This repo contains the code for an impedance calculator for a coplanar waveguide, implemented with the help of a plain vanilla neural network. This was originally a project idea that I had for a project for my microwaves assignment, but I didn't end up submitting it. I finished up the code a bit later.
As we know, a neural network can learn the properties of a function during training.
The theory for the coplanar waveguide is given on Microwaves101. They also have an online calculator to find the parameters of the waveguide. However, this website gives the formulae to calculate the impedance of the waveguide, alongwith a calculator.
Since this is a regression problem, where the network needs to learn elliptical functions, it was essential to generate a dataset to train the network. I generated the test set using the CPW in-built functions in the RF Toolbox in Matlab. The code for the same is in the Generate.m file.
I've also uploaded the ZData.csv file for the dataset.
Following packages are required:
python
TensorFlow
Keras
scikit-learn
numpy
pandas
tabulate
Code is pretty much self-explanatory. Just change the Predictions and epochs values to save your results in different files.
In general, results improved as number of epochs was increased. The network performace could be better with a deeper network. Average error is the average of the difference between the predicted and actual values of the impedance.
| Epochs | Average Error |
|---|---|
| 100 | 7.49 |
| 500 | 1.27 |
| 1000 | 0.57 |
This was inspired by this paper. However, the source code for the same was not released so I created the network using Keras.