Re: [Perc] Double PERC draft 5 and webrtc audio FEC

Cullen said:

"Let me just ignore for a second that we disagree on best way to do RED
when using SRTP (never mind PERC).  My read of draft-ietf-rtcweb-fec was
that 2198 is a "MAY" be used but that the "RECOMMENDED" approach for opus
in hybrid mode (which is preferred for WebRTC) was just to use the FEC that
is build into Opus.  All the data I have seen suggests that opus FEC
outperforms RED in pretty much all ways RED typically gets used so this
seems somewhat reasonable."

[BA]  Opus FEC can't handle burst loss, but RED can.  So if you have burst
loss (quite likely on wireless networks), RED can substantially outperform
Opus FEC.  I realize this would involve use of RED in an *atypical* way,
though.

Also, it's important to compare Opus FEC against concealment.  When that is
done, its benefits do not stand out.

On Thu, Jun 29, 2017 at 3:43 PM, Cullen Jennings <fluffy@iii.ca> wrote:

>
> > On Jun 29, 2017, at 1:46 PM, Sergio Garcia Murillo <
> sergio.garcia.murillo@gmail.com> wrote:
> >
> > Hi all,
> >
> > In the new double encryption draft that Cullen has just submitted, it is
> says:
> >
> > 7.3.  FEC
> >
> >    The algorithms recommended in [I-D.ietf-rtcweb-fec] for audio work
> >    with no additional considerations.
> >
> > But in ietf-rtcweb-fec it recommends to use rfc 2198 RTP payload for
> redundant audio data. The problem is that RED only protects the payload,
> and any RTP header extension will not be protected. That will prevent RED
> to work HBH as the receiver end will not be able to recover the OHB and
> perform the inner crypto on the recovered packet.
> > Best regards
> >
>
> Let me just ignore for a second that we disagree on best way to do RED
> when using SRTP (never mind PERC).  My read of draft-ietf-rtcweb-fec was
> that 2198 is a "MAY" be used but that the "RECOMMENDED" approach for opus
> in hybrid mode (which is preferred for WebRTC) was just to use the FEC that
> is build into Opus.  All the data I have seen suggests that opus FEC
> outperforms RED in pretty much all ways RED typically gets used so this
> seems somewhat reasonable.
>
> But that said, if you think the draft should mention RED, glad to add it.
> Plus the sentence in the draft probably needs to explain what works means
> in this context.
>
>
> > Sergop
> >
> > _______________________________________________
> > Perc mailing list
> > Perc@ietf.org
> > https://www.ietf.org/mailman/listinfo/perc
>
> _______________________________________________
> Perc mailing list
> Perc@ietf.org
> https://www.ietf.org/mailman/listinfo/perc
>