Research On Automatic Waveform Modulation Recognition

The identification,analysis and process of communication signals can benefit greatly information process and application,which is of great significance in military reconnaissance,electronic countermeasures,wireless network security,machine artificial intelligence and other aspects.The modulation type of communication signal is one of its important technical characteristics.And basis can be provided to further analysis and process of signal by modulation type recognition.as an intermediate step between signal detection and demodulation in non-cooperative communication,signal modulation pattern recognition is a key step in signal analysis and process.With the development of communication technology,the types of communication signals are becoming more and more diverse,and the signal environment is becoming more and more complex.Therefore,finding an efficient and sustainable automatic signal recognition technology has become a research focus.Multi-fractal spectrum features can be used to realize waveform recognition effectively.Firstly,a discrete sequence of signal data is obtained by sampling and processing the received signal.The sequence is then placed into a set of points in an mdimensional Euclidean space.The correlation integral is obtained by analyzing the distance distribution of each point in this set.The slope of the approximate straight line segment of the correlation integral curve represents the multi-fractal spectrum characteristics of the signal,and different slopes represent different signals.Simulation results show that this feature can be used to distinguish modulation signals effectively.By extracting the instantaneous characteristics of signal,the waveform recognition can also be realized by using neural network.Firstly,the digital modulation signal is generated and labeled to obtain the sample database.Then,five instantaneous characteristics of the signal are calculated and stored in the characteristic matrix as the training data set of BP neural network.Finally,a forward neural network is established and used for recognition after training.Experimental results show that the system can achieve more than 94% recognition accuracy under the condition of better SNR.The combination of deep learning and signal modulation recognition can improve the efficiency of communication signal recognition.In this paper,the software radio platform is combined with deep learning.The software radio platform is used to generate and transmit signals,which is collected labeled and stored by the receiver to build a waveform library.Then,a deep learning recognition model consisting of two convolution layers(Conv)and a fully-connected layer is designed.The size of the convolution kernel of the first convolutional layer is 2*5 while the size of the convolution kernel of the second convolutional layer is 1*5.Secondly,the fullconnected layer outputs a 6-bit vector which represents the probability of 6 modulation signal categories.This model can be used for waveform recognition after training the data set generated by the software radio platform.Experimental results show that this method can achieve 98% recognition accuracy under the condition of better SNR.