Hybrid Hierarchical Models for ISAC Predictions with Wireless Links

Utilizing data from communication networks for short-term predictions is crucial for mitigating disruptions and enhancing reliability. These networks can also improve environmental sensing by serving as sensors. The emerging field of Integrated Sensing and Communication (ISAC) is key to next-generation networks, combining sensing and communication to perform under diverse conditions. In this study, we focus on rainfall, a significant source of signal attenuation, that influences the operations of commercial microwave links (CMLs), and therefore can be sensed and predicted, showcasing the dual benefits of opportunistic ISAC. Recent advancements highlight the potential of machine learning (ML) in analyzing time series patterns. However, the superiority of ML models for forecasting under data constraints remains inconclusive, while these models often face limitations due to data availability and interpretability, and are sensitive to variations in input data. To overcome this, we propose a hybrid hierarchical forecasting model (HHFM) that integrates model-based time series approaches with Reccurent Neural Networks (RNNs) models, enhancing performance in predicting the signals through a dynamic learning environment. We provide a comprehensive evaluation using real-world measurements from operational communication networks in Sweden, showcasing the benefits of the HHFM for both standard model-based and RNN components in real-time forecasting, enabling an adaptable prediction framework that relies solely on network measurements without the need for external datasets.