Network coding with multi-generation mixing
Mohammed Halloush, Hayder Radha · 2008
Connectivity, losses, and buffering are factors that directly affect the performance of network coding. These factors affect the ability of intermediate nodes to generate useful encodings; these are encodings that contribute in propagating and recovering data at receiver node(s). It has been shown that in particular scenarios network coding performance can be improved by simply increasing the generation size (k); however, this leads to increasing transmission overhead, encoding complexity and (more importantly) buffer sizes at intermediate nodes. In this paper, we propose a new network coding approach where we employ Multi-Generation Mixing (MGM). MGM eliminates the need to increasing buffer sizes while improving the performance of network coding. Under MGM, we define a mixing set of size m generations that can be network coded (mixed) together. Within each MGM mixing set, a new set of generation packets are mixed with previously transmitted (network coded) generations. This generalized approach provides a great deal of resilience against losses when compared with traditional generation-based network coding. Our analysis of the performance of MGM-based network coding demonstrates significant reduction in required overhead for a given recovery performance at the receivers. We also illustrate the performance of MGM using extensive simulations that provide a useful insight into the viability of MGM-based network coding.