Handset Hardware Evolution

3G handset hardware allows us to capture voice (audio bandwidth), image and video
bandwidth, and application bandwidth. These are typically multiplexed into multiple
traffic streams that may be separately modulated onto multiple OVSF code streams
over the physical layer (radio air interface). The choice of image processor (CCD or
CMOS) dictates the dynamic range of the image or video stream and other qualities
such as color depth and resolution. In addition, the accurate representation of fastmoving
action requires a reasonably fast frame rate. Therefore, the hardware dictates
our bandwidth quantity and quality requirements.
3G handset hardware generates bursty bandwidth. Voice, image, and video
encoders have historically been constant rate encoding devices but are increasingly
moving to become variable rate to accommodate the varying amount of entropy and
redundancy in the source-coded information. In addition, voice, image, video, and
data is being multiplexed together in the encoder. Intuitively you might think this
would help to average out some of the burstiness. In practice, peaks of information
energy still need to be accommodated. These peaks can either be dealt with by allocating
additional bandwidth (bandwidth on demand) or by buffering to smooth out the
bit rate. Bursty bandwidth can always be turned into constant rate bandwidth by
buffering. The cost is the additional memory needed, delay, and delay variability.
As we will see in later chapters, conversational rich media exchanges are relatively
intolerant to delay and delay variability. The best option from an application point of
view is to make the delivery bandwidth dynamically responsive to the application
bandwidth required, remembering that delivery bandwidth is a summation of radio
bandwidth and network bandwidth.
Our handset hardware has captured and described the time domain and frequency
domain components of our speech, image, and video waveforms. It is the job of the
physical layer to preserve these time domain and frequency domain properties. The
physical layer includes the radio link and the network, as well as another radio link to
the other side of the network if we are talking about handset-to-handset communication.
On the receive side, it is the job of the hardware components to rebuild and reconstruct
the original signal (composed of audio, image, and video waveforms),
preferably without noticeable loss of quality.
Loss of quality can be caused by a poor, inconsistent radio channel, a badly designed
receiver, a poorly designed decoder, or, for image and video, display and display driver
constraints. Bandwidth quality in this context is an end-to-end concept that encompasses
every hardware component involved in the duplex simultaneous process of
send and receive. As a result, the quality of our MPEG-4 encoder/decoder has a direct
impact on perceived image and video quality, and the quality of our voice codec
(encoder/decoder) has a direct impact on perceived voice quality.
With image and video, there is not much point in transmitting a 24-bit color depth,
30 frame per second video stream if we only have a display capable of supporting 12
bits × 12 frames per second. Bandwidth quality, therefore, becomes a balancing act.
Where do we put our processing power?
Adding MIPS to a voice codec improves quality and reduces the bit rate needed
from the radio channel but increases codec processor overhead—and introduces delay.
The same principle applies even more so to image and video encoders/decoders. We
could transmit a video stream at 12 frames a second and use rendering and interpolation
in the decoder to double the frame rate—a perceived quality improvement traded
against an increase in processor power.
Bandwidth quality comes with a cost and power budget price tag. As processor
costs and processor power budgets improve, quality improves. However, we also need
to deliver consistency. A poor-quality channel that is consistent may often be perceived
as being better quality than a better-quality channel that is inconsistent. (We remember
the bad bits.)
The idea of the variable-rate encoder is to deliver constant-quality source coded
voice, image, and video (the coding rate changes, not the quality). The idea of having
adaptive radio bandwidth that codes out the fast fading on the channel is to deliver
constant quality. 138