Internet Engineering Task Force M. Reha Civanlar INTERNET-DRAFT Glenn L. Cash File: draft-civanlar-bmpeg-01.txt Barry G. Haskell Expire in six months AT&T Labs-Research February, 1997 RTP Payload Format for Bundled MPEG Status of this Memo This document is an Internet-Draft. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet- Drafts as reference material or to cite them other than as ``work in progress.'' To learn the current status of any Internet-Draft, please check the ``1id-abstracts.txt'' listing contained in the Internet- Drafts Shadow Directories on ftp.is.co.za (Africa), nic.nordu.net (Europe), munnari.oz.au (Pacific Rim), ds.internic.net (US East Coast), or ftp.isi.edu (US West Coast). Distribution of this memo is unlimited. Abstract This document describes a payload type for bundled, MPEG-2 encoded video and audio data to 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 is to be used when its advantages are important enough to sacrifice the modularity of having separate audio and video streams. A technique to improve packet loss resilience based on "out-of-band" transmission of MPEG-2 specific, vital information is described also. 1. Introduction draft-civanlar-bmpeg-01.txt [Page 1] INTERNET-DRAFT RTP Payload Format for Bundled MPEG February, 1997 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. RFC 2038 [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 messages on the client side. 2. Provides implicit synchronization of audio and video. The server need not do anything else (e.g. generate RTCP packets) for this purpose. This is particularly convenient when the A/V data is stored in an interleaved format at the server and no stream other than the bundled A/V is to be transmitted during the session. 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 draft-civanlar-bmpeg-01.txt [Page 2] INTERNET-DRAFT RTP Payload Format for Bundled MPEG February, 1997 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. 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 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 the rules described in [3] with the addition of the following: each packet must contain an integral number of video slices 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. 2.1. RTP Fixed Header for BMPEG Encapsulation The following RTP header fields are used: Payload Type: A distinct payload type number should be assigned to BMPEG. M Bit: Set for packets containing end of a picture. draft-civanlar-bmpeg-01.txt [Page 3] INTERNET-DRAFT RTP Payload Format for Bundled MPEG February, 1997 timestamp: 32-bit 90 kHz timestamp representing transmission time of the MPEG picture and is monotonically increasing. 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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |MBZ|R|N| P | Audio Length | Audio Offset | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ MBZ: Reserved for future use (2 bits). They must be set to zero now. R: Redundant audio (1 bit). Set if the audio frame contained in the packet is a repetition of the last audio frame. 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. P: Picture type (2 bits). I (0), P (1), B (2). Audio Length: (10 bits) Length of the audio data in this packet in bytes. Audio Offset: (16 bits) The offset between the audio frame and the start of the video segment in this packet in number of audio samples. 3. Out-of-band Transmission of the "High Priority" Information In MPEG encoded video, loss of the header information, which includes sequence, GOP, and picture headers, and the corresponding extensions, causes severe degradations in the decoded video. When possible, dependable transmission of the header information to the receivers can improve the loss resiliency of MPEG video significantly [4]. RFC 2038 describes a payload type where the header information can be repeated in each RTP packet. Although this is a straightforward approach, it may increase the overhead. The "data partitioning" method in MPEG-2 defines the syntax and semantics for partitioning an MPEG-2 encoded video bitstream into "high priority" and "low priority" parts. If the "high priority" (HP) part is selected to contain only the header information, it is less than two percent of the video data and can be transmitted before the start of the draft-civanlar-bmpeg-01.txt [Page 4] INTERNET-DRAFT RTP Payload Format for Bundled MPEG February, 1997 real-time transmission using a reliable protocol. In order to synchronize the HP data with the corresponding real-time stream, the initial value of the timestamp for the real-time stream may be inserted at the beginning of the HP data. Alternatively, the HP data may be transmitted along with the A/V data using layered multimedia transmission techniques for RTP [5]. 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 can simply be given to the video decoder which will normally repeat missing pixels from a previous picture. The next audio frame must be delayed by the duration of the lost video segment. If the received new data after a loss is from the next picture 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 available from previously received HP data. In both cases audio needs to be delayed properly. If data for more than one picture is lost and HP data is not available, resynchronization to a new video sequence header is advisable. In all cases of large packet losses, if the HP data is available, appropriate portions of it 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] H. Schulzrinne, S. Casner, R. Frederick, V. Jacobson, "RTP: A Transport Protocol for Real-Time Applications," draft-civanlar-bmpeg-01.txt [Page 5] INTERNET-DRAFT RTP Payload Format for Bundled MPEG February, 1997 RFC 1889, January 1996. [2] ISO/IEC International Standard 13818; "Generic coding of moving pictures and associated audio information," November 1994. [3] D. Hoffman, G. Fernando, S. Kleiman, V. Goyal, "RTP Payload Format for MPEG1/MPEG2 Video," RFC 2038, October 1996. [4] M. R. Civanlar, G. L. Cash, "A practical system for MPEG-2 based video-on-demand over ATM packet networks and the WWW," Signal Processing: Image Communication, no. 8, pp. 221-227, Elsevier, 1996. [5] M. F. Speer, S. McCanne, "RTP Usage with Layered Multimedia Streams," Internet Draft, draft-speer-avt-layered-video-02.txt, December 1996. Author's Address: M. Reha Civanlar Glenn L. Cash Barry G. Haskell AT&T Labs-Research 101 Crawfords Corner Road Holmdel, NJ 07733 USA e-mail: civanlar|glenn|bgh@research.att.com draft-civanlar-bmpeg-01.txt [Page 6]