Network Working Group M. Civanlar
Request for Comments: 2343 G. Cash
Category: EXPerimental B. Haskell
AT&T Labs-Research
May 1998
RTP Payload Format for Bundled MPEG
Status of this Memo
This memo defines an Experimental Protocol for the Internet
community. This memo does not specify an Internet standard of any
kind. Discussion and suggestions for improvement are requested.
Distribution of this memo is unlimited.
Copyright Notice
Copyright (C) The Internet Society (1998). All Rights Reserved.
Abstract
This document describes a payload type for bundled, MPEG-2 encoded
video and audio data that may be used with RTP, version 2. Bundling
has some advantages for this payload type particularly when it is
used for video-on-demand applications. This payload type may be used
when its advantages are important enough to sacrifice the modularity
of having separate audio and video streams.
1. IntrodUCtion
This document describes a bundled packetization scheme for MPEG-2
encoded audio and video streams using the Real-time Transport
Protocol (RTP), version 2 [1].
The MPEG-2 International standard consists of three layers: audio,
video and systems [2]. The audio and the video layers define the
syntax and semantics of the corresponding "elementary streams." The
systems layer supports synchronization and interleaving of multiple
compressed streams, buffer initialization and management, and time
identification. RFC2250 [3] describes packetization techniques to
transport individual audio and video elementary streams as well as
the transport stream, which is defined at the system layer, using the
RTP.
The bundled packetization scheme is needed because it has several
advantages over other schemes for some important applications
including video-on-demand (VOD) where, audio and video are always
used together. Its advantages over independent packetization of
audio and video are:
1. Uses a single port per "program" (i.e. bundled A/V). This may
increase the number of streams that can be served e.g., from a VOD
server. Also, it eliminates the performance hit when two ports are
used for the separate audio and video streams on the client side.
2. Provides implicit synchronization of audio and video. This is
particularly convenient when the A/V data is stored in an
interleaved format at the server.
3. Reduces the header overhead. Since using large packets increases
the effects of losses and delay, audio only packets need to be
smaller increasing the overhead. An A/V bundled format can provide
about 1% overall overhead reduction. Considering the high bitrates
used for MPEG-2 encoded material, e.g. 4 Mbps, the number of bits
saved, e.g. 40 Kbps, may provide noticeable audio or video quality
improvement.
4. May reduce overall receiver buffer size. Audio and video streams
may experience different delays when transmitted separately. The
receiver buffers need to be designed for the longest of these
delays. For example, let's assume that using two buffers, each with
a size B, is sufficient with probability P when each stream is
transmitted individually. The probability that the same buffer size
will be sufficient when both streams need to be received is P times
the conditional probability of B being sufficient for the second
stream given that it was sufficient for the first one. This
conditional probability is, generally, less than one requiring use
of a larger buffer size to achieve the same probability level.
5. May help with the control of the overall bandwidth used by an
A/V program.
And, the advantages over packetization of the transport layer streams
are:
1. Reduced overhead. It does not contain systems layer information
which is redundant for the RTP (essentially they address similar
issues).
2. Easier error recovery. Because of the structured packetization
consistent with the application layer framing (ALF) principle, loss
concealment and error recovery can be made simpler and more
effective.
2. Encapsulation of Bundled MPEG Video and Audio
Video encapsulation follows rules similar to the ones described in
[3] for encapsulation of MPEG elementary streams. Specifically,
1. The MPEG Video_Sequence_Header, when present, will always be at
the beginning of an RTP payload.
2. An MPEG GOP_header, when present, will always be at the
beginning of the RTP payload, or will follow a
Video_Sequence_Header.
3. An MPEG Picture_Header, when present, will always be at the
beginning of a RTP payload, or will follow a GOP_header.
In addition to these, it is required that:
4. Each packet must contain an integral number of video slices.
It is the application's responsibility to adjust the slice sizes and
the number of slices put in each RTP packet so that lower level
fragmentation does not occur. This approach simplifies the receivers
while somewhat increasing the complexity of the transmitter's
packetizer. Considering that a slice can be as small as a single
macroblock, it is possible to prevent fragmentation for most of the
cases. If a packet size exceeds the path maximum transmission unit
(path-MTU) [4], this payload type depends on the lower protocol
layers for fragmentation although, this may cause problems with
packet classification for integrated services (e.g. with RSVP).
The video data is followed by a sufficient number of integral audio
frames to cover the duration of the video segment included in a
packet. For example, if the first packet contains three 1/900
seconds long slices of video, and Layer I audio coding is used at a
44.1kHz sampling rate, only one audio frame covering 384/44100
seconds of audio need be included in this packet. Since the length of
this audio frame (8.71 msec.) is longer than that of the video
segment contained in this packet (3.33 msec), the next few packets
may not contain any audio frames until the packet in which the
covered video time extends outside the length of the previously
transmitted audio frames. Alternatively, it is possible, in this
proposal, to repeat the latest audio frame in "no-audio" packets for
packet loss resilience. Again, it is the application's responsibility
to adjust the bundled packet size according to the minimum MTU size
to prevent fragmentation.
2.1. RTP Fixed Header for BMPEG Encapsulation
The following RTP header fields are used:
Payload Type: A distinct payload type number, which may be dynamic,
should be assigned to BMPEG.
M Bit: Set for packets containing end of a picture.
timestamp: 32-bit 90 kHz timestamp representing sampling time of
the MPEG picture. May not be monotonically increasing if B pictures
are present. Same for all packets belonging to the same picture.
For packets that contain only a sequence, extension and/or GOP
header, the timestamp is that of the subsequent picture.
2.2. BMPEG Specific Header:
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
P NMBZ Audio Length Audio Offset
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
MBZ
P: Picture type (2 bits). I (0), P (1), B (2).
N: Header data changed (1 bit). Set if any part of the video
sequence, extension, GOP and picture header data is different than
that of the previously sent headers. It gets reset when all the
header data gets repeated (see Appendix 1).
MBZ: Must be zero. Reserved for future use.
Audio Length: (10 bits) Length of the audio data in this packet in
bytes. Start of the audio data is found by suBTracting "Audio
Length" from the total length of the received packet.
Audio Offset: (16 bits) The offset between the start of the audio
frame and the RTP timestamp for this packet in number of audio
samples (for multi-channel sources, a set of samples covering all
channels is counted as one sample for this purpose.)
Audio offset is a signed integer in two's complement form. It allows
a ~ +/- 750 msec offset at 44.1 KHz audio sampling rate. For a very
low video frame rate (e.g., a frame per second), this offset may not
be sufficient and this payload format may not be usable.
If B frames are present, audio frames are not re-ordered along with
video. Instead, they are packetized along with video frames in
their transmission order (e.g., an audio segment packetized with a
video segment corresponding to a P picture may belong to a B
picture, which will be transmitted later and should be rendered at
the same time with this audio segment.) Even though the video
segments are reordered, the audio offset for a particular audio
segment is still relative to the RTP timestamp in the packet
containing that audio segment.
Since a special picture counter, such as the "temporal reference
(TR)" field of [3], is not included in this payload format, lost GOP
headers may not be detected. The only effect of this may be
incorrect decoding of the B pictures immediately following the lost
GOP header for some edited video material.
3. Security Considerations
RTP packets using the payload format defined in this specification
are subject to the security considerations discussed in the RTP
specification [1]. This implies that confidentiality of the media
streams is achieved by encryption. Because the data compression used
with this payload format is applied end-to-end, encryption may be
performed after compression so there is no conflict between the two
operations.
This payload type does not exhibit any significant non-uniformity in
the receiver side computational complexity for packet processing to
cause a potential denial-of-service threat.
A security review of this payload format found no additional
considerations beyond those in the RTP specification.
Appendix 1. Error Recovery
Packet losses can be detected from a combination of the sequence
number and the timestamp fields of the RTP fixed header. The extent
of the loss can be determined from the timestamp, the slice number
and the horizontal location of the first slice in the packet. The
slice number and the horizontal location can be determined from the
slice header and the first macroblock address increment, which are
located at fixed bit positions.
If lost data consists of slices all from the same picture, new data
following the loss may simply be given to the video decoder which
will normally repeat missing pixels from a previous picture. The next
audio frame must be played at the appropriate time determined by the
timestamp and the audio offset contained in the received packet.
Appropriate audio frames (e.g., representing background noise) may
need to be fed to the audio decoder in place of the lost audio frames
to keep the lip-synch and/or to conceal the effects of the losses.
If the received new data after a loss is from the next picture (i.e.
no complete picture loss) and the N bit is not set, previously
received headers for the particular picture type (determined from the
P bits) can be given to the video decoder followed by the new data.
If N is set, data deletion until a new picture start code is
advisable unless headers are made available to the receiver through
some other channel.
If data for more than one picture is lost and headers are not
available, unless N is zero and at least one packet has been received
for every intervening picture of the same type and that the N bit was
0 for each of those pictures, resynchronization to a new video
sequence header is advisable.
In all cases of heavy packet losses, if the correct headers for the
missing Pictures are available, they can be given to the video
decoder and the received data can be used irrespective of the N bit
value or the number of lost pictures.
Appendix 2. Resynchronization
As described in [3], use of frequent video sequence headers makes it
possible to join in a program at arbitrary times. Also, it reduces
the resynchronization time after severe losses.
References
[1] Schulzrinne, H., Casner, S., Frederick, R., and V. Jacobson,
"RTP: A Transport Protocol for Real-Time Applications", RFC1889,
January 1996.
[2] ISO/IEC International Standard 13818; "Generic coding of moving
pictures and associated audio information," November 1994.
[3] Hoffman, D., Fernando, G., Goyal, V., and M. Civanlar, "RTP
Payload Format for MPEG1/MPEG2 Video", RFC2250, January 1998.
[4] Mogul, J., and S. Deering, "Path MTU Discovery", RFC1191,
November 1990.
Authors' Addresses
M. Reha Civanlar
AT&T Labs-Research
100 Schultz Drive
Red Bank, NJ 07701
USA
EMail: civanlar@research.att.com
Glenn L. Cash
AT&T Labs-Research
100 Schultz Drive
Red Bank, NJ 07701
USA
EMail: glenn@research.att.com
Barry G. Haskell
AT&T Labs-Research
100 Schultz Drive
Red Bank, NJ 07701
USA
EMail: bgh@research.att.com
Full Copyright Statement
Copyright (C) The Internet Society (1998). All Rights Reserved.
This document and translations of it may be copied and furnished to
others, and derivative works that comment on or otherwise explain it
or assist in its implementation may be prepared, copied, published
and distributed, in whole or in part, without restriction of any
kind, provided that the above copyright notice and this paragraph are
included on all such copies and derivative works. However, this
document itself may not be modified in any way, such as by removing
the copyright notice or references to the Internet Society or other
Internet organizations, except as needed for the purpose of
developing Internet standards in which case the procedures for
copyrights defined in the Internet Standards process must be
followed, or as required to translate it into languages other than
English.
The limited permissions granted above are perpetual and will not be
revoked by the Internet Society or its successors or assigns.
This document and the information contained herein is provided on an
"AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING
TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING
BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION
HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF
MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
