Convolutional Radio Modulation Recognition Networks

In an arxiv pre-publication report out today, Johnathan Corgan and I study the adaptation of convolutional neural networks to the task of modulation recognition in wireless systems.   We use a relatively simple two layer convolutional network followed by two dense layers, a much smaller network than required for tasks such as ImageNet/ILVC.


We demonstrate that blind time domain feature learning can perform extremely well at the task of modulation classification, achieving a very high accuracy rate on both clean and noisy data sets.


As we compare the classifier performance across a wide range of signal to noise ratios, we demonstrate that it outperforms a number of more traditional expert classifiers using zero-delay cumulant features by a large margin.


While this is preliminary work, we think the results are exciting and that many additional promising results will come from the marriage of software radio and deep learning fields.

For much more detail on these results, please see our paper!

6 thoughts on “Convolutional Radio Modulation Recognition Networks

  1. Dear author,
    My name is Yanlun Wu and I am a graduate student in Electrical Engineering at the University of China, Chengdu. I found your paper on “Convolutional Radio Modulation Recognition Networks” very insightful and would like to replicate the results you have reported. I have found the dataset and your GitHub code but it is not the same with your paper, your paper used 96000 examples for training and 64000 for test, the classifier performance vs SNR is higher than the public code in GitHub. If possible, could you please share the source code with me?

  2. The approach and results are impressive. However, have you compare it with feature based method such as those using cyclic cumulants or the likelihood based method? Meanwhile, if it is possible to share your matlab simulation codes?

  3. Did the Keras code you used for training on this work get posted to github recently? I recall seeing something about it at GRCon. What’s the URL?

  4. 1.Could you give a accurate result with numbers, but not the figure 7 in your paper?Because I want to compare your result with mine .
    2.Could you explain the sentence”After training, we achieve roughly a 87.4% classification accuracy across all signal to noise ratios on the test dataset” in your paper? is 87.4% average accuracy at all SNR?

  5. Hi Prof. O’Shea,
    Our group is doing similar research on AMC using neural network. We managed to reproduce similar accuracy result using CLDNN (We discussed about this in Asilomar Conference). But our classification result for QAM signals are not good, while I see in the confusion matrix that you managed to achieve perfect classification between QAM16 and QAM64. Could you please let us know the network architecture you used for the result you presented in the paper?

Leave a Reply

Your email address will not be published. Required fields are marked *