| Accurate identification of Line-of-Sight (LoS) and Non-Line-of-Sight (NLoS) propagation is essential for robust millimeter-wave (mmWave) communication in beyond-5G
and 6G networks, where narrow-beam transmissions are highly sensitive to blockage and
environmental variability. The evolution toward 6G networks requires ultra-reliable operation
in mmWave and sub-terahertz bands, where propagation is fundamentally constrained by
high path loss and susceptibility to blockage. Consequently, accurate prediction of LoS
and NLoS conditions is critical for effective beam management and link adaptation. This
thesis proposes a real-world LoS/NLoS prediction framework based on the DeepSense 6G
dataset and investigates the benefits of transformer-based channel embeddings generated by
the Large Wireless Model (LWM). A comprehensive evaluation is conducted using a broad
set of machine learning (ML) algorithms, including K-Nearest Neighbors (KNN), Support
Vector Machines (SVM), Random Forest (RF), eXtreme Gradient Boosting (XGBoost),
Categorical Boosting (CatBoost), Light Gradient Boosting Machine (LightGBM), and a
Multi-Layer Perceptron (MLP). All models are evaluated using LWM-based classification
embeddings. Experimental results demonstrate that LWM embeddings substantially enhance
separability between LoS and NLoS links, enabling models to capture complex multipath
characteristics that are not effectively represented by raw features. The MLP achieves the
best performance, with 98.3% accuracy and an Area Under the Curve (AUC) exceeding
0.996, effectively capturing nonlinear patterns in dynamic 6G environments. The proposed
LWM-based prediction framework outperforms state-of-the-art methods by more than 9%,
15%, and 12% in LoS, NLoS, and average accuracy, respectively. These findings highlight
the importance of foundation-model-based representations and real-world data for reliable
link-state prediction in next-generation wireless systems.
Keywords: mmWave, LoS/NLoS prediction, 6G networks, DeepSense, foundation models,
Large Wireless Model, Machine Learning |