Optimization of Beyond Diagonal RIS: A Universal Framework Applicable to Arbitrary Architectures
This work provides a general solution for BD-RIS optimization, addressing the lack of a universal framework for different architectures, which is important for advancing RIS technology in wireless communications.
The paper proposes a universal optimization framework for beyond-diagonal reconfigurable intelligent surfaces (BD-RISs) that works for arbitrary circuit architectures, achieving better trade-off between performance and computational efficiency than existing methods in MU-MISO systems.
Reconfigurable intelligent surfaces (RISs) are envisioned as a promising technology for future wireless communication systems due to their ability to control the propagation environment in a hardware- and energy-efficient way. Recently, the concept of RISs has been extended to beyond-diagonal RISs (BD-RISs), which unlock the full potential of RISs thanks to the presence of tunable interconnections between RIS elements. While various algorithms have been proposed for BD-RIS optimization, they mainly focus on specific architectures whose scattering matrices exhibit very special structures. A universal optimization framework that can accommodate different BD-RIS circuit topologies is still lacking. In this paper, we bridge this research gap by proposing an architecture-independent framework for BD-RIS optimization, with the main focus on sum-rate maximization and transmit power minimization in multiuser multi-input single-output (MU-MISO) systems. Specifically, we first incorporate BD-RIS architectures into the models by connecting the scattering matrix with the admittance matrix and introducing appropriate constraints to the admittance matrix. The formulated problems are then solved by our custom-designed partially proximal alternating direction method of multipliers (pp-ADMM) algorithms. The pp-ADMM algorithms are computationally efficient, with each subproblem either admitting a closed-form solution or being easily solvable. Simulation results demonstrate that the proposed approaches achieve a better trade-off between performance and computational efficiency compared to existing methods.