Compression Levels and Applications

6.7 Compression Levels and Applications

The MPEG-2 compression system is certainly very efficient and offers excellent results. But it is perhaps even more important to stress that it is inherently a very flexible system offering a number of different combinations resulting in different bit rates and stream structures. The same set of basic techniques can be used for a very wide gamut of applications from video conferencing, through digital broadcast, to the most demanding television production applications. MPEG-2 is therefore systematized in a number of profiles and levels. The profiles represent a given subset, or tool kit, of the complete MPEG-2 repertory, while the levels define constraints influencing the resolution of the picture. In principle, a decoder designed for a given level should also be capable of decoding the lower levels.

In addition to high, main, and low levels and profiles, the MPEG standard defines a scaleable profile. The coder used for scaleable profiles produces two sepa- rate bitstreams. One is a regular MPEG stream, which can be decoded as such by a standard decoder ensuring a given quality level. The other is a special stream, which cannot be decoded individually but, if received by a special, more com- plex, and expensive decoder and added to the standard stream, will enhance the quality of the decoded signal by increasing its resolution or the signal-to-noise ratio.

The most widely used MPEG 2 option is the Main Profile@Main Level (MP@ML), since it is part of practically all television transmission applications and it constitutes the preferred source-coding method for current standard-definition digital television transmission standards (DVB∗ATSC∗∗). Its basic parameters are

• 720 pixels per line, up to 576 (626/50) or 480 (525/60) lines • 4:2:0 sampling before coding

• DCT coding

• adaptive frame and field motion prediction • I-, P-, and B-frames structure

• 15 Mbps maximum bit rate

The ATSC standards for HDTV are also based on MPEG-2 compression schemes but, considering the higher resolution that has to be handled, it was

∗_{DVB (Digital Video Broadcasting) is a standard developed in Europe but used worldwide for broad-}

casting digital television signals over terrestrial and satellite transmitters as well as for distribution over cable systems. It was recently enhanced to encompass transmission to handheld devices as well as transmission of HDTV signals.

∗∗_{ATSC (Advanced Television Systems Committee) is the name of a standard developed in the United}

States for broadcasting digital television signals over terrestrial transmitters. It is also the name of the body that codified it.

necessary to use the Main Profile@High Level (MP@HL) resulting in a bit rate of about 19 Mbps.

The intrinsic qualities of MPEG-2, its flexibility, and the quality of decoded pictures prompted broadcasting professionals to envisage its usage throughout the broadcasting chain—from acquisition through postproduction to transmis- sion. However, as stated above, with its 15 Mbps, MP@ML could not respond to the most important requirements of broadcasting production. To overcome these limitations and at the same time make it possible to use MPEG-2 through- out the whole broadcasting chain, a new profile was proposed and eventually standardized—the 4:2:2 Profile@Main Level (422P@ML), also called professional, or studio profile. Its main parameters are defined as

• 720 pixels per line, by 608 (625) or 512 (525) lines • 4:2:2 or 4:2:0 sampling before coding

• DCT coding

• adaptive frame and field motion prediction • I-, B-, and P-frames allowed

• 50 Mbps maximum coding rate

Although all types of frames (I, B, and P) are legal in the 422P@ML it should be stressed that due to editing and switching requirements only two versions of that standard are generally encountered: one with GOP of only one frame (obviously an I-frame) and the other with two frames (I and B).

It is important to stress that the MPEG coding schemes were initially devel- oped for transmission applications. Consequently, regardless of the level and profile used, the MPEG encoders are considerably more complex and more expen- sive than the corresponding decoders, which is an excellent solution for the point-to-multipoint transmissions in which a limited number of encoders can send signals to innumerable decoders. In television production applications, however, this particular feature of MPEG does not offer any advantage whatsoever.

6.8 DV Compression

The main incentive to develop the DV compression scheme came from the area of consumer video recording and editing. A consortium of 10 mainly Japanese companies agreed to develop a common standard for such applications, thus creating a level playing field for all manufacturers and offering to users the long- awaited full compatibility of all the competing products on the market. During the development process a number of other companies joined the consortium thus ensuring from the onset very large support for the standard.

6.8 DV Compression 89

That newly developed recording standard, named originally “DVC” (digital videocassette, shortened later to “DV”), also included a compression scheme. Basically it was an intraframe compression based on chrominance subsampling and DCT coding, resulting in a constant bit rate of 25 Mbps, which proved to be extensible to 50 Mbps and over. At the same time, since it was developed for videotape-recording applications, the standard responded to the specific demands of that domain:

• the compression had to be limited to the rejection of spatial redundancy and entropy coding;

• the resulting bit rate had to be constant;

• each frame had to have the same number of bytes.

These essential characteristics make DV compression attractive for professional applications, and several contemporary broadcast-quality recording formats are based on it (DVCPRO 25, 50, and 100, and DVCAM).

The DV compression system is not radically different from MPEG-2 and consists of the following operations (see Figure 6.5):

• raster transformation (for 25 Mbps) • shuffling

• DCT coding • entropy coding

The first operation—the raster transformation—is a filtering procedure that consists of transforming a standard 4:2:2 signal to a 4:2:0 or a 4:1:1 subsampled one.

Entropy Quantization Buffer DCT

Scaling Analysis Quantizer control SDI IN Raster transf. 4:2:2-4:2:0 Intra-frame shuffling OUT

As in MPEG-2 compression, the picture is divided into a number of smaller blocks. These blocks are then shuffled to redistribute complex image information that is usually concentrated in several distinct parts of the picture over the entire picture area. As a result the compression process is never overloaded with an excessive amount of data, and the more and less complex blocks are compressed together, which can ensure a more consistent quality level after compression.

The next step is the same DCT coding described above in the explanation of the MPEG compression method, followed by the quantization process. However, the complexity of the video signal is not constant but varies depending on the content of each frame. If a static quantization table is used, the net result of the compression process will be a variable bit rate—higher for more complex pictures and lower for the simpler ones. Such a variable bit rate is acceptable if the compressed signal has to be stored on a DVD or a hard disk, but it is inadequate or even unacceptable if the signal has to be transmitted over a channel of a given bandwidth or to be recorded on a videotape.

For videotape recording it is necessary to ensure that the compressed signal to be recorded has a constant bit rate. To fulfill that demand it is necessary to act upon the quantization process and make it variable. Such a control can be achieved either through a feedback control system used in MPEG-2 or a feedforward

control system.

The MPEG-2 feedback control system (see Figure 6.6a) consists of compressing the incoming signal (through DCT coding, quantization and variable length coding), then to buffer the compressed signal in order to extract the informa- tion on the bitrate variations and feedback it to the quantizer, obtaining as a final result the desired constant bitrate.

The bit-rate control through a feedforward system used in the DV compression scheme is represented on Figure 6.6b. The incoming signal is DCT coded and the coding product is analyzed. The resulting information is forwarded to the quan- tizer to regulate dynamically the coding so that a constant bit rate is ensured already at the input of the variable length coder. This method requires a consid- erably higher processing power since it is based on the analysis of noncompressed signals in real time.

The last step, entropy coding, is again practically identical to the process described above in the Section 6.5.