Re: Help understanding Arith coding

hi!

some remarks about the implementation you used, concerning the entropy
limit:

1. it uses a static model, this means, first it counts the bytes,
builds its model and then compresses the bytes, so the model has to be
saved to the compressed file.

2. since you "only" use finite precision integer math the output
becomes inaccurate (and size increases) due to rounding errors.

3. 16 bit integers are used for the arithmetic registers, this means
the frequency counts have to fit in a 14 bit integer. if you got a
stream larger than 2^14 bytes, the model dosn`t exactly represent the
symbol distribution, it is only a rough approximation.

greets

.