Code for "Towards the Automatic Modulation Classification with Adaptive Wavelet Network".
Jiawei Zhang, Tiantian Wang, Zhixi Feng, and Shuyuan Yang
Xidian University
We conducted experiments on three datasets, namely RML2016.10a, RML2016.10b, and RML2018.01a.
| dataset | modulation formats | samples |
|---|---|---|
| RML2016.10a | 8 digital formats: 8PSK, BPSK, CPFSK, GFSK, PAM4, 16QAM, 64QAM, QPSK; 3 analog formats: AM-DSB，AM-SSB，WBFM | 220 thousand (2×128) |
| RML2016.10b | 8 digital formats: 8PSK, BPSK, CPFSK, GFSK, PAM4, 16QAM, 64QAM, QPSK; 3 analog formats: AM-DSB，WBFM | 1.2 million (2×128) |
| RML2018.01a | 19 digital formats: 32PSK, 16APSK, 32QAM, GMSK, 32APSK, OQPSK, 8ASK, BPSK, 8PSK, 4ASK, 16PSK, 64APSK, 128QAM, 128APSK, 64QAM, QPSK, 256QAM, OOK, 16QAM; 5 analog formats: AM-DSB-WC, AM-SSB-WC, AM-SSB-SC, AM-DSB-SC, FM, | 2.5 million (2×1024) |
The datasets can be downloaded from the DeepSig. Please extract the downloaded compressed file directly into the ./data directory, and keep the file name unchanged. The final directory structure of ./data should is shown below:
data
├── GOLD_XYZ_OSC.0001_1024.hdf5
├── RML2016.10a_dict.pkl
└── RML2016.10b.dat
We provide pre-trained models on three datasets, which can be downloaded from Google Drive or Baidu Netdisk. Please extract the downloaded compressed file directly into the ./checkpoint directory.
- Python >= 3.6
- PyTorch >=1.7
This version of the code has been tested on Pytorch==1.8.1.
Run the following commands to train the AWN. (<DATASET> in {2016.10a, 2016.10b, 2018.01a}).
python main.py --mode train --dataset <DATASET>
The YAML configs for three datasets lies in ./config.
After running the command, a new directory <DATASET>_$ will be created in the ./training directory, and ./models, ./result, ./log directories will be created under ./<DATASET>_$. The trained model will be saved to ./models, and the training logs will be saved to ./log. The loss, accuracy, and learning rate changes during training and validation will be plotted together in ./result.
After the training is completed, an testing on the test set will be performed automatically, which can be referred to in the Evaluation section.
Run the following command to evaluate the trained AWN:
python main.py --mode eval --dataset <DATASET>
After running the command, a new directory <DATASET>_$ will be created in the ./inference directory, which is consistent with the Training section. The overall accuracy, macro F1-score, and Kappa coefficient on the test set will be displayed in the terminal. The test logs will be saved to ./log, and the accuracy curve with respect to SNRs and confusion matrix will be saved to ./result.
If you have further analysis needs, we recommend modifying the Run_Eval() function to directly save the raw data, such as Confmat_Set.
We provide an additional mode to visualize the decomposition of the feature maps by the adaptive lifting scheme, which can be called by the following command:
python main.py --mode visualize --dataset <DATASET>
Similar to Evaluation, the plotted figures are stored in ./result in the form of .svg.
Some of the code is borrowed from DAWN. We sincerely thank them for their outstanding work.
This code is distributed under an MIT LICENSE. Note that our code depends on other libraries and datasets which each have their own respective licenses that must also be followed.
Please consider citing our paper if you find it helpful in your research:
@ARTICLE{10058977,
author={Zhang, Jiawei and Wang, Tiantian and Feng, Zhixi and Yang, Shuyuan},
journal={IEEE Transactions on Cognitive Communications and Networking},
title={Towards the Automatic Modulation Classification with Adaptive Wavelet Network},
year={2023},
doi={10.1109/TCCN.2023.3252580}
}
Contact at: zjw AT stu DOT xidian DOT edu DOT cn