Communication Signal Recognition Based On Machine Learning

Communication signal modulation identification is a basic component of military jammers,threat analysis systems,and surveillance systems.It also has important civilian applications,including intelligent communication systems such as software radios and cognitive radios.If the modulation type of a signal is unknown and it is applied to a mismatched demodulator,the information content of the signal may be destroyed,making the communication signal useless.This has a negative impact on the deciphering process.Therefore,modulation identification is the key to selecting a suitable demodulator.Once the classifier correctly identifies the modulation type,it can perform signal demodulation and then perform other information extraction and other operations.Therefore,the modulation recognition of communication signals based on machine learning is studied in this paper.The main research work of this paper is summarized as follows:1.Commonly used communication signal recognition features are described and derived in detail,such as instantaneous parameter features,higher-order cumulant features,time-frequency features,and cyclic stationary spectral features,and these features are simulated on the MATLAB platform as the signal-to-noise ratio changes.Because a single higher-order cumulant feature is affected by the signal energy,the higher-order cumulant features are combined to form a new feature.In order to reduce the execution time of the entire recognition system,a principal component analysis method commonly used in dimensionality reduction is introduced to optimize the system.The time-frequency image and the cyclic spectrum top view are combined into one image,and the image is processed by wavelet denoising,and then the image is used as an input feature.Three commonly used machine learning methods are described and derived in detail,including decision trees,support vector machines,and convolutional neural networks.2.A communication signal recognition system based on decision tree is proposed.This system combines time domain features and high-order cumulant features,and uses the combined features as the input data of the decision tree and the decision tree as the classifier.The system recognizes seven communication signals of 2ASK,4ASK,2FSK,4FSK,4PSK,8PSK,and 16 QAM with an accuracy rate of more than 90% under in 2d B.The system is compared with the identification system of manually established decision tree under the same condition,it is found that the recognition rate of the system proposed in this paper is higher.3.A communication signal recognition system based on support vector machine is proposed.The system uses support vector machine as a classifier,and the cyclic domain features of dimensionality reduction after principal component analysis are used as the input data of support vector machine.Experimental simulations show that the system can correctly identify 5 communication signals 2ASK,2FSK,4FSK,4PSK,and 2PSK when the signal-to-noise ratio is greater than-2d B.The recognition rate is higher than 90%,and the system is compared with the signal recognition method based on machine learning proposed in the existing literature under the same condition.The comparison results show that the features proposed in this paper are more effective in improving the accuracy of signal recognition.4.A communication signal recognition system based on convolutional neural network is proposed.The time-frequency graph and cyclic spectrum are used as the input of the convolutional neural network to transform the modulation recognition problem into an image recognition problem.Experimental simulations show that when the signal-to-noise ratio is greater than-4d B,the recognition rate of {2FSK,4FSK,8FSK,2PSK,4PSK}signals is greater than 90%,The recognition rate is higher than the system using the time-frequency graph and as a feature alone.