A Generalized Beyond-Diagonal RIS Architecture: Theory and Design via Structure-oriented Symmetric Unitary Projection
Beyond-diagonal reconfigurable intelligent surface (BD-RIS), which enables advanced wave control through interconnection of RIS elements, is gaining growing recognition as a promising technology for 6G and beyond. However, the enhanced flexibility of BD-RIS in controlling the phase and amplitude of reflected signals comes at the cost of high circuit complexity. To further explore the trade-off between circuit complexity and performance, in this paper, we propose a generalized BD-RIS architecture named cluster-connected RIS. Specifically, the proposed cluster-connected RIS offers a unified framework that generalizes existing BD-RIS architectures—including single-connected, fully-connected, group-connected, tree-connected (arrowhead), forest-connected (arrowhead), and stem-connected RISs—as special cases. This framework enables a much more flexible trade-off between circuit complexity and system performance than existing ones. Based on the proposed BD-RIS architecture, we introduce a novel and generalized structure-oriented symmetric unitary projection method for designing the scattering matrix across all BD-RIS configurations. This method is effectively applied to solve the sum channel gain maximization problem and other utility-based optimization problems. Numerical results demonstrate that the cluster-connected RIS enlarges the performance–complexity trade-off range. Furthermore, the proposed projection-based algorithms demonstrate high efficiency, achieving performance comparable to conventional alternative optimization-based approaches while significantly reducing computational complexity. This highlights their promising applicability in practical BD-RIS designs. © 2002 IEEE.