Direction-of-Arrival Estimation by Metasurface Assisted by Deep Learning for IoT Systems

We propose to merge reconfigurable intelligent surface (RIS) and deep learning concepts to build a platform capable of estimating the direction-of-arrival (DOA) of incoming signals rapidly and accurately. The RIS is exploited as an electronically scanned virtual parabolic reflector with a coarse scanning step of 5°. A radio-frequency (RF) power detector placed at the focal point records the power levels of the incoming signals for each scan step and the data is fed into a pretrained multilayer neural network to deduce the arriving angle with a precision less than 1°. Measurements are performed to validate the approach.