High-definition encoders: They're not created equal
Aldon Caron
10/30/1999 World Broadcast News
Copyright 1999 by Intertec Publishing Corporation, a PRIMEDIA Company. All rights reserved.
Digital television (DTV) is the driving force toward the emergence of high-definition (HD) TV broadcasting around the world. HD encoders are the latest technological advancements in the family of MPEG -2 products that provide manufacturers with the challenge of developing products to meet the higher resolution requirements. While interim solutions that meet the needs of system field trials and early HD broadcasts have been available, the longevity of these first-generation encoders is yet to be determined.
The next generation of HD encoders allows greater compression efficiency, improved video quality and enhanced flexibility with an eye toward the future. Broadcasters have to choose between the older systems that meet the basic requirements of today and the new systems that provide an upgrade path for the future.
HDTV & MPEG encoders Today's faster encoder engines can handle about 270Mb of input video data per second. While this is adequate for encoding D1 video with just one encoder engine, HDTV requires up to 1.5Gb of input video to be processed per second. The solution is to use more than one compression engine in the HD encoder, as was the case when the first generation of encoder silicon was produced several years ago for standard-definition (SD) encoding.
The maximum bitrate is the major determining factor in final picture quality. In the U.S., the Federal Communications Commission (FCC) has set the maximum bandwidth for HDTV broadcasting to 19.4Mb/s of compressed data for terrestrial use. The Advanced Television Systems Committee (ATSC) has issued various digital video formats, per Standard A/53, clarifying the specific parameters of the digital video signal.
Freedom to choose The question of adaptability is critical when discussing HD video formats. Future formats could change; the FCC did not lock the U.S. TV broadcasters into a specific format when they issued their DTV mandate. The FCC did not make Table 3 of the Standard A/53 mandatory, although many believe it to be the case. Basically, the FCC said, "Let the market select the best format." In short, the issue of formats is far from settled, especially with non-traditional broadcasters choosing non-Table 3 high-definition formats for their satellite-delivered services.
At this time, there are several HDTV frame formats. The most commonly discussed are the 1080i (interlaced) and 720p (progressive) formats listed by ATSC in Table 3, with other future proposals under discussion. Next, there are multiple selections of the frame rates. The ATSC list specifies 23.976Hz, 24Hz, 29.97Hz, 30Hz, 59.94Hz and 60Hz. While it is fairly clear that the 1080i and 720p HD formats will play predominate roles, it is unclear what other formats may come to the forefront. Additionally, Table 3 lists 480i and 480p SDTV frame formats, with a similar variety of frame rates.
Economic factors will play a key role in the selection of which formats survive and which fade into memory for those "good-old-day" discussions. These factors are yet to be played out. In the interim, broadcasters all over the world are expected to spend millions of dollars equipping new DTV stations and facilities.
To obtain high-quality encoded video, an HD encoder must be designed to use the available bandwidth efficiently. An overall systematic approach with appropriate encoding sophistication, instead of a brute force approach, becomes necessary for superior picture and audio quality.
MPEG : Video to digital bits MPEG is the definition of the compression data bitstream to be delivered to the decoder along with a decoder reference model. The standard defines exactly how to decode a signal. It is intently neutral on how bitstream encoding is to be accomplished or how to implement a decoder. This neutrality continues to free the designers of MPEG encoders to innovate as required, unencumbered by outdated technology. This has lead to a dramatic evolution of MPEG encoders with increased video quality and features. The evolution can be characterized as progressions from a brute force method to a more highly refined compression finesse.
In the early days of MPEG , the prototypes were large boxes filled with hand-built circuit boards. Big and cumbersome, these prototypes were little more than proofs of concept. However, they successfully simulated the overall concept and provided the initial development of special custom silicon circuits that drove down costs to acceptable levels. Each generation was faster and more capable than the previous, with the number of compression chips needed to encode D1-quality video falling from 14 to only one.
Quality first, efficiency next Early encoders were manufactured with a focus on picture quality. Using faster processors to handle more bits per second enabled encoders to make the video quality as lifelike as possible. Once the high picture quality was achieved, the emphasis shifted to bit-rate efficiency, reducing the number of bits per second required to obtain a given quality level through refinements in techniques for motion estimation and quantization.
MPEG uses discrete cosine transform (DCT) compression techniques for spatial, or intra-frame, compression. However, the removal of temporal or inter-frame picture information provides the most significant bit-rate reduction. Through a technique known as motion estimation, an encoder sends informationabout picture changes and not the complete picture. This permits the decoder to create a new video frame by manipulating information it already has in its memory. Imagine a scene of a hot air balloon floating along a mountain. Once the decoder has a complete frame of that scene, it can recreate subsequent frames simply by moving the balloon over the original mountain image.
The MPEG process compresses a complete video frame into what is called an I-frame every half-second using DCT compression. This picture contains, within itself, all the necessary information for a decoder to reproduce that video frame. Typically, a DCT-compressed picture gives only about 10:1 compression. Because I-frames consist only of DCT-compressed video data, they are bit-intensive.
For the rest of the half-second, the encoder creates a datastream consisting mostly of motion vectors that instruct the decoder how to move the information within the I-frame to recreate the associated video frames. This is done with a minimum of additional DCT compressed data. Because motion vectors often give the equivalent of more than 100:1 compression, these frames, called P (predictive) and B (bi-directional) frames, are much less bit-intensive. When looking at the bitrate actually required by MPEG to encode video on a frame-by-frame basis, peaks of high activity are seen about every half second, with a much lower bitrate used in between.
The variability for the required bitrates for I-frames as compared with B-and P-frames was well known, but in the early years of MPEG development, there was little that could be done. The limitations of the technology required the MPEG designer to implement decoders and encoders that processed video at constant bitrates. This led to undesirable compromises. A bitrate that is too low causes noticeable artifacts, called macroblocks, in the video during high-action scenes. A higher overall bitrate reduces artifacts and improves the video image produced for high-action scenes. However, it is an uneconomical use of bits on low activity scenes. These limitations have been addressed more recently by the development of variable bit-rate (VBR) encoding. A VBR encoder adjusts the bitrate according to the requirements of the source video.
Force, not finesse In some ways, the development of HD encoders has led to a repetition of the development cycle of SD encoders. As shown in Figure 1 on page 52, many of the first-generation HD encoders simply divided the HD screen into large tiles and then encoded each tile individually.
The overall logic diagram for this class of HD encoders is shown in Figure 2 on page 54. The incoming video frame is divided much the same way it is for a video wall so that the individual incoming HDTV frame now consists of six separate SDTV frames, which then are encoded separately. Finally, the six separate streams are recombined into one compliant MPEG -2 stream.
Obviously, this is a brute force solution, but it is a perfectly compliant one. In retrospect, it was probably a necessary step, simply because SD encoder engines were the only available technology at the time this class of HD encoders was designed. However, these encoders are neither efficient in the use of bandwidth nor adaptive. Several issues surface when using the technique of dividing the HD picture into six tiles:
* Motion estimation allows greater efficiencies in compression by sending picture changes, rather than sending the entire new picture. When moving an object across multiple tiles, motion vector information is lost, and the object must be re-encoded as the new tile is entered. Returning to the example of a hot air balloon floating alongside a mountain, the image of the balloon must be downloaded in the MPEG stream as DCT compressed data whenever the balloon crosses a tile boundary. This requires more bandwidth utilization than a motion estimation technique.
* Because there usually is more action in one part of a frame, simply dividing the available bandwidth by six artificially constrains the available bitrate to the part with the most action, which leads to noticeable artifacts such as macro-blocking.
* Another approach is to provide statistical multiplexing to the output from the six encoders, each of which is operating as a VBR encoder. In theory, this reduces the number of bits required to encode a scene.
The next-generation HD encoder Fortunately, in the case of the MPEG encoder, there is a flexible alternative that has been designed to overcome the shortcomings of the earlier models. This alternative is the MediaView MV400. The MV400 is an HD MPEG -2 encoder in a 1-rack-unit-high chassis. It provides superior image processing techniques and supports multiple formats.
Designed to exceed broadcaster expectations, the MV400 provides a choice of 1080i, 720p and 480p high-definition and enhanced definition DTV formats supporting ATSC (American), DVB (European) and ISDB (Japanese) standards. In addition to supporting today's formats, software upgrades will enable the MV400's multiformat architecture to support future formats and to provide further performance enhancements.
DiviCom engineering has enhanced the way HD images are processed by using unique image division techniques. With its exclusive MotionTrack feature, the MV400 processes the entire video image uniformly, bypassing the inherent architectural limitations that existed in first-generation encoders.
The foundation for this capability is found in the MV400 hardware architecture. Unlike first-generation HD encoders that group together individual encoder systems, the MV400 integrates multiple C-Cube DVxpert II encoding engines on a single module linked via a high-speed bus. This architecture enables inter-encoder communication and increases the MV400's ability to handle demanding, high action, complex video content. The end result is better video quality with less effort.
Instead of tiling the input video, this new breed of encoder divides the incoming video into strips across the frame, as shown in Figure 3 on page 56, in a technique we call "image slicing."
While this might seem just a variation from the tile arrangement shown in Figure 1 on page 52, there is a profound difference. The earlier generation of HD encoders was akin to having a car with six separate 1-cylinder engines combined together in a common transmission, while the new generation is quite analogous to a multicylinder engine. The newer image slicing design is much easier to control and enhance. This is better illustrated in the block diagram of this class of HD encoder as shown in Figure 4 on page 56.
HD MPEG encoder checklist The latest advances in technology provide a choice to those who need high-definition encoders. Below is a list of questions to consider when selecting an HD encoder:
* Is the encoder a first-generation interim solution or a next-generation encoder that addresses the broadcasting needs of the future? * Does the encoder comply only with ATSC Table 3, or are additional HD formats that are in use today supported? * Will there be software upgrades to enable future performance enhancements and additional video formats? * Are advanced motion estimation and rate control techniques used to ensure that the entire video image is processed uniformly? * Is inverse telecine mode (removal of 3:2 pull-down) supported to conserve bandwidth on film-based material? * What are the physical and environmental characteristics? * Are Dolby Digital audio encoding and closed caption insertion integrated to reduce requirements for external devices? * Is the manufacturer dedicated to developing MPEG encoder solutions? * Is there an SNMP-based network management system to provide centralized control of the entire encoding system -- SD, HD and data systems? |
