Research And Implementation On Communication Signal Modulation Recognition Based On Neural Network

Modulation pattern recognition of the communication signal refers to a process in which the communication receiver judges the received signal modulation mode without knowing the modulation mode of the transmission signal.In non-cooperative communication scenarios such as communication detection,electronic countermeasures,and signal supervision,and the design of general receivers for cognitive radio systems,the modulation recognition of communication signals is critical step because it is closely related to the subsequent signal demodulation process.Therefore,modulation recognizer is very critical components in both military and civilian communications systems.The recognition of modulation modes is usually divided into manual and automatic recognition.Munual recognition has low efficiency and accuracy because it depends on the experience of the inspectors.Therefore,the automatic recognition with higher efficient realization and accuracy obviously have more research value.The automatic modulation recognition technology of communication signals is mainly divided into two categories,one is a method based on Bayesian decision theory,and the other is a method based on statistical machine learning theory.Although the former has complete theory,it has poor generality and high implementation complexity,so it is not practical.The latter implements simple procedures and is suitable for general communication signal modulation recognition.We mainly studies the modulation recognition technology of communication signals based on machine learning in this thesis.There are so many algorithm models in machine learning theory,while artificial neural network is a classic one of them.we constructs two kinds of general communication signal modulation pattern classifiers based on neural network using its universal approximation in this thesis,one is shallow neural network modulation recognizer based on manual feature engineering,the other is deep neural network modulation recognizer based on end to end model.In the shallow neural network modulation recognizer,we extracts spectral features and cumulant features of communication signals as input,uses the rectified linear unit as activation function,and uses cross entropy as loss function.In the deep neural network modulation recognizer,we constructs convolutional neural network and recurrent neural network to modulation recognition for communication signals.In this thesis,we implement a neural network automatic modulation recognizer cooperating with the CPU and GPU computing platforms,and verify the recognition accuracy rate performance.Experiments show that the communication signal modulation recognizer based on artificial neural network has good classification accuracy performance on both training set and test set.