Proactive Clustering of Base Stations in 5GC-RAN using Cellular Traffic Prediction

Abstract

The rapid growth in mobile network traffic and dynamic user mobility patterns have propelled network operators toward the Cloud-Radio Access Network (C-RAN) to reduce operational costs and improve service quality. C-RAN handles the traffic and mobility issues in a centralized manner by segregating the central units (CUs) from the distributed units (DUs) in a shared CU pool. The ability of C-RAN to map multiple DUs to the same CU allows optimal coverage with high multiplexing gains, using the least number of CUs. However, dynamically mapping DUs to CUs is not trivial since the network traffic and mobility patterns are difficult to predict. This paper presents a two-phase framework for an optimal city-wide C-RAN network. In the first phase, we propose to use the ConvLSTM model, which simultaneously learns the hidden spatial and temporal dependencies in a real-world dataset and makes accurate traffic forecasts for a future duration of time. In the second phase, we use the predicted traffic from the first phase to develop a proactive optimal DU-CU clustering scheme that is cost-effective and meets quality objectives. We first formulate an optimization problem, and later, to reduce the computational complexity of the optimization, we propose a lightweight heuristic algorithm. Finally, we evaluate the performance of our prediction model and the mapping scheme using a two-month real-world mobile network dataset of Milan, Italy. Based on simulation results of phase one, we observe the ConvLSTM model, when deployed in a C-RAN architecture, outperforms existing state-of-the-art prediction models with up to 26% better RMSE (Root Mean Square Error) and up to 36% better MAPE (Mean Absolute Percentage Error) values. Similarly, in phase two, our simulation results show that compared to reactive threshold-based clustering, proactive clustering can reduce the average number of active CU servers by up to 18% every 10 minutes without overloading.