Rob wrote:
Of course, the headers of IPv4 are already considered big in the
amateur packet radio world. At the datarates typically used, one
wants to conserve on header size.

Yes, some have tried using Van Jacobsen header prediction to cut down
on the size of the actual transmitted header - a rather nice solution,
but not really workable in a lossy datagram world.

So far the best method seems to be (using a separate pid) an adaptation
of the SLIP with compression and header prediction over a connected-mode
(i.e, reliable stream). This adds packet framing and reduces the overhead
a bit, and at 9600 bps or faster, is actually slightly faster than a
carrier pigeon. It also places the burden of retries on the air link
where it belongs, rather than end-to-end, which is clearly the wrong
place to do it.

A number of F-S-U folks have been doing some very interesting work in
real high-speed packet (230 kb/s or faster, some multi-Mb/s), and
such links would easily support digital voice without much need for
super-bandwidth-conserving codecs.

GSM and similar codecs are now easily available in software
implementations; these work quite well at an air rate of 9600 or
sometimes less. Some are free of significant encumbrances that
would hinder widespread usage.

In the land mobile radio regime, the ultimate goal is to reduce the
consumed bandwidth so as to stack more customers in the same band.
Hams have no such need; if anything we have (shhh!) more bandwidth than
we need on the VHF and up bands, so using extremely narrowband codecs
just isn't something we need to do.
- Brian