Modulation recognition method of satellite communication based on CLDNN model

Modulation recognition as an important part of signal processing has been widely used in the field of satellite communication. Since the present existing modulation recognition method is still requires manual processing by professionals, automatic modulation recognition (AMR) is proposed. It has adaptive modulation capabilities to sense and learn environments and make corresponding adjustments. In this paper, we proposes a deep learning classification algorithm called Convolutional Long-Short Term Deep Neural Network (CLDNN) to implement AMR. This model integrates architectures of Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM) and deep neural networks (DNN) model. It was trained on the RadioML2016.10a dataset that composed of eleven commonly used modulation modes with different signal to noise ratio(SNR). The signal was generated in a real system using GNU radio to classify modulation. We adopt dropout instead of pooling operation to achieve higher recognition accuracy. In addition, comparisons with Residual Neural Network(ResNet) and Visual Geometry Group(VGG) models are presented. Experimental results demonstrate the significant performance advantage and application feasibility of the CLDNN model for AMR.

Automated summary

Automatic modulation recognition (AMR) using deep learning algorithms improves recognition accuracy by adapting to different modulation environments. The CLDNN model, trained on a dataset with various modulation modes and SNRs, shows significant performance advantages in comparison to other models such as ResNet and VGG.