Bi-directional ANN framework for forward and inverse design of microstrip antenna in RF applications
Year: 2026
Type: article
Abstract: In this paper, a bidirectional machine learning (ML)-based framework is introduced to predict the performance parameters such as \(|S_{11}|\) , VSWR and radiation pattern in two planes (XZ and YZ), radiation efficiency, and total efficiency, as well as to estimate the design variables including the patch and feed line dimensions (L, W, L1, and W1) and the dielectric constant ( \( \varepsilon _r \) ) of a quarter-wave transformer (QWT)-microstrip patch (MSP) antenna at an operating frequency of 3 GHz. A Forward Artificial Neural Network (F-ANN) is developed to predict performance parameters based on given design variables. At the same time, an inverse artificial neural network (I-ANN) is implemented to estimate design variables from the desired performance parameter. This technique involves designing, training, and optimizing an MSP antenna in real time. The F-ANN architecture has five input features and one output, whereas the I-ANN has three input features and five outputs. Model evaluation is done using the R-squared (R 2 ) and mean squared error (MSE). The F-ANN and I-ANN achieve R 2 values of 0.99 and minimal MSE values. The estimated design variables from inverse modeling are further re-verified using retro simulation and compared with simulated and predicted results; the developed models \(|S_{11}|\) and VSWR parameters are also validated with measured performance. Both F-ANN and I-ANN predict and estimate parameters with higher accuracy. As a comparative model, the prediction of the ANN framework is compared with two other regression-based models, namely Random Forest (RF) and Support Vector Machine (SVM). All the developed models are able to predict the performance with good prediction accuracy.
Source: Scientific Reports
Authors: Gayatri Routhu, Manas Sarkar, Rupesh Kumar
Language: English
Topic: Antenna Design and Analysis
Subfield: Aerospace Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7225005651