Metasurfaces (MTSs) are increasingly emerging as enabling technologies to meet the demands for multi-functional, small form-factor, efficient, reconfigurable, tunable, and low-cost radio-frequency (RF) components because of their ability to manipulate waves in a sub-wavelength thickness through modified boundary conditions. They enable the design of reconfigurable intelligent surfaces (RISs) for adaptable wireless channels and smart radio environments, wherein the inherently stochastic nature of the wireless environment is transformed into a programmable propagation channel. In particular, space-limited RF applications, such as communications and radar, that have strict radiation requirements are currently being investigated for potential RIS deployment. The RIS comprises sub-wavelength units or meta-atoms, which are independently controlled and whose geometry and material determine the spectral response of the RIS. Conventionally, designing RIS to yield the desired EM response requires trial and error by iteratively investigating a large possibility of various geometries and materials through thousands of full-wave EM simulations. In this context, machine/deep learning (ML/DL) techniques are proving critical in reducing the computational cost and time of RIS inverse design. Instead of explicitly solving Maxwell's equations, DL models learn physics-based relationships through supervised training data. The ML/DL techniques also aid in RIS deployment for numerous wireless applications, which requires dealing with multiple channel links between the base station (BS) and the users. As a result, the BS and RIS beamformers require a joint design, wherein the RIS elements must be rapidly reconfigured. This chapter provides a synopsis of DL techniques for both inverse RIS design and RIS-assisted wireless systems.
翻译:元表层(MTS)日益成为能够满足多功能、小形式因素、高效、可重新配置、可调制、可调制和低成本无线电频率(RF)组件需求的辅助技术,因为能够通过修改边界条件在亚波长厚度范围内操纵波浪;能够设计可调适无线频道和智能无线电环境的可重新配置智能表面(RIS),其中无线环境固有的随机性质被转化成一个可编程的传播频道;特别是,目前正在对具有严格辐射要求的通信和雷达等空间限制的RF应用程序进行可能部署的调查;RIS包括亚波长单元或元原子部分,这些单元或元原子部分通过经修改的边界条件来独立控制,其地理测量和材料决定了RIS的光谱反应;《公约》为产生理想的EM反应而设计RIS需要反复地研究,通过数千个全波流的EM模拟,对各种地貌和材料的大规模可能性进行测试;在这方面,对具有严格辐射要求的通信/持续学习(ML) 和D-D-DL 数据库数据库数据库数据库数据库数据库数据库数据库数据库中的关键设计、数据库数据库数据库数据库数据库数据库数据库设计需要通过大量数据流路路路路路路路路路路路路段的计算。