| In order to transfer different types of communication signals quickly,efficiently and accurately,the transmitter can use different types of modulation methods.Therefore,identifying the modulation method of the received signals accurately becomes an important prerequisite for the next step in the processing of communication signals.With the rapid development of communication technology,the communication environment has become more and more complex,communication signals have become more diverse and complex,and various types of modulation methods have also emerged.In the context of some communication applications,the receiver will receive some communication signals that use an unknown modulation method.Therefore,how to identify the unknown modulation type signal effectively becomes a major research topic in the field of modulation recognition.In the traditional modulation recognition technology with feature-based,the characteristics mainly come from the expert features extracted by human beings.Manual extraction is not only time-consuming and inefficient but also has poor robustness.With the successful application of deep learning in the fields of speech recognition and image processing,researchers have paid more attention to the characteristic learning ability of deep learning.In view of the problems faced in the field of modulation recognition and the superior learning ability of deep learning,this thesis proposes a signal recognition method for unknown modulation types in two different environments.The main research contents are as follows:First,this thesis addresses the situation where there are a large number of labeled data sets in the database.An unknown modulation type signal recognition method based on supervised learning classification model and dimension reduction algorithm is proposed.The method establishes the network structure diagram of the Convolutional Neural Network(CNN)classification model,inputs the I/Q component of the known modulation type signal,and selects appropriate network parameters to train the CNN classification model.Perform feature learning on the received signal sample set with unknown modulation type by using the trained CNN classification model,that feature can characterize the inherent information of the sample signal is obtained.Finally,reduce and visualize the high-dimensional features obtained by using different dimensionality reduction algorithms,according to the visualized effect map,selecting the low dimensional feature data obtained by the algorithm with the best dimension reduction effect,and the recognition rate of the unknown modulation type signal is calculated based on the distance formula.The simulation results show that this method can identify signals of unknown modulation type effectively.Second,in view of the difficulty in obtaining labeled data,this thesis proposes a signal recognition method based on the combination of Long Short-term Memory(LSTM)prediction model and Gaussian distribution model.This method designs the network structure diagram of the LSTM prediction model,inputs the amplitude of the known modulation type signal,and uses different network parameters to realize the construction and training process of the LSTM layer.The error is the difference between the predicted value and the true value of the LSTM model,the error distribution of the signal data set of the known modulation method is estimated,and the error is used as the eigenvalue of the Gaussian distribution detection model to realize the recognition of the unknown modulation type signal.LSTM model has strong learning ability for sequence data,the modulated signal dataset in this thesis belongs to a kind of sequence data,so the model can well identify the unknown modulation type of signal,and the simulation results prove that the model is also applicable for small sample data sets. |
