Q1: Does the highlighted line indicate the worst case of ADC's gain error when using an internal reference?

A1)Yes, this is the worst case number of LSBs that gain error will introduce into the ADC result.  Since gain error is a multiplier, this error will only happen at the full scale of 4096.  So, if a customer is convering half range(1.65V) then worst case gain error would be +/-30LSBs.  

It may be more helpful to think of this as a percent error, ~+/-1.5% of the expected result could be the error term for gain.

Q2: Does the gain error has any relation with temperature or other things? If so, is there any graph to show these characteristic?

A2) Yes, I would say that the first order factor of gain error change is with temperature.  This is indicated with the temperature co-efficient listed for the internal reference of 50ppm/deg C.  If we back out the 50ppm/degC number this translates into approximately 0.825% of movement of the gain error over the full operating temperature range of the device.  During production we attempt to center the gain error at 30C as best possible.  So the gain error at 30C(commonly referred to as "room temperature"), should be the best case; and move from there.  We do not have a graph we can share as to average gain errors over temp; but customer could use the above to back calculate what the error term might be at their temp range.

Q3: Does this chart indicates gain error is the biggest error contributor?

A3)Yes, the gain error is the dominant error term for this ADC.  Not only is the offset smaller; but using the internal offset calibration it can effectively be brought down the the +/-4LSBs shown.  Devin has a good E2E explaining the total unadjusted error here: https://e2e.ti.com/support/microcontrollers/c2000/f/171/t/587441?tisearch=e2e-sitesearch&keymatch=faq%3Atrue&keyMatch=ADC%20calibration&tisearch=Search-EN-Everything

Q4: Do you have any suggestions to reduce this error?

A4)I beleive in the above link there is mention of both the use of an external ADC VREFHI, as well as calibration.  Using an external VREFHI removes the temperature co-efficient from the equation, although the temperature co-efficient of said reference would need to be considered.  To furthrer reduce the gain error; the external reference could also be sampled and then the gain term corrected in SW.  This would induce some SW overhead to each ADC sample, unlike the offset calibration which is done by internal HW and had no impact on the sample rate.

Best,

Matthew