Other Standards

MPEG-4 is not the only standard. Microsoft has its own Windows Media Player.
MPEG-4 does have, however, reasonably wide industry support (including support
from Microsoft) and builds on earlier work with MPEG-2 and MPEG-3 (audio encoding).
One of the problems with these compression standards is that they are optimized
to improve storage bandwidth efficiency and are sometimes rather suboptimum when
used in variable quality and occasionally discontinuous transmission channels, for
example, wireless. MPEG-4 does try to take into account the idiosyncrasies of the radio
physical layer. It has also been absorbed into DivX, the PC industry standard for
downloadable video, and by Apple in their QuickTime product, so it has some crossindustry
adoption.
Figure 4.6 shows the functional diagram for the Amphion MPEG-4 decoder.
The input to the Amphion video decoder core is a compressed MPEG-4 video
stream; the data rate is variable from extremely low rates of several kbps up to a maximum
defined by the MPEG-4 profile or that possible on the transport stream. For
example, the MPEG-4 Simple profile enables up to 384 kbps while the Advanced Simple
profile provides up to 8 Mbps (four times the maximum bit rate available from the
present W-CDMAphysical layer). Higher profiles and bit rates support image scalability,
the ability to scale image resolution and frame rate for a given variable channel
bandwidth. The MPEG-4 standard supports a wide range of resolutions up to and
beyond that of HDTV (high definition television).
The Amphion hybrid architecture of hardware accelerators plus control software on
a microcontroller, typically an ARM microprocessor, enables an efficient partition and
acceleration of data intensive tasks while maintaining general sequencing and control
tasks (such as error resilience) in software. The video bit stream processor extracts variable
length symbols from the compressed serial stream, often applying Huffman and
run-length decoding for downstream texture decoding and motion compensation. The
pixel generation core performs inverse scan, ACDC differential prediction, quantization
and discrete cosine transforms on texture coefficients from the video bit-stream
processing unit. The image post processor and picture out control does post processing
and filtering to take out blockiness and compression artifacts, and then finally colorspace
conversion and display output timing. Not shown in the functional diagram is
the motion compensation accelerator which handles the pixel reference reads, reconstructions
and write-backs to frame memory.
Power consumption is much reduced—to below 15 milliWatts—by implementing
decompression in a hybrid (i.e., hardware-software) solution because this approach
significantly reduces not only the need for processor program and data RAM, but also
the overall clock rate required for decoding. Additionally, the main processor can be
made available for other tasks such as speech and audio decoding or demux functions.
The hardware accelerators can support resolutions and frame rates much higher than
any processor-based implementation and thus higher quality video can be supported.
The design challenge for both hardware- and software-based MPEG-4 encoders/
decoders is to deliver the functionality needed to support different visual and audio
quality metrics: color depth, frame rate, and aspect ratio for imaging, frame rate for
video and audio fidelity, which can then be mapped onto a quality-based billing
metric. We discuss quality-based billing in more detail in Chapter 8. Video processing
will also support 3D effects (for interactive games), which involves the convergence of
MPEG-4 and VRML (the IETF’s Virtual Reality Modeling Language). The work groups
have taken to describing this as visual information engineering. The importance of the
MPEG-4 encoder to us is that it effectively defines uplink offered traffic by taking in the
imaging and video bandwidth generated by the CMOS or CCD imaging platforms
together with other audio and data inputs. 122