Year: 2025
Type: article
Abstract: Objectives: Designing and optimizing a compact 2.4 GHz patch antenna through AI-based models in predicting return loss (S11) for facilitating effective performance estimation in IoT applications. Methods: Parametric analysis was performed to evaluate the effects of slot/feed dimensions on antenna performance. A bespoke dataset of 55,053 samples was created via CST simulations. Various regression models (e.g., XGBoost, KNN, MLP) were trained and compared for S11 prediction. Data preprocessing, train/test splitting, and statistical metrics (MAE, RMSE, R2) were utilized to determine the most accurate model for performance prediction and inverse optimization. Findings: The suggested AI-enhanced method greatly curtailed design time and enhanced prediction accuracy as compared to conventional parametric sweeps. KNN and MLP models performed better than XGBoost with RMSE of 0.42 and 0.51, and R2 of 0.9866 and 0.9803 respectively, for maintaining high correlation between predicted and simulated S11 values. Final antenna design was achieved for resonance at 2.45 GHz with a return loss of less than - 40 dB, showing perfect impedance matching. Sensitivity analysis revealed patch width and feed length as most dominant features. The correlation matrix pointed out that return loss relies on complicated nonlinear relationships, validating the application of AI models. These findings vary from typical findings through the provision of an interpretable and efficient solution based on machine learning. In comparison to previous research, our methodology ensures quicker optimization and the capability to carry out inverse design, rendering it suitable for time-critical, miniaturized IoT devices. The amalgamation of AI hence presents a scalable and precise avenue to antenna engineering. Novelty: Unlike previous studies limited to small datasets or single algorithms, this work benchmarks six ML algorithms on over 55,000 simulated antenna samples. The results show that KNN and MLP achieved the highest prediction accuracy (R2 > 0.98), outperforming traditional regression and advanced ensemble methods such as XGBoost, making it a highly scalable and efficient tool for inverse antenna design for IoT applications. Keywords: Patch Antenna, Return Loss Prediction, Machine Learning, IA, IoT Applications
Authors: Ameni Mersani, Kawther Mekki, Omran Ncibi
Language: English
Topic: Antenna Design and Analysis
Subfield: Aerospace Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4414101926