Page 248 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

Section 9.8. Interactive Techniques 225

9.8.1 Automatic Repeat Request (ARQ)
In this technique, when an error is detected, the decoder automatically requests
the encoder to retransmit the damaged data. When this ARQ is received, the
encoder retransmits the requested data. Usually, this retransmission is repeated
until either the requested data is correctly received or a predetermined number
of retransmissions is exceeded.
Typically, when a decoder sends an ARQ, it waits for the arrival of the
requested data before resuming normal operation. This introduces delays that
may not be acceptable in real-time applications like mobile video communi-
cation. To overcome such delays, Wang and Zhu [179] proposed a technique
called retransmission without waiting. In this technique, instead of waiting
for the arrival of the requested data, the damaged video part is concealed
and normal decoding operation is then resumed. A trace of the a ected pels
and their associated coding information is recorded until the arrival of the
requested data. This error trace, along with the received data, is then used to
correct the a ected pels. Another technique proposed in Ref. 179 is the mul-
ticopy retransmission. In this technique, multiple copies of the damaged data
are sent in each single retransmission trial. This reduces the required number
of retransmissions and, consequently, reduces delays.

9.8.2 Error Tracking
When feedback information is received, the encoder can reconstruct the error
propagation process. In other words, the encoder can track the error prop-
agation from the original occurrence up to the current frame. A number of
techniques can then be used to utilize this error trace, as discussed next.

9.8.2.1 INTRA Refresh Based on Feedback
Based on the error trace, areas in the current frame that would have been
predicted from a ected pels in the reference frame are INTRA encoded. This
is illustrated in Figure 9.9. Figure 9.9(a) shows the spatial and temporal prop-
agation in a sequence of frames due to an error in frame n. In Figure 9.9(b)
a feedback message arrives at the encoder before the time to encode frame
n + d. The encoder tracks this error and the a ected pels from frame n up
to frame n + d − 1. During the encoding process of the current frame, n + d,
blocks that would have been predicted from a ected pels in the reference
frame, n + d − 1, are encoded in INTRA mode to stop error propagation to
the next frame, n + d +1.
There are two main drawbacks to this approach. First, a perfect reconstruc-
tion of error propagation is a computationally complex process. Second, in
cases of high error rates, INTRA refresh can result in a signi cant loss in