Unknown Modulation Recognition Technology Based On Artificial Neural Network And SVDD

As an indispensable part of the current communication process,signal modulation always has a great impact on the overall performance and transmission capacity of the entire communication system.With the development of modern wireless communication technology,the types and methods of modulated signals are constantly increasing,and it becomes more and more difficult to distinguish modulated signals in the communication environment.Therefore,modulation recognition,a method of judging and classifying un-known modulation signals for modulation types,has great research significance in various non-cooperative communication scenarios such as electronic reconnaissance,electronic countermeasures,and spectrum detection,and affects the actual performance of subse-quent demodulation of communication signals and communication parameter extraction.However,with the continuous increase of the types and modulation modes of communi-cation signals,and the increasingly complex communication environment,the traditional modulation recognition method,which only has a good performance for a few modula-tion signals,can no longer meet the needs of current users.Aiming at this problem,this paper proposes an open-set recognition system for unknown modulation signals based on neural networks and support vector data domain description methods.Through a selec-tive screening of the complex modulation signal sets,a few interested modulation signal types can be classified to improve the processing performance of subsequent modulation recognition.Different from the traditional modulation recognition method of artificial feature ex-traction of the modulation signal,this paper chooses the method of neural network with excellent extraction performance and better model universality to complete the process of feature extraction and compression.In this paper,two types of neural networks are con-structed for feature extraction of modulated signals.One of them is the one-dimensional convolutional neural network,whose operation of row convolution of one-dimensional data is very suitable for extracting relevant features of time series.For the other kind of neural network,we adopt the stack self-encoding network,which is similar to the pro-cess of encoding and decoding.It can compress the data features while retaining the core features that can restore the input data itself as much as possible.On the basis of extracting the feature of the modulation signal,this paper constructs a single classification model for the modulation signal based on the method of support vector data domain description.By experimenting with different model parameters,kernel functions and positive and negative sample set construction methods,this paper realizes the classification model of the multi-type modulation signal open-set recognition system,and analyzes the influence of different system parameters on the model performance,and finally gives a reasonable classification method of positive and negative modulated signal sample class.