Abstract:
Integrated Sensing and Communications (ISAC) shares radio resources, evolving from existing to 6G cellular networks. However, due to the highly dynamic nature of 6G, traditional ISAC models fail. This thesis investigates whether an AI-based framework can help converge ISAC by using an experimental simulation-driven methodology. A synthetic dataset containing 10,000 6G ISAC scenarios has been created in various environments. A selection of 7 different machine learning techniques were used to assess their ability to determine an accurate prediction of the Sensing Range from the communication parameters (Sensing Accuracy/Sensing Range, Data Rate/Sensing Range and Spectral Effciency/Sensing Range). The outcome presents a fundamental trade-off between data rate and the Sensing Range through an inverse-read correlation coeffcient of 0.38. Based upon the use of more advanced set techniques (e.g. ensemble models), XGBoost (R² = 0.60, MAE = 4.24 m) outperformed all previous linear based established models (R² = 0.37, MAE = 5.25 m). Thus providing conclusive evidence for the complexity of non-linear characteristics in relation to ISAC systems. This research provides a single empirical framework driven by AI and demonstrates that XGBoost provides an accuracy and interpretability necessary to operate and manage intelligent 6G networks; thereby flling the gap between the theoretical building blocks of ISAC and the practical AI implementation required by ISAC for 6G.
