Test procedure.
Nucleo-H753zi with "arduino style" shield installed.
Hardware includes DAC80501 to generate linear ramp 512 points, than OP amp to scale up and convert SE to Diff.
ADC2 runs in Diff mode, channel 3 (PA.6 & PA.7 -> SB 33/ 35 modified),
clock 10.5 MHz, S/H 16.5 clocks, oversampling 256, than software oversampling x16 - total 4096.
Linear regression done on the same nucleo, data plotted on TFT ILI9486 .
Scale in 16-bits, meaning all screen 32 = 5 LSB, or 0.001611328 V, where LSB = 50.354 uV.
Results: Linearity is exceptionally good, error less than 1 LSB (limited by 16-bits DAC). Left side slightly distorted by OP amp not capable to get 6.4 mV to ground.
Last image shows intentionally introduced distortion to calibration factor N-4, value changed from 1FF to 1A0, for verification.
