Optimization of the Mobile Radio Network through Classification and Individual Treatment of Users with Fixed Trajectories

The increasing demand for higher data throughput in wireless networks drives the ongoing development of mobile radio technologies. The first part of this thesis focuses on wireless channel prediction using ray tracing. In addition to the models employed, the necessary environmental modeling is described in detail. Due to the high computational cost of ray-based prediction, several acceleration techniques are evaluated. A novel method tailored for map-based predictions is developed, and the use of GPUs is investigated for further efficiency gains.
To validate the ray tracing results, two approaches are presented: comparison with a public dataset at 947 MHz and a custom measurement campaign at 2.1 GHz. In both cases, the results confirm the reliability of the ray tracer, with deviations mainly due to incomplete environmental modeling.
The second part of the thesis addresses the optimization of beamforming in dynamic scenarios. Determining ideal radiation patterns typically requires significant signaling overhead, especially in environments with high-mobility users like on highways. The proposed system exploits the fact that such users follow predictable paths, resulting in distinct characteristics detectable from the base station's perspective.
Machine learning is used to classify these users, followed by an individual angle estimation to determine suitable beam directions. Compared to conventional fixed-beam management, this approach significantly improves performance. With a mean estimation error of less than 1°, beamforming gain improves in over 95% of cases, reaching a mean gain increase of 2.33 dB for an 8x8 antenna array.