Lists Home |
Date Index |
> From: Mike Champion [mailto:firstname.lastname@example.org]
>So, my somewhat ignorant question is: why would these WBXML folks go to
>the trouble of defining yet another compression scheme for wireless XML?
There are both theoretical and practical reasons for reconsidering
compression in wireless environments. All have to do with the interaction
between compression and error correction. Virtually all digital wireless
systems employ some form of Reed-Solomon based forward error correction.
For those not familiar, this involves coding a field of n information
elements as n+p information elements such that when those n+p elements are
transmitted if some number greater than n elements arrive, then the message
can be reconstructed. Almost all systems employ this coding over bit
fields. And some wireless applications employ it over packet fields in
addition to whatever bit level correction is used.
It has been observed (and I believe proved as well) that for transmissions
on a randomly noisy channel, that transmission with forward error correction
is closer to the Shannon limit for that noisy channel. In other words one
comes closer to the total information carrying capacity of the channel by
adding information, when that channel is noisy!
Compression on a clear channel assumes that by coding using a fewer bits is
optimal. But in the noisy wireless environment we've demonstrated a case
where more bits is more efficient (in the information theoretical sense).
Compression and error correction are often treated as discrete steps. One
might find greater overall efficiencies if these bit-adding, and
bit-reducing steps were treated holistically.
On the practical side, there are several different types of Reed-Solomon
error correction. Some will guarantee to deliver a correct message or fail.
Others make a best effort, but will indicate whether a field might be
incorrect. Applications that can tolerate some loss of data, or require as
much of a message as it can reliably receive, might not want a compression
method that require the entire stream to reliably decode it.
That's a long answer to why one would want to reconsider compression for
wireless. From a not particularly thorough reading of WBXML, I'm not sure
that it's trying to deal with these issues. Maybe it should be.