In this section we discuss the performance of the adaptive quantizer
and on-line channel estimator.
Figure 10-Figure 12
show the convergence
of the linear LRI learning
to the optimal quantizer when pe=10-1.
It is seen that p1(n) corresponding to the
optimal quantizer converges to 1 and p2(n) and p3(n)
converge to zero for all the three sequences. Both p2(n) and p3(n)
have the same behavior as n varies. This is why we see only one dotted
curve in the figures.
We also observe that increasing the value of the reward parameter results in
a faster learning.
We find similar convergence results for other
bit error probabilities and other video sequences too.
Figure 13 shows the reconstructed frames of the
Missamerica sequence when pe=10-1. Clearly, the optimal
quantizer when M=2 results
in a decrease in the number of corrupted blocks.
A similar behavior is seen in Figure 14 when
p2=10-2.
This performance can be further enhanced by using additional
error protection.
