EE Seminar: Rain Detection and Estimation Using Recurrent Neural Network and Commercial Microwave Links
Speaker: Hai Victor Habi
M.Sc. student under the supervision of Prof. Hagit Messer-Yaron
Wednesday, November 13th, 2019 at 15:30
Room 011, Kitot Bldg., Faculty of Engineering
Rain Detection and Estimation Using Recurrent Neural Network and Commercial Microwave Links
Abstract
A novel method for environmental monitoring suggested by Messer et al. in 2006, involving existing commercial microwave links used in the backhaul communication links, for the sake of precipitation monitoring. This method is founded on traditional signal processing and the Power-Law approximation.
In this work, we introduce a rain detection (wet-dry classification) and estimation method based on recurrent neural network using commercial microwave links. We analyze three aspects of rain estimation algorithms: performance, robustness and complexity, and compare between a Power-Law based method and RNN methods.
Using actual measurements, we show that the power-law based methods are more robust while the RNN methods are more accurate, when properly trained. Also, we introduce a Time Normalization (TN) layer for controlling the trade-off between performance and robustness of RNN methods. We analyzed and draw conclusions based on actual measurements from CMLs.