The existence of layers in the traditional network architecture facilitates the network design by modularizing it and thus enabling isolated design of the different layers. However, due to the inherent coupling and interactions between these layers, their isolated design often leads to suboptimal performance. On the other hand, the recent popularity of realtime multimedia applications has pushed the boundaries of layered designs. Cross-layer network design provides opportunities for significant performance improvement by selectively exploiting the interactions between layers, and therefore has attracted a lot of attention in recent years.

Realtime multimedia applications are characterized by their delay-sensitivity and distortion-tolerance. The focus of this thesis is on Source Coding for Delay-Sensitive Distortion-Tolerant data. In particular, we notice that even though using longer descriptions for source symbols results in smaller distortion for each particular symbol, it also increases the delay experienced in the network, which in turn causes information loss for a delay-sensitive source, and therefore, increases the overall distortion of the received message. In this thesis we investigate this trade-off across the layers by considering two different problems.

In the first problem, we focus on a single source-destination pair to exploit the interconnection between Source Coding, traditionally a presentation layer component, and Parallel Routing, a network layer issue. We use a Distortion Measure that combines signal reconstruction fidelity with network delay. We minimize this measure by jointly choosing the Encoder Parameters and the Routing Parameters. We look at both single-description and multiple-description codings and perform numerical optimizations that provide insight into design tradeoffs which can be exploited in more complex settings.

We then investigate the problem of finding minimum-distortion policies for streaming delay-sensitive distortion-tolerant data. We use a cross-layer design which exploits the coupling between the presentation layer and the transport and link layers. We find an optimum transmission policy for error-free channels, which is independent of the particular form of the distortion function when it is convex and decreasing. For a packet-erasure channel, we find computationally efficient heuristic policies which have near optimal performance.