Audio/Video Transport (AVT) Working Group Minutes
Reported by Colin Perkins
The AVT working group met once at the 64th IETF meeting (November 2005,
Vancouver). Subjects under discussion included media type registration
rules, RTP and RTCP XR MIBs, codec control messages, the scalable video
codec, a framework for multiple description encoding, and RTP payloads
for Speex, Vorbis, Theora, VC-1 and EVRC. The meeting was chaired by
Colin Perkins and Magnus Westerlund.
Introduction and Status Update
The chairs introduced the meeting with the usual status review. The
group has had three RFCs published since the last meeting (RFCs 4175,
4184 and 4247), has a further 11 drafts in the RFC editor queue, and
three more at various points in the IESG review process. The revised
H.263+ and telephony tones/events formats have recently completed WG
last call and need minor changes as a result of comments received.
The following draft are believed almost ready for WG last call: MIME
type registration for H.224, RTP/SAVPF profile, RTCP SSM extensions,
revised H.261 payload format, reclassification of RFC 2190 as historic,
the ULP payload format, and MIDI payload format and implementers guide.
Review of these draft is solicited.
The chairs reviewed the milestones, noting we are slightly behind
schedule. The chairs also noted that there is no draft that defines
RTP over TLS, and apparently no interest in the work item. If there
is no volunteer to proceed with this, we will consider removing that
item from the working group's charter. On a related note, Steve
Casner asked is there was interest in RTP over DTLS? Colin Perkins
noted that this subject has been mentioned informally, but there is
no active work at present (although it can be expected to fit into
the charter should there be interest).
MIME Type Registration of RTP Payload Formats
Steve Casner asked for input regarding the text in section 2.1 of the
draft. Ralph Giles and Roni Even asked if payload formats exist that
don't follow these rules, and if these rules are intended to apply to
all formats, not just those defined in RFC 3555. Steve clarified that
the rules in section 2.1 of the draft document existing practice, and
are believed consistent with existing media type registrations of RTP
payload formats. There were no other comments or objections.
Steve noted that several people have questioned if the draft should
include updated media type registrations, and asked for input on this
subject. Roni Even noted that we're moving the registrations into new
drafts as we take the payload formats to draft standard, and that this
is easier if we separate out the registrations from 3555bis. Stephan
Wenger agreed, but suggested that trivial registrations might remain
in the 3555bis draft. Conclusion: most media type registrations will
be removed from the 3555bis draft.
RTCP XR MIB and RTP MIB Updates
Alan Clark discussed the RTCP XR MIB, summarising the changes since the
last version, and describing the proposed alignment between the RTCP XR
MIB and the newly updated RTP MIB. The proposed alignment uses a common
Session Table between the two MIBs, and there was some discussion about
how this is to be achieved, with Dan Romascanu offering suggestions for
the mechanics of aligning the two MIBs.
Magnus Westerlund expressed concern that the new Session Table includes
RTP flows as its basic object, rather than RTP sessions. This led Steve
Casner and Colin Perkins to wonder if the new structure would limit the
monitoring of multicast sessions? The existing RTP MIB allows monitoring
of the full range of RTP sessions, losing that ability is not desirable.
Alan is expecting to have a revised RTCP XR MIB and an initial draft of
the updated RTP MIB by mid-December, and will circulate on the list for
comments. Dan Romascanu asked if the aim is to progress these together,
and Alan clarified that is the intent.
Codec Control Messages
Magnus Westerlund discussed codec control messages, outlining changes
since the previous version of the draft and the open issues.
The temporal-spatial trade-off message is now acknowledged, informing
receivers that the sender has chosen a specific trade-off. This is more
feedback than provided by existing systems, but we might chose to add a
reason code to the response, to inform receivers why the sender chose a
trade off that might not match their preference. Stephan Wenger noted
that he doesn't see a need for this. Colin Perkins noted that he would
prefer a model where the sender announces the trade-off it has chosen,
rather than acknowledging specific requests, since this allows senders
to make unsolicited changes, and better conforms to a "hint" model for
the control messages.
Regarding the TMMA feedback message, there is presently no signal back
to the receivers of the current maximum bit rate in force - should one
be added? Stephan Wenger and Roni Even discussed use of this feature,
with Stephan outlining possible use cases and Roni noting that he did
not find them persuasive; there was no real conclusion.
Colin Perkins noted that the draft assumes a particular design of MCU,
which makes each leg of the call a separate RTP session, and that this
doesn't match the RTP translator/mixer model described in RFC 3550. To
avoid confusion, Colin requested that the next version of the draft be
updated to include some discussion of the MCU design model chosen, and
how it differs from RFC 3550.
The consensus of the room was to accept this as a working group draft,
although some push-back was noted.
Scalable Video Codec (SVC) Payload Format
Stephan Wenger discussed signalling layered coding structures and the
SVC payload format. He began with a discussion of scalable and layered
coding, explaining possible topologies and requirements, and outlining
how layered codec streams have previously been handled in AVT and
Leaving aside the details of how scalable coding works, Stephan focussed
on the options for signalling layer dependencies: 1) layer dependencies
are signalled using a media-level SDP attribute that is independent of
the codec; 2) layer dependencies are signalled using codec specific SDP
attributes, in the native syntax of the payload format; or 3) dependency
information is sent in-band within the RTP payload format. It is desired
to keep signalling and media separate, so the third option is discounted.
The group should consider if we want a generic signalling method, perhaps
with limited capabilities, or a codec specific signalling method that can
express complex dependencies.
Stephan also briefly outlined some other properties of the SVC extension
to H.264, and the initial proposal for an RTP payload format. There is a
desire to keep this format aligned as much as possible with RFC 3984, in
particular to ensure the base layer can decoded by an unmodified RFC 3984
decoder. Colin Perkins expressed his support for the aim of aligning the
format with RFC 3984, emphasising the desire for backwards compatibility.
Stephan noted that the draft is presented to AVT very early, to allow
coordinated standardisation, and asked the group to consider it as a
work item. Colin Perkins noted that this is, in principle, appropriate
for AVT, but deferred the decision to accept the draft after a poll of
the room showed that only a few people had read it.
Multiple Description Standard-Compatible Framework
Andrea Vitali described the concept of multiple description coding for
error resilience and scalability, and outlined proposed SDP attributes
that can be used to signal various multiple description coding schemes.
The proposed attributes are an example of general SDP signalling, much
as described earlier in the session by Stephan Wenger, and demonstrate
the complexity of defining codec independent signalling for multiple
Magnus Westerlund asked if the transforms described are standardised
anywhere? No, this is a proposal for standard SDP signalling, not an
attempt to standardise the set of transforms. Magnus expressed some
concern that the flexibility was inappropriate, and that it might be
better to define a standard set of transforms elsewhere, which could
be signalled within SDP.
Stephan Wenger noted that there were useful parts to the signalling,
but agreed with Magnus that signalling filter coefficients in SDP is
not appropriate. Colin Perkins and Steve Casner agreed: this is at a
level below that where AVT and MMUSIC usually work.
There is clearly some interest in multiple description and scalable
coding, but no consensus on this particular approach. Discussion of
the alternatives and trade-offs on the mailing list is encouraged.
RTP Payload Formats for Speex, Vorbis and Theora
Ralph Giles discussed the RTP payload formats for Speex, Vorbis and
Theora. The Speex format is straight-forward: Ralph requested review
of the draft, but didn't devote extensive time to it.
The aim is to closely align the Vorbis and Theora formats, and to this
end the Theora format is being held while details of the Vorbis format
are resolved. The issue with both formats is that there is no standard
codebook, and the codec requires a stream-specific codebook to be sent
before decoding can start. Discussion focussed on the need for separate
codebooks, and options for transferring them.
Colin Perkins suggested Ralph consider the IMG work in MMUSIC as an
alternative for distributing code books, in addition to using HTTP.
Ross Finlayson asked how many codebooks are being used with Vorbis, and
if it is possible to standardise a set for streaming. There are only a
small number, but the set of codebooks used changes over time, as the
codec is improved, so flexibility is desired. Stephan Wenger seconded
the desire for flexibility.
Magnus Westerlund and Steve Casner noted that well-known codebooks can
be identified by their URI, or that a longer hash can be used a unique
identifier, avoiding the need for a registry of codebooks. Colin Perkins
suggested the Vorbis community could manage a portion of their URL space
in a manner that allows persistent caching of codebooks, or for off-line
distribution of well-known codebooks.
The Vorbis payload format was previously accepted as an AVT work item,
although it was never submitted as a working group draft; there are no
objections to taking it now. The Theora draft is appropriate as a work
item, but we cannot formally accept it until a draft is available for
RTP Payload Format for VC-1
Anders Klemets discussed the RTP payload format for VC-1. He noted that
an IPR disclosure for the payload format has been submitted to the IETF.
The codec specification is available as SMPTE-421M, and DVB-H expressed
interest in referencing this format.
Anders outlined the changes since the previous draft, and discussed the
open issues. The main issue is the need for offer/answer considerations
in the draft, but Colin Perkins, Roni Even and Magnus Westerlund noted
that this was essential, and other drafts missing offer/answer section
need to be updated.
Stephan Wenger, Ross Finlayson and Ralph Giles noted concern about IPR
on the payload format, and asked for clarification of licensing terms.
There was some considerable discussion on this subject, with the group
expressing a general preference for royalty free licensing, but noting
that there are existing AVT RFCs with RAND terms.
RTP Support for Enhancements to EVRC Family Codecs
Qiaobing Xie discussed enhancements to the payload formats for EVRC
family codecs, noting that 3GGP2 is working on extensions to the codec.
The draft defines a payload format for EVRC-B, a complementary bundled
format, and an update to RFC 3558 with DTX support. Steve Casner asked
if there are interoperability issues with the extra parameters defined
for this format? Perhaps - need to investigate.
The consensus of the room was that this is appropriate as an AVT work
- + -