Machine Learning-Assisted Multi-User Frequency Selective Beam Steering with a Reconfigurable Intelligent Surface in the Ka-Band

The phase shift and attenuation of a wave reflected from a reconfigurable intelligent surface (RIS) is dependent on the frequency of the wave. We exploit this dispersion in a wireless communication scenario to deflect a normally incident wave to multiple clients across different locations, each served at its unique carrier frequency. A Machine Learning model has been implemented to optimize the signal strength for all users simultaneously. The approach is experimentally verified for the case of two spatially separated users receiving signals at 27 GHz and 31 GHz. Here, we show that the electric field magnitude of the signal at each location can be increased by at least 5.5 dB and up to 28.4 dB (on average 17.15 dB) compared to signals originating from reflections of a flat aluminum plate of similar size as the RIS. The effect could be observed for deflection angles from 10° to 45°.