Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Statistical Multiport-Network Modeling and Efficient Discrete Optimization of RIS

Cheima Hammami, Luc Le Magoarou, Philipp del Hougne

TL;DR

The paper tackles the challenge of optimizing reconfigurable intelligent surfaces under mutual coupling and 1-bit programmability by introducing a reverberation-chamber inspired statistical generator for multi-port-network parameters with a controllable MC knob. It then compares a range of optimization methods, from model-agnostic dictionary search and coordinate descent to model-based TABP and GA, across MC regimes using a SISO channel-gain objective. The key finding is that coordinate descent on a 1-bit RIS with non-negligible MC achieves the best balance of performance, speed, and memory, while model-based TABP performs well only when a calibrated model is available. These insights provide practical guidance for RIS prototype optimization and extend to other reconfigurable wave systems, with potential extensions to broadband scenarios and richer correlations.

Abstract

This Letter addresses the physics-consistent optimization of reconfigurable intelligent surfaces (RISs) with mutual coupling (MC) and 1-bit-programmable RIS elements. This combination of constraints is typical of current prototypes but unexplored in theoretical work. First, we present a simple statistical generator for multiport-network-theory (MNT) parameters of rich-scattering, RIS-parametrized channels. We account for reciprocity, passivity, and coherent backscattering; then, we add a simple hyper-parameter to control the MC strength. Second, we benchmark model-agnostic (dictionary search, coordinate descent, genetic algorithm) and model-based (temperature-annealed back-propagation) strategies under varying MC, with and without intelligent initialization. Except when MC is negligible, coordinate descent with random initialization offers the best trade-off in performance, runtime, and memory. Our insights can guide wireless practitioners who optimize RIS prototypes and other reconfigurable wave systems.

Statistical Multiport-Network Modeling and Efficient Discrete Optimization of RIS

TL;DR

The paper tackles the challenge of optimizing reconfigurable intelligent surfaces under mutual coupling and 1-bit programmability by introducing a reverberation-chamber inspired statistical generator for multi-port-network parameters with a controllable MC knob. It then compares a range of optimization methods, from model-agnostic dictionary search and coordinate descent to model-based TABP and GA, across MC regimes using a SISO channel-gain objective. The key finding is that coordinate descent on a 1-bit RIS with non-negligible MC achieves the best balance of performance, speed, and memory, while model-based TABP performs well only when a calibrated model is available. These insights provide practical guidance for RIS prototype optimization and extend to other reconfigurable wave systems, with potential extensions to broadband scenarios and richer correlations.

Abstract

This Letter addresses the physics-consistent optimization of reconfigurable intelligent surfaces (RISs) with mutual coupling (MC) and 1-bit-programmable RIS elements. This combination of constraints is typical of current prototypes but unexplored in theoretical work. First, we present a simple statistical generator for multiport-network-theory (MNT) parameters of rich-scattering, RIS-parametrized channels. We account for reciprocity, passivity, and coherent backscattering; then, we add a simple hyper-parameter to control the MC strength. Second, we benchmark model-agnostic (dictionary search, coordinate descent, genetic algorithm) and model-based (temperature-annealed back-propagation) strategies under varying MC, with and without intelligent initialization. Except when MC is negligible, coordinate descent with random initialization offers the best trade-off in performance, runtime, and memory. Our insights can guide wireless practitioners who optimize RIS prototypes and other reconfigurable wave systems.

Paper Structure

This paper contains 12 sections, 4 equations, 3 figures, 2 algorithms.

Table of Contents

  1. Introduction
  2. System Model
  3. High-Fidelity Physics-Consistent MNT Model
  4. Low-Fidelity Affine Models
  5. Statistical MNT Model Parameter Generation
  6. MC and Binary-Tunability Aware Optimization
  7. Dictionary Search (DS)
  8. Coordinate Descent (CD)
  9. Temperature-Annealed Back-Propagation (TABP)
  10. Genetic Algorithm (GA)
  11. Performance Evaluation
  12. Conclusion

Figures (3)

  • Figure 1: MNT system model of an RIS-parametrized radio environment.
  • Figure 2: SISO channel gain $|S_{21}|^2$ for the considered methods as a function of $M$ (horizontal axis) and $\mu_\mathrm{n}$ (panels), averaged over 1500 realizations.
  • Figure 3: Impact of $M$ and $\mu_\mathrm{n}$ on the methods' convergence.