Research on topology control method of multi-hop wireless communication network based on graph neural network

Multi-hop wireless communication networks occupy an irreplaceable position in many application areas by virtue of the advantages of network flexibility, easy deployment, low cost and expansion strength. In order to further optimize its network topology and improve the destructive resistance to malicious attacks, this paper uses the GCN model to learn the features and representations of the nodes in the graph, extracts the local and global features of the nodes, and uses a more complex function multilayer perceptron to fuse the features of the nodes, topology, etc., and optimizes the topology control method by judging the probability of the existence of each edge in the network topology. Simulation experiments on the power control and allocation performance of GCNTO are conducted under cellular network conditions, and the final convergence result of GCNTO exceeds 96% of the optimization result of the FP algorithm under different hyperparameters and number of training samples, as well as under the conditions of scaling up and scaling down the area and increasing and decreasing the number of users, which demonstrates the excellent convergence and scalability. After the model training of the two types of interference channels in advance, GCNTO is able to realize the accurate dynamic capture of topology and node information for untrained user channels and maintain the ideal data rate.