Channel Capacity Prediction Using Point of Interest for Design and Operation Support of Network

Numerous small base stations (BSs) have been deployed in a 5G radio access network. In such an ultra-dense network, mobile operators need to maintain the performance levels at their required levels while handling the inter-cell interference and increasing power consumption. For improving network operation or design, such as BS deployment, channel capacity for a system model is an important metric often calculated as throughput. However, the appropriate assumption of user equipment (UE) location and the consideration of the radio wave propagation characteristics unique to the area are required for a practical simulation. To address this issue, in this study, we propose an artificial intelligence that estimates an area channel capacity using the relative location of BS and point of interest (POI) as the feature. In detail, a convolutional neural network is used to model the relationship between the area channel capacity and the location of BS and POI in the area. The model does not require the UE location assumption and the definition of radio wave propagation. We regard POI as a related metric for propagation characteristics and as alternative information to UE location. The real-world data publicly available are used, and data augmentation is performed to improve prediction accuracy. As a result, in our model, the coefficient of determination in prediction was 33% higher than that of the existing method, where the number of POIs and BSs was used as input. In addition, our model with location input enables the prediction of the channel capacity of an arbitrary BS arrangement. To leverage this, the areas where BS sleep control can save energy were estimated as an example of application.