Semi-Supervised Machine Learning Applications in RAN Design: Towards Data-Driven Next Generation Cellular Networks

Abstract

The explosive growth of mobile internet services and demand for data connectivity boosts the innovation and development in Radio Access Network (RAN) to define how next generation mobile networks will look like. Continuous improvement in existing RAN is crucial to meet very strict speed and latency requirements by different mobile applications with minimum investments. Exploiting the advancement in Machine Learning and AI-driven algorithms is essential to tackle these challenges in different functions within the RAN domain. In this paper we surveyed how to leverage different clustering algorithms to understand underlying community structures within RAN and what benefits those insights could bring to serve different use cases in next generation networks. Finally, the paper proposes a clustering based framework to solve RAN Tracking Area (TA) planning problem using both mobile users data and base stations geographical locations aiming to reduce network signaling overhead. Live network dataset extracted from operational mobile operator used to assess results of different popular clustering techniques. Results showed potential reduction of 20.3% in TA signaling overhead compared to a baseline of current network configuration.