Radio environment map generation and performance evaluation in heterogeneous environments

The objective of this thesis is the creation of a Radio Environment Map (REM) \cite{Zhao2006RADIOEM} in environments with heterogeneous propagation characteristics because of it is more likely to reflect real-world conditions accurately. Moreover, the scarcity of concrete studies on this topic in the literature highlights a gap that needs to be addressed. This aim necessitates two primary areas of investigation: firstly, the simulation of REMs based on specified parameters, and secondly, the reconstruction of REMs from a limited set of measurements derived from the simulated REM. Consequently, the resulting system developed within this thesis will enable the prediction of a REM for a given area utilizing only a limited number of measurements taken within that area, without defining its propagation characteristics. The simulation of the REM will be based on a scenario wherein the environment contains regions with differing propagation characteristics. This allows for the creation of realistic environments and their subsequent simulation. The output of this initial scenario comprises over forty thousand REMs, each with distinct characteristics. The scenario for REM reconstruction is based on performing a limited number of measurements on the generated REMs, and subsequently utilizing these measurements to reconstruct the entire REM. Our proposed methodologies fall under two primary categories: \textit{Single-Point Estimation} and \textit{Full-REM Estimation.} Within each category, machine learning models employing artificial neural networks with varying architectures were developed, and their respective success rates were investigated.