Applying MPEG Standards

Which brings us to the MPEG standards. Existing MPEG codecs are relatively straightforward
constant-rate block encoders. An MPEG-2 encoder, for example, takes a 16 x 16
pixel block (macroblock) and codes the motion differences on a block-by-block basis. In
HDTV, a 1080-line picture has 1920 pixels per line subdivided down into macroblocks.
It will be a little while before we have high-definition digital TV in a handset. Atypical
digital TV decoder has nine or more parallel decoders running at 100 MHz producing
20 billion operations per second (BOPS) consuming 18 W of power! We are,
however, beginning to see similar techniques being used, albeit on a more modest
scale, in digital cellular video compression.
Video encoders today are typically constant rate. This makes them easier to manage
over the physical and transport layer, but it means they are less efficient than if they
were variable rate, that is, like the adaptive multirate vocoder or SMR vocoder. The
SMR vocoder adapts to the dynamic range of the audio waveform. The same principle
can apply to video encoders.
Consider that any source-coded content, whether audio or video, consists of entropy,
unpredictable or novel material, and redundancy. An ideal compressor would separate
out entropy and redundancy perfectly but would be infinitely complex and would have
infinite processing delay. Entropy and redundancy ratios are constantly changing. Ideally,
the video encoder rate would vary as the amount of entropy increases and
decreases. A person jumping up and down will have high entropy (and a low coding
rate); a person standing still will have low entropy (and a low coding rate).
A variable-rate video encoder would ideally be matched to a variable-rate radio
layer and network layer physical channel. The objective from a user’s perspective is to
have constant quality. Consider as an example DVB/DVD (digital video broadcasting
and digital video/versatile disc). In DVB/DVD a complex scene yields a fast encoding
rate, a simple scene yields a slow encoding rate (see Figure 7.2).
In 3GPP1 it has generally been considered that variable-rate differential encoding
was suboptimal for wireless because of the variability of the radio channel. Constantrate
coding schemes not using differencing, such as H320, were considered to be more
suitable. However, as we discussed in Chapter 1, the idea of a 3G 5 MHz channel is to
use power control to track out the fast fading�"turning our variable quality channel
into a constant-quality channel (see Figure 7.3).
We can move from constant-rate variable-quality bandwidth to variable-rate
constant-quality bandwidth, but this has to include both radio and network bandwidth
consistency. We would argue this points the way toward future MPEG-4 evolution.
The Motion Picture Expert Group (MPEG) was founded in 1993. This makes it
young in terms of telecom standards and old in terms of Internet standards. It was
originally focused on producing a standard for noninteractive (simplex) video compression
but was extended, as MPEG-4 and MPEG-5, to include the manipulation,
management, and multiplexing of multimedia content. MPEG proposals tend to be initiated
by the broadcast or content producing industry but end up as ISO standards and
ITU recommendations. They start in a different place than telecom standards but end
up at the same place.

MPEG-1 covers
CD-ROM storage, MPEG-2 covers DVB and DVD, MPEG-2—Layer 3 (unofficially but
widely known as MPEG-3) covers audio streaming, MPEG-4 adds video streaming
(and quite a lot else), MPEG-5 covers multiple viewing angles, MPEG-7 addresses content
identification, and MPEG-21 defines—or will define—network quality requirements,
content quality, and conditional access rights. MPEG-21 is described as a
“multimedia umbrella standard.”
The main purpose of MPEG-3 is to improve storage compression efficiency—
although, as a consequence, it also reduces delivery bandwidth requirements. An
uncompressed 5-minute song creates a 50-Mbyte file that is compressed down to a
5-Mbyte MPEG-3 file. MPEG-3 is a sub-band compression technique; dividing audio
bandwidth into 32 sub-bands that are each separately encoded. It helps fit an hour of
MPEG-3 music onto a 64-Mbyte memory card or (back to our hard disk!) 150 CDs on
an 8-Gbyte hard disk.
MPEG-4 adds video to produce a combined audio/video encoding/decoding standard.
In Chapter 4 we describe MPEG-4 as presently implemented—that is, a block
coding scheme in which a discrete cosine transform takes time domain information
into the frequency domain to exploit macroblock-by-macroblock and image-to-image
redundancy. The DCT is precisely prescribed in the standard, as are the multiplexing
of the audio and video streams. Other processing tasks are vendor-specific—for example,
the preprocessing, motion estimation, compensation and rate control in the
encoder, error control and error concealment, and post-processing in the decoder (the
implementation of coding noise reduction).
This vendor differentiation is probably not good news for network designers needing
to deliver a consistent user experience, as this is going to vary between codecs—
particularly when one vendor’s codec needs to talk to another vendor’s codec.
Realistically this will have to be resolved by the vendors. At present, most of the proprietary
solutions are constant-rate variable-quality.