Lossy transmission over fading and packet erasure channels.

摘要

Recent demand for multimedia applications for wireless channels and packet networks has motivated the design of various transmission strategies for such channels. Multimedia sources are loss-tolerant and the design objective is to optimize the overall quality of the received signal, as opposed to the traditional measure of bit error probability. A joint source-channel coding approach to this problem is taken in this thesis. We start with packet transmission and propose low-complexity optimization techniques for progressive transmission over packet networks, wireless channels or a combination of the two.; We then investigate the problem of lossy transmission over multiple-antenna flat fading channels. The distortion minimization problem is considered in an information theoretical setting based on the rate-distortion function of the source and the outage capacity of the channel. The case where only the receiver is aware of the channel state information is studied first. We propose low-complexity optimization techniques for layered transmission with time-sharing and superposition coding strategies and show the equivalence of the time-sharing strategy to the corresponding problem for packet erasure networks.; In the next step, we consider a scenario where a low-rate feedback channel provides channel state information to the transmitter. We show the effectiveness of low-rate noiseless feedback in reducing the distortion. The equivalence of K-level noiseless feedback without power adaptation and K-layer superposition coding without feedback is shown in terms of the distortion exponent. Finally, we consider the case where the feedback channel itself suffers from fading and propose a transmission scheme that exploits the unreliable feedback information. We also show that a system with noisy feedback has the same distortion exponent as a single-layer no-feedback scheme. An efficient numerical technique for designing quantized feedback systems is also presented.