[vorbis-dev] RFC draft for Vorbis over RTP

[Tor-Einar Jarnbjo]

Gregory Maxwell wrote:

>With most FEC implimentations the extra data isn't just 'extra'. Though I
>suppose you could make one RTP include the bare minimum needed for
>decode.. Really if you go totally gunho with FEC it would be advantagious
>to break up the vorbis packets and transmits the differnt layers of the
>cascade with differing amounts of FEC protection.
>
I took a quick glance at the papers you referenced and this piece of 
source code (or rather a Java port of it):

http://info.iet.unipi.it/~luigi/fec.html#fec.ps

If you throw k packets into the encoder and get n>k packets out, the 
packets 0 to k-1 is just a copy of the original packets and all the 
redundancy is placed in the packets k to n-1. It's still possible to 
take _any_ k packets out of the encoded bundle and reconstruct the 
original packets. I must admit that I didn't understand the mathematical 
theory behind this and it truly amazed me a little bit, but it wouldn't 
be very difficult to send the redundancy packets in a separate stream, 
letting a client with a good network connection just receive the regular 
packets and allow other clients to dynamically subscribe to different 
streams with different levels of FEC. I bet that implementing such a 
transport protocol is less work than writing an RFC compliant 
description of the FEC algorithms used :)

>Sometimes the long delay might be acceptable (esp in the case were the
>client caches headers and only one set of headers is ever used by the
>station)... 
>
I'm sorry, but I can't agree with you on this. I image some 
non-technical radio station employee testing such a client. He starts 
it, and after 15-20 seconds, there's still no sound playing. He'll ditch 
it and go back to the Windows Media Player.

Tor

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'vorbis-dev-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.