Skip to main content

A practical guide to encoders

The most obvious engineering goal of an encoder is to get the best picture and audio quality per bit used. The other significant objective is to keep down the total cost of ownership. Factors that can affect the latter are the upfront cost, ease of use and maintenance, reliability, rack density, and power consumption.

Bandwith comparisons

There is always a trade off between good and cheap. Operators will make their choices based on how they value their bandwidth.

Cable companies have traditionally seen their bandwidth as cheap, though HD may change that. Satellite operators can usually put a tangible cost on their bandwidth. Terrestrial and DSL operators have hard limits on their total bandwidth, and it's no coincidence that they often lead the way in coding efficiency. Although ATSC will be MPEG-2-based for the foreseeable future, many of the terrestrial systems being launched use MPEG-4. Some of the more mature terrestrial operators are upgrading to MPEG-4, for example France and Spain, to allow HD and multiple streams on one line.

Interfaces

It is often useful to think about the interfaces of a system. The inputs to an encoder went from baseband video supplied as analog NTSC to SDI and now HD-SDI.

On the output side, IP transport is taking over from ASI, mainly driven by the lower hardware costs. (See Figure 1.) IP transport allows headends to be physically distributed. For IPTV networks, encoders do not usually work through a mux but provide a complete single program transport stream at their output. (See Figure 2 on page 66.)

Another important interface is statistical multiplex (statmux) control. This is always a proprietary mechanism. Encoders make predictions about their bandwidth requirements for the next GOP and send it to the statmux controller. The controller then sends out bandwidth allocation messages to all the encoders in the group.

Some vendors combine the statmux controller with the mux. This might be a factor to keep in mind if you want to choose compression and mux vendors separately.

Statmux can make a significant contribution to coding efficiency, but not all network topologies support it. For example, a DSL only carries one stream. In this case, variable bit rate can be used, and the troughs can provide space for time-insensitive data, such as e-mail.

Table 1. Typical MPEG-2 and MPEG-4 rates. These will depend on many factors and should be treated with caution.
Typical Rates (Mb/s) MPEG-2 MPEG-4 525 2.5Mb/s to 3Mb/s 1Mb/s to 1.5Mb/s 720p 12Mb/s 6Mb/s to 8Mb/s 1080i 16Mb/s 8Mb/s to 10Mb/s

In the United States, audio has mainly been Dolby AC-3. In markets influenced by DVB, MPEG Layer II has been more dominant. MPEG-4 includes the Advanced Audio Coding (AAC) standard, which gives about a 50 percent improvement over Layer II.

Trends

The most important trend in the industry is that the huge base of MPEG-2 STBs is not going away anytime soon, so most DTV signals will use this standard in the future. The exceptions to this are where there are green field developments — mainly HD and IPTV.

Many operators are upgrading their MPEG-2 systems with the most modern encoders to free up bandwidth, often for new HD services. Since the first generation of MPEG-2, there has been a significant and steady improvement in encoding technology — about 50 percent since 1999.

MPEG-4 currently offers a 30 percent to 50 percent improvement over MPEG-2, and we can expect improvements at a faster rate than MPEG-2, as there is now a greater body of knowledge in the industry. (See Table 1 on page 66.)

The main areas of improvement have been:

  • Better coding decisionsAn example is knowing which algorithm to use under particular conditions.
  • Better video preconditioningAn example is softening the picture during high-speed motion.
  • Statistical multiplexingAn example is the improved bandwidth requirement prediction to allow better bandwidth allocation decisions. The bandwidth allocation algorithms have also been improved and now treat low-rate encoders less harshly in percentage terms. This follows the observation that viewers' perception of quality is based on the worst quality and not the average quality.

Some MPEG-4 research has fed back into MPEG-2 developments. One example is adaptive GOP. In most MPEG-2 systems, the GOP structure was kept as IBBP. During MPEG-4 work, it was discovered that B frames do not work well during high-speed motion. This is because they are largely based on motion vectors, which become large during high-speed motion. The solution is to change the GOP structure to IPPP during high-speed motion. This is possible because MPEG-2 does not mandate the GOP structure, so MPEG-2 STBs will decode alternative structures.

In the VC-1 versus MPEG-4 debate, most DTV operators have stuck to the open standard and gone with MPEG-4. The consensus now seems to be that both standards offer similar performance. This is not surprising because the fundamental algorithms are similar.

Architecture

Most encoder manufacturers use digital signal processors or general-purpose microprocessors, with field programmable gate arrays to support hardware acceleration. (See Figure 3 on page 68.) Algorithms are downloadable. This means, for example, that a VC-1 encoder can be converted into an MPEG-4 encoder by a download. Processing power is a constraint, of course, so it is not usually possible to change an MPEG-2 encoder into an MPEG-4 encoder by download alone. Most vendors supply encoders in a 1RU package, though some vendors use a blade architecture to give high rack densities.

Quality assessment

There are several tools on the market that offer objective picture quality assessment. These tools can be useful for monitoring trends where a baseline has been established, e.g. automatic monitoring of output.

It is harder to find evidence that these tools can help in the selection of encoders. What matters is the viewers' perception of the audio and video quality. The ITU-R suggests a subjective test, such as using a panel of viewers in recommendation BT.500. This involves the viewers rating compressed video against the uncompressed source. The sequence is randomized so the viewers don't know which is which. In practice, other factors such as price, quality of support and ease of integration are also significant in the choice

David Short is a technical architect working on the design of new DTV systems. He also is a member ofBroadcastProjects.com, an alliance of independent consultants.