Research On Applications Of Machine Learning In Modulation Recognition Of Communication Signals

Signal modulation identification is an intermediate link between signal detection and demodulation in digital receivers.It has been widely used in military and civil fields such as communication countermeasures,spectrum detection and management,and information security.In recent years,with the rapid developments of wireless communication technologies,the communication environment is becoming more and more complex and the modulation pattern of communication signals is increasingly diversified,which undoubtedly increases the difficulty of modulation recognition.Therefore,to develope new and efficient modulation recognition methods has become a hot issue concerned by relevant scholars.Machine learning technology,especially the deep learning that has developed rapidly in recent years,has made breakthrough progresses in many applications such as image recognition and text classification.In this paper,based on machine learning,the common deep learning frames combined with the other signal processing methods are utlitized to develop the automatic modulation classification for the five kinds of 5G signals recommended in the 3GPP R15 release.Main tasks of the thesis as follows:(1)Aiming at the problems that signal modulation recognition requires both prior information of signals and manual selection of complex features under non-cooperative conditions,an improved signal modulation recognition algorithm based on Alex Net convolutional neural network is proposed.For recognizing the five commonly used modulation signals(i.e.?/2-BPSK,QPSK,16 QAM,64QAM,and 256QAM),the constellation diagram is selected as the input feature of the Alex Net network to construct a recognition classification algorithm.Simulation results show that compared with the existing Alex Net-based modulation recognition algorithm using the scatter diagram feature of signals,the proposed algorithm has a higher recognition accuracy rate for higher-order QAM signals.(2)A modulation recognition algorithm based on variational mode decomposition(VMD)and VGGNet neural network is proposed.First,the signal is denoised by using the VMD decomposition-based preprocessing alogirthm.Accordingly,the constellation features of the preprocessed signals are extracted,and the VGGNet neural network is adopted to construct a recognition classifier for signal modulation recognition.Simulation results show that under the same conditions,the performances of the VGGNet-based algorithm combined with the VMD denoising is better than that of the algorithm without using VMD denoising.(3)A modulation recognition algorithm based on empirical mode decomposition(EMD)and hybrid neural network is proposed.First,EMD decomposition and denoising preprocessing is performed on received signals,and the first-level decomposition energy characteristics of EMD is defined to distinguish MPSK signals and MQAM signals.Then the high-order cumulant features and the constellation features are defined and the hybrid network fusion of the BP neural network and Res Net neural network are used to construct a hybrid neural network to realize the modulation recognition of signals.Experimental results show that under the same signal-to-noise ratio,the recognition performances of the proposed algorithm is better than those of the other two algorithms proposed in chapter 2 and chapter 3 of the thesis.