Pattern Reconfigurable Intelligent Surfaces to Enable "circuits-in-the-air" for Wireless Radios

The growth in spatial and spectral degrees of freedom (DoF) in next-generation wireless systems has placed increasing demands on transceiver hardware, particularly in wideband and multi-antenna scenarios. While fully digital and hybrid beamforming architectures are limited by cost and power consumption, transmissive Reconfigurable Intelligent Surfaces (RIS) offer a promising low-power alternative. In this work, we use a transmissive RIS with novel pattern reconfigurability for developing radios that can support multiuser communication systems. To jointly optimize the digital precoder and the RIS phase and pattern parameters, we propose an algorithm that combines conventional optimization approach with a Deep Reinforcement Learning (DRL) algorithm to reduce overall complexity. Simulation results demonstrate the convergence of the proposed method and over 15% gain with 2-bit pattern control in Rayleigh fading channels compared to RIS without pattern control. We use our proposed approach to examine the bit allocation between phase and pattern parameters under constrained bit budget for RIS control. Our findings highlight a prioritization of phase control and demonstrate potential additional capacity gains through pattern reconfigurability.