Graph Embedding-Based Wireless Link Scheduling With Few Training Samples

Link scheduling in device-to-device (D2D) networks is usually formulated as a non-convex combinatorial problem, which generally NP-hard and difficult to get the optimal solution. Traditional methods solve this problem are mainly based on mathematical optimization techniques, where accurate channel state information (CSI), obtained through estimation feedback, needed. To overcome high computational complexity of traditional eliminate costly stage, machine leaning (ML) has been introduced recently address wireless link problems. In article, we propose novel graph embedding method for D2D networks. We first construct fully-connected directed network, each pair node while interference links among pairs edges. Then compute low-dimensional feature vector graph. The process distances both communication links, therefore without requiring CSI. By utilizing multi-layer classifier, strategy can be learned supervised manner results node. also an unsupervised train further reinforce scalability develop K-nearest neighbor representation reduce complexity. Extensive simulation demonstrates that proposed near-optimal compared with existing state-of-art but only hundreds training network layouts. It competitive terms generalizability more complicated scenarios.