Performance Analysis of Divided Random Linear Fountain

The random linear fountain (RLF) is an efficient form of fountain coding with an expected overhead of only 1.6 packets. Because of increasing computational complexity, however, it cannot be directly used for large message sizes. In this paper, we study the performance penalty of dividing the data into k parts which are coded using the RLF. To ease the performance problem, we propose the use of macropackets which are generated by using LT-coding over the different parts of divided RLF. We calculate the decoding probabilities for k = 2 and 3 and compare the results to the overhead obtained using the combination of divided RLF and data carousel. The results indicate that while the use of macropackets improves the performance over the divided RLF, still better performance is obtained by data carousel in the case of low channel loss rate.