Large file distribution using efficient generation-based network coding

To distribute a large amount of data through a network, much coordination information, such as packet scheduling and ARQ feedback, is needed between nodes. In this paper, we propose a network coding based scheme to ratelessly distribute data without any coordination between network nodes. To overcome the cubic decoding complexity of network coding, we follow a generation-based strategy, which groups the source packets into subsets called generations and only permits coding among packets belonging to the same generation. A maximum local potential innovativeness strategy is proposed to schedule transmissions of generations at intermediate nodes, where only local buffer information is needed. The transmission overhead at sink nodes, which is defined as the number of extra packets required for successful decoding caused by the non-optimal scheduling, is reduced by using overlapping generations where one source packet may be present in multiple generations. The and-or tree analysis technique is applied to analyze the performance of the overlapping generation-based network coding scheme and used to determine the optimal amount of overlap needed when constructing generations. The overall design is shown to achieve promising throughput rate while maintaining low decoding complexity.