Research On Intelligent Identification Technology For The Modulation Modes Of Communication Signals

In order to improve the transmission efficiency and reliability of communication system,different modulation modes are used to convert the modulation signals into modulated signals,which are suitable for the transmission in communication channels.In the informative battlefield,modulation recognition technology is widely used in satellite measurement and control countermeasures,reconnaissance jamming subsystem,missile,airborne,shipborne and ground communication interception receiver,etc.In order to intercept communication intelligence,the modulation modes of communication signals should be identified first,so as to correctly demodulate and carry out subsequent information analysis and processing.Based on the project from a certain research institute,which is named "XX Intelligent Processing Technology",this thesis studies the intelligent identification technology for the modulation modes of communication signals.The main work of this thesis is as follows:(1)One effective feature vector is constructed to realize the modulation recognition algorithm based on artificial features.Three new features based on signal power spectrum and high power spectrum are proposed.Together with the instantaneous features in time domain,the features in transform domain and the features of high-order cumulants,these features form one effective feature vector of 20 dimensions.Then,the four-layer fully connected BP neural network is used to classify the feature vector.The recognition accuracy of 14 kinds of collected real signals is 97.5%,and when SNR is no less than-5dB,the recognition accuracy of 18 kinds of simulation signals is higher than 90%.(2)Based on the theory of deep learning,a kind of end-to-end modulation recognition is carried out on the time-frequency graphs of these signals to be identified.Firstly,these signals are preprocessed by continue wavelet transform,grey scale processing and the bicubic interpolation method to obtain the time-frequency graphs of fixed size.Then,with the help of AlexNet model and the migrated Inception-ResNet-v2 model,features are automatically extracted by network,so as to realize the modulation recognition based on convolutional neural networks.Finally,the effectiveness of the algorithm is verified by using the simulation signals under different SNR and the real signals.(3)The combined model is designed by online splicing of artificial features and automatic features.With the migrated Inception-ResNet-v2 model,automatic features are extracted from the time-frequency graphs.After the automatic features and artificial features are spliced,two full connection layers are used to complete the classification task.In this way,it improves the existing model and obtains one combined model with better recognition effect.The recognition accuracy of 14 kinds of collected real signals is 99.19%,and when SNR is no less than-5dB,the recognition accuracy of 18 kinds of simulation signals is higher than 92%.(4)The model based on triplet network is designed to improve the existing model.With the help of the migrated Inception-ResNet-v2 model,one appropriate feature extraction sub-network can be built,and the similarity of the input time-frequency graphs can be evaluated by the euclide distance of feature vectors.By designing a reasonable sample decision method,the recognition of these existing signals and new signals can be realized,so the scalability of the model is also enhanced.The model can accurately identify 14 kinds of collected real signals with an accuracy of nearly 100%,and it can also effectively recognize 3 kinds of new real signals.The research work and achievements of this thesis fully meet the requirements of the project on intelligent identification technology for the modulation modes of communication signals,and also plays an important role in the smooth implementation and final acceptance of this project.