From: gretzteam on
Hi,
My understanding is that there are two valid ways to specify fixed-point
wordlength in a CIC decimating filter:

1) Use the maximum width everywhere and let overflow happen...this yield an
EXACT result. Bmax = (N*log2(M*R) + Bin)

2) IF we are not going to use every output bits, we can get rid of some
bits along the way. When using the Hogenauer formula to calculate each
section's length, we make sure that the error induced will be smaller than
the truncation we'll do at the end.

Now, when comparing (1) and (2) running in parallel, I do get 1 lsb error
(-1 lsb, 0 and +1 lsb) at the output length level. Is this to be expected?
I was hoping to have them match exactly.

Thanks!