| The properties of random linear network coding make it an attractive proposition for the implementation of scalable multicasting on erasure channels. Motivated by this application, this thesis studies network coding within a minimal feedback framework and addresses some barriers to the practical implementation of network coding. The foundation of this work is a framework for evaluating the impact of generation-based network coding on throughput, which leads to the first formulation of throughput-motivated guidelines for the selection of network coding parameters. Building on this framework, a novel network coding method is proposed wherein limited inter-generation mixing is allowed. This new method is shown to be a useful tool for increasing the throughput achievable with generation-based network coding.The tradeoff between encoding complexity and throughput is also studied. Two encoding schemes are compared it is shown that due to the peculiarities of network coding, non-recursive coding schemes achieve considerably higher throughput than recursive coding schemes with comparable complexity. Furthermore, complexity can be further reduced by replacing multiplicative operations with shifting the impact on throughput is negligible. |
