Abstract
We analyze and compare different methods for handling the mutual coupling in RIS-aided communication systems. A new mutual coupling aware algorithm is derived where the reactance of each element is updated successively with a closed-form solution. In comparison to existing element-wise methods, this approach leads to a considerably reduced computational complexity. Furthermore, we introduce decoupling networks for the RIS array as a potential solution for handling mutual coupling. With these networks, the system model reduces to the same structure as when no mutual coupling were present. Including decoupling networks, we can optimize the channel gain of a RIS-aided SISO system in closed-form which allows to analyze the scenario under mutual coupling analytically and to draw connections to the conventional transmit array gain. In particular, a super-quadratic channel gain can be achieved which scales as $N^4$ where $N$ is the number of RIS elements.
Introduction
Reconfigurable Intelligent Surfaces (RIS) are a promising technology for 6G, typically modeled as arrays of independent reflecting elements. However, to maintain a compact form factor, elements are often spaced closely together, leading to significant mutual coupling. This electromagnetic interaction means the state of one element influences its neighbors, complicating the optimization of the RIS configuration.
Ignoring these effects results in inaccurate performance models and suboptimal beamforming. Conversely, existing coupling-aware optimization methods are often computationally prohibitive ($O(N^4)$ complexity) or prone to getting trapped in local optima due to the non-convex nature of the coupled system's optimization landscape.
Methodology: Efficient Algorithms and Decoupling Networks
This research proposes two primary solutions to address the challenges of mutual coupling.
1. $O(N^3)$ Element-Wise Algorithm
The authors developed a numerical algorithm that updates the reactance of each RIS element successively. By leveraging the matrix inversion lemma, the need for full matrix inversions at each step is eliminated, reducing computational complexity from $O(N^4)$ to $O(N^3)$. This provides an exact solution with the same efficiency as previous approximation-based methods.
2. Hardware-Based Decoupling Networks
A more radical approach involves introducing a "decoupling network"—a lossless, reciprocal multiport network—between the RIS antennas and their adjustable loads.
As shown in Figure 1, this network mathematically diagonalizes the system's impedance matrix. This transformation effectively "decouples" the elements, returning the system to a structure identical to one without mutual coupling. This allows for global optimization of phase shifts in closed-form, bypassing the non-convexity issues of the original coupled system.
Key Results and Insights
The decoupling network approach not only simplifies optimization but also reveals untapped potential in RIS performance.
Overcoming Local Optima
Iterative algorithms like gradient descent often fail in highly coupled scenarios (small element spacing) because they converge to local maxima. The decoupled approach, by providing a closed-form global optimum, significantly outperforms these methods.
In Figure 4, for $N=4$ elements at spacings below $0.25\lambda$, the "Decoupled" method achieves the theoretical maximum gain ($N^4=256$), while other iterative methods (Z-OPT, Gradient) plateau at much lower values, having failed to find the global optimum.
Super-Quadratic Array Gain
A major theoretical finding is the existence of "super-quadratic" array gain. While standard RIS models predict a gain scaling with $N^2$, the authors prove that in "end-fire" scenarios (where signals propagate along the array plane), the gain can scale as $N^4$ as element spacing $d \to 0$. This effect is most pronounced in lossless systems; however, the presence of Ohmic losses in real-world hardware will dampen this gain as spacing decreases.
Conclusion and Future Work
While decoupling networks offer a path to global optimality and extreme gains, their physical implementation is challenging. A fully connected network's complexity scales as $O(N^2)$, which is impractical for large surfaces.
Future research is directed toward partially connected decoupling networks. By only accounting for coupling between nearest neighbors, hardware complexity could be reduced to $O(N)$ while still capturing the bulk of the performance benefits offered by mutual coupling awareness.
Summary updated