Capacity Prediction for Wireless Networks Based on Convolutional Neural Network

Abstract

The deployment of a wireless network greatly affects the system capacity, which is difficult to be optimized with massive devices and complicated propagation environment. The machine learning tools, e.g., the convolution neural network (CNN), can extract the implicit features of the network deployment and provide directions for the capacity optimization of wireless networks. In this paper, we generate the artificial data based on a practical wireless network model for datasets acquisition, and propose an efficient approach for the capacity prediction of wireless network based on the CNN. In particular, the deployment of access points in a wireless network is regarded as 2-dimensional matrix, which is the input of the neural network. Then, the CNN is used to handle the matrices and output numeric for the capacity prediction. The impacts of different parameters and architectures of CNN on the predictive accuracy are evaluated. Our results demonstrate the accuracy and robustness of the proposed prediction approach.