A Neural Network Based Conflict-Graph Construction Approach for Ultra-Dense Networks

Abstract

Ultra-dense networks (UDNs) have been widely regarded as a promising technology to meet higher requirements of the fifth generation (5G) network. However, densely deployed femtocells bring an unprecedented challenge of importing severe co-channel interference (CCI), which greatly limits the performance of the network. Therefore, interference management in UDNs is particularly important. The conflict-graph is widely recognized as the representation of underlying interference constraints of the network. Most prior studies establish conflict-graphs based on accurate geographical distance information which is usually unavailable by the network operators in practice. A more practical neural network based conflict-graph construction approach, which utilizes up-link signal to interference plus noise ratio (SINR) data and up-link resource block (RB) allocation data, is proposed in this paper. These data are used to train a neural network for predicting the up-link SINR under single interfering user conditions. Utilizing the predicted SINR data as the edge weights of the constructed conflict-graph considers both the influence of inter-user interference and the interference tolerance ability of each user, which better reflects the resource reuse conflict between users. Users can collaborate and coordinate their resource allocation strategies according to the constructed conflict-graph so as to mitigate severe CCI in the network. Therefore, the proposed approach facilitates the realization of the intelligent resource allocation optimization. Furthermore, the proposed conflict-graph construction approach achieves quite a high accuracy, which has been verified by simulation results.