Re: [codec] Suggested summary...

["Christian Hoene" <[e-mail]>]

Hi Stephen,

 

 

From: stephen botzko [mailto:[e-mail]] 
Sent: Friday, July 02, 2010 3:26 PM
To: Christian Hoene
Cc: Herve Taddei; [e-mail]
Subject: Re: [codec] Suggested summary...

 

I understand the 256 sample size (I did the math also!), though it is perfectly straightforward to adapt transforms to other sizes.  G.719 for instance uses a 960 sample DCT transform for 20 ms frame size.  There are many other examples that could be named.  I don't think the sound card argument is all that relevant, since that is a normal place to put a buffer anyway, and there is no reason why that buffer needs to be related to the frame size.

[Christian Hoene] If it comes to ultra low delay, the buffering at the sound card becomes important and must be optimized. Some sound cards (and their DMA controller) have restrictions on the size of blocks (called periods in ALSA). Having “strange” frame sizes will increase the latency because of re-framing.

Perhaps more importantly, I do not think it is appropriate to work backwards from technology contributions to create the requirements.  

[Christian Hoene] You are right. However, the requirements have to be realistic. Looking at existing technologies helps you to know what is possible at the moment.

The normal way in the IETF is to work forward from use cases and problem statements.  IMHO working forward is much more likely to get good results, and it is fair to all (past and future) contributors.

BTW, personally I do not place a very high priority on distributed internet music ensemble performances.  While I am ok if that happens to fall out of the design, it possibly has a lot of other implications beyond delay.  I would happily trade outstanding packet loss performance for music ensembles if it came down to it, and I suspect many other VOIP folks would concur.  Is there a more compelling  (maybe less fun) use case you can reference in the low-delay requirement?



[Christian Hoene] Musical telepresence is a good use case because it is the upper limit of quality. A monaural audio transmissions can hardly be better. It is also a good marketing argument: The IWAC codec provides the best audio quality at lowest latency.  Also, this use-case is unique because  here is not no other standardized audio codec out here supporting it.

Please have a look on the following PhD to understand the requirements of musical telepresence better:
http://www.itm.uni-luebeck.de/users/carot/Docs/dissertation_AC.pdf

Christian