Fast Compressed Sensing Beam Training in RIS-assisted MISO Systems

This paper is dedicated to the problem of Beam Training (BT) in Reconfigurable Intelligent Surface (RIS)assisted systems. There is a lack of BT solutions for the case when both the RIS and the user positions are unknown. Furthermore, the most popular solution (hierarchical search) requires frequent feedback from the user. We propose a joint BS and RIS scanning, which provides stable Compressed Sensing (CS) Channel Acquisition (CA) independent from RIS and user angular positions. Simulation results show that proposed method, while using feedback only once, is competitive with the hierarchical scheme. Another experiment ensures reduced dependence on initialization: optimized measurements provide up to 3 times less deviation of successful recovery probability in comparison with random measurements (10% and 30% respectively).