A novel design of RIS unit cell with ML-PSO based varactor diode tuning for optimal reflection characteristics

Abstract

This paper introduces a novel reconfigurable intelligent surface (RIS) unit cell capable of frequency-dependent dynamic phase and polarization control. The structure integrates four SMV series varactor diodes, enabling continuous tuning of both phase and axial ratio across multiple frequency bands without requiring physical modification. Full-wave simulations confirm the reconfigurability of the structure across multiple frequency bands. The proposed unit cell demonstrates wideband phase tunability, ranging from $-78.66°$ to $134.31°$ at 3.2 GHz, $-146.87°$ to $-177.571°$ at 4.7 GHz, $-158.54°$ to $-192°$ at 5.0 GHz, $-163.41°$ to $239.73°$ at 5.15 GHz and $-156.99°$ to $+179.9°$ at 6.75 GHz. The reflection amplitude remains high (better than –3 dB) across the lower bands (3.2–5.15 GHz), whereas certain states at 6.75 GHz exhibit increased loss, primarily due to tuning asymmetry. At 3.2 and 6.75 GHz, the unit cell demonstrates significant polarization agility, where the axial ratio varies from 1 dB to 65 dB under different biasing conditions, allowing transitions between circular, elliptical, and linear polarization states. Electromagnetic (EM) behavior is validated through co-simulation, and an equivalent circuit model is developed to replicate the varactor-based tuning response. Thereby, a machine learning (ML)-Particle Swarm Optimization (PSO)-based framework is proposed to optimize the reflection characteristics by tuning the input to the varactor diodes. The proposed methodology leverages ML models, including Random Forest, XGBoost, Gradient Boost, Neural Networks, and Circular Regression, to estimate the values of optimal fitness functions. Thereby, the best-performing ML model is utilized to solve a multi-objective optimization problem using PSO, leading to a performance improvement of phase of 16.78%, 8.77%, 7.3%, 3.31% and 1.6% at each frequency respectively. Consequently, the proposed integration of EM simulation, circuit modeling, and data-driven optimization offers a scalable design methodology for RIS, which has been proven with a design of 4 × 3 RIS array and validated with fabrication and testing in an anechoic chamber.