| Automatic modulation recognition is a key technology in cognitive radio system.Its purpose is to detect and analyze the modulation types of signal for demodulation and other operations.With the continuous development of wireless communication system,the modulation methods of signal become more and more diverse.The traditional automatic modulation recognition technology has been difficult to apply to complex and diverse modulation signals,and also difficult to deal with the impact of complex channel environment.The automatic modulation recognition method based on deep learning can learn features from signal and carry out automatic modulation recognition,which has become an important research direction.Based on CLDNN(Convolution LSTM Deep Neural Network),this thesis studies the application of deep learning in the field of modulation recognition of communication signal,the main works are as follows:On the basis of CLDNN,this thesis proposes a flexible network architecture: DenseMix CLDNN(Dense-Mix Convolution LSTM Deep Neural Network).By introducing dense connection to solve the problem that CLDNN can not fully extract the correlation features between I component and Q component in IQ signal,and introducing mix convolution to enable Dense-Mix CLDNN to extract multi-scale features.Data augmentation technology is applied to modulation recognition research based on deep learning,and four data augmentation algorithms are designed for communication signal: Random Erasing,Down Sample/Up Sample,Random Shift,Time-Domain Inverse.In this thesis,Tensor Flow deep learning framework is used to build Dense-Mix CLDNN,70% of the data in RML2016.10 a is selected as the training set,and 30% of the data is used as the test set for simulation experiment.In the experiments of 11 modulation signals with 20 different SNR,Dense-Mix CLDNN can achieve good recognition performance,and the network structure has good design flexibility.Compared with CLDNN and two-layer LSTM,it has a certain improvement on recognition accuracy.Using the data after augmented by the data augmentation algorithm to train Dense-Mix CLDNN can further improve its recognition performance. |
