Deep Neural Network Architectures for Modulation Classification (1712.00443v3)

Abstract: In this work, we investigate the value of employing deep learning for the task of wireless signal modulation recognition. Recently in [1], a framework has been introduced by generating a dataset using GNU radio that mimics the imperfections in a real wireless channel, and uses 10 different modulation types. Further, a convolutional neural network (CNN) architecture was developed and shown to deliver performance that exceeds that of expert-based approaches. Here, we follow the framework of [1] and find deep neural network architectures that deliver higher accuracy than the state of the art. We tested the architecture of [1] and found it to achieve an accuracy of approximately 75% of correctly recognizing the modulation type. We first tune the CNN architecture of [1] and find a design with four convolutional layers and two dense layers that gives an accuracy of approximately 83.8% at high SNR. We then develop architectures based on the recently introduced ideas of Residual Networks (ResNet [2]) and Densely Connected Networks (DenseNet [3]) to achieve high SNR accuracies of approximately 83.5% and 86.6%, respectively. Finally, we introduce a Convolutional Long Short-term Deep Neural Network (CLDNN [4]) to achieve an accuracy of approximately 88.5% at high SNR.

• The paper introduces advanced DNN models, including CNN, ResNet, DenseNet, and a novel CLDNN, for autonomous modulation classification achieving up to 88.5% accuracy.
• It details enhancements with a deep architecture featuring four convolutional and two dense layers that overcome CNN limitations such as vanishing gradients.
• Utilizing the RadioML2016.10b dataset, the research effectively addresses real-world channel imperfections in wireless signal processing.

Deep Neural Network Architectures for Modulation Classification

This paper explores the application of deep learning techniques to wireless signal modulation recognition, an essential task for both signal detection and demodulation in wireless communication systems. As traditional modulation recognition methods often rely heavily on potentially inaccurate prior knowledge of signal and channel parameters, the authors address this limitation by using deep neural networks (DNNs) to facilitate autonomous modulation recognition.

The research builds upon a convolutional neural network (CNN) framework previously shown to outperform expert-based approaches, achieving approximately 75% accuracy in modulation recognition. By optimizing this framework, the authors enhance the architecture with four convolutional layers and two dense layers, reaching 83.8% accuracy at high signal-to-noise ratio (SNR). Additional architectures utilizing Residual Networks (ResNet) and Densely Connected Networks (DenseNet) yield accuracies of 83.5% and 86.6%, respectively. Notably, the paper introduces a Convolutional Long Short-term Deep Neural Network (CLDNN), which achieves 88.5% accuracy in high SNR scenarios.

The research utilizes the RadioML2016.10b dataset, which encompasses ten modulation types including eight digital and two analog modulations. This dataset accounts for real-world channel imperfections, providing a robust foundation for testing the aforementioned neural network architectures.

Highlighting the limitations of CNNs, the paper points to issues like vanishing gradients and accuracy degradation beyond certain network depths. To mitigate these, the authors implement ResNet and DenseNet architectures, which introduce shortcut paths to enhance feature propagation. The CLDNN further advances this field by exploiting the complementarity of CNNs, Long Short-Term Memory (LSTM) networks, and fully connected DNNs, benefiting from LSTM's ability to process time-dependent sequences.

Despite these advances, the paper acknowledges certain areas of misclassification within the CLDNN results, particularly between modulation types that are similar, such as QAM16/QAM64 and WBFM/AM-DSB. These challenges highlight the need for further optimization and possibly preprocessing of inputs to improve classification accuracy.

In summary, this research demonstrates significant improvement in modulation type recognition using advanced neural network architectures. The findings underscore the potential of integrating deep learning techniques in communication systems, facilitating more autonomous and accurate modulation recognition processes. Future work could explore the scalability of these architectures and their applicability to broader communication scenarios, as well as further refinement to address remaining misclassification issues.