Optimization of Reconfigurable Intelligent Surface Codebooks Using a Mutual Coupling Aware CNN Model of Scattered Fields

Abstract

Digital reconfigurable intelligent surfaces (RISs) consist of unit cells that can operate in a finite number of discrete states, typically controlled by switching circuits. The selection of the state of each RIS cell, commonly referred to as the “codebook,” can be made similar to array design, assuming that the far field of the RIS is a weighted combination of the far fields of individual cells. However, the full-wave analysis of even a 1-bit RIS geometry reveals that mutual coupling between cells strongly affects the scattered fields of the RIS, creating a wide range of realizable scattering patterns. We show that a computationally efficient, mutual coupling aware model of RIS scattered fields (empowered by a convolutional neural network) enables the exploration of these patterns by an optimizer, to meet design objectives beyond those achievable by standard codebook optimization approaches. Moreover, we demonstrate the design of 1-bit RIS codebooks that are free of spurious (“quantization”) lobes—a problem that was previously resolved only by modifying the RIS unit cells, increasing their complexity.