Prediction of mmWave BS Layout by DNN with Sub-6GHz Band Features

Abstract

Due to the different propagation characteristics such as large path loss and susceptibility to blockage, the coverage of mmWave base stations (BSs) is more irregular and complex than that of the legacy wireless frequency band. As a consequence, the traditional network planning method will not satisfy the demand for mmWave system. To address this issue, this paper proposes a predicting algorithm for the layout of mmWave BSs leveraging the Sub-6GHz band. Specifically, the mapping relationship between the channel information in the Sub-6GHz band and the layout of mmWave BSs is utilized. To approximate this implicit and highly nonlinear mapping relationship, a deep neural network (DNN) is employed, leveraging the strong approximation capability of DNN. To train the DNN by the training data set, an integer programming (IP) algorithm is used as a supervisor, which obtains the BSs layout outcomes by solving an optimization problem. Moreover, to improve the generalization and scalability performance of the proposed scheme in large-scale scenarios, the candidate BS locations are clustered by environment features in addition to the wireless features. Simulation results show that the proposed algorithm can achieve a near-optimal network coverage even only Sub-6 GHz channel information is used. And the time consumption is less than 0.3% compared with the reference algorithm.