Research On Deep Modulation Recognition

Modulation recognition aims to recognize the modulation format of the received signals and has shown great potential in both civilian and military wireless communications.Early works dedicated in likelihood based methods which is motivated by its optimality in Bayesian sense when the channel model and parameters are known to the receiver.However,likelihood based methods suffers from great complexity and relies on prior information of signal model,feature-based methods emerge as an alternative way for modulation recognition,providing a suboptimal performance and significantly reducing the complexity of modulation recognition algorithms,but hand-craft features need to be carefully designed in feature-based methods,which limits the modulation recognition performance in real applications.Due to the great success of deep learning in computer vision and natural language processing,deep learning based methods have been applied to modulation recognition and has achieved better performance compared with traditional methods.In this thesis,we give two points of analysis on deep modulation recognition.At first,we give a research on the robustness of deep modulation recognition in different environment.Deep modulation recognition has demonstrated high classification accuracy when a neural network is trained on largescale datasets.However,when applied in a different radio environment,it's performance can be significantly degraded due to the change of signal distributions.In this thesis,we propose incorporating adversarial domain adaptation method to adapt the modulation recognition to an unknown environment.Results show that,the proposed method can significantly improve the modulation recognition performance and achieve comparable performance with that of the model trained with sufficiently large labeled data in some environment.Then we give a research on the deep modulation recognition for multiple receive antennas.Most existing research of modulation recognition focuses on single receive antenna,with the development of modern communication systems,the need of high speed and large capacity in wireless communication makes it common for a radio receiver to equipped with multiple receive antennas.In this thesis,two end-to-end feature learning deep architectures are introduced for modulation recognition with multiple receive antennas.(1)Model based on multi-view convolutional neural network,by treating signals from different receive antennas as different views of a 3D object and designing the location and operation of view-pooling layer that are suitable for feature fusion of multi-antenna signals,multi-view convolutional neural network can be used for modulation recognition with multiple receive antennas.(2)Weight-learning convolutional neural network,which uses a weight-learning module to automatically learn the weights for feature combing of different receive antennas to adapt to feature combining with different signal characteristics.Results show that both end-to-end feature learning deep architectures outperform the existing algorithm,and the proposed weightlearning convolutional neural network achieves the best performance.