1、SMPTE 381M-2005 Page 1 of 33 pages Table of contents 1 Scope 2 Normative references 3 Glossary of acronyms, terms and data types 4 Introduction 5 Mapping MPEG into the generic container 6 KLV coding of MPEG elements 7 SMPTE label for essence container identification 8 Essence descriptors for MPEG ma
2、ppings Annex A Index tables for MPEG mappings Annex B Identifying MPEG picture types Annex C Requirements for mapping MPEG into MXF Annex D Guidance for mapping new or non-ISO defined essence in MPEG streams into MXF Annex E Bibliography 1 Scope This standard specifies the mapping of MPEG streams, a
3、s identified by an ISO 13818-1 stream_id value, into the MXF generic container. This includes, but is not limited to, ISO 13818-2 MPEG video stream, ISO 11172-2 MPEG video streams, ISO 13818-3 MPEG audio streams, and ISO 11172-3 MPEG audio streams. Provision is made for mapping the content with or w
4、ithout the ISO 13818-1 PES layer. Provision is also made for single stream mapping and synchronized multi-stream mapping. The document outlines synchronization requirements for the MXF encapsulation of co-timed MPEG essence streams. This standard defines SMPTE universal labels to be used to uniquely
5、 identify specific MPEG implementations. This standard also provides for clip wrapping a single program MPEG program stream or transport stream multiplex. Support for multi-program MPEG multiplexes is not provided in this standard. Support for transport streams where some of the transport stream pac
6、kets have been removed is not provided in this standard. This standard does not map MPEG PSI or DVB-SI or ATSC-PSIP to any metadata scheme. In order to achieve interoperability within any given operational pattern, restrictions may be placed on the way in which this essence container can be implemen
7、ted. The reader is advised to carefully study the appropriate operational pattern document before implementation. 2 Normative references The following standards contain provisions which, through reference in this text, constitute provisions of this standard. At the time of publication, the editions
8、indicated were valid. All standards are subject to revision and Page 1 of 33 pages SMPTE 381M-2005 Copyright 2005 by THE SOCIETY OF MOTION PICTURE AND TELEVISION ENGINEERS 595 W. Hartsdale Ave., White Plains, NY 10607 (914) 761-1100 Approved February 15, 2005 SMPTE STANDARD for Television Material E
9、xchange Format (MXF) Mapping MPEG Streams into the MXF Generic Container SMPTE 381M-2005 Page 2 of 33 pages parties to agreements based on this standard are encouraged to investigate the possibility of applying the most recent editions of the standards indicated below. SMPTE 336M-2001, Television Da
10、ta Encoding Protocol using Key-Length-Value SMPTE 377M-2004, Television Material Exchange Format (MXF) File Format Specification SMPTE 379M-2004, Television Material Exchange Format (MXF) Generic Container SMPTE 385M-2004, Television Material Exchange Format (MXF) Mapping SDTI-CP Essence and Metadat
11、a into the MXF Generic Container SMPTE RP 210, Metadata Dictionary Registry of Metadata Element Descriptions SMPTE RP 224, SMPTE Labels Registry ISO 13818 Parts 1, 2, 3, 7, Information Technology Generic Coding of Moving Pictures and Associated Audio Information: Systems, Video, Audio, Advanced Audi
12、o Coding (AAC) 3 Glossary of acronyms, terms and data types The full glossary of terms and acronyms used in the MXF specification is given in the MXF file format specification. Additional terms are given in the glossaries in SMPTE 379M and ISO 13818 (the MPEG2 specification). Terms are not repeated
13、here to avoid any divergence of meaning. AFD: Active format descriptor, defined by ETSI ETR154 a parameter which defines the aspect ratio of the active content within the MPEG coded frame; e.g., 4:3 content within a coded 16:9 rectangle where black bars left and right are part of the MPEG coded sign
14、al. BWF: Broadcast wave audio format (see SMPTE 382M in annex E) ES: Elementary stream GOP: Group of pictures SPTS: Single program transport stream PES: Packetized elementary stream PMT: Program map table PSM: Program stream map VCD: Video CD VOB: Video object (DVD term) 4 Introduction In this intro
15、duction, it is assumed that a general understanding of MPEG video compression and the meaning of acronyms such as GOP, B-frame, and PES are understood. These acronyms are well defined in the normative references and will not be repeated here. 4.1 Basics of wrapping: Frames and access units (informat
16、ive) The MPEG video specifications enable the compression of picture-based video. These pictures may be frames or fields. The MPEG specification provides signaling for the number of pixels in the picture, the field / frame rate and other parameters including the aspect ratio of the pixels. Many of t
17、hese parameters will be SMPTE 381M-2005 Page 3 of 33 pages copied into the MXF essence descriptor. In order to ensure interoperability, it is important to ensure that MPEG header information can be found when decoding a KLV wrapped MPEG file. For this reason, the text in section 5 refers to access u
18、nits (AUs) rather than pictures. The full definition of an access unit (AU) is given in ISO 13818-1 and is summarized below for MPEG video and MPEG audio: Audio: 1 AU = The coded representation of an audio frame Video: 1 AU = The coded data for a picture and any stuffing that follows it, up to but n
19、ot including the start of the next access unit. If a picture is not preceded by a group_start_code or a sequence_header_code, the access unit begins with a picture_start_code. If a picture is preceded by a group_start_code or a sequence_header_code, the access unit begins with the first byte of thes
20、e start codes. If it is the last picture preceding a sequence_end_code in the bit stream all bytes between the last bytes of the coded picture and the sequence_end_code (including the sequence_end_code) belong to the access unit. 4.2 Timeline of the MXF tracks representing MPEG content (informative)
21、 Many MXF files containing MPEG content will contain a picture track for the MPEG video data and sound tracks for any audio data. Many professional applications will use uncompressed audio which will have a simple timing relationship with the MPEG video. When an MPEG transmission or distribution fee
22、d is captured, the MXF representation may be more complicated. There is no intention within this mapping document to be able to exactly recreate a program stream or transport stream from its demultiplexed MXF components. 4.2.1 Creating an MXF file from an MPEG transport stream or an MPEG program str
23、eam Consider an MPEG transport stream which contains compressed audio tracks and compressed video with other PES streams such as subtitles. It is likely that these PES streams may be bursty in nature; i.e., there may be many minutes of picture and sound with no other PES content present in the strea
24、m. These same issues also apply to program streams. Capture option 1: Capture the entire transport stream or program stream and wrap in a single KLV packet (as outlined in section 6.3.5). This may be optimal when the stream is likely to be processed or used as a transport stream or program stream, r
25、espectively. Capture option 2: Capture a partial transport stream (i.e., a stream where elements within the selected duration of the original stream have been removed) or a partial stream program stream (i.e., where elements within the selected duration of the original stream have been removed). The
26、se scenarios are outside the scope of this standard. Capture option 3: Demultiplex the components of the captured transport stream or program stream into its elementary streams or PES streams and maintain the relative timing information by creating a valid MXF header on the fly. There are a few issu
27、es here: 1. In general the desired start frame is not the first presentation frame in a closed GOP sequence. The first “I“ frame or sufficient frames before the desired frame (in the case where there are no I frames) must be captured to pre-charge the decoder. 2. The compressed audio AU duration is
28、rarely the same as the video AU duration. To capture without recoding the audio requires calculating the temporal audio-video offset by inspection of PTS values and using these to create the MXF tracks with appropriate origin values. 3. PES data elements may have AU durations unrelated to the audio
29、or video AUs. For example data such as ETSI EN300374 subtitle bitmaps may be delivered in the stream well before they are required to be displayed. A capture device may not know that the data is subtitles, only that the PTS is some time in the distant future. The MXF track which defines the subtitle
30、 shall be able to offset the origin of the track by a sufficiently large value to correct the timing. SMPTE 381M-2005 Page 4 of 33 pages 4. Any PES data elements must be preserved with their PES headers which may contain extra information which is unknown to the capture device. 5. When demultiplexin
31、g takes place, it is assumed that preservation of the MPEG timing is sufficient and that preservation of the data in a form which allows streaming with the small MPEG buffers is not required. There is no buffer model for MXF streaming and, as a result, the streaming of the data in MXF form may have
32、a higher latency than the streaming as a transport stream or program stream. It also means that re-creation of the transport stream or program stream will require an MPEG multiplexer with knowledge of the buffer models of all the tracks (possibly even the private PES ones) in the stream. The exact r
33、elative timing relationship between the ES and PES components of the program stream or transport stream depends on the original time base of the video stream, the sampling rate of the audio, the compression scheme used by the audio, and the PES strategy employed by the original multiplexer. These fa
34、ctors allow the relative timing relationships between the MXF tracks to be defined. 4.3 MXF tracks (normative) In order to be able to represent the captured MPEG data in an MXF file, it must be possible to describe the start point of the track data, as well as the “in“ and “out“ points of the desire
35、d picture, sound, and PES data. This is shown in figure 1. Picture Sound PES Data GOP precharge GOP overrun Audio access unit size difference Private PES data delivery / PTS difference Desired ”In point“ Desired “Out point“ Audio access unit size difference Figure 1 MXF track timing for MPEG generic
36、 containers The example shown in figure 1 shows the relationship between the desired start and end points of a stored generalized MPEG segment and the actual data start and end points which would be found in the stored essence. The track and sequence items within the header must describe accurately
37、the synchronization scenario above. The picture track shows that the desired start frame does not coincide with the first frame of an MPEG random access point. There is picture data between the first byte of stored data and the first byte of the desired picture information. In the MXF picture track,
38、 there shall be a positive origin value which shall correspond to the difference in position of the first picture stored in the generic container and the position of the desired start picture of the essence. Caution is advised, however, when higher operational patterns are used with the MPEG generic
39、 container mapping. In an OP1a file, the material package shall have the same start position and duration as the top-level file package. Unwanted frames which were captured to allow decoding of the desired frames are, therefore, hidden using the mechanism described above. This is still a valid OP1a
40、file. SMPTE 381M-2005 Page 5 of 33 pages In an operational pattern 3x file, editing is allowed; i.e., the material package may reference any portion of a top-level file package track. In this case, the desired start point of the file package track may be the very first image in the container, even w
41、hen it is not decodable. The desired start position of the material package track may then be offset relative to the start of the top-level file package track so that a decodable portion of the essence is referenced. If an MXF encoder captures an MPEG long GOP stream and is unable to determine the c
42、orrect value of origin, then a decoder shall assume that the desired start point of the track is the first frame of the track. The first frame of the track may not be decodable and/or displayable. An MXF decoder should exhibit some consistent behavior under these circumstances; e.g., by displaying b
43、lack frames until a decodable frame is reached. The audio track shows that the decoded audio sample which should start the captured clip does not coincide with the first sample of the MPEG coded audio frame in which it lies. Compressed audio frames shall be decoded in order to obtain the desired aud
44、io sample within them. In the MXF sound track, there shall be a positive origin value which shall correspond to the difference in position of the first stored audio sample in the compressed audio frame and the position of the desired audio start sample. To achieve this, the edit rate of the sound tr
45、ack (e.g., 1/48 kHz) may be different to the edit rate of the picture track (e.g., 1/29.97 Hz). It is important to note that in order to construct an index table, all the indexed tracks shall have the same edit rate as described in the MXF format document. The PES data track shows that there may be
46、PES data in the file which is delivered a long time before it needs to be applied. The origin and edit rate of the data track shall be set to ensure that the data is delivered synchronously with the picture and sound information which had the same PTS values in the original MPEG stream. It is the in
47、tention of this standard that the above scenarios can be calculated by using the PTS/DTS values in each captured stream. 4.4 External audio (informative) In many applications involving MPEG video MXF files, the audio may be located in an external file as described in SMPTE EG 41. When the audio is o
48、rganized in large chunks / frames, then the timing guidelines above must still be respected. Note that when the audio is external, there will be at least two file packages. 4.5 Wrapping options (normative) 4.5.1 Overview The MXF generic container consists of one or more contiguous KLV wrapped conten
49、t packages. The goal is that each content package should have the same duration and should contain the picture and sound data for that portion of the timeline. When long GOP MPEG video is used, this becomes difficult to achieve. If we choose to wrap every access unit of long GOP MPEG video then we discover that the frames are being wrapped in transmission order rather than display order. If we interleave uncompressed audio sample in the same content package, we will not be re-ordering the audio samples to match the video. In addition, if the video i
