Predictive Forward- Reverse-Hybrid Modeling of Microstrip Antenna Using Machine Learning Based Regression and Classification Techniques
Loading...
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Department of Electronics and Communication Engineering, NITS
Abstract
This Ph.D. thesis presents the design and development of a new class of printed microstrip
antennas optimized for space and satellite communication applications. With the rapid
advancement of intelligent computational techniques, particularly machine learning (ML),
conventional approaches to antenna design, analysis, and optimization have been significantly
improved. In this framework, the thesis proposes an integrated ML-based modeling strategy that
combines forward, inverse, and hybrid methodologies utilizing both regression and
classification algorithms. The research focuses on accurately predicting key electrical
parameters—such as resonance, cut-off, and notched frequencies—and estimating their
corresponding geometrical configurations for various antenna architectures, including single
port, dual-port, and four-port models. These antenna designs are tailored to operate efficiently
across the S-band to X-band frequency spectrum, meeting the high-performance requirements of
contemporary satellite communication systems. The forward modelling phase, multiple
supervised regression algorithms-such as Support Vector Regression (SVR), Gradient Boosted
Decision Trees (GBDT), Ridge, Lasso, Elastic Net, and Deep neural networks (DNNs)-are
implemented to predict frequency characteristics from antenna geometries. Conversely, in the
reverse modelling approach, the algorithms predict optimal geometrical parameters from known
electrical specifications. Hybrid models are further developed by integrating forward and
reverse predictions to improve reliability and design efficiency.
For antennas incorporating U-shaped slots and Electromagnetic Band Gap (EBG) structures,
the ML models effectively learn the complex nonlinear relationships between inputs and
outputs. Models are evaluated using data generated from antenna simulator in Ansys HFSS, and
in selected cases, validated through fabricated prototypes. Deep learning architectures- artificial
neural networks (ANN), convolution neural network (CNN), and recurrent neural network
(RNN)-are employed to model more complex multi-output problems involving dual- and quad
port antennas, achieving notched frequency responses up to 8.2 GHz. Additionally, the thesis
explores classification tasks for recognizing antenna types such as circular, triangular, and
rectangular patches using Random Forest and K-Nearest Neighbors (KNN) classifiers. The
classifiers achieve accuracies exceeding 97.9%, confirming the robustness of the proposed
models.
The performance of all machine learning models is meticulously assessed using a
comprehensive set of evaluation metrics. For regression tasks, metrics such as Mean Absolute
Error (MAE), Root Mean Square Error (RMSE), Mean Square Error (MSE), R² score, Mean
Squared Logarithmic Error (MSLE), and Root Mean Squared Logarithmic Error (RMSLE) are
employed. For classification models, evaluation is conducted using accuracy, precision, recall,
F1-score, Matthews Correlation Coefficient (MCC), and Cohen’s kappa, ensuring a rigorous
and holistic performance analysis.
The results indicate that ensemble methods and deep learning architectures significantly
outperform conventional algorithms, confirming the effectiveness of machine learning in
predictive antenna modeling. This study introduces a novel data-driven approach that
accelerates the antenna design cycle and improves the functional performance of microstrip
antennas, particularly for satellite communication applications spanning low, medium, and
geostationary orbits.