Maybe you can use 256 samples.
You only have to add the highbytes of the samples!


(a1+a2+...+a256)/256 is the same as: a1/256+a2/256+...+a256/256

The problem is rounding! I would add the MSB of the low-bytes and divide this sum by 2 and add it to the first result.

Just think a little about it !

($1270+$1280+$1290+...)/256 = $12+$12+$12+...
and: (0+1+1+...)/2=1.0
so you get: $36+1.0= 55 !!!