AI-Driven Optimization of Wave-Controlled Reconfigurable Intelligent Surfaces

Abstract: A promising type of Reconfigurable Intelligent Surface (RIS) employs tunable control of its varactors using biasing transmission lines below the RIS reflecting elements. Biasing standing waves (BSWs) are excited by a time-periodic signal and sampled at each RIS element to create a desired biasing voltage and control the reflection coefficients of the elements. A simple rectifier can be used to sample the voltages and capture the peaks of the BSWs over time. Like other types of RIS, attempting to model and accurately configure a wave-controlled RIS is extremely challenging due to factors such as device non-linearities, frequency dependence, element coupling, etc., and thus significant differences will arise between the actual and assumed performance. An alternative approach to solving this problem is data-driven: Using training data obtained by sampling the reflected radiation pattern of the RIS for a set of BSWs, a neural network (NN) is designed to create an input-output map between the BSW amplitudes and the resulting sampled radiation pattern. This is the approach discussed in this paper. In the proposed approach, the NN is optimized using a genetic algorithm (GA) to minimize the error between the predicted and measured radiation patterns. The BSW amplitudes are then designed via Simulated Annealing (SA) to optimize a signal-to-leakage-plus-noise ratio measure by iteratively forward-propagating the BSW amplitudes through the NN and using its output as feedback to determine convergence. The resulting optimal solutions are stored in a lookup table to be used both as settings to instantly configure the RIS and as a basis for determining more complex radiation patterns.

• The paper demonstrates an AI-driven framework that leverages neural networks and genetic algorithms to predict and optimize radiation patterns in RIS.
• The methodology employs a multilayer perceptron optimized with a genetic algorithm to address complex electromagnetic modeling challenges.
• The integrated use of simulated annealing and adaptive lookup tables accelerates beamforming, achieving improved signal-to-leakage-plus-noise ratio (SLNR).

AI-Driven Optimization of Wave-Controlled Reconfigurable Intelligent Surfaces

Introduction

This paper presents an AI-driven approach to optimize wave-controlled Reconfigurable Intelligent Surfaces (RIS) using machine learning techniques. The methodology focuses on creating an input-output mapping between the biasing standing wave (BSW) amplitudes and the resulting radiation pattern sampled by the RIS, utilizing a neural network (NN) optimized by a genetic algorithm (GA). This approach circumvents the inherent complexities of modeling RIS behavior, such as device non-linearities and frequency dependencies.

RIS and Signal Models

The studied architecture implements a wave-controlled RIS utilizing varactor diodes controlled via biasing transmission lines. The system design, which accommodates electromagnetic and circuit model challenges, involves uniform linear arrays of RIS elements powered by BSWs. These BSWs are sampled to provide DC bias voltages that alter the reflection coefficients of the RIS elements. Signal models used in this paper include line-of-sight (LoS) channels, catering to both desired signals and noise. The signal-to-leakage-plus-noise ratio (SLNR) serves as the key performance metric, guiding the optimization.

Machine Learning Model Design

The NN model proposed in this paper is a Multilayer Perceptron (MLP) optimized using a genetic algorithm to minimize the mean squared error (MSE) between the predicted and sampled radiation patterns. The NN is tasked with generalizing the complex nonlinear relationship between the BSW amplitudes and the resulting power pattern. Extensive random dataset generation aids in training the NN, emphasizing input diversity within realistic operational limits.

Genetic Algorithm for NN Optimization

The genetic algorithm optimizes the MLP architecture by evolving candidate solutions based on their performance in predicting accurate radiation patterns. The GA framework incorporates selection, crossover, and mutation mechanisms to converge to an optimal model architecture capable of universal function approximation across the diverse range of input data.

Simulated Annealing for Optimization

Simulated Annealing (SA) is employed for offline optimization of beam patterns through the NN. Given a desired radiation pattern, SA iteratively adjusts the BSW amplitudes by controlling the temperature parameter to enhance the SLNR. By combining NN inference with SA, the approach enables configuration of complex radiation patterns without requiring channel state information (CSI).

Lookup Table for Efficient Beamforming

An adaptive lookup table stores optimized BSW configurations, expediting the configuration of standard and complex beam patterns. This approach leverages precomputed solutions to accelerate convergence and provide baselines for new optimizations. This method improves SLNR results by adapting to both direct and interpolated power directionality requirements.

Implications and Future Directions

The novel integration of NN, GA, and SA in the proposed RIS optimization framework demonstrates significant potential for enhancing wireless communication systems. It alleviates the dependency on detailed electromagnetic modeling and extensive environmental characterization. Future work could explore the extension of this methodology to other deployments and explore different machine learning architectures to further improve efficiency and adaptability in dynamic environments.

Conclusion

The AI-driven optimization framework for wave-controlled RIS addresses critical challenges in metasurface deployment by leveraging data-driven strategies. The combination of NN-based modeling and metaheuristic optimization provides a robust mechanism for configuring RIS, evidencing superior performance, particularly with the integration of a lookup table for complex configurations. The proposed method exemplifies a scalable and adaptive approach for real-time wireless network optimization using AI technologies.