Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Antenna Near-Field Reconstruction from Far-Field Data Using Convolutional Neural Networks

Sahar Bagherkhani, Jackson Christopher Earls, Franco De Flaviis, Pierre Baldi

TL;DR

Reconstructing near-field distributions from far-field data is an ill-posed inverse problem because evanescent components are not present in the far field. The authors train a CNN to learn FF-NF mappings from a large HFSS-generated dataset of a $4\times4$ microstrip patch antenna array at 30 GHz, with NF and FF grids defined by $Z_{NF}=4.5\lambda$ and $Z_{FF}=25\lambda$ and phase excitations sampled over $4^{16}$ configurations. The model achieves a training MSE of 0.0199 and a test MSE of 0.3898 (NRMSE ≈ $8.66\times 10^{-5}$), with 10-fold cross-validation showing stable performance and inference times under one minute. This data-driven approach provides a practical alternative to analytical FF-NF transforms for antenna diagnostics and electromagnetic interference analysis, potentially accelerating computational EM workflows.

Abstract

Electromagnetic field reconstruction is crucial in many applications, including antenna diagnostics, electromagnetic interference analysis, and system modeling. This paper presents a deep learning-based approach for Far-Field to Near-Field (FF-NF) transformation using Convolutional Neural Networks (CNNs). The goal is to reconstruct near-field distributions from the far-field data of an antenna without relying on explicit analytical transformations. The CNNs are trained on paired far-field and near-field data and evaluated using mean squared error (MSE). The best model achieves a training error of 0.0199 and a test error of 0.3898. Moreover, visual comparisons between the predicted and true near-field distributions demonstrate the model's effectiveness in capturing complex electromagnetic field behavior, highlighting the potential of deep learning in electromagnetic field reconstruction.

Antenna Near-Field Reconstruction from Far-Field Data Using Convolutional Neural Networks

TL;DR

Reconstructing near-field distributions from far-field data is an ill-posed inverse problem because evanescent components are not present in the far field. The authors train a CNN to learn FF-NF mappings from a large HFSS-generated dataset of a microstrip patch antenna array at 30 GHz, with NF and FF grids defined by and and phase excitations sampled over configurations. The model achieves a training MSE of 0.0199 and a test MSE of 0.3898 (NRMSE ≈ ), with 10-fold cross-validation showing stable performance and inference times under one minute. This data-driven approach provides a practical alternative to analytical FF-NF transforms for antenna diagnostics and electromagnetic interference analysis, potentially accelerating computational EM workflows.

Abstract

Electromagnetic field reconstruction is crucial in many applications, including antenna diagnostics, electromagnetic interference analysis, and system modeling. This paper presents a deep learning-based approach for Far-Field to Near-Field (FF-NF) transformation using Convolutional Neural Networks (CNNs). The goal is to reconstruct near-field distributions from the far-field data of an antenna without relying on explicit analytical transformations. The CNNs are trained on paired far-field and near-field data and evaluated using mean squared error (MSE). The best model achieves a training error of 0.0199 and a test error of 0.3898. Moreover, visual comparisons between the predicted and true near-field distributions demonstrate the model's effectiveness in capturing complex electromagnetic field behavior, highlighting the potential of deep learning in electromagnetic field reconstruction.

Paper Structure

This paper contains 11 sections, 3 figures.

Table of Contents

  1. Introduction
  2. Far-Field to Near-Field Transformation
  3. Electromagnetic Modeling and Dataset Generation
  4. Architecture of the Convolutional Neural Network
  5. Convolutional Layer
  6. ReLU Activation Function
  7. Average Pooling Layer
  8. Linear Layer
  9. Model Training
  10. Results
  11. Conclusion

Figures (3)

  • Figure 1: Antenna array with near-field and far-field sampling grids used for dataset generation ($Z_{\text{NF}}=4.5\lambda$, $Z_{\text{FF}}=25\lambda$).
  • Figure 2: Comparison of predicted and ground truth total near-field magnitude distributions for three representative examples. Each row corresponds to one example, with the ground truth shown in the left column and the CNN-predicted field in the right.
  • Figure 3: Error bar resulting from 10-fold cross-validation.