Modulation Recognition Method Of Communication Signal Based On Multi-Mode Deep Learning

Signal modulation recognition technology has numerous civic and military applications.The number of radiation sources such as radar,communication,navigation,and electronic warfare equipment is increasing in today’s informationized battlefield,the modulation forms are becoming more and more diverse,and the signal density is increasing,making the battlefield’s electromagnetic environment more complicated.The traditional signal modulation recognition technology has been unable to adapt.Therefore,this article will study the robust feature extraction,fusion and recognition technology of complex communication modulation signals.In this paper,an Alexnet network and complex-valued neural network based on deep learning are proposed.At the same time,the technology of multi-modal feature fusion and collaborative fusion architecture is adopted to fuse the multi-modal information of the Contour Stellar Image(CSI)domain and signal I / Q waveform domain,so as to realize signal modulation recognition.The main work contents are as follows:(1)Considering the complementarity between different modes and the importance of mode fusion,we extract multimodal information from the original modulation signal.The first mode is the CSI domain of the signal,and the second mode is the signal I / Q waveform domain.The two modes are extracted by using Alexnet networks and deep complex-valued networks respectively,and the image and waveform features are combined to form a joint feature representation,so as to obtain a more accurate and robust signal modulation recognition scheme.T-SEN algorithm is used to visualize the extracted features.(2)Considering that the prediction results of each mode will be different,a signal modulation recognition method based on multi-modal model depth fusion framework is proposed to represent the fusion objectives of the multi-modal feature fusion model,the CSI domain and the I / Q waveform domain with a loss function,so as to maintain the similarity structure between modes and within modes,and make various separated modes cooperate with each other under constraints,It helps to maintain the unique characteristics and exclusivity of a single mode.Simulation results show that the identification accuracy of this method is better than that of single-mode identification method and multi-mode collaborative fusion framework.(3)Three modulated signal datasets with different sampling points are established by simulation software to measure the performance of our proposed model.The simulation results show that the identification accuracy of the proposed method is better than that of the single-mode identification method and the method without multi-mode collaborative fusion framework.