Abstract
We consider the transmission over an unknown frequency-selective
channel of two independent sources with different application-layer
characteristics: one source (such as voice) has a low information rate
with a strict delay constraint; the other (such as data) has a high rate
but without any delay constraints. We assume that, due to the delay
requirements, the low-rate source must be decoded first and below a
certain error probability. To exploit the different application layer
decoding requirements for a better physical channel utilization, we
study a communication system in which pilot symbols are not
present and the low-rate information is decoded non-coherently. The
decoded low-rate codewords are then used for channel estimation to
facilitate coherent decoding of the high-rate source. For a fixed
detection error probability of the low-rate source, we derive
achievable rate expressions for the high-rate source. We demonstrate
a convergence behavior of the achievable rate of the high-rate source
as the decision error probability of the low-rate source goes to zero.
Numerical results show that the achievable rate of the high-rate
source converges to that achievable by a training-based scheme at
moderate decision error levels.