A Universal Transformer-Based Algorithm for Rain Estimation Using Commercial Microwave Links Across Different Sampling Intervals

While the use of signal level measurements in commercial microwave links (CMLs) for rain estimation is widely spreading, each of the various existing algorithms is designed for a specific measurement sampling protocol. In this letter, we present, for the first time, a universal algorithm for extracting rainfall estimates collected under various operational sampling protocols. The proposed algorithm is a transformer-based neural network trained with a protocol-aware training objective. Using the proposed universal transformer, we empirically compare the performance of rainfall estimation based on various sampling rates. The results provide a comprehensive empirical quantified assessment of these protocols, demonstrating the robustness of the proposed model across different sampling intervals and identifying the sampling-rate saturation point in terms of estimation accuracy. Moreover, an empirical study shows that the proposed processor is also effective when applied to sampling protocols on which it has not been trained, making it most attractive for dynamic technologies.