Integrated RNNs for Rainfall Sensing with Wireless Communication Networks

Incorporating physical characteristics effectively into dynamic time series data is vital for improving machine learning models. Our research focuses on utilizing Wireless Communication Links (CMLs) as sensors for rainfall prediction. We explore the advantages of incorporating the physical attributes of CMLs into the learning mechanism, aiming to enhance the adaptability and applicability of the sensing capabilities of communication networks. We investigate these benefits through two distinct approaches for embedding sensor features in Integrated Recurrent Neural Networks (I-RNNs). Aligned with the opportunistic use of communication networks within the Integrated Sensing, Communication, and Computation (ISCC) framework, our research aims to optimize the integration of sensor capabilities. We conduct a comprehensive analysis with real-world measurements from operational CMLs, including unique data from smart cities, aiming to enhance sensing strategies in both existing and emerging landscape of Beyond 5G/6G systems in the ISCC. Our results demonstrate the value of incorporating static information into RNNs for sensor differentiation, thereby enhancing the accuracy, generalization, and robustness of weather sensing with communication networks.