The fraction of machine-to-machine traffic carried by satellite networks is increasing, and an efficient delivery is required in order to enable a large set of applications that can benefit from the advantages provided by the use of satellites. This work analyses the use of random linear network coding techniques in land mobile satellite channels to reliably deliver machine-to-machine traffic to mobile nodes in urban areas. The considered scenario takes into account a cooperative coverage extension in land mobile satellite vehicular networks, where the use of random linear network coding techniques can remove the need for any fixed equipment on the ground at least in urban environments, where the density of mobile nodes is typically high.
KEYWORDSgossiping, M2M, network coding, RLNC, satellite, VANET
INTRODUCTIONThe application of network coding (NC) to communications networks is relatively recent and dates back to year 2000. 1 Since then, NC has shown great potentials in correcting random packet errors, erasures, and errors introduced by malicious nodes, making it a powerful tool to achieve efficient network service delivery and network reliability. In the seminal work by Ahlswede et al, 2 NC is illustrated by the butterfly network example: A data source delivers a stream of messages to two receivers; all links can deliver up a single message per unit of time. In the example, the middle link represents a bottleneck, and to overcome that, the authors propose to send a combination of the messages instead of plain packets. Two incoming messages a i and b i are XORed, a i ⊕ b i , then the coded packet a i ⊕ b i is sent in the middle link, reaching the two receivers in a single unit of time. The two receivers can reconstruct the whole piece of information by XORing the network coded packet and a i (or b i ), which comes on a separate link. In this way, the throughput of the network is increased, because less packet transmissions are needed to deliver the whole information block. Linear NC (LNC) is introduced by Li et al 3,4 : the idea behind LNC is to create combinations of packets by taking coefficients from a Galois field of size q = 2 x .The original messages can be retrieved, at destination, by performing Gaussian elimination on the matrix generated from the LNC coefficients of the packets. Considering k source packets, a so-called generation worth of k input symbols, n ⩾ k output symbols are generated by the coding procedure.If at least k independent symbols received at destination, then decoding is possible. Random linear network coding (RLNC) has been proposed in Ho et al 5 and Chou et al 6 in 2003: when using RLNC, the coefficients are randomly drawn from the finite field. According to Ho et al, 5 the failure probability when relying on RLNC depends on the size q of the finite field. The larger q, the lower the failure probability, but at the price of a larger packet overhead and larger lookup tables* to be maintained in memory. 7 Anyway, it should be noted that a larger q provides a lower energy per bit cost...