I'm sorry. I wished I could talk about these things in my mother language...


It depends on what you mean by "loose". If you generate the mean from these 256 samples and throw away the 256 samples afterwards, then you have done filtering and decimation. Of course, if the peak sample is thrown away together with the other 255 samples, then it's away. But you still have it taken into account in your mean. (I assume by "mean" you mean the "average" of the 256 samples?) The question is, is it enough to have the peak considered in your average only or do you need it as actual maximum sample.

Why is it of benefit at all to do decimating and filtering?
In a typical mixed analog/digital board you have two kinds of noise, the correlated and the uncorrelated noise. I guess these expressions sound entirely wrong for an english native speaker, so let me explain what I mean:

When using analog to digital or digital to analog converters there is always a certain time window, where the conversion is highly sensitive to noise and glitches on ground or elsewhere, which can considerably warp the result of conversion. By correlated noise I mean noise which always takes place at the same moment relative to the start of conversion. Correlated noise falling into the sensitive time period will always detoriate the conversion in the same manner and there's nearly no way to eliminate it. You could eventually set the micro into a sleep or idle mode to smooth the board and keep the noise at a minimum. But doing any subsequent digital low pass filtering will hardly have the desired effect on this noise.

Uncorrelated noise on the other hand is noise coming from other sources on the board, where the timing is not correlated to the start of conversion. This noise can now be eliminated or at least decreased by an oversampling and subsequent decimating! Means, all you need to do is to sample your signal more often then needed from the Shannon-Nyquist sampling theorem, to do some low pass filtering afterwards (by a moving average over the too many samples, for instance) and to throw away the too many samples by substituting them by the average or mean, as you would say.

This methode is nearly the same as the one Michael mentioned in his post:



Kai