adc accuracy

[Yano]

Hi, We have an nRF51244 and are using a single ADC in 10bit mode with a 1.8V supply (smooth regulated from battery) wired in LDO mode. The problem is our readings are all high by a few mV. Looking deeper we see we are always a few counts higher than calculated. Measurements we done with a regulated Keithley 2601A attached to the ADC input through an RC filter (100nF / 1K0). We eventually attached the ADC line directly to GND and we still get a count of 3 or 4. Is this normal and do we just have to calibrate each ADC in production? Thanks

[Stefan Birnir Sverrisson]

Hi Yano

Yes, this offset error of the ADC is within specification. The spec allows up to 2% offset error and up to 2% gain error additionally. To correct any potential offset and gain error, you will need to do that in production. This thread has a guide on how to do that.

However, any impedance on the ADC input pin will cause error in the reading. If you use prescaler 1/1 for the configuration of the ADC, then you can expect error of 1kohm/130kohm = 0.77% error. The error could be for 10 bit resolution 1024*0.0077 = 8 bits. However, if you add a capacitor between the ADC input and ground, that should eliminate most of the error. It will however not do that with your filter of 100nF/1kohm as the capacitor would need larger capacitance to keep the voltage steady for 68us, which is the time it takes to sample with 10-bit resolution. The impedance of the ADC input is given in table 313 in the nRF51 series Reference Manual v3.0. A guide on how to calculate capacitor size for a voltage divider is given here.