Classifying different signals modulations into their right modulations using different baseline classifiers and a CNN model
DeepSig Dataset: RadioML 2016.04B
Download Link: opendata.deepsig.io/datasets/2016.10/RML2016.10b.tar.bz2
Description: A synthetic dataset, generated with GNU Radio [1], consisting of 11 modulations. This is a variable-SNR dataset with moderate LO drift, light fading, and numerous different labeled SNR increments for use in measuring performance across different signal and noise power scenarios.
- It has 1,200,000 samples
- Each sample is presented using two vectors each of them has 128 elements.
- The data is split into 70% for training/validation and 30% for testing.
- 5% of the training and validation dataset for validation.
4 different features spaces are used and results are compared
- Raw time series as given (two channels)
- First derivative in time (two channels)
- Integral in time (two channels)
- Combinations of 1,2 and 3. (More channels)
- Logistic Regression Classifier
- Decision Tree
- Random Forest
- A fully connected dense layer:
- Non-linear function: Relu
- Optimizer: ADAM
- Early stopping
Architecture: [2]
- Input : 2 x 128
- Conv Relu : 64 x (1 x 3)
- Conv Relu : 16 x (2 x 3)
- Dense Relu : 128
- Dense Softmax : 10
- Output : 10
[1] T. O’shea, N. West. “Radio Machine Learning Dataset Generation with GNU Radio”, https://pubs.gnuradio.org/index.php/grcon/article/download/11/10/
[2] T. O’Shea, J. Corgan, and T. Clancy. “Convolutional Radio Modulation Recognition Networks” https://arxiv.org/pdf/1602.04105.pdf