Data-Driven Radio Planning and Cellular Network Optimization

Abstract Integrating AI into wireless network design and management is essential for creating self-sustaining 6G networks. A key challenge is the development of automated network procedures with minimal human intervention, leveraging real-time monitoring data for immediate feedback. These advancements promote data-driven decision-making but pose risks related to data availability, safety, and the black-box nature of learning algorithms. This cumulative thesis proposes and evaluates novel procedures and algorithms for data- driven radio planning and cellular network optimization, addressing practical challenges in applying learning-based methods on real-world deployments. It emphasizes the utility of monitoring data and the integration of model-based and model-free methods, ensuring the scalability and safety of adaptive network procedures across diverse environments. The first part of the thesis explores the application of deep learning to radio propagation modeling in live cellular networks. The first paper presents a novel network planner, trained on drive-test measurements and a 3D city model, to predict signal strength in urban environments. The second paper extends the analysis to crowdsourced data, addressing the variability due to the uncontrolled collection process through Bayesian learning for uncertainty-aware prediction. Both approaches demonstrate improved accu- racy over traditional methods and exhibit physically sound propagation mechanisms. Then, the second part focuses on data-driven online network optimization, addressing scalability and safe exploration challenges. The first paper introduces a throughput model that integrates monitoring data for scalable optimization via gradient descent. The second paper extends the framework by a Monte Carlo tree search agent, which tightly integrates domain knowledge from a digital twin, guiding it towards safe and accelerated exploration. This hybrid approach outperforms both purely model-free and model-based methods, showing robust performance also under severe model mismatch. Finally, the third part investigates the use of crowdsourced measurements to infer the network topology. Such a publicly available “network twin” can then support applications like performance map generation or fingerprinting-based localization. Here, the first paper introduces a Bayesian model for base station triangulation using Timing Advance measurements, while the second one extends this to infer sector antenna orientations. Overall, the thesis thus demonstrates the potential of learning-based methods across various use-cases, backed by measurement-based evaluations in real-world networks.