Research On Digital Modulation Recognition Technology Based On Multi-feature Fusion

With the rapid development of communication technology,the communication environment has become more and more complex and the modulation of signals has become more and more diversified.At the same time,digital modulated signals have become an indispensable part of modern communication field due to their advantages such as high accuracy and reliability.In non-collaborative communication,signal modulation identification is an indispensable process before the receiver demodulates the received signal.However,traditional modulation recognition methods are often too complex,require a large amount of a priori knowledge and have difficult thresholds to measure.Existing modulation recognition methods based on deep learning have good feature extraction capability,but due to their single feature,the lack of feature characterization capability leads to poor recognition results when facing multiple modulation methods.Therefore,this thesis investigates the modulation recognition technology for 4ASK,8ASK,2FSK,4FSK,BPSK,QPSK,16 QAM and 64 QAM,a total of 8 modulation signals,with the aim of improving the recognition accuracy of the system.The innovation points and research contents of this thesis mainly include the following two aspects:(1)In this thesis,an improved Inception-V4 digital modulation-based recognition model is designed.The feature sequences extracted from the received signals are used as the input of the improved Inception-V4 network,and the deep feature information of the feature sequences is extracted in parallel by asymmetric convolution of different sizes to maximize the InceptionV4 classification performance,thus improving the single feature recognition rate.In this thesis,five feature sequences of digital modulated signals are selected and the modulation recognition models corresponding to different feature sequences are obtained by training.The modulation recognition performance and the average recognition rate of different feature sequence models under different signal-to-noise ratios are analyzed,and the feature sequence with better recognition performance is selected as the input of the multi-feature fusion network model.Through experimental analysis,three feature sequences,IQ,instantaneous phase and amplitude and power spectral density,have better recognition effect.(2)To address the problem that the existing modulation recognition technology has a single input feature,resulting in low modulation recognition accuracy,this thesis proposes a digital modulation recognition method based on multi-feature fusion,constructs a multi-feature fusion network model(MFFNet)based on the modified modulation recognition model of InceptionV4,and selects the above three feature sequences with better recognition performance as the input of the MFFNet model.Next,a multi-feature fusion attention network model(MFFANet)is designed by adding a channel attention mechanism to MFFNet.The modulated recognition performance of the two models under different signal-to-noise ratios is analyzed to compare the effect of the presence or absence of the attention mechanism on the recognition accuracy,and finally the recognition effects of this method are compared with Res Net50 and LSTM network models based on the same data set.The experimental results show that the recognition rate of MFFANet reaches 80% when SNR=0d B,and the recognition rate of MFFANet is about 5%higher than that of MFFNet,and the recognition effect of MFFANet is also better than that of Res Net50 and LSTM.