thank you for your answer..
I do 20 measure every sampling so I think I hope to reduce the effect of INL.
In input I have a regulated supply voltage and I do a test measure from 0 to 2500mV, I use an Agilent 6dgt multimeter to verify the value. I take the measure with reference P0_7 where I have a TI REF3325.
I do single ended measure, so with 12bit I have 11bit ( 0 to 2047), correct? If I consider the 0 I have 2048 levels.
I post the measure I have done:
| Vinput | ADC register | ADC register ( expected ) |
| 4,8 | 0,0 | 3,9 |
| 15,5 | 6,0 | 12,7 |
| 21,5 | 11,0 | 17,6 |
| 244,0 | 194,0 | 199,9 |
| 311,0 | 253,0 | 254,8 |
| 399,0 | 325,0 | 326,9 |
| 514,0 | 421,0 | 421,1 |
| 601,0 | 494,0 | 492,3 |
| 722,0 | 594,0 | 591,5 |
| 819,0 | 674,0 | 670,9 |
| 912,0 | 752,0 | 747,1 |
| 1030,0 | 850,0 | 843,8 |
| 1121,0 | 925,0 | 918,3 |
| 1229,0 | 1015,0 | 1006,8 |
| 1319,0 | 1089,0 | 1080,5 |
| 1662,0 | 1373,0 | 1361,5 |
| 1700,0 | 1405,0 | 1392,6 |
| 1706,0 | 1410,0 | 1397,6 |
| 1805,0 | 1492,0 | 1478,7 |
| 1928,0 | 1594,0 | 1579,4 |
| 2129,0 | 1761,0 | 1744,1 |
| 2226,0 | 1841,0 | 1823,5 |
| 2350,0 | 1944,0 | 1925,1 |
| 2421,0 | 2003,0 | 1983,3 |
| 2470,0 | 2043,0 | 2023,4 |
| 2475,0 | 2047,0 | 2027,5 |
The expected value is ADCvalue*2500/2048.
If you take a look to the results, we have an offset that variable from about 4mV to 20mV.
With a software compensation from these values I have obtain better results, but I would be sure this is a "normal" behavior.
thank you
Regards,
Emanuele