5G Throughput Prediction For 28 GHz Channels Using Physical Space Information

Recent advances in wireless communication technology such as fifth-generation (5G) have enabled the creation of various novel applications. As a result, a large number of devices are now being connected to mobile networks, and mobile traffic is increasing year by year. Although the use of the millimeter-wave (mmWave) bands is a promising approach to increasing the capacity of mobile networks, there are many challenges to use mmWave bands. The link quality (LQ) of mmWave wireless links is impacted by the surrounding objects. Therefore, in order to stably utilize mmWave bands, we believe that it is necessary to predict future LQ predictions and adaptively control wireless communications. In this paper, we evaluated the throughput prediction methods using physical space information of the target UE and surroundings in a commercial 5G network. The evaluation entails measuring the throughput in an actual indoor environment where both the target UE and surrounding objects are moving. To create the huge dataset necessary to allow the moving terminal holder and surrounding pedestrian (objects) to be modelled, we develop two autonomous humanoid robots and make one move so as to block the LOS of the other robot, which is the UE holder. The experiments shows that our proposed method using physical space information yields a 57.5 % improvement in prediction accuracy at the 50th percentile absolute error value over a naive prediction model that uses past throughput information.