Re: [[oc] Huffman coding]

[Graham Seaman <graham@seul.org>]

On Thu, 24 Oct 2002, Andras FERENCZ wrote:

> - at encoding you must parse the data twice, to build the statistics and the
> dictionary and for the second time to do the encoding itself (i.e. streaming
> is excluded); for the decoding part if you got the dictionary first, the
> decoding could be implemented on the fly 
> - you have to pass the encoded data AND the dictionary

adaptive huffman coding often compresses better and doesn't need to send a
dictionary or do 2 passes (google, there's lots on it). It's often
better for continuous data streams. Don't know about hardware
implementation, though. For static huffman compression guess you'd do
a precalculation of the codes assuming (hopefully) the statistics of your
data were fixed, so you wouldn't need the tree generation in hardware, or
to send the dictionary. If your data varies too much, you're going to
need the tree generation whether adaptive or static.

Graham

--
To unsubscribe from cores mailing list please visit http://www.opencores.org/mailinglists.shtml