Broadband service providers need an answer to satellite’s aggressive high definition (HD) rollout strategy. A key component to that answer is finding the available bandwidth to support a large rollout of HD services. For the capacity challenged operator, the inability to support a full channel lineup of bandwidth-heavy HD content in an increasingly competitive environment has set the stage for a subscriber retention crisis.
Consider an average cable operator running a state-of-the-art MPEG-2 digital simulcast plant. The operator has upgraded its facility to support 860 MHz and has allocated approximately 300 MHz (550 to 860) or 50 quadrature amplitude modulation (QAM) channels of that bandwidth to digital services. At 256-QAM, our operator can fit three HD services into a single QAM channel. So in order to match the forward-looking claims of 150 HD channels (see sidebar on page 78) of the satellite provider, our operator would have to devote its entire digital spectrum to HD (50 QAM channels x three HD services per QAM channel). For an operator committed to offering standard definition (SD) digital video services, music-choice packages, voice over Internet protocol (VoIP), video on demand (VOD) and high-speed data, this plan clearly will not fly.
Bandwidth-constrained cable service providers must determine how they will meet this growing consumer demand for HD before it reaches critical mass. Operators are currently exploring a host of options, including MPEG-4 compression, which will address this concern. Four options One option that has been around since the introduction of digital services is to reclaim analog bandwidth. Operators can deploy three channels of HD for every one channel of analog SD. The downside is that basic analog subscribers still make up a substantial part of the operators’ customer and revenue base, and they don’t like to have their channels taken away.
A second option and particularly "hot topic" is switched digital video (SDV). SDV enables our operator to take advantage of the old 80/20 rule by reassigning the bandwidth from programs that no one watches to programs that someone watches. However, as most HDTV set owners will attest, HD programs are not the programs that no one watches and are therefore less likely to give up their bandwidth in a SDV environment. A system with any number of HD channels is less likely to recognize enough of a bandwidth benefit from SDV to solve the problem completely.
A third and well-known option is node splitting. Node splitting will be an effective part of the bandwidth expansion plan especially when paired with the new class of dense edge QAM devices. By reducing the number of homes passed but maintaining the number of QAM channels per node, operators can free up substantial bandwidth, especially for switched services such as VOD and "long-tail" SDV content. For HD broadcast channels that are "always on," node splits have some of the same limitations as SDV in that the "on" programs are duplicated on either side of the node, reducing any bandwidth efficiency. As detailed in the March issue of Communications Technology, node splitting is an effective bandwidth management tool, but can be costly, complex and not a universal remedy for providing HD services in a bandwidth-constrained network.
A fourth approach that complements switched digital rollouts and all-digital conversion is utilizing more efficient compression standards, specifically MPEG-4. (See Figure 1.) MPEG-4 uses half the bandwidth of today’s MPEG-2 standard, yielding twice the number HD programs over the same network. That bandwidth efficiency has not gone unnoticed. DirecTV’s aggressive HD rollout plan is based on MPEG-4 technology, as is the new AT&T U-verse offering. Recently a number of cable operators have also started MPEG-4 trials. Deployed base Like all new technologies, MPEG-4 adoption presents its own set of challenges. For the cable operator, the key issue is the distribution of MPEG-4 content to a deployed base of MPEG-2 set-top boxes. Thanks to cable’s active deployment of interactive applications over set-tops, this challenge is minimized.
Interactivity is a pillar of the cable industry’s response to the satellite competitive threat. Not only does interactivity enable such applications as VOD or Time Warner Cable’s "StartOver," but it also provides the foundation on which an MPEG-4 deployment could be based. In a VOD or SDV environment, the set-top tells the headend which HD program it needs. If it also happens to mention that it’s one of the new boxes with HD MPEG-4 capability, the headend is free to send HD programming in MPEG-4.
This approach works perfectly for narrowcast applications like VOD, but unless all households downstream from the node are MPEG-4 enabled, SDV doesn’t provide any additional benefit, because popular or viewed programs would have to be simulcast in both MPEG-2 and MPEG-4. However, if HD penetration or demand is high enough downstream from a particular node, the operator can chose to upgrade all set-tops downstream from that particular node and then make use of the SDV infrastructure to roll out an expanded tier of MPEG-4 HD service one node at a time.
Deployment scenarios that follow this model will be primarily based on the demographic trends of the served neighborhood. Birds of a feather flock together, and those households with HDTV sets and the willingness to sign up for a premium and expanded tier of HD service are largely grouped together. By defining nodes by demographic as well as geographic profiles, cable operators can take advantage of these similar flocks.
The eventual market shift toward OpenCable Application Platform (OCAP) further supports and facilitates this approach. When consumers are able to purchase their own set top box/DVR combos from electronics retailers, those households with HDTV and more sophisticated home theater capabilities are the same households that are more likely to spend the additional dollars to purchase units with the advanced HD MPEG-4 feature sets and the longer DVR storage time that MPEG-4 provides. Transcoding But how does the content get turned into MPEG-4? Broadcasters transmit their HD content in MPEG-2 and have not yet recognized any strong incentive to upgrade their systems to MPEG-4. Converting the HD MPEG-2 programs into MPEG-4 requires the operator to transcode the video into the new advanced format. Operators’ recent experiences with transcoding show that video quality takes a hit when the HD content is transcoded to MPEG-4, suffering up to 1 dB of quality degradation, a significant and visible loss. The reason for this loss is that although the algorithms are similar, they’re not similar enough.
MPEG-4 encoding is more efficient on several levels. The MPEG-4 motion estimation has been expanded to include additional block shapes and sizes, and multiple reference frames can be used to predict a macroblock. A new intraframe spatial prediction mode has also been introduced that has no correspondence to the MPEG-2 coding modes. The discrete cosine transform (DCT) used in MPEG-2 has been replaced by a smaller integer transform, and VideoLAN client (VLC) entropy coding has been augmented with optional, context-adaptive binary arithmetic coding (CABAC). The MPEG-4 standard also allows a filter in the encoding loop that helps mitigate encoding artifacts, such as blocking at low encoding rates. All of these improvements allow MPEG-4 to maintain image quality at significantly lower bit rates, typically about half the rate of MPEG-2.
They both use block-based motion compensated prediction, quantized transform coding of residuals, and entropy coding, but MPEG-4 introduces basic differences in these tools along with additional modes of operation. These differences have led vendors to develop basic decode and re-encode solutions that are prone to suffer from what is referred to as "generational loss." (See Figure 2.) Video is encoded in frames, and not all frames are treated equally. Some frames (I-frames) are completely independent pictures with all the supporting data, while other frames (P and B frames) are differential updates from other reference frames. The I and P frames are encoded at a higher quality since they are used as reference frames. While this works well with one generation of coding, if a subsequent generation of encoding tries to make an MPEG-4 I-frame out of an MPEG-2 B frame, the results can be disappointing. For obvious reasons, one of the lowest hanging fruits for more efficient transcoding is to re-use the frame-types from the MPEG-2 content. MPEG-2 I-frames on the way in are converted to MPEG-4 I-frames on the way out.
Figure 3 is a quality measurement across roughly 2 seconds of video. The tall peaks in the chart relate to I-frames, the shorter peaks in the chart relate to P-frames, and the valleys in the chart relate to B-frames. For the most efficient transcoding operation, output MPEG-4 I-frames that need the highest quality must be derived from the input MPEG-2 I frames. Other aspects of the original MPEG-2 HD stream can also be reused. One of the key differences between MPEG-2 and MPEG-4 is the blocks on which motion estimation is based. MPEG-2 employs 16×16 (16×8 in the case of field) blocks for motion estimation, while in MPEG-4 various block shapes are used. There is no guarantee that the MPEG-4 encoders will use the same block shapes when converting from MPEG-2 to MPEG-4, but when they do, information such as motion vectors can be re-used to improve the quality of the transcoding. When different block shapes are used, the input MPEG-2 blocks and their attributes can still be, albeit less, useful information for more efficient MPEG-4 transcoding. The bottom line is to use as much of the original information from the original encoding process as possible to obtain the best possible video quality from transcoding and to reduce the computational complexity of transcoding.
Figure 4 demonstrates that different block sizes and shapes, combined with the in-loop filtering included in MPEG-4, have a noticeable difference in video quality, even at the same peak signal to noise ratio (PSNR). It is also possible, in some cases, to improve the quality of the HD program by transcoding. This is especially true when the original encoding was performed by a poorer quality encoder or when the bit rate was so aggressively low that it created compression artifacts in the stream. When overly aggressive or lackluster encoding has occurred, clues remain behind that a smart transcoding platform can exploit to improve quality.
For example, when images are overly compressed, an artifact commonly referred to as "macro-blocking" is visible. Macro-blocking looks like a tiling of the image and is the result of a larger than appropriate quantizer step-size. The quantizer step-sizes are clues that the smart transcoder can look for and then optimize around. If the transcoder sees quantizer step-sizes over a certain value, it knows that there is likely an accompanying and visible macro-blocking artifact. This implies that more filtering should be applied to that part of the image than to others to remove the first-generation compression artifact. Conclusion The demand is here: Consumers want their HDTV, and service providers who can supply the programming will get the business. Of the four options detailed here, MPEG-4 is the unifying element of the overall bandwidth solution, working in conjunction with SDV, node splitting and analog reclamation to create the necessary bandwidth to support HD migration. The transcoding component, while not perfect, is the best and most cost-efficient option for cable operators seeking to integrate MPEG-4 into their bandwidth strategy. Smart technology that re-uses information and clues from the input HD source, converts to MPEG-4 and maintains video quality, allows the operator to address HD demand now and remain flexible to adapt to changes in compression formats going forward. Chris Gordon is senior director, product management, for EGT. Reach him at firstname.lastname@example.org.