Research On Modulation Recognition Algorithm Based On Deep Learning

Modulation recognition is an important research area of communication systems.Modulation recognition technology has been widely used in civil and military wireless communications,such as signal monitoring,information interception,interference identification and electronic warfare.And Modulation recognition is the basis of cognitive radio,spectrum sensing and other fields,so the research of this technology has never stopped and people pay more and more attention.The performance of most existing automatic modulation recognition algorithms is highly dependent on the choice of key features and classifiers.Whether the classifier matches the key features directly affects the effect of modulation recognition performance.In recent years,with the advent of the era of big data and computing resources are getting cheaper,deep learning technology has been continuously developed.Deep learning is a data-driven model,which learns complex feature representation from massive data directly.There are two typical deep learning networks: Convolutional networks and Long Short Term Memory which have achieved unprecedented results in multiple areas.In this paper,we resort to a deep learning approach to improve the robustness for modulation recognition.We consider the problem of automatic modulation recognition for either digital or analogue modulation types.The receiver does not have any knowledge about the modulation type of the received signal,and the objective of this paper is to develop a deep learning approach that can automatically recognize the modulation type of the received signal.With regard to the application of deep learning in modulation recognition,researchers have previously tried to use a convolutional neural network structure to identify modulated signals,the results show that the algorithm has a better classification effect than traditional classifiers.However,modulation recognition algorithms based on convolutional neural networks have not been able to make a good use of the timing characteristics of signals.Specifically,to efficiently explore the temporal and spatial correlation,we construct a deep neural network consisting of a convolutional neural network followed by a long short-term memory as the classifier.Firstly,the convolutional neural network is used to extract the spatial characteristics of the signal,and then the temporal correlation of the signal is extracted by the long short term memory networks.Our experimental results show that the proposed network architecture achieves a significantly improved compare to a convolutional neural network architecture that exploits only the spatial correlation of the signal and the traditional machine learning algorithm support vector machine.Besides,in order to reduce the parameters of the deep learning network,we proposed the back propagation algorithm based on iteratively reweighted least square for parameters updates in the networks.Experiments show that the proposed algorithm can get a more sparse parameters and at the same time the performance almost remain the same,which implies that this algorithm can reduce the calculation amount of the parameters in the networks and has application value to a certain extent.