SOLVED: How can I reduce ADC jitter

[Ioannis]

Lets clarify something Robert please.

Do you expect to have an ADC reading that will be absolutely stable on every pot position? Say that a user can set 1023 positions on the pot. Do you expect to have solid 1023 readings?

This cannot be done no matter what. On an absolutely perfect circuit you will get +/- 1 digit because of the way ADC works along with every kind of noise present (wire induced, ground loops, resistors noise, shot noise, Johnson noise and so many other kinds).

The ADC steps are 5/1023= 4.8876mV. This means that if ADC input is over that step, you get a count of 1. But if one of the above noise or more than one is present, you may get 0 or even 2 as a result, if the noise adds another 4.88mv.

Hope this is clear and helps you not chasing the impossible.

Any kind of averaging helps but because of the above, it does not eliminate the small jitter that you should accept.

Averaging with Moving average routine limits the maximum number of counts. The one Richard posted on #57 goes up to 1004 after about 30 samples. Darrels Taylor Oversampling routine does also suffer of this kind of problem. Well gives from 8bit ADC a 16bit result but you have to pay in time...

Best keep a simple average of 10 samples and get done with it.

Ioannis

[Demon]

Lets clarify something Robert please.

Do you expect to have an ADC reading that will be absolutely stable on every pot position? Say that a user can set 1023 positions on the pot. Do you expect to have solid 1023 readings? ...

No, I was expecting a reading ranging between 0 and 1023, with 256 near 1/4 rotation (from available travel or course), 512 midway and 768 about 3/4 through.

[Ioannis]

Yes, you can expect that, depending on the pot of course.

But the readings will have a play around the specific point, because of what I explained.

Ioannis

[richard]

But the readings will have a play around the specific point

agreed , at the 1/4 turn posn the raw reading expected WHEN vref == vdd and vdd is a well filtered stable supply would be 256 + or - 1 count. if you want better than that use either a 12 bit adc or averaging or oversampling. none of which come without a cost.


This all becomes a ridiculous overkill if the "pot" for instance represents say the 'CHOKE' position when it is displayed as 0-100 in 5% steps.

i have asked several times now what are your expectations ?