Abstract
Reconfigurable intelligent surfaces (RIS) have been proposed to extend the coverage of wireless communication signals at mm-wave frequencies. Here, we designed and fabricated a reconfigurable intelligent surface (RIS) for multi-user frequency selective beam steering (MU-FSBS) that achieves higher channel capacity than traditional time division multiple access (TDMA) techniques in multi-user communication scenarios. MU-FSBS is enabled by a varactor diode-tuned RIS that allows to steer several beams in the Ka-band independently at the same time. For such targeted multi-frequency beam steering, we implemented a customized neural network-based machine learning architecture specifically designed to optimize the bias voltage patterns of the RIS. As an experimental demonstration, we first accomplished independent beam steering of normally incident beams at 27 GHz and 31 GHz to deflection angles between 10[Formula: see text] and 45[Formula: see text] in a defined plane. Secondly, we compared the achievable channel capacities of the proposed MU-FSBS approach with those of TDMA, finding an average increase of approximately 50% in channel capacity at a fixed SNR of 20 dB. At an SNR of 60 dB, MU-FSBS even demonstrated a remarkable 84% increase in channel capacity compared to TDMA.