Serial digital interface (SDI) refers to a family of video interfaces standardized by SMPTE. For example, ITU-R BT.656 and SMPTE 259M define digital video interfaces used for broadcast-grade video. A related standard, known as high-definition serial digital interface (HD-SDI), is standardized in SMPTE 292M; this provides a nominal data rate of 1.485 Gbit/s.
An emerging interface, commonly known in the industry as dual link HD-SDI and consisting essentially of a pair of SMPTE 292M links, is standardized in SMPTE 372M; this provides a nominal 2.970 Gbit/s interface used in applications (such as digital cinema or HDTV 1080P) that require greater fidelity and resolution than standard HDTV can provide. A more recent interface, 3G-SDI, consisting of a single 2.970 Gbit/s serial link, is standardized in SMPTE 424M that will replace the dual link HD-SDI.
These standards are used for transmission of uncompressed, unencrypted digital video signals (optionally including embedded Audio and/or Time Code) within television facilities; they can also be used for packetized data. They are designed for operation over short distances (less than 300m. with coaxial cable); due to their high bitrates they are inappropriate for long-distance transmission. SDI and HD-SDI are currently only available in professional video equipment; various licensing agreements, restricting the use of unencrypted digital interfaces to professional equipment, prohibit their use in consumer equipment. There are various mod kits for existing DVD players and other devices, which allow a user to add a serial digital interface to these devices.
The various serial digital interface standards all use (one or more) coaxial cables with BNC connectors, with a nominal impedance of 75 ohms. This is the same type of cable used in analog video setups, which potentially makes for easier upgrades (though higher quality cables may be necessary for long runs at the higher bitrates). The specified signal amplitude at the source is 800 mV (±10%) peak-to-peak; far lower voltages may be measured at the receiver owing to attenuation. Using equalisation at the receiver, it is possible to send 270 Mbit/s SDI over 300 metres without use of repeaters, but shorter lengths are preferred. The HD bitrates have a shorter maximum run length, typically 100 meters.
Uncompressed digital component signals are transmitted. Data is encoded in NRZI format, and a linear feedback shift register is used to scramble the data to reduce the likelihood that long strings of zeroes or ones will be present on the interface. The interface is self-synchronizing and self-clocking. Framing is done by detection of a special synchronization pattern, which appears on the (unscrambled) serial digital signal to be a sequence of ten ones followed by twenty zeroes (twenty ones followed by forty zeroes in HD); this bit pattern is not legal anywhere else within the data payload.
|Standard||Name||Bitrates||Example Video Formats|
|SMPTE 259M||SD-SDI||270 Mbit/s, 360 Mbit/s, 143 Mbit/s, and 177 Mbit/s||480i, 576i|
|SMPTE 344M||540 Mbit/s||480p, 576p|
|SMPTE 292M||HD-SDI||1.485 Gbit/s, and 1.485/1.001 Gbit/s||720p, 1080i|
|SMPTE 372M||Dual Link HD-SDI||2.970 Gbit/s, and 2.970/1.001 Gbit/s||1080p|
|SMPTE 424M||3G-SDI||2.970 Gbit/s, and 2.970/1.001 Gbit/s||1080p|
Several bit rates are used in serial digital video:
SMPTE 292M defines an optical interface as well as an electrical one; this interface is widely considered to be obsolete. An 8-bit parallel digital interface is defined by ITU-R Rec. 601; this is also obsolete (however, many clauses in the various standards accommodate the possibility of an 8-bit interface).
In SD and ED applications, the parallel data format is defined to 10 bits wide, whereas in HD applications, it is 20 bits wide, divided into two parallel 10-bit datastreams (known as Y and C). The SD datastream is arranged like this:
whereas the HD datastreams are arranged like this:
For all serial digital interfaces (excluding the obsolete composite encodings), the native color encoding is 4:2:2 YCbCr format. The luminance channel (Y) is encoded at full bandwidth (13.5 MHz in 270 Mbit/s SD, ~75 MHz in HD), and the two chrominance channels (Cb and Cr) are subsampled horizontally, and encoded at half bandwidth (6.75 MHz or 37.5 MHz). The Y, Cr, and Cb samples are co-sited (acquired at the same instance in time), and the Y' sample is acquired at the time halfway between two adjacent Y samples.
In the above, Y refers to luminance samples, and C to chrominance samples. Cr and Cb further refer to the red and blue "color difference" channels; see Component Video for more information. This section only discusses the native color encoding of SDI; other color encodings are possible by treating the interface as a generic 10-bit data channel. The use of other colorimetry encodings, and the conversion to and from RGB colorspace, is discussed below.
Video payload (as well as ancillary data payload) may use any 10-bit word in the range 4 to 1019 (004 to 3FB in hexadecimal) inclusive; the values 0-3 and 1020-1023 (3FC - 3FF) are reserved and may not appear anywhere in the payload. These reserved words have two purposes; they are used both for #synchronization packets and for #ancillary data headers.
A synchronization packet (commonly known as the timing reference signal or TRS) occurs immediately before the first active sample on every line, and immediately after the last active sample (and before the start of the horizontal blanking region). The synchronization packet consists of four 10-bit words, the first three words are always the same--0x3FF, 0, 0; the fourth consists of 3 flag bits, along with an error correcting code. As a result, there are 8 different synchronization packets possible.
In the HD-SDI and dual link interfaces, synchronization packets must occur simultaneously in both the Y and C datastreams. (Some delay between the two cables in a dual link interface is permissible; equipment which supports dual link is expected to buffer the leading link in order to allow the other link to catch up). In SD-SDI and enhanced definition interfaces, there is only one datastream, and thus only one synchronization packet at a time. Other than the issue of how many packets appear, their format is the same in all versions of the serial-digital interface.
The flags bits found in the fourth word (commonly known as the XYZ word) are known as H, F, and V. The H bit indicates the start of horizontal blank; and synchronization bits immediately preceding the horizontal blanking region must have H set to one. Such packets are commonly referred to as End of Active Video, or EAV packets. Likewise, the packet appearing immediately before the start of the active video has H set to 0; this is the Start of Active Video or SAV packet.
Likewise, the V bit is used to indicate the start of the vertical blanking region; an EAV packet with V=1 indicates the following line (lines are deemed to start at EAV) is part of the vertical interval, an EAV packet with V=0 indicates the following line is part of the active picture.
The F bit is used in interlaced and segmented-frame formats to indicate whether the line comes from the first or second field (or segment). In progressive scan formats, the F bit is always set to zero.
In the high definition serial digital interface (and in dual-link HD), additional check words are provided to increase the robustness of the interface. In these formats, the four samples immediately following the EAV packets (but not the SAV packets) contain a cyclic redundancy check field, and a line count indicator. The CRC field provides a CRC of the preceding line (CRCs are computed independently for the Y and C streams), and can be used to detect bit errors in the interface. The line count field indicates the line number of the current line.
The CRC and line counts are not provided in the SD and ED interfaces. Instead, a special ancillary data packet known as an EDH packet may be optionally used to provide a CRC check on the data.
Each sample within a given datastream is assigned a unique line and sample number. In all formats, the first sample immediately following the SAV packet is assigned sample number 0; the next sample is sample 1; all the way up to the XYZ word in the following SAV packet. In SD interfaces, where there is only one datastream, the 0th sample is a Cb sample; the 1st sample a Y sample, the 2nd sample a Cr sample, and the third sample is the Y' sample; the pattern repeats from there. In HD interfaces, each datastream has its own sample numbering--so the 0th sample of the Y datastream is the Y sample, the next sample the Y' sample, etc. Likewise, the first sample in the C datastream is Cb, followed by Cr, followed by Cb again.
Lines are numbered sequentially, starting from 1, up to the number of lines per frame of the indicated format (typically 525, 625, 750, or 1125). Determination of line 1 is somewhat arbitrary; however it is unambiguously specified by the relevant standards. In 525-line systems, the first line of vertical blank is line 1, whereas in other interlaced systems (625 and 1125-line), the first line after the F bit transitions to zero is line 1.
Note that lines are deemed to start at EAV, whereas sample zero is the sample following SAV. This produces the somewhat confusing result that the first sample in a given line of 1080i video is sample number 1920 (the first EAV sample in that format), and the line ends at the following sample 1919 (the last active sample in that format). Note that this behavior differs somewhat from analog video interfaces, where the line transition is deemed to occur at the sync pulse, which occurs roughly halfway through the horizontal blanking region.
Link numbering is only an issue in dual-link interfaces. The first link (the primary) link, is assigned a link number of 1, subsequent links are assigned increasing link numbers; so the second (secondary) link in a dual-link system is link 2. The link number of a given interface is indicated by a VPID packet located in the vertical ancillary data space.
Note that the data layout in dual link is designed so that the primary link can be fed into a single-link interface, and still produce usable (though somewhat degraded) video. The secondary link generally contains things like additional LSBs (in 12-bit formats), non-cosited samples in 4:4:4 sampled video (so that the primary link is still valid 4:2:2), and alpha or data channels. If the second link of a 1080P dual link configuration is absent, the first link still contains a valid 1080i signal.
In the case of 1080p60, 59.94, or 50 Hz video over a dual link; each link contains a valid 1080i signal at the same field rate. The first link contains the 1st, 3rd, and 5th lines of odd fields and the 2nd, 4th, 6th, etc. lines of even fields, and the second link contains the even lines on the odd fields, and the odd lines on the even fields. When the two links are combined, the result is a progressive-scan picture at the higher frame rate.
Like SMPTE 259M, SMPTE 292M supports the SMPTE 291M standard for ancillary data. Ancillary data is provided as a standardized transport for non-video payload within a serial digital signal; it is used for things such as embedded audio, closed captions, timecode, and other sorts of metadata. Ancillary data is indicated by a 3-word packet consisting of 0, 3FF, 3FF (the opposite of the synchronization packet header), followed by a two-word identification code, a data count word (indicating 0 - 255 words of payload), the actual payload, and a one-word checksum. Other than in their use in the header, the codes prohibited to video payload are also prohibited to ancillary data payload.
Specific applications of ancillary data include embedded audio, EDH, VPID and SDTI.
In dual link applications; ancillary data is mostly found on the primary link; the secondary link is to be used for ancillary data only if there is no room on the primary link. One exception to this rule is the VPID packet; both links must have a valid VPID packet present.
Both the HD and SD serial interfaces provide for 16 channels of embedded audio. The two interfaces use different audio encapsulation methods--SD uses the SMPTE 272M standard, whereas HD uses the SMPTE 299M standard. In either case, an SDI signal may contain up to sixteen audio channels (8 pairs) embedded 48 kHz, 24-bit audio channels along with the video. Typically, 48 kHz, 24-bit (20-bit in SD) PCM audio is stored, in a manner directly compatible with the AES3 digital audio interface. These are placed in the (horizontal) blanking periods, when the SDI signal carries nothing useful, since the receiver generates its own blanking signals from the TRS.
In dual-link applications, 32 channels of audio are available, as each link may carry 16 channels.
As the standard definition interface carries no checksum, CRC, or other data integrity check, an EDH (Error Detection and Handling) packet may be optionally placed in the vertical interval of the video signal. This packet includes CRC values for both the active picture, and the entire field (excluding those lines at which switching may occur, and which should contain no useful data); equipment can compute their own CRC and compare it with the received CRC in order to detect errors.
EDH is typically only used with the standard definition interface; the presence of CRC words in the HD interface make EDH packets unnecessary.
VPID (or video payload identifier) packets are increasingly used to describe the video format. In early versions of the serial digital interface, it was always possible to uniquely determine the video format by counting the number of lines and samples between H and V transitions in the TRS. With the introduction of dual link interfaces, and segmented-frame standards, this is no longer possible; thus the VPID standard (defined by SMPTE 352M) provides a way to uniquely and unambiguously identify the format of the video payload.
The active portion of the video signal is defined to be those samples which follow an SAV packet, and precede the next EAV packet; where the corresponding EAV and SAV packets have the V bit set to zero. It is in the active portion that the actual image information is stored.
Several color encodings are possible in the serial digital interface. The default (and most common case) is 10-bit linearly sampled video data encoded as 4:2:2 YCbCr. (YCbCr is a digital representation of the YPbPr colorspace). Samples of video are stored as described above. Data words correspond to signal levels of the respective video components, as follows:
Note that the scaling of the luma and chroma channels is not identical. The minimum and maximum of these ranges represent the preferred signal limits, though the video payload may venture outside these ranges (providing that the reserved code words of 0 - 3 and 1020 - 1023 are never used for video payload). In addition, the corresponding analog signal may have excursions further outside of this range.
The dual-link and 3 Gbit/s interfaces additionally support other color encodings besides 4:2:2 YCbCr, namely:
If an RGB encoding is used, the three primaries are all encoded in the same fashion as the Y channel; a value of 64 (40 hex) corresponds to 0mV, and 940 (3AC hex) corresponds to 700mV
12-bit applications are scaled in a similar fashion to their 10-bit counterparts; the additional two bits are considered to be LSBs.
For portions of the vertical and horizontal blanking regions which are not used for ancillary data, it is recommended that the luma samples be assigned the code word 64 (40 hex), and the chroma samples be assigned 512 (200 hex); both of which correspond to 0 mV. It is permissible to encode analog vertical interval information (such as vertical interval timecode or vertical interval test signals) without breaking the interface, but such usage is nonstandard (and ancillary data is the preferred means for transmitting metadata). Conversion of analog sync and burst signals into digital, however, is not recommended--and neither is necessary in the digital interface.
The various versions of the serial digital interface support numerous video formats.
In addition to the regular serial digital interface described here, there are several other similar interfaces which are similar to, or are contained within, a serial digital interface.
There is an expanded specification called SDTI (Serial Data Transport Interface), which allows compressed (i.e. DV, MPEG and others) video streams to be transported over an SDI line. This allows for multiple video streams in one cable or faster-than-realtime (2x, 4x,...) video transmission. A related standard, known as HD-SDTI, provides similar capability over an SMPTE 292M interface.
The SDTI interface is specified by SMPTE 305M. The HD-SDTI interface is specified by SMPTE 348M.
The standard SMPTE 349M: Transport of Alternate Source Image Formats through SMPTE 292M, specifies a means to encapsulate non-standard and lower-bitrate video formats within an HD-SDI interface. This standard allows, for example, several independent standard definition video signals to be multiplexed onto an HD-SDI interface, and transmitted down one wire. This standard doesn't merely adjust EAV and SAV timing to meet the requirements of the lower-bitrate formats; instead, it provides a means by which an entire SDI format (including synchronization words, ancillary data, and video payload) can be encapsulated and transmitted as ordinary data payload within a 292M stream.
The G.703 standard is another high-speed digital interface, originally designed for telephony.