Study On Digital Signal Modulation Recognition Based On Deep Learning

Modulation recognition technology refers to the technique of judging the modulation mode and other parameters of the signal by analyzing and studying the received signal samples in the absence of prior knowledge.As a research hotspot in the field of communication,modulation recognition technology is widely used in many military and civilian communication fields such as signal monitoring,spectrum management,and electronic countermeasures.At present,modulation recognition research mainly adopts a combination of feature extraction and pattern classification.With the rapid development of artificial intelligence technology,deep learning theory is increasingly applied to the field of modulation recognition.Compared with the traditional modulation recognition algorithm,the deep learning-based modulation recognition algorithm can automatically learn more complex feature representations from the signal data,and achieve better recognition results.This paper has made the following two parts research on the application of deep learning technology in the field of modulation recognition.In the first part,the modulation recognition of FSK signals is studied based on singular value denoise and convolutional neural networks.First,the samples of the three received signals(2FSK,4FSK,8FSK)are denoised according to the singular entropy theory;then,cyclic spectrum feature maps of the signal samples after noise reduction are calculated;after that,cyclic spectrum feature maps are processed to remove redundancy of information;finally,the convolutional neural network is used to further features extraction and classification.In the part of simulation analysis,the recognition rates of the three algorithms are compared:algorithm proposed in this part,algorithm with cyclic spectrum feature map and convolutional neural network,algorithm with cyclic spectrum slice vector and neural network.The result shows that,algorithm proposed by the paper achieves higher recognition rate under the condition of low SNR,and the recognition rate is close to 100%under the signal-to-noise ratio of-lOdB.In the second part,this paper studies the digital signal modulation recognition based on Auto-encoding network in MIMO system.In the space division multiplexing MIMO system,the complex channel environment poses a great challenge to the research of modulation recognition.Firstly,cumulants characteristics of five modulation signals(2PSK,4PSK,8PSK,16QAM,32QAM)are extracted;secondly,Auto-encoding network is used to process cumulants for getting new low-dimensional nonlinear features;after that,neural network is used as classifier.In the part of simulation analysis,the recognition rates obtained by four algorithms are compared:the algorithm proposed by the paper,algorithm directly using 12-dimensional cumulant,algorithm using manual selection of 3D cumulant and algorithm using feature processed by PCA.The result shows that,the proposed algorithm in this part can achieve much higher recognition rate compared with other algorithms,and the recognition rate is close to 100%with a signal-to-noise ratio of OdB.