Efficient wireless coverage maps using sparse Gaussian processes : Exploring applications of Gaussian processes in mobile cellular networks and real-world train connectivity scenarios

Abstract

This thesis provides a comprehensive understanding of Gaussian Processes and their possible applications in the field of wireless networks with a specific focus on railway communication scenarios. It highlights the potential of Gaussian Processes in addressing various challenges in wireless communications, like improving network efficiency and performance but also predicting and modeling certain Key Performance Indicators. Gaussian Processes are a powerful tool that offers a flexible and non-parametric approach to capturing complex patterns and uncertainties. While Gaussian Processes are versatile and sophisticated models, their computational demands can pose challenges, particularly for large data sets, making them computationally unfeasible in such cases. To address calability challenges, we explore Sparse Gaussian Processes to mitigate computational complexity while maintaining predictive performance. We provide an overview of existing Sparse Gaussian Process models, compare selected methods, and benchmark their performance. We then delve into specific potential use cases of Gaussian Processes within wireless networks, such as traffic prediction, localization, trajectory planning for Unmanned Aerial Vehicles, signaling reduction, and performance evaluation. Lastly, we discuss the specific scenario of railway connectivity which poses its own challenges. A measurement campaign onboard a train, and an accompanying statistical analysis of the observed Reference Signal Receive Power (RSRP) values along the train route connecting Vienna and Innsbruck are conducted. We derive a Path Loss model specific to the railway connectivity scenario. Additionally, we utilize Gaussian Processes to estimate spatial correlations between measured RSRP values, enhancing our understanding of wireless propagation effects that occur onboard trains.