Physical layer optimization for transportation of scalable media over AWGN channels

Abstract: This thesis presents research results on how the physical layer can aid in the delivery of scalable media. The thesis focuses on wireless links (point-to-point and broadcast) due to their lower capacity. To avoid unnecessary delays and system complexity the physical layer is assumed to cooperate with the source coder and the rest of the network at arm's length. For point-to-point links this means that only an indication of the scalability of a stream needs to be available to the wireless link and that current channel conditions (signal to noise ratio) are available. In cases where the source coder generates progressively coded data packets the thesis shows the potential in optimal truncation of packets to allow the stream to adapt to current channel conditions with a minimum delay. It is also shown for Additive White Gaussian Noise (AWGN) channels (with slow fading allowed) that closed form expression for expected error free length and its variance are available if no coding is used. Results from optimization of received expected error free length are included for both cases with and without channel coding. Another benefit of the truncation method presented in the thesis is that it has the potential to be usable on both bottlenecks in wireless links and in congested network routers. Also considered are Broadcast systems. For layered multiresolution broadcast systems it is shown that for a fixed transmission energy there are in some cases a potential benefit in optimizing the energy distributions among the layers. The main result is a method and some analytical models that allow the complexity of the optimization of such a system to be reduced from multidimensional optimization problem to a number of simpler line searches. Further an example system shows that real systems, based on advanced signal propagation model and population model, may in some cases be well approximated with the simple analytical models presented. The theoretical results from the optimal truncation method and the layered broadcastmodels are then combined to show the potential for theoretical optimization of unequal error protection for a speech codec application. Here a data independent metric (expected error free length) inspired the use of a metric based on perceptual metric (PESQ MOS - used to measure speech quality) into a simple data independent metric (Cumulative PESQ MOS). The thesis further shows that the new metric may be used to simplify theoretical optimization of unequal error protection and thereby provide guidance on how to design unequal error protection schemes.