Research On Communication Signal Modulation Recognition Based On Deep Learning

Communication signal modulation identification is a process of identifying the modulation mode of the blind signala that received by a receiver.Modulation identification has a very wide range of application requirements in current radio communication.For example,in spectrum management,modulation identification research can enhance the management and use of frequency band resources.In cognitive radio,it can help to efficiently use spectrum resources.In the field of military communications,it has important significance for military battlefield applications such as communication investigation and electronic countermeasures.Therefore,no matter in civil or military fields,modulation identification is a very hot research direction.Modulation recognition usually has two modes: manual classification and automatic recognition based on feature extraction.The automatic identification method is also the machine learning method we often mention.The deep learning method is also a branch of the machine learning method.As a powerful branch of machine learning,deep learning is the layered intelligent learning method closest to the human brain.It can represent complex functional relationships by establishing a hierarchical model structure similar to the human brain and breaking through the limitations of shallow learning.The input data is extracted layer by layer from the bottom layer to the high layer,and finally the modeling of the complex application scenario is completed.In this paper,the application of deep learning algorithm in the field of modulation identification is a fusion of new technology and traditional research direction,which has important significance.In this paper,the convolutional neural network structure is used as the modulation recognizer model,the Rectified Linear Unit is used as the activation function,the cross-entropy is used as the loss function,and Adam is us-ed as the optimization method to implement the modulation recognition algorithm based on deep learning.Firstly,for the convolutional neural network structure used in this paper,we make modulated signal samples.The samples were produced using MATLAB scientific calculation software,which moves all modulated signal frequencies to the intermediate frequency to normalize the frequency and normalizes the signal amplitude with a zero-mean normalization algorithm.At different signal-to-noise ratios we generate enough data assamples.The construction and training of the network model is carried out in the deep learning framework called Keras,which is based on Tensorflow for computing backend.The optimal model parameters are obtained after comparing the performance of different convolutional layers and networks with different input lengths.At the end of the paper,the optimal model parameters are obtained based on experimental experience and the performance of the satellite modulation category and the ultrashort wave modulation category are tested under different carrier-to-noise ratios.Experiments show that the modulation recognizer based on convolutional neural network has good recognition performance for various modulation signals.Compared with the traditional modulation recognition method,it has the advantages of automatic extraction feature,good recognition performance and can recognize many types of modulation,which is of great value to the method innovation in the field of modulation recognition.