Modeling and performance analysis of random linear network coding in multiple unicast transmissions

Abstract

Random linear network coding has been considered as one of the most promising technologies to improve the performance of communication system such as throughput and delay. The operation of random linear network coding is mainly made by coding coefficients and coded packets. The coded packet is a combination of the coding coefficients and uncoded packets (or original packets). In the environment where there are a source and multiple receivers, the source generally transmits the coded packets and the related coding coefficient at the same time. Then, receivers have to obtain the coded packets and the coding coefficients together. If a receiver does not obtain the coded packets or coding coefficients, the receiver fails to receive that packet.

In this thesis, by using the characteristics of random linear network coding, a variety of random linear network coding schemes are researched in multiple unicast transmissions. At first, related works with random linear network coding are researched and the concept of multiple unicast transmission is examined. Then, in time varying channels, opportunistic scheduling schemes based on random linear network coding are presented in case of single frame and multiple frames, respectively, to improve outage probability and delay. In case of single frame, we examine the detailed operational procedure of random linear network coding and study a new random linear network coding scheme to minimize an outage probability by opportunistically transmitting the coded packet according to the result of the coding coefficient transmission. In case of multiple frames, we study the effect of random linear network coding by considering two things: the first is frame length according to usage of feedback and the second is practical modulation level such as binary phase shift keying (BPSK) and M-ary phase shift keying (MPSK). Random linear network coding schemes are proposed to improve overall success probability and reduce total symbol lengths. We also analyze the performance of random linear network coding when channel information is delayed due to mobility and time difference between channel estimation and data transmission, and provide proper opportunistic scheduling schemes in outdated channel state information.

Besides, a combination of random linear network coding and non-orthogonal multiple access is researched to improve system performance in case of 2 multiplexing levels such as near group and far group. Basically, the operation of source is composed of two things: one is coded packet generation from uncoded packets for random linear network coding and the other is superposed packet generation from coded packets in a power domain for non-orthogonal multiple access. The operation of receiver is also composed of two things: one is successive interference cancellation for non-orthogonal multiple access and the other is Gauss-Jordan elimination process for random linear network coding. In this case, we research total packet success probability in 4 cases: non-orthogonal multiple access, orthogonal multiple access, non-orthogonal multiple access with random linear network coding and orthogonal multiple access with random linear network coding. Finally, we further research on random linear network coding based on non-orthogonal multiple access in case of general multiplexing levels. There are trade-off relationships between higher multiplexing levels and lower multiplexing levels in terms of two things: the first thing is an amount of interference and the second thing is the number of required packets for decoding successfully. We study an average and a variance of the expected transmission times for completing decoding to evaluate random linear network coding based on non-orthogonal multiplexing in case of general multiplexing levels by considering the scheduling scheme that makes multiplexing groups from entire network according to the difference of channel gains.