AVR Freaks

Look here for an example code
http://maxembedded.com/2011/06/2...

Do

and

do the same thing?

Do

and

have the same meaning?

It is a better function, as it is flexible enough to select a MUX channel... and the waiting_for_DAC_ready loop does not return early;

Suppose your average value is 1 and the input becomes zero; the first value your linear, infinite response filter will give you will be (1-0.125), the second (1-0.125)** 2, the Nth (1-0.125)**N (and it will never reach zero, as it has an infinite memory. Computations are straigtforwards if you modify .125 (used to avoid a division).
OTOH, if you have a 8/10 samples buffer, you can do a finite memory filter: with trivial means, the output will decay linearly towards zero, and the true value will be reached within 8 steps.... (it wonot be reached with an infinite response filter)...

You can also do non linear filtering; Arduino sodar library newping (http://code.google.com/p/arduino... )does median filtering (instead of the average, they compute the median with a 2N+1 buffer: it needs to sort the incoming buffer). This resists to outliers, untill there are more than 50% of them...

ex median (0,1,2,444,5,6,7 ) is 5 -average would be greater than 60-
If one feels/bets there are less outliers -say 1%-, a solution is to compute the maximum and the mininimu from the buffer and to compute a mean from the remaining values -ie, the values whose positions are not the two extremes : this formulation is tricky, but allows for multiple maxima, say...- .

ex : trimmed_mean(0,1,2,444,5,6,7) is 4.2

If your avr has an usart and you have time and skills to operate it, you may transmit the values to your PC, plot them and (with Excel-maybe-, R, octave/matlab) find out the best filter you need (if any)

If you want to compute an ordinary average, just change:
avg +=((current_value-avg)>>3);
into
avg +=((current_value)>>3); (and the >>3 should be put outside the loop :it is faster and avoids rounding issues)

If you want to compute an autorecursive filter, and get its results after 8 samples,
the function prototype mightbe:
uint16_t adc_filter( uint16_t previous_average)
its internal value avg should not be reset to zero, but to pervious_average
int16_t avg=0; becomes
int16_t avg= previous_average;
and its invocation would look like:
avg=filter(avg);

Bob: my first name is Denis ; there is no superiority -nor inferiority- of finite response filters vs infinite response (I am more accustomed to the 1rst varity, as it is easier to remove/hide outliers and as they do not introduce weird behaviors for georeferenced data(2D, on limited domains -data are expensive to gather-).

There are two questions which might been answered before coding -cutting and pasting code-:
a) is digital filtering necessay (as it amounts to simulating a RC input filter upstreams the ADC, recursive infinite response is redundant with analog filtering: if one fears it eats cpu cycles, analog filtering does not use **any** cycle and is , at least when prototyping, easier to implement)

b) is root mean square equivalent to linear means?? (and is root mean square -in the original post- necessary?)