Page 87 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

64 Chapter 3. Video Coding: Standards

frame. In general, it is advisable to use the Improved PB-frames mode instead
of the original PB-frames mode.

3.4.6.8 Reference Picture Selection Mode (Annex N)

In normal operation, a picture is temporally predicted from the most recently
decoded picture. The reference picture section (RPS) mode, however, allows
temporal prediction from pictures other than the most recently decoded one.
Thus, in this mode, both the encoder and the decoder use more than one
picture memory. As discussed in Chapter 6, this method belongs to a class
of motion estimation and compensation techniques called multiple-reference
motion-compensated prediction. The information to signal which picture is
selected for prediction is included by the encoder in the encoded bitstream.
However, the strategy used by the encoder to select this picture is not subject
for standardization.
This mode can be used to improve the performance of video communication
over error-prone channels. In normal operation, if part of the reference picture
is lost due, for example, to a transmission error, then this error will propagate
to and severely degrade the quality of future pictures. In this mode, however,
the encoder may switch to another reference picture to suppress the temporal
error propagation due to interframe coding.
In order to utilize this mode, the encoder needs to have some knowledge
about the conditions of the channel and the outcome of the decoding process
(e.g., which parts of the reference picture have been decoded in error). One
way to achieve this is to utilize a backward (feedback) channel. This mode
has two back-channel mode switches that de/ne whether a backward channel
is used and what kind of messages are returned on that backward channel from
the decoder. Together, the two switches de/ne four basic methods of operation:
NEITHER (no backward messages), ACK (acknowledgment messages only),
NACK (negative acknowledgment messages only), and ACK+NACK (both
acknowledgment and negative acknowledgment messages). There are also two
methods of operation in terms of the channel for backward channel messages.
The /rst method is the Separate Logical Channel mode, where back-channel
data is delivered through a separate logical channel in the multiplex layer of
the system, whereas the second method is the VideoMux mode, where back-
channel data for received video is delivered within the forward video data of
a video stream of encoded data.

3.4.6.9 Temporal, SNR, and Spatial Scalability Mode (Annex O)
Scalability implies that a bitstream is composed of a base layer and one
or more associated enhancement layers. The base layer is separately decod-
able. The enhancement layers can be decoded in conjunction with the base