Reciprocal Beyond-Diagonal Reconfigurable Intelligent Surface (BD-RIS): Scattering Matrix Design via Manifold Optimization

This paper proposes a low-complexity manifold optimization framework that enforces reciprocity constraints on Beyond-Diagonal Reconfigurable Intelligent Surfaces (BD-RIS) to maximize sum-rate performance, demonstrating superior results compared to state-of-the-art approaches.

The Gist

Here is an explanation of the paper, translated from academic jargon into everyday language using analogies.

The Big Picture: The "Smart Mirror" Problem

Imagine you are in a large, crowded city with tall buildings. You are trying to talk to a friend on the other side of a massive skyscraper. Your voice (the radio signal) hits the building and bounces off, or worse, gets blocked completely. You can't hear each other.

Now, imagine you have a Reconfigurable Intelligent Surface (RIS). Think of this as a giant, high-tech "smart mirror" mounted on the side of a building. It doesn't amplify your voice (it's passive, like a real mirror); instead, it can instantly change its shape to catch your voice and bounce it perfectly toward your friend.

For a long time, these mirrors were "dumb." They could only adjust each tiny square on the mirror independently. It was like having a wall of thousands of tiny, independent mirrors that couldn't talk to each other. This worked okay, but it wasn't very efficient.

The New Idea: The "Teamwork" Mirror

This paper introduces a new, smarter version called a Beyond-Diagonal RIS (BD-RIS).

Instead of every tiny square acting alone, the new mirror allows the squares to talk to each other.

• The Old Way (Diagonal): If you tap one square, only that square moves.
• The New Way (Beyond-Diagonal): If you tap one square, it can coordinate with its neighbors to move in a synchronized wave. This creates a much more powerful and precise "beam" of signal.

The Catch:
While this teamwork makes the signal stronger, it creates a massive engineering headache. To make these squares talk to each other without needing expensive, power-hungry electronics (like amplifiers), the mirror must follow a strict rule: Reciprocity.

Reciprocity is like a conversation. If Person A speaks to Person B, Person B must hear them exactly as they were spoken. In physics, this means the mirror's "scattering matrix" (the mathematical map of how it bounces signals) must be symmetric. If the mirror bounces a signal from the left to the right, it must bounce a signal from the right to the left in the exact same way.

The Problem: The "Impossible" Math Puzzle

The researchers wanted to design this mirror to maximize the total internet speed (sum-rate) for everyone in the city.

1. The Goal: Make the mirror bounce signals so perfectly that everyone gets the fastest possible connection.
2. The Constraint 1 (Symmetry): The mirror must be reciprocal (symmetric).
3. The Constraint 2 (Unitary): The mirror must be "lossless." It can't absorb the signal; it must reflect 100% of the energy (like a perfect mirror).

Mathematically, trying to satisfy both constraints at the same time while maximizing speed is like trying to solve a puzzle where the pieces keep changing shape. The math is "non-convex," which is a fancy way of saying the landscape is full of hills and valleys, and it's very easy to get stuck in a small valley (a local optimum) thinking you've reached the top, when the real peak is far away.

The Solution: "Manifold Optimization" (The Hiker's Map)

The authors didn't try to force the mirror into a shape it didn't want to be. Instead, they used a technique called Manifold Optimization.

The Analogy:
Imagine you are a hiker trying to find the highest peak in a mountain range, but you are only allowed to walk on a specific, winding trail (the manifold).

• Standard Math: Would try to walk in a straight line, hit a cliff, bounce back, and try again. It's messy and inefficient.
• Manifold Optimization: Realizes you are on the trail. It calculates the steepest path along the curve of the trail itself. It respects the shape of the mountain.

In this paper, the "trail" is the shape of the Symmetric Unitary Matrices. The authors developed a new algorithm that walks along this specific mathematical trail, ensuring the mirror stays symmetric and lossless the whole time, while climbing toward the highest possible speed.

The Secret Sauce: "Fractional Programming"

To make the math easier, they used a trick called Fractional Programming.

• The Problem: The speed formula is a fraction (Signal / Noise + Interference). These are hard to optimize directly.
• The Trick: They transformed the fraction into a simple quadratic equation (like a parabola). This made the "hill" smooth and easy to climb. They also added a "penalty" term to the equation. If the mirror started to lose its symmetry, the penalty would push it back onto the correct path.

The Results: Why It Matters

The researchers tested their new "Smart Mirror" design against the current best methods (State-of-the-Art).

1. Faster Speeds: Their mirror consistently provided higher internet speeds (sum-rate) than the old methods.
2. Works Everywhere: It worked well whether the mirror was fully connected (every square talks to every other) or partially connected (groups of squares talk to each other).
3. Hardware Friendly: Because they enforced the symmetry rule mathematically, the resulting design is actually possible to build with cheap, passive hardware. You don't need expensive active components to make it work.

Summary in One Sentence

The authors created a new mathematical "GPS" that guides the design of a super-smart, cooperative signal mirror, ensuring it stays physically realistic (symmetric) while climbing to the absolute peak of internet speed performance.

Technical Summary

Here is a detailed technical summary of the paper "Reciprocal Beyond-Diagonal Reconfigurable Intelligent Surface (BD-RIS): Scattering Matrix Design via Manifold Optimization."

1. Problem Statement

The paper addresses the challenge of maximizing the sum-rate in a multi-user multiple-input single-output (MU-MISO) wireless communication system aided by a Reciprocal Beyond-Diagonal Reconfigurable Intelligent Surface (RBD-RIS).

• Context: Traditional Diagonal RIS (D-RIS) architectures have limited degrees of freedom (DoF) because their scattering matrices are diagonal. BD-RIS architectures overcome this by allowing interconnections between elements, resulting in full or block-diagonal scattering matrices. However, BD-RIS introduces complex constraints: the scattering matrix must be unitary (to ensure lossless operation) and symmetric (to ensure reciprocity in passive hardware).
• The Core Challenge: The joint optimization of the transmit beamforming and the BD-RIS scattering matrix is a non-convex problem. Specifically, designing the scattering matrix under the dual constraints of unitarity (Stiefel manifold) and symmetry (symmetric matrix manifold) is difficult. Existing methods often rely on non-reciprocal designs (increasing hardware complexity) or use penalty-based approaches that are computationally expensive or fail to strictly enforce symmetry during the optimization trajectory.
• Goal: To design a low-complexity, physically realizable (reciprocal) scattering matrix that maximizes the system sum-rate without requiring active components or magnetic biasing.

2. Methodology

The authors propose a novel optimization framework that combines Fractional Programming (FP) with Manifold Optimization (MO) to solve the scattering matrix design problem.

A. Problem Reformulation

1. Fractional Programming (FP): The original sum-rate objective (a sum of logarithms of Signal-to-Interference-plus-Noise Ratios) is non-convex. The authors apply the Lagrangian Dual Transform (LDT) and Quadratic Transform (QT) to convert the objective function into a quadratic form. This allows the problem to be handled more efficiently.
2. Handling Reciprocity (Symmetry): Instead of projecting onto the intersection of the unitary and symmetric manifolds (which is geometrically complex), the authors incorporate the symmetry constraint as a quadratic regularization term (penalty) into the objective function. The term $\nu \|\Theta - \Theta^T\|_F^2$ penalizes deviations from symmetry.
3. Handling Unitarity: The unitary constraint ($\Theta \Theta^H = I$) is handled by treating the scattering matrix as a point on the Stiefel manifold.

B. Optimization Algorithm

The authors develop a Conjugate Gradient Ascent (CGA) algorithm tailored for the Riemannian manifold:

• Manifold Geometry: The optimization operates on the Stiefel manifold. The algorithm uses retraction (mapping a tangent vector back to the manifold via QR decomposition) and tangent projection (projecting the Euclidean gradient onto the tangent space of the manifold).
• Closed-Form Gradients: A key contribution is the derivation of closed-form expressions for the Riemannian gradient of the regularized, FP-transformed objective function. This avoids the need for numerical differentiation and significantly reduces computational overhead.
• Post-Processing: While the penalty term encourages symmetry during iteration, the final step involves a specific projection (based on Singular Value Decomposition) to strictly enforce both unitary and symmetric constraints on the resulting matrix blocks.

C. Architecture Support

The method is general and supports three BD-RIS architectures:

1. Single-Connected: Diagonal matrix (standard RIS).
2. Group-Connected: Block-diagonal matrix (intermediate complexity).
3. Fully-Connected: Full matrix (maximum DoF).

3. Key Contributions

• Novel Manifold Optimization Technique: The paper proposes the first method (to their knowledge) that addresses the dual-manifold nature of reciprocal BD-RIS design by embedding the symmetry constraint into the objective function while optimizing over the Stiefel manifold for unitarity.
• Closed-Form Gradients: The authors derived explicit closed-form gradients for the MU-MISO case, enabling the implementation of a highly efficient Conjugate Gradient Ascent algorithm. This facilitates hardware-friendly implementation (dedicated DSP/VLSI) due to the deterministic nature of the operations.
• Hardware Compatibility: By enforcing reciprocity (symmetry) and unitarity, the proposed design is compatible with passive, lossless, and low-cost hardware implementations, avoiding the need for active non-reciprocal components.
• Scalability: The algorithm is designed to scale efficiently with the number of RIS elements and users, offering lower computational complexity growth compared to state-of-the-art (SotA) alternatives like pp-ADMM.

4. Simulation Results

The proposed method was evaluated against two State-of-the-Art (SotA) benchmarks:

1. Interference Nulling (SotA [20]): A passive Zero-Forcing approach.
2. pp-ADMM (SotA [33]): A joint scattering and beamforming design using Alternating Direction Method of Multipliers.

Key Findings:

• Sum-Rate Performance: The proposed method consistently outperforms both SotA benchmarks across all connectivity architectures (Single, Group, and Fully-connected) and channel conditions (Rayleigh and Rician fading).
• Robustness: The performance gain is particularly notable in the Group-Connected architecture (e.g., group size 2 or 4), where the proposed method achieves significantly higher sum-rates (up to an order of magnitude in high-power regimes for specific configurations) compared to SotA methods.
• Convergence: The algorithm converges reliably. While fully-connected architectures require more iterations (approx. 600) than single-connected ones (approx. 100), the convergence is monotonic and stable.
• Scalability: As the number of Reflective Elements (REs) increases, the proposed method maintains its performance advantage, demonstrating scalability.

5. Significance

This work bridges the gap between theoretical optimization and practical hardware implementation for BD-RIS.

• Practicality: By enforcing reciprocity, the solution is directly applicable to passive, low-cost hardware, which is crucial for the widespread deployment of RIS technology.
• Efficiency: The use of manifold optimization with closed-form gradients provides a computationally efficient alternative to heavy iterative methods like ADMM, making real-time or near-real-time configuration feasible.
• Performance: It demonstrates that directly optimizing for sum-rate (rather than decoupled interference nulling) yields superior system performance, validating the necessity of advanced signal processing in BD-RIS design.

In conclusion, the paper presents a robust, mathematically rigorous, and practically viable framework for designing reciprocal BD-RIS scattering matrices, setting a new standard for sum-rate maximization in next-generation wireless networks.