Research On Feature Extraction And Recognition Technology Of Communication Signal

With the improvement of communication technology,modulation methods have become complex and multiple.In order to make better use of spectrum resources and better cope with the diversification of communication systems,the feature extraction algorithm and modulation recognition technology of communication signals will be studied in depth in this paper.The main research results are as follows:In this paper,CNN and RNN are used for modulation recognition firstly;then more complex CLDNN and MCLDNN are imported to extract the spatial characteristics and temporal correlation contained in communication signals.Then MCLDNN model is optimized by determining the optimal values of learning rate,batch?size,number of convolution kernels and optimization algorithm.The simulation results show that the recognition effect of the MCLDNN model is the best.Comparing with other networks,the accuracy of modulation recognition is increased by 2% to 10%.In this paper,high-order cumulant features and envelope features of the communication signals are constructed,and new features based on frequency domain and high power spectrum are designed,and an improved method of calculating the mean values of signal features piecewise is also proposed.At the same time,SVM multi-classifier,Ada Boost classifier and decision tree classifier for modulation recognition of communication signals are constructed.The simulation results show that the new features can distinguish the modulation types obviously;the method which calculates the mean values of signal features piecewise can greatly improve overall effect of modulation recognition;decision tree classifier is more suitable for modulation recognition of communication signals.Moreover,the classifier is not sensitive to noise,so that the recognition accuracy is also very high while the signal-to-noise ratio is low.In this paper,the methods of Finetune and network feature re-extraction are combined in MCLDNN model,and the model is applied for modulation recognition.The simulation results show that using Finetune to transfer the weights of neural network can improve the recognition accuracy of modulated signals which is under low signal-to-noise ratio,and improve the stability of network;network feature re-extraction can effectively identify modulated signals that are prone to being confused;the combination of Finetune and network feature re-extraction can improve the recognition accuracy of network models,and to a certain extent,can solve the problem of poor recognition rate of modulated signals disturbed by noise and modulated signals that easily confused.