Re: [video-codec] Comments on draft-maxwell-videocodec-requirements-00

[Sergio Garcia Murillo <[e-mail]>]

El 05/12/2012 0:46, Ali C. Begen (abegen) escribió:
>
> -----Original Message-----
> From: "Timothy B. Terriberry" <[e-mail]>
> Date: Tuesday, December 4, 2012 6:42 PM
> To: "[e-mail]" <[e-mail]>
> Subject: Re: [video-codec] Comments
> on	draft-maxwell-videocodec-requirements-00
>
>> Ali C. Begen (abegen) wrote:
>>>> This is something I hope rmcat will explore in more detail, though I
>>>> recognize it's hard to evaluate. One idea that occurs to me is actually
>>>> using VBR video and padding it with FEC data to get something closer to
>>>> CBR. That might actually be a better use of the bits.
>>> I doubt it. If a frame (or slide) has less bits due to VBR encoding,
>>> that
>>> means it is easier to encode it, which implies it is less important than
>>> the one that has more bits. FEC'ing these frames will not make the
>>> stream
>>> CBR.
>> Well, the idea would be to put the redundant information from the larger
>> (more important) frames in space you didn't use in the smaller frames,
>> to protect against lost packets in those frames.
> That is possible but that will introduce potentially long delays (which we
> don't want in video-codec).
>
Redundant information is important for congestion control algorithms. 
For example the google draft:

http://tools.ietf.org/html/draft-alvestrand-rtcweb-congestion-03

Expects a packet lost rate >10% before reducing the video rate. 
Providing this redundancy in the codec instead
of using FEC could have some benefits.

Also, I think that for videoconferencing, what is most important is not 
CBR vs VBR but to be able to limit the
maximum size of an encoded frame (i.e. something like constrained VBR).

Best regards
Sergio