Unicast is the dominant form of transmission on LANs (local area networks) and the Internet today. Therefore, schemes that can better utilize network resources to serve multiple unicast connections have many potential applications. Intra-session network coding is well understood today, achievable rates are known and coding schemes designed to achieve those rates are available. In contrast, finding the optimal network coding strategy for the inter-session scenario is arguably the most important open research problem in the network coding community.;This thesis considers the problem of network coding for multiple unicast connections in networks represented by directed acyclic graphs. The interference alignment techniques, originally developed for wireless interference channels, are extended to the linear network coding setup and its performance in this setup is analyzed. We describe two general approaches (namely, coding at the edge or in the middle of the network) and we also discuss a closed form method (namely, eigenvector method) for one of the approaches. We also revisit the more general symbol extension method of the coding at the edge approach and try to understand the relation between the feasibility conditions and the network structure. Further, we also compare alignment to alternate approaches. For three unicast sessions with min-cut one, we show that whenever alignment is possible, alternative approaches can also achieve half the min-cut. However, for more than three sessions and/or for min-cut per session greater than one, we show examples where alignment is necessary.
展开▼
