MetaFAP: Meta-Learning for Frequency Agnostic Prediction of Metasurface Properties

Rafid Umayer Murshed 1 {}^{\textbf{1}}, Md Shoaib Akhter Rafi 2 {}^{\textbf{2}}, Sakib Reza 1 {}^{\textbf{1}}, Mohammad Saquib 1 {}^{\textbf{1}}, and Ifana Mahbub 1 {}^{\textbf{1}} · 2025

Read MetaFAP: Meta-Learning for Frequency Agnostic Prediction of Metasurface Properties by Rafid Umayer Murshed 1 {}^{\textbf{1}}, Md Shoaib Akhter Rafi 2 {}^{\textbf{2}}, Sakib Reza 1 {}^{\textbf{1}}, Mohammad Saquib 1 {}^{\textbf{1}}, and Ifana Mahbub 1 {}^{\textbf{1}} (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-14866.txt.

Introduction

Rafid Umayer Murshed and co-authors wrote a 2025 paper on predicting how smart surfaces behave across radio bands. Such surfaces steer wireless waves and could reshape next generation networks. Their response shifts with frequency, which breaks designs tuned for one band when bands change on the fly.

Classic simulations are slow, and plain deep learning tends to overfit or carry poorly to new bands. Their frame learns from many frequency tasks at once, so it can adapt to fresh bands with scant data. Errors fell by a large factor against past methods, and each forecast lands in under a millisecond. The authors pitch the tool for real time tuning of these surfaces.

Catalog record

CREATORRafid Umayer Murshed 1 {}^{\textbf{1}}, Md Shoaib Akhter Rafi 2 {}^{\textbf{2}}, Sakib Reza 1 {}^{\textbf{1}}, Mohammad Saquib 1 {}^{\textbf{1}}, and Ifana Mahbub 1 {}^{\textbf{1}}
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE61 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-14866

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

MetaFAP: Meta-Learning for Frequency Agnostic Prediction of Metasurface Properties Rafid Umayer Murshed 1 {}^{\textbf{1}}, Md Shoaib Akhter Rafi 2 {}^{\textbf{2}}, Sakib Reza 1 {}^{\textbf{1}}, Mohammad Saquib 1 {}^{\textbf{1}}, and Ifana Mahbub 1 {}^{\textbf{1}} Abstract Metasurfaces, and in particular reconfigurable intelligent surfaces (RIS), are revolutionizing wireless communications by dynamically controlling electromagnetic waves. Recent wireless communication advancements necessitate broadband and multi-band RIS, capable of supporting dynamic spectrum access and carrier aggregation from sub-6 GHz to mmWave and THz bands. The inherent frequency dependence of meta-atom resonances degrades performance as operating conditions change, making real-time, frequency-agnostic metasurface property prediction crucial for practical deployment. Yet, accurately predicting metasurface behavior across different frequencies remains challenging. Traditional simulations struggle with complexity, while standard deep learning models often overfit or generalize poorly. To address this, we introduce MetaFAP ( Meta -Learning for F requency- A gnostic P rediction), a novel framework built on the meta-learning paradigm for predicting metasurface properties. By training on diverse frequency tasks, MetaFAP learns broadly applicable patterns. This allows it to adapt quickly to new spectral conditions with minimal data, solving key limitations of existing methods. Experimental evaluations demonstrate that MetaFAP reduces prediction errors by an order of magnitude in MSE and MAE while maintaining high Pearson correlations. Remarkably, it achieves inference in less than a millisecond, bypassing the computational bottlenecks of traditional simulations, which take minutes per unit cell and scale poorly with array size. These improvements enable real-time RIS optimization in dynamic environments and support scalable, frequency-agnostic designs. MetaFAP thus bridges the gap between intelligent electromagnetic systems and practical deployment, offering a critical tool for next-generation wireless networks. \corresp Corresponding author: Rafid Umayer Murshed (email: [email protected]). \authornote {IEEEkeywords} Deep Learning, Frequency-Agnostic Prediction, Meta-Learning, Metasurface, Reconfigurable Intelligent Surfaces, Reflectance, Transmittance 1 Introduction \IEEEPARstart Metasurfaces have emerged as a transformative technology in wireless communications, offering unprecedented control over the radio environment through the dynamic manipulation of electromagnetic waves [ 1 ] . In particular, reconfigurable intelligent surfaces (RIS) exploit sub-wavelength engineered meta-atoms to tailor the phase, amplitude, and polarization of incident signals [ 2 , 3 ] . This capability enables critical applications—including passive beamforming, interference suppression, energy-efficient propagation, and enhanced physical-layer security—that are essential for meeting the stringent performance requirements of both 5G Advanced and emerging 6G networks [ 4 , 5 ] . Despite the significant benefits, designing an effective RIS remains challenging due to the complex electromagnetic (EM) behavior of individual meta-atoms. The EM response—characterized by reflectance, transmittance, and absorptivity—is highly sensitive to variations in geometry, material properties, incident angles, and operating frequencies [ 6 ] . Traditional design methodologies rely heavily on full-wave EM simulations and analytical circuit models [ 7 ] . While full-wave solvers offer high accuracy by rigorously solving Maxwell’s equations, they are computationally expensive and impractical for exhaustive exploration or real-time optimization [ 3 ] . On the other hand, Analytical models offer faster evaluations by approximating metasurfaces as transmission line circuits or impedance sheets. However, they rely on simplifying assumptions that limit their accuracy over wide operating ranges, especially when incident angles and frequencies vary [ 6 , 7 ] . To mitigate these challenges, recent research has increasingly turned to data-driven approaches using machine learning (ML) and deep learning (DL) to develop surrogate models that can rapidly predict RIS behavior [ 8 ] . Supervised neural networks have been successfully employed to approximate the nonlinear mapping from metasurface configurations (including geometry, materials, and bias states) to their EM responses [ 9 ] . These networks deliver predictions in milliseconds and significantly accelerate the design cycle. Nevertheless, these conventional ML/DL methods are highly data-dependent and tend to perform poorly when extended beyond the frequency bands or design conditions present in their training sets. For example, a model trained on a narrow frequency band may fail to accurately predict metasurface behavior when applied to an unseen frequency range, a critical shortcoming in dynamic wireless environments. The demand for broadband or multi-band functionality further exacerbates this issue. While many metasurfaces are traditionally engineered for a single frequency band, next-generation applications increasingly require RIS that operate efficiently across diverse spectral regimes. These regimes range from sub-6 GHz to mmWave and even THz bands [ 10 ] . This expanded frequency coverage is crucial for advanced wireless systems that utilize dynamic spectrum access, carrier aggregation, and spectrum sharing to meet ever-growing capacity and latency requirements [ 11 , 1 ] . However, each meta-atom’s resonance is frequency dependent. A design optimal at 5 GHz may underperform at 28 GHz. This frequency shift may cause unpredictable phase errors and reduce beamforming gains. [ 9 ] . Moreover, real-time reconfiguration is becoming indispensable as RIS-based networks must adapt to changing user demands, environmental conditions, and regulatory constraints [ 12 , 13 ] . In such scenarios, redesigning or recalibrating a metasurface is impractical. Full-wave simulations are computationally expensive. This is especially true for large-scale deployments with hundreds or thousands of meta-atoms [ 14 ] . Instead, predictive models must be capable of extrapolating beyond the conditions observed during training or calibration. Achieving this level of frequency agility calls for fast, accurate, and robust modeling techniques. These techniques must generalize to unseen frequency bands, which ensures RIS operates effectively across wide bandwidths without frequent retraining. Recent deep-learning design methods show promise over limited spectral ranges. However, their accuracy drops outside the trained distribution. This highlights the need for more flexible, frequency-agnostic approaches [ 9 , 15 ] . Motivated by these challenges, our work explores the potential of meta-learning—“learning to learn”—to overcome the generalization limitations of conventional models. Meta-learning frameworks, such as Model-Agnostic Meta-Learning (MAML) [ 16 , 17 ] , aim to learn an initialization that can be rapidly adapted to new tasks with only a few training examples. In our context, each frequency band or RIS configuration is treated as a distinct task. It enables the meta-learned model to be fine-tuned quickly to accommodate new spectral conditions with minimal additional data. This approach endows our predictive model with “prior knowledge” that enhances its ability to generalize beyond the training distribution—a crucial capability for real-time network adaptation where rapid reconfiguration is imperative. In dynamic scenarios involving user mobility or rapid environmental changes, an RIS that can adjust its electromagnetic response nearly in real time can significantly enhance link reliability and throughput [ 11 ] . Moreover, by reducing the need for extensive field measurements and iterative tuning, our method offers a scalable, frequency-agnostic solution for next-generation wireless environments. In summary, our work bridges the gap between high-fidelity EM modeling and practical RIS design by integrating meta-learning into the predictive framework. To the best of our knowledge, this is the first study to deploy a meta-learning-based approach for metasurface property prediction. By leveraging MAML to model the reflectance and transmittance characteristics of RIS elements across diverse frequency ranges, our approach accelerates the design cycle and reduces computational overhead. It also robustly generalizes to unseen spectral conditions. This paves the way for adaptive and scalable RIS deployments in dynamic wireless networks [ 1 , 11 , 16 ] . The primary contributions of this paper are summarized as follows: 1. We introduce MetaFAP, the first meta-learning framework leveraging Model-Agnostic Meta-Learning (MAML), specifically

Read the complete text