Accurate origin-destination traffic prediction (ODTP) is a persistent problem in network management. Due to the structural nature of networks, the spatial correlations are critical for effective perdition. In this paper, the ODTP problem is built onto the graph domain by mapping OD traffic into graph-structured data involving the topology information. Moreover, due to the flow characteristics of OD traffic, the complex spatial-temporal (ST) correlations should not be limited at one-hop neighbors or consecutive time steps. To benefit from this observation, we propose a novel graph learning based method, called multi-scale spatial-temporal graph neural network (MSTNN). In MSTNN, spatial and temporal extractors are designed to capture multi-scale spatial and temporal correlations. Specifically, the spatial extractor employs the graph attention mechanism to capture the time-varying spatial correlations of nodes and their multi-hop neighbors. In the temporal extractor, we design gated dilated convolution layers with different dilation factors, each of which represents a different granularity of time, to exploit the multi-scale temporal correlations in both adjacent and non-adjacent time steps. After cascading the spatial and temporal extractors, the multi-scale ST correlations are weighted fused. Simulations on two real-world datasets show that MSTNN outperforms existing approaches that work well in various prediction tasks.

Random linear network coding (RLNC) is a promising network coding solution when the network topology information is not fully available to all the nodes. However, in practice, nodes have partial knowledge of the network topology information. Motivated by this, we investigate the performance of RLNC and obtain different upper bounds on the failure probability of RLNC for the network constrained by different partial network topology information. These upper bounds not only improve the existing ones in the literature, but also show that the partial network topology information can bring benefits to the performance analysis of RLNC. On the other hand, it is observed that if more network topology information can be utilized, tighter upper bounds can be obtained, as expected. The upper bounds on two classical networks are compared for demonstration. To obtain a deeper understanding about the performance of RLNC, the asymptotic behavior of RLNC as the field size goes to infinity is also investigated.

Batched sparse (BATS) code is a promising technology for reliable end-to-end transmission in multi-hop wireless networks. One main research topic for BATS code is how to design an optimal inner code that is typically random linear network code. In this paper, this issue is focus on the number of transmissions from an end-to-end perspective. The problem is formulated as a mixed integer nonlinear programming (MINLP) problem with the objective of minimizing the total number of transmissions from source to destination. Subsequently, the inherent properties of inner codes are exploited to relax the integer restrictions by the means of the regularized incomplete beta function. As a result, a new nonlinear programming (NLP) problem is constructed. Solving the NLP problem provides a valid lower bound on the optimal solution, and, hence, is used as the performance measure for our heuristic. Furthermore, a centralized approximation approach is developed to solve our MINLP problem efficiently. The numerical results demonstrate that all solutions developed in the paper are near-optimal with a guaranteed performance bound.

In conventional distributed storage exact repair problems, all sources are reconstructed when the decoder has access to a certain number of encoders (disks). So, the underlying reconstruction network is equivalent to a single-source problem. This paper considers a variant of the exact repair problem, where the underlying reconstruction network is the independent distributed source coding problem, a type of multi-source problem. As the first non-trivial case with two sources and three encoders, the storage-repair tradeoff regions are proved for all the 33 instances, and it is shown that binary codes are optimal.

In this paper, we propose an encoded packet-assisted retransmission, namely EAR, approach that use opportunistic network coding (ONC) to improve the bandwidth efficiency for unicast sections in wireless networks. In existing ONC schemes, network coding opportunity are not always fully exploited due to the lack of understanding of the use of the encoded packets receivers cannot decode. To overcome this limitation, EAR develops an improved opportunistic listening mechanism that allows receivers to store all heard undecodable ONC packets and to periodically report the reception status to the sender. The sender then can directly XOR these encoded packets with other packets to exploit the potential coding opportunities of them. To further maximize the utilization of coding opportunity in the network, EAR introduces a greedy coding algorithm that constructs a simple scheduler. The scheduler prioritizes packets according to their contribution to the overall transmission performance. Simulation results show that, EAR can significantly reduce the total number of retransmissions and provide more efficient bandwidth utilization, compared with existing schemes.