This repository hosts a simple demonstration of a deep learning approach for the inverse design of patch antennas. The goal is to explore energy-efficient designs and to significantly reduce simulation cost compared to conventional methods.
The following papers have been published in relation to this repository:
Additionally, we have another video that demonstrates the inverse design of reflective metasurfaces. You can watch it here.
The information about the Python version, interpretation of the prediction results, and the simulation validation section can be found in the issues.
The reference design for this project is from Professor K. L. Wong's book "Compact and Broadband Microstrip Antennas", page 237, featuring a U-slot patch antenna, as shown below for reference.
Necessary comments have been added to the code.
To configure training and prediction for CPU or GPU, modify line 19 in train.py
and predict.py
:
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
for CPU-based training and prediction.os.environ["CUDA_VISIBLE_DEVICES"] = "0"
for GPU-based training and prediction. (Requires additional installation of CUDA and cuDNN.)
When using CPU, adjust the count
value at line 168 in predict.py
to control the frequency of saving predictions (dictated by available memory). The total number of prediction samples is approximately 2 million.
For GPU usage, employ predict-gpu.py
to significantly enhance the prediction output rate, offering a 500x speedup compared to the CPU version.
The requirements.txt
file has been updated to include only four essential libraries for this project:
- matplotlib==3.8.4
- numpy==1.26.4
- tensorflow-gpu==2.10.0
- scipy==1.13.0
To ensure wider accessibility, the train.py
and predict.py
scripts have been updated to support training using only the CPU. This change allows users without a GPU to run the training and prediction processes. The code now defaults to CPU training, making it convenient for users who prefer or need to use CPU resources.
Please note that while tensorflow-gpu
is still listed as a requirement, the updated code will gracefully fallback to CPU training if a GPU is not available.
This repository presents a novel inverse design methodology for patch antennas using deep learning techniques. The project encompasses several key components:
- Data Generation: Generation of a comprehensive dataset for training the deep learning model.
- Data Preprocessing: Preprocessing of the dataset to ensure high-quality input for the network.
- Network Training & Testing: Implementation and training of the deep learning model, followed by rigorous testing to validate its performance.
- Prediction: Utilization of the trained model to predict optimal patch antenna designs.
- Inverse Design: Process of designing patch antennas based on the predictions from the deep learning model.
The version of python is 3.9.18.
The detailed steps for the environmental configuration will be updated soon. Please stay tuned for the latest information.
To get started with this simple demonstration of inverse design for patch antennas, follow these steps:
-
Clone the Repository Clone this repository to your local machine using the following command:
git clone https://github.com/youxch/Inverse-Design-of-Patch-Antennas.git
-
Install Dependencies Install all necessary dependencies by running:
pip install -r requirements.txt
-
Data Preparation The training dataset is prepared through automated simulations controlled by code, resulting in the
train_data.txt
file which is properly formatted with the structural parameters and simulation results for patch antennas. -
Model Training Run the
train.py
script to preprocess the training data and train the Multilayer Perceptron (MLP) network:python train.py
The trained model weights will be saved as
saved-model-5000.h5
, with ‘5000’ indicating the number of iterations or a training metric. -
Model Prediction and Inverse Design Use the
predict.py
script to make predictions and perform inverse design.python predict.py
or
python predict-gpu.py
Ensure that the trained model weights file saved-model-5000.h5 is available before running predictions.
For any questions or suggestions, please open an issue or directly contact the maintainers.
This project is licensed under the MIT License - see the LICENSE.md file for details.
If you use the code provided in this repository, please cite the following papers:
-
You, X. C., & Lin, F. H. (2024). Inverse Design of Reflective Metasurface Antennas Using Deep Learning from Small-Scale Statistically Random Pico-Cells. Microwave and Optical Technology Letters, 2024, 66(2), e34068.
-
You, X. C., & Lin, F. H. (2023). Energy Efficient Design of Low-Profile Wideband Microstrip Patch Antennas Using Deep Learning. In 2023 International Conference on Microwave and Millimeter Wave Technology (ICMMT), Qingdao, China, 2023, pp. 1-3.
Thank you for your cooperation in acknowledging the original work.