Embedded Intelligent Reconfigurable Metasurface Architectures for Adaptive Full-Duplex RF Front-End Systems

Abstract

Radio-frequency (RF) front-end systems that are full-duplex provide the opportunity to ameliorate spectral efficiency, where simultaneous transmission and reception via the same spectral band is possible; these possibilities are harshly restricted due to strong self-interference (SI), dynamic wireless environment, and hardware reconfigurability constraints. This paper suggests an embedded intelligent reconfigurable metasurface architecture to adaptive full-duplex RF front-end systems, in which metasurface control can be modelled by a deep reinforcement learning (DRL)-based control framework. The suggested solution makes metasurface structure and RF cancelation a sequential decision-making task, ensuring the embedded DRL agent will cooperate to optimise SI and signal-to-interference-plus-noise ratio (SINR) with real hardware conditions, such as discrete phase resolution, low update rates, and energy, among others. An architecture based on hardware awareness is designed, combining the configurable metasurface, RF front-end and an inference engine with embedded DRL that is friendly to FPGA/ SoC-dominated devices. The large-scale simulation and hardware-in-the-field testing shows that the proposed DRL-enabled metasurface control performs better in terms of both SI suppression and faster response to channel variations and minimised reconfiguration overhead in comparison with a static and optimization-based and non-learning adaptive baseline control. The findings have been used to emphasise the suitability of learning-controlled reconfigurable metasurfaces as useful and scalable method towards the next generation intelligent full-duplex RF front end system.