Consider the problem of transmitting K samples of a
discrete-time
i.i.d Gaussian source in N uses of an additive white Gaussian
noise (AWGN) channel to two users with different signal-to-noise
ratios (SNRs). Let D1 and D2 be the distortion in reconstructing
the source at the two receivers. We are interested in
finding good
broadcast strategies and in determining the region (or atleast
good bounds on the region) of pairs of distortion (D1,D2)
that are
simultaneously achievable. For N not equal to K, determining the
exact region is still open.
We show that joint source channel coding using hybrid digital
analog (HDA) codes can be significantly better than broadcasting
using purely digital techniques. After a brief summary of
the work
that has been done for the case of N > K, we will focus on the
case when N < K. We show how coding with side information can be
useful for this problem and in particular show layered coding
strategies which outperform purely digital techniques. A
particularly novel feature is the use of HDA version of Costa's
writing on dirty paper.
We will summarize some existing results, present new results and
discuss specific open problems in this area.
Host: Ashu Sabharwal |
