Radio Feature Recognition Based On Deep Learning

The extraction of radio signal features and the identification of modulation have always been an important content for spectrum management and electronic detection.With the continuous increase of social communication requirements,there are more and more wireless communication methods.Not only the radio signals themselves but also the spatial radio environment become more and more complex and variable.Then,there are many other interference signals in the space,and the signal-to-noise ratio is relatively low.Therefore,it is very difficult to extract and identify radio signal features.In this paper,the method of deep learning is used to study the identification of radio signal modulation method from the aspects of cyclic spectrum feature and time domain feature.The semisupervised method is used to reduce the dependence of deep learning on labeled samples andimprove the applicability of deep learning in radio signal feature extraction and modulation mode identification.The specific research work is as follows:1.The cyclical periodicity of wireless communication signals is studied.The cyclic spectrum features of various modulation methods are analyzed.The effective features are automatically selected through deep learning,and the modulation modes are identified.The simulation and measured data processing results show that the proposed method has a significant improvement in anti-noise performance compared with the traditional artificial feature extraction.2.The multi-layer perceptron is used to identify the time domain signals of different modulation methods.It is finally proved that the effective recognition of radio signals by multi-layer perceptrons requires a signal-to-noise ratio of more than 15 dB,and the recognition accuracy can reach over 90% in this SNR range,,and with the increase of signal to noise ratio,finally reach 95%.3.The feature extraction of wireless modulated signals is performed by convolutional neural network.The unique noise reduction characteristics of convolutional neural networks can neutralize the interference of noise on feature points.The experimental results show that to achieve 90% feature recognition rate,only the signal-to-noise ratio is above 10 dB.4.a semi-supervised method by combining supervised learning and unsupervised learning mode cooperative training is proposed.The experimental results show that the method has a great independence on the marked data and has better applicability in radio signal recognition.