Wireless Link Scheduling for D2D Communications with Graph Embedding Technique

Link scheduling for device-to-device (D2D) communications is usually formulated as an NP-hard non-convex combinatorial problem, which is difficult to get the optimal solution. Traditional methods are mainly based on mathematical optimization techniques with the help of accurate channel state information (CSI), which is costly to obtain. In this paper, we propose a graph embedding based method to achieve link scheduling without CSI for D2D communications. We first construct a fully-connected directed graph for the D2D network, and then compute a low-dimensional feature vector for each node in the graph based on the distances of both communication and interference links. Finally, a scheduling strategy can be learned based on the graph embedding results by utilizing a multi-layer classifier. Extensive simulation demonstrates that the proposed method is near-optimal compared with the existing state-of-art methods and only needs hundreds of training network layouts. It is also competitive in terms of scalability and generalizability to more complicated scenarios.