Deep neural network architectures for modulation classification
2017 51st Asilomar Conference on Signals, Systems, and Computers
In this work, we investigate the value of employing statistical machine learning in general and deep learning in particular for the task of wireless signal modulation recognition. Recently in O'Shea & Corgan (2016), 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
... eeds that of expert-based approaches. We tested the architecture of O'Shea & Corgan (2016) and found it to achieve an accuracy of approximately 75% of correctly recognizing the modulation type. We 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, He et al. (2015) ) and Densely Connected Networks (DenseNet, Huang et al. (2016)) and achieve high SNR accuracies of approximately 83.5% and 86.6%, respectively. We achieve the best accuracy of approximately 88.5% at high SNR by applying a Convolutional Long Short-term Deep Neural Network (CLDNN, Sainath et al. (2015)) to the modulation classification task. We then focus on the modulation types of QAM16 and QAM64 that were not well learned by neural networks and explore different statistical machine learning methods using expert features to classify them. We achieve an accuracy of 72 % in classifying QAM16 and QAM64 signals at high SNR using the combination of time and a high-order cumulant as expert feature.