SDI video uses no compression, it simply moves an astounding number of bits at a very high speed. For fun, you might want to challenge the matrix, Y, Pb, Pr system as a form of compression, but that is nit-picking.
You will find digital stuff specified in megaBITS and megaBYTES. There are 8 bits in a byte, and things that store and transfer whole bytes (parallel) use the BYTE unit. Many devices stream the data one bit at a time (serial). Serial streams require extra bits in each byte, like start bits, stop bits, and error-checking bits, for the bits to be reassembled into bytes, so the relationship is not simply 8 to 1. To make matters worse, the abbreviation "Mb" is used for "megaBITS", and "MB" is used for megaBYTES. That's a very subtle difference for two terms where one is about about ten times the size.
Digital video is usually quoted in serial megaBITS per second, and computer hard drives are specified in mega (and giga) BYTES. 25 megaBITS/second (the DVcam rate) will fill about 2.5 megaBYTES of hard drive every second.
SDI blows an amazing 270 million bits per second down the wire in order to deliver high-quality, non-degrading pictures and sound to the various pieces of equipment in a television plant. Devices like production switchers and framesyncs work internally in parallel words, but input and output SDI with no degradation or compromise. Tape recorders and disk-based video servers that can record and play pure SDI are rare, huge, and very expensive, due to the frightening amount of data generated by the SDI.
Most recorders and all of our digital broadcast methods need to use some form of mathematical compression to be practical. JPEG-style compression is at the root of most, if not all, compressed video.
|
|
|
JPEG compression reduces data by first dividing the image into 8 by 8 pixel blocks, then performing a mathematical fancy-trick called a "discrete cosine transform" (DCT), as seen in this formula:
... yeah, well maybe the discrete cosine transform is a little "out of the scope of this tutorial", as it is said. It is interesting to look at the blow-up of Mona's smile. This was a fairly low-quality JPEG image that looks horrible in close-up, but not too bad in the intended size.
This type of DCT based compression can be adjusted from mild to severe, with better images generating a higher bit-rate.
Decent-looking pictures can be made with bit-rates reduced up to a factor of ten, but we try to make sure that our pictures never go through the mathematical process more than once. Converting from one compressed format to another does cause a further loss in quality.
The DV formats (DVcam/ DVCpro) compress each video frame using the above techniques, adapting the amount of compression to keep the bits flowing at a rate of 25 megabits/second. As well, the Y:Pb:Pr ratio is 4:1:1 There is a form of DVCpro that runs at 50mb/second.
DV and similar compressed digital tape formats treat each frame as a complete picture. This makes editing and switching simple... a cut can be made from source to source at the end of any frame.
Is explained in the next section. The above forms of compression are "spatial"... they concern themselves with the stuff in one, still image. Stop and compare the differences in adjacent frames. Here is a reference clip. Now look at the differences between successive frames.
MPEG is an an acronym for the Motion Pictures Experts Group. If editing and switching the picture stream is not a priority, compression can be applied across many frames. MPEG sends one complete DCT compressed frame, then only updates the bits of a picture that have moved. Think of a still graphic, or a locked-off shot of someone talking.... why keep sending the background thirty times a second? Don't look to closely at the graphic above, it is not meant to make the intricacies of MPEG math suddenly become clear! MPEG uses three different types of frame:
I-Frames are complete, stand-alone pictures (Intra-frame).
P-frames are predicted from frames that came before.
B-frames are bi-directionally predicted from frames occurring before and after.
If your head hasn't caught fire yet, you may be wondering how B-frames can work. How can a picture be based on one that hasn't occurred yet? MPEG coders and decoders store many frames at once (on the fly!) to do the comparisons. There is a significant delay of up to several seconds through an MPEG coder.
MPEG uses Groups Of Pictures, or GOP's (pronounced "gop"... go ahead, it's fun). Each GOP starts with a stand-alone I-Frame, and is followed by as many B and P frames as possible. A common GOP is 18 frames long. A scene-change in the video forces a new GOP to begin even if the sequence isn't complete.
MPEG is difficult (maybe impossible) to edit and switch, because of the long sequence of pictures. Edits can only be made right before an I-Frame, or a bunch of random "chicklets" (DCT pixel-blocks) occurs until the next sequence can be built. One exception is BetaSX, an MPEG recording format. BetaSX uses a two frame GOP I-B-I-B-I-B sequence. Lots of I-Frames = lots of places to make an edit.
HOME | BACK | HOW IT WORKS | TIMING | RECORDING | EDITING & TIMECODE | AUDIO | COMPUTERS
