Research On Modulation Recognition Of Wireless Signals Based On Deep Learning

The wireless modulation recognition refers to a technique that the receiver recognizes the modulation type of unknown transmitted signal.Modulation recognition technology plays an important role in both civil and military fields.With the continuous development of mobile communication systems,new technologies such as multi-carrier and multi-antenna make the scene of modulation signal recognition more complicated.Traditional modulation signal recognition methods based on signal statistical characteristics are difficult to apply in complicated communication scene.Applying deep learning algorithms to modulation signal recognition and overcoming the shortcomings of traditional methods is the main content of the paper.The main contents of this paper include the following aspects:Firstly,the modulation recognition algorithms based on three typical neural network structures of CNN,RNN and FNN are compared and analyzed.Considering that all algorithms are in the same magnitude of computational complexity,the modulation recognition rate of different model underlying structures is analyzed.The results show that the CNN model has better recognition ability.The hyper-parameters of CNN model are further analyzed,including the influence of different network layers and different convolution kernels on the modulation recognition rate.The results show that the number of convolution kernels has an important influence on the modulation recognition rate.Secondly,the modulation recognition technology in OFDM-IM system is studied in this paper.This paper introduces the principle of subcarrier index modulation of OFDM-IM.For the detection of OFDM-IM and the identification of communication parameters in non-cooperative communication,a joint algorithm activation subcarrier number of detection and modulation identification based on Inception convolutional neural network is proposed.The simulation analysis proves that the proposed neural network structure converges quickly and is robust under noise interference.Finally,a method based on blind source separation and CNN is proposed for the identification of modulation methods in non-cooperative MIMO communication systems.In non-cooperative MIMO system,the number of transmit antennas and the channel matrix are unknown.This method estimates the number of transmitting antennas by the minimum description length algorithm and separates the aliased received signals into independent matrices by joint approximate diagonalization of eigen matrices algorithm.The CNN model is used to estimate from the aliasing matrix.Modulation method.Finally,the effectiveness of the proposed algorithm is verified by simulation.In summary,the deep learning-based modulation recognition technology proposed in this paper provides a high-accuracy,low-complexity recognition method for in complicated communication scenarios.