Physical Space Information Preprocessing for 5G Throughput Prediction on 28 GHz Channels

The use of millimeter wave (mmWave) bands is a promising approach to increase the capacity of mobile networks. However, the mm Wave link quality (LQ) is strongly affected by the condition of the line of sight (LOS) path. In order to stably utilize mm Wave bands, an effective solution is to predict future LQ and adaptively control wireless communication. LQ prediction based on Deep Neural Network (DNN) using physical space information, such as the position information of user equipment (UE), is effective to enable the proactive control. On the other hand, huge training data is required to provide the prediction model. In a supervised learning approach, the prediction models must be generated for all the base stations with which the UE communicates. Thus, it is a key to generate the prediction models for different base stations using no or less amount of the training data. In this paper, we propose a method for preprocessing physical space information to reduce the amount of training data when the UE accesses to the different base station. The proposal normalizes the position and direction information of UE by using the position of the base station. To evaluate the effectiveness of the proposed method, we conducted experiments using an automated humanoid robot in a commercial 5G network. Experiment results show that the proposed method reduces the root-mean-square-error (RMSE) of the differences between the predicted and actual throughputs by 35.1 and 51.0% at the different base station when using no samples and 100 samples as training data, respectively, compared to without physical space information preprocessing.