Microstrip Patch Antenna Modeling With Knowledge-Based Neural Networks

Date

2023-8-01

Author

Saçın, Ekin Su

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

251
views

514
downloads

Cite This

Designing and optimizing microwave devices require significant computational resources, particularly when dealing with intricate stack-up structures. Utilizing machine learning techniques based on neural networks as surrogate models for microwave device design and optimization offers the advantage of reducing computational resource usage. In this thesis, a spectral transposed convolutional neural network was developed for the design of a microstrip patch antenna on a package. The neural network was trained to predict the S11 parameters and gain of the antenna as a function of its geometrical parameters and material properties. The neural network provides scalable frequency resolution by upsampling the output dimension through transposed convolution layers. The antenna was intentionally designed without a predefined feeding network to allow the integration of versatile feeding techniques based on specific criteria. Weights of the neural network are updated using the Huber loss function, while the relative squared error (RSE) was adopted as the evaluation metric. To further enhance the performance of the neural network with a smaller dataset, knowledge-based regularization methods were implemented during the training process. These methods include integrating derivatives and spectral domain representation of the signal into the loss function, and magnitude regularization to minimize passivity violation. Moreover, the loss function of the neural network was improved by incorporating the cavity model of the patch antenna, allowing accurate prediction of the antenna performance while considering resonant frequencies of the relevant modes. This work demonstrated the remarkable capability of neural networks to accurately predict patch antenna performance.

Subject Keywords

Neural Network, Spectral Transposed Convolutional Neural Network, Microstrip Patch Antenna, Knowledge-Based Regularization, Cavity Model

URI

https://hdl.handle.net/11511/104880

Collections

Graduate School of Natural and Applied Sciences, Thesis