================================== NUT Open Container Format 20061104 ================================== Intro: ====== NUT is a Free multimedia container format for storage of audio, video, subtitles and related user defined streams, it provides exact timestamps for synchronization and seeking, is simple, has low overhead and can recover in case of errors in the stream other common multimedia container formats are AVI, OGG, Matroska, MP4, MOV ASF, MPEG-PS, MPEG-TS Features / goals: (supported by the format, not necessarily by a specific implementation) Simple use the same encoding for nearly all fields simple decoding, so slow CPUs (and embedded systems) can handle it Extendible no limit for the possible values of all fields (using universal vlc) allow adding of new headers in the future allow adding more fields at the end of headers Compact ~0.2% overhead, for normal bitrates index is <100kb per hour a usual header for a file is about 100 bytes (audio + video headers together) a packet header is about ~1-5 bytes Error resistant seeking / playback without an index headers & index can be repeated damaged files can be played back with minimal data loss and fast resync times The spec is frozen. All files following spec will be compatible unless the spec is unfrozen. Definitions: ============ MUST the specific part must be done to conform to this standard SHOULD it is recommended to be done that way, but not strictly required keyframe A keyframe is a frame from which you can start decoding, a more exact definition is below The nth frame is a keyframe if and only if frames n, n+1, ... in presentation order (that are all frames with a pts >= frame[n].pts) can be decoded successfully without reference to frames prior n in storage order (that are all frames with a dts < frame[n].dts) if no such frames exist (for example due to use of overlapped transforms like the MDCT in an audio codec) then the definition shall be extended by dropping n out of the set of frames which must be decodable, if this is still insufficient then n+1 shall be dropped, and so on until there is a keyframe every frame which is marked as a keyframe MUST be a keyframe according to the definition above, a muxer MUST mark every frame it knows is a keyframe as such, a muxer SHOULD NOT analyze future frames to determine the keyframe status of the current frame but instead just set the frame as non keyframe (FIXME maybe move somewhere else?) pts presentation time of the first frame/sample that is completed by decoding the coded frame. dts of a frame is the time when it is input into a synchronous 1-in-1-out decoder Syntax: ======= Since NUT heavily uses variable length fields, the simplest way to describe it is using a pseudocode approach. Conventions: ============ The data types have a name, used in the bitstream syntax description, a short text description and a pseudocode (functional) definition, optional notes may follow: name (text description) functional definition [Optional notes] The bitstream syntax elements have a tagname and a functional definition, they are presented in a bottom up approach, again optional notes may follow and are reproduced in the tag description: name: (optional note) functional definition [Optional notes] The in-depth tag description follows the bitstream syntax. The functional definition has a C-like syntax. Type definitions: ================= f(n) (n fixed bits in big-endian order) u(n) (unsigned number encoded in n bits in MSB-first order) v (variable length value, unsigned) value=0 do{ more_data u(1) data u(7) value= 128*value + data }while(more_data) s (variable length value, signed) temp v temp++ if(temp&1) value= -(temp>>1) else value= (temp>>1) b (binary data or string, to be use in vb, see below) for(i=0; i 4096) header_checksum u(32) packet_footer reserved_bytes checksum u(32) [Note: in index packet, reserved_bytes comes before index_ptr] reserved_headers while(next_byte == 'N' && next_code != main_startcode && next_code != stream_startcode && next_code != info_startcode && next_code != index_startcode && next_code != syncpoint_startcode){ packet_header packet_footer } Headers: main_header: version v stream_count v max_distance v time_base_count v for(i=0; i0) tmp_pts s if(tmp_fields>1) tmp_mul v if(tmp_fields>2) tmp_stream v if(tmp_fields>3) tmp_size v else tmp_size=0 if(tmp_fields>4) tmp_res v else tmp_res=0 if(tmp_fields>5) count v else count= tmp_mul - tmp_size for(j=6; j>=1 n=j if(type){ flag= x & 1 x>>=1 while(x--) has_keyframe[n++][i]=flag has_keyframe[n++][i]=!flag; }else{ while(x != 1){ has_keyframe[n++][i]=x&1; x>>=1; } } for(; j65536 then max_distance MUST be set to 65536 This is also half the max frame size without a checksum after the frameheader. max_pts_distance (v) max absolute difference of pts of new frame from last_pts in the timebase of the stream, without a checksum after the frameheader. A frame header MUST include a checksum if abs(pts-last_pts) is strictly greater than max_pts_distance. Note that last_pts is not necessarily the pts of the last frame on the same stream, as it is altered by syncpoint timestamps. SHOULD NOT be higher than 1/timebase stream_id (v) Stream identifier stream_id MUST be < stream_count stream_class (v) 0 video 1 audio 2 subtitles 3 userdata Note: the remaining values are reserved and MUST NOT be used a demuxer MUST ignore streams with reserved classes fourcc (vb) identification for the codec example: "H264" MUST contain 2 or 4 bytes, note, this might be increased in the future if needed the id values used are the same as in avi, so if a codec uses a specific fourcc in avi then the same fourcc MUST be used here time_base_nom (v) / time_base_denom (v) = time_base the length of a timer tick in seconds, this MUST be equal to the 1/fps if FLAG_FIXED_FPS is set time_base_nom and time_base_denom MUST NOT be 0 time_base_nom and time_base_denom MUST be relatively prime time_base_denom MUST be < 2^31 examples: fps time_base_nom time_base_denom 30 1 30 29.97 1001 30000 23.976 1001 24000 There MUST NOT be 2 identical timebases in a file. There SHOULD NOT be more timebases than streams. time_base_id (v) index into the time_base table MUST be < time_base_count convert_ts To switch from 2 different timebases, the following calculation is defined: ln = from_time_base_nom*to_time_base_denom sn = from_timestamp d1 = from_time_base_denom d2 = to_time_base_nom timestamp = (ln/d1*sn + ln%d1*sn/d1)/d2 Note: this calculation MUST be done with unsigned 64 bit integers, and is equivalent to (ln*sn)/(d1*d2) but this would require a 96bit integer compare_ts Compares timestamps from 2 different timebases, if a is before b then compare_ts(a, b) = -1 if a is after b then compare_ts(a, b) = 1 else compare_ts(a, b) = 0 Care must be taken that this is done exactly with no rounding errors, simply casting to float or double and doing the obvious a*timebase > b*timebase is not compliant or correct, neither is the same with integers, and a*a_timebase.num*b_timebase.den > b*b_timebase.num*a_timebase.den will overflow. One possible implementation which shouldn't overflow within the range of legal timestamps and timebases is: if (convert_ts(a, a_timebase, b_timebase) < b) return -1; if (convert_ts(b, b_timebase, a_timebase) < a) return 1; return 0; msb_pts_shift (v) amount of bits in lsb_pts MUST be <16 decode_delay (v) size of the reordering buffer used to convert pts to dts codecs which don't support b frames normally use 0 mpeg1/mpeg2 style codecs with b frames use 1 h264 style b pyramid uses 2 h264 and future codecs might need values >2 audio codecs generally use 0 (we aren't aware of any which doesn't but its theoretically possible that one exists which needs it >0) decode_delay MUST NOT be set higher than necessary for a codec. stream_flags (v) Bit Name Description 1 FLAG_FIXED_FPS indicates that the fps is fixed codec_specific_data (vb) private global data for a codec (could be huffman tables or ...) if a codec has a global header it SHOULD be placed in here instead of at the start of every keyframe the exact format is specified in the codec spec for H.264 the NAL units MUST be formatted as in a bytestream (with 00 00 01 prefixes) codec_specific_data SHOULD contain exactly the essential global packets needed to decode a stream, more specifically it SHOULD NOT contain packets which contain only non essential metadata like author, title, ... it also MUST NOT contain normal packets which cause the reference decoder to generate any specific decoded samples the encoder name and version, shall be considered essential as it is very usefull to workaround possible encoder bugs frame_code (f(8)) frame_code is an 8-bit field which exists before every frame, it can store part of the size of the frame, the stream number, the timestamp and some flags amongst other things, what is not directly stored in it but is needed is stored in various fields immediately after it the values stored in it can be found in the main header the value 78 ('N') is forbidden to ensure that the byte is always different from the first byte of any startcode a muxer SHOULD mark 0x00 and 0xFF as invalid to improve error detection flags[frame_code], frame_flags (v) Bit Name Description 0 FLAG_KEY if set, frame is keyframe 1 FLAG_EOR if set, stream has no relevance on presentation. (EOR) 3 FLAG_CODED_PTS if set, coded_pts is in the frame header 4 FLAG_STREAM_ID if set, stream_id is coded in the frame header 5 FLAG_SIZE_MSB if set, data_size_msb is at frame header, otherwise data_size_msb is 0 6 FLAG_CHECKSUM if set then the frame header contains a checksum 7 FLAG_RESERVED if set, reserved_count is coded in the frame header 12 FLAG_CODED if set, coded_flags are stored in the frame header. 13 FLAG_INVALID if set, frame_code is invalid. EOR frames MUST be zero-length and must be set keyframe. All streams SHOULD end with EOR, where the pts of the EOR indicates the end presentation time of the final frame. An EOR set stream is unset by the first content frames. EOR can only be unset in streams with zero decode_delay . FLAG_CHECKSUM MUST be set if the frame's data_size is strictly greater than 2*max_distance or the difference abs(pts-last_pts) is strictly greater than max_pts_distance (where pts represents this frame's pts and last_pts is defined as below). last_pts the timestamp of the last frame with the same stream_id as the current if there is no such frame between the last syncpoint and the current frame then the syncpoint timestamp is used, see global_key_pts stream_id[frame_code] (v) if FLAG_STREAM_ID is not set then this is the stream number for the frame following this frame_code if FLAG_STREAM_ID is set then this value has no meaning MUST be <250 data_size_mul[frame_code] (v) if FLAG_SIZE_MSB is set then data_size_msb which is stored after the frame code is multiplied with it and forms the more significant part of the size of the following frame if FLAG_SIZE_MSB is not set then this field has no meaning MUST be <16384 data_size_lsb[frame_code] (v) the less significant part of the size of the following frame this added together with data_size_mul*data_size_msb is the size of the following frame MUST be <16384 pts_delta[frame_code] (s) MUST be <16384 and >-16384 if FLAG_CODED_PTS is set in the flags of the current frame then this value MUST be ignored, if FLAG_CODED_PTS is not set then pts_delta is the difference between the current pts and last_pts reserved_count[frame_code] (v) MUST be <256 data_size the size of the following frame data_size= data_size_lsb + data_size_msb*data_size_mul; coded_pts (v) if coded_pts < (1< pts=0 frame lsb_pts=3 -> pts=3 frame lsb_pts=1 -> pts=1 frame lsb_pts=2 -> pts=2 ... keyframe msb_pts=257 -> pts=257 frame lsb_pts=255 -> pts=255 frame lsb_pts=0 -> pts=256 frame lsb_pts=4 -> pts=260 frame lsb_pts=2 -> pts=258 frame lsb_pts=3 -> pts=259 all pts's of keyframes of a single stream MUST be monotone dts decoding timestamp the dts of a frame is the timestamp of the first sample which is output by a decoder when it is fed with the frame, note that the data output is not necessarily what is coded in the frame, but may be data from previous frames dts is calculated by using a decode_delay+1 sized buffer for each stream, into which the current pts is inserted and the element with the smallest value is removed, this is then the current dts this buffer is initialized with decode_delay -1 elements Pts of all frames in all streams MUST be bigger or equal to dts of all previous frames in all streams, compared in common timebase. (EOR frames are NOT exempt from this rule) Dts of all frames MUST be bigger or equal to dts of all previous frames in the same stream width (v) / height (v) width and height of the video in pixels MUST be set to the coded width/height, MUST NOT be 0 sample_width (v) /sample_height (v) (aspect ratio) sample_width is the horizontal distance between samples sample_width and sample_height MUST be relatively prime if not zero both MUST be 0 if unknown otherwise both MUST be non zero colorspace_type (v) 0 unknown 1 ITU Rec 624 / ITU Rec 601 Y range: 16..235 Cb/Cr range: 16..240 2 ITU Rec 709 Y range: 16..235 Cb/Cr range: 16..240 17 ITU Rec 624 / ITU Rec 601 Y range: 0..255 Cb/Cr range: 0..255 18 ITU Rec 709 Y range: 0..255 Cb/Cr range: 0..255 samplerate_nom (v) / samplerate_denom (v) = samplerate the number of samples per second, MUST NOT be 0 crc32 checksum Generator polynomial is 0x104C11DB7. Starting value is zero. checksum (u(32)) crc32 checksum checksum is calculated for the area pointed to by forward_ptr not including the checksum itself (from first byte after the packet_header until last byte before the checksum). for frame headers the checksum contains the framecode byte and all following bytes up to the checksum itself header_checksum (u(32)) checksum over the startcode and forward pointer Syncpoint tags: --------------- back_ptr_div16 (v) back_ptr = back_ptr_div16 * 16 + 15 back_ptr must point to a position within 16 bytes of a syncpoint startcode. This syncpoint MUST be the closest syncpoint such that at least one keyframe with a pts lower or equal to the original syncpoint's global_key_pts for all streams lies between it and the current syncpoint. A stream where EOR is set is to be ignored for back_ptr. global_key_pts (t) After a syncpoint, last_pts of each stream is to be set to: last_pts[i] = convert_ts(global_key_pts, time_base[id], time_base[i]) global_key_pts MUST be bigger or equal to dts of all past frames across all streams, and smaller or equal to pts of all future frames. Index tags: ----------- max_pts (t) The highest pts in the entire file syncpoints (v) number of indexed syncpoints syncpoint_pos_div16 (v) offset from beginning of file to up to 15 bytes before the syncpoint referred to in this index entry. Relative to position of last syncpoint. has_keyframe indicates whether this stream has a keyframe between this syncpoint and the last syncpoint. keyframe_pts The pts of the first keyframe for this stream in the region between the 2 syncpoints, in the stream's timebase. (EOR frames are also keyframes) eor_pts Coded only if EOR is set at the position of the syncpoint. The pts of that EOR. EOR is unset by the first keyframe after it. index_ptr (u(64)) Length in bytes of the entire index, from the first byte of the startcode until the last byte of the checksum. Note: A demuxer can use this to find the index when it is written at EOF, as index_ptr will always be 12 bytes before the end of file if there is an index at all. Info tags: ---------- stream_id_plus1 (v) Stream this info packet applies to. If zero, packet applies to whole file. chapter_id (s) Id of chapter this packet applies to. If zero, packet applies to whole file. Positive chapter_id's are real chapters and MUST NOT overlap. Negative chapter_id indicate a sub region of file and not a real chapter. chapter_id MUST be unique to the region it represents. chapter_id n MUST not be used unless there are at least n chapters in the file chapter_start (t) timestamp of start of chapter chapter_len (v) Length of chapter in same timebase of chapter_start. count (v) number of name/value pairs in this info packet type for example: "UTF8" -> string or "JPEG" -> JPEG image "v" -> unsigned integer "s" -> signed integer "r" -> rational Note: nonstandard fields should be prefixed by "X-" Note: MUST be less than 6 byte long (might be increased to 64 later) info packet types the name of the info entry, valid names are "Author" "Description" "Copyright" "Encoder" the name & version of the software used for encoding "Title" "Cover" (allowed types are "PNG" and "JPEG") image of the (CD, DVD, VHS, ..) cover (preferably PNG or JPEG) "Source" "DVD", "VCD", "CD", "MD", "FM radio", "VHS", "TV", "LD" Optional: appended PAL, NTSC, SECAM, ... in parentheses "SourceContainer" "nut", "mkv", "mov", "avi", "ogg", "rm", "mpeg-ps", "mpeg-ts", "raw" "SourceCodecTag" the source codec id like a fourcc which was used to store a specific stream in its SourceContainer "CaptureDevice" "BT878", "BT848", "webcam", ... (more exact names are fine too) "CreationTime" "2003-01-20 20:13:15Z", ... (ISO 8601 format, see http://www.cl.cam.ac.uk/~mgk25/iso-time.html) Note: do not forget the timezone "Keywords" "Language" ISO 639 and ISO 3166 for language/country code something like "eng" (US English), can be 0 if unknown and "multi" if several languages see http://www.loc.gov/standards/iso639-2/englangn.html and http://www.din.de/gremien/nas/nabd/iso3166ma/codlstp1/en_listp1.html the language code "Disposition" "original", "dub" (translated), "comment", "lyrics", "karaoke" Note: if someone needs some others, please tell us about them, so we can add them to the official standard (if they are sane) Note: nonstandard fields should be prefixed by "X-" Note: names of fields SHOULD be in English if a word with the same meaning exists in English Note: MUST be less than 64 bytes long value value of this name/type pair stuffing 0x80 can be placed in front of any type v entry for stuffing purposes except the forward_ptr and all fields in the frame header where a maximum of 8 stuffing bytes per field are allowed Structure: ---------- the headers MUST be in exactly the following order (to simplify demuxer design) main header stream_header (id=0) stream_header (id=1) ... stream_header (id=n) headers may be repeated, but if they are, then they MUST all be repeated together and repeated headers MUST be identical Each set of repeated headers not at the beginning or end of the file SHOULD be stored at the earliest possible position after 2^x where x is an integer and the file end, so the headers may be repeated at 4102 if that is the closest position after 2^12=4096 at which the headers can be placed Note: this allows an implementation reading the file to locate backup headers in O(log filesize) time as opposed to O(filesize) headers MUST be placed at least at the start of the file and immediately before the index or at the file end if there is no index headers MUST be repeated at least twice (so they exist three times in a file) there MUST be a sync point immediately before the first frame after any headers Index: ------ Note: with realtime streaming, there is no end, so no index there either Index MAY only be repeated after main headers. If an index is written anywhere in the file, it MUST be written at end of file as well. Info: ----- If a info packet is stored anywhere then a muxer MUST also store an identical info packet after every main-stream-header set If a demuxer has seen several info packets with the same chapter_id and stream_id then it MUST ignore all but the one with the highest position in the file demuxers SHOULD NOT search the whole file for info packets demuxer (non-normative): ------------------------ in the absence of a valid header at the beginning, players SHOULD search for backup headers starting at offset 2^x; for each x players SHOULD end their search at a particular offset when any startcode is found (including syncpoint) Seeking without an index (non-normative): ----------------------------------------- A. backward seeking 1. perform a binary search on the syncpoint timestamps finding the one which is largest and <= the target timestamp B. forward seeking 1a. perform a binary search on the syncpoint timestamps finding the one which is smallest and >= the target timestamp 1b. perform a binary search on the syncpoint back pointers finding the smallest one which has a back ptr >= the position of what was found in 1. 2. follow the back pointer to the corresponding syncpoint Seeking with an index (non-normative): -------------------------------------- The demuxer only has to find the appropriate keyframe in the index and start demuxing from the previous syncpoint Note, more complicated seeking methods exist which are capable of quickly seeking to the optimal point in the presence of an index even if only a subset of all streams is active A muxer SHOULD place syncpoints so that that simple low complexity seeking works with fine granularity, that is syncpoints should be placed prior to keyframes instead of non keyframes and with high enough frequency (once per second unless there are no keyframes between this and the previous syncpoint) encoders SHOULD place keyframes so that the number of points where all streams have a keyframe at the same time is maximized, this ensures that seeking (complicated or not) does not need to demux and decode significant amounts of data to reach a point where a presentable frame for each stream is available after seeking Semantic requirements: ====================== If more than one stream of a given stream class is present, each one SHOULD have info tags specifying disposition, and if applicable, language. It often highly improves usability and is therefore strongly encouraged. A demuxer MUST NOT demux a stream which contains more than one stream, or which is wrapped in a structure to facilitate more than one stream or otherwise duplicate the role of a container. any such file is to be considered invalid. for example vorbis in ogg in nut is invalid, as is mpegvideo+mpegaudio in mpeg-ps/ts in nut or dvvideo + dvaudio in dv in nut Sample code (Public Domain, & untested): ======================================== typedef BufferContext{ uint8_t *buf; uint8_t *buf_ptr; }BufferContext; static inline uint64_t get_bytes(BufferContext *bc, int count){ uint64_t val=0; assert(count>0 && count<9); for(i=0; ibuf_ptr++); } return val; } static inline void put_bytes(BufferContext *bc, int count, uint64_t val){ uint64_t val=0; assert(count>0 && count<9); for(i=count-1; i>=0; i--){ *(bc->buf_ptr++)= val >> (8*i); } return val; } static inline uint64_t get_v(BufferContext *bc){ uint64_t val= 0; for(; space_left(bc) > 0; ){ int tmp= *(bc->buf_ptr++); if(tmp&0x80) val= (val<<7) + tmp - 0x80; else return (val<<7) + tmp; } return -1; } static inline int put_v(BufferContext *bc, uint64_t val){ int i; if(space_left(bc) < 9) return -1; val &= 0x7FFFFFFFFFFFFFFFULL; // FIXME can only encode up to 63 bits currently for(i=7; ; i+=7){ if(val>>i == 0) break; } for(i-=7; i>0; i-=7){ *(bc->buf_ptr++)= 0x80 | (val>>i); } *(bc->buf_ptr++)= val&0x7F; return 0; } static int64_t get_dts(int64_t pts, int64_t *pts_cache, int delay, int reset){ if(reset) memset(pts_cache, -1, delay*sizeof(int64_t)); while(delay--){ int64_t t= pts_cache[delay]; if(t < pts){ pts_cache[delay]= pts; pts= t; } } return pts; } Authors: ======== Folks from the MPlayer developers mailing list (http://www.mplayerhq.hu/). Authors in alphabetical order: (FIXME! Tell us if we left you out) Beregszaszi, Alex (alex@fsn.hu) Bunkus, Moritz (moritz@bunkus.org) Diedrich, Tobias (ranma+mplayer@tdiedrich.de) Felker, Rich (dalias@aerifal.cx) Franz, Fabian (FabianFranz@gmx.de) Gereoffy, Arpad (arpi@thot.banki.hu) Hess, Andreas (jaska@gmx.net) Niedermayer, Michael (michaelni@gmx.at) Shimon, Oded (ods15@ods15.dyndns.org)