Electromagnetic Interference Signal Detection And Recognition System Based On USRP And Deep Learning

Electromagnetic interference(EMI)is a major obstacle to the electronics working well.The problem of EMI is becoming more and more serious,especially in today's intricate space electromagnetic environment filled with various electromagnetic waves.Therefore,EMI is an urgent problem to be solved,to ensure the machine work normally and signal reliable transmission in the complex electromagnetic environment.However,the collection,detection,analysis and identification of interference signals are vital foundations and prerequisites.Based on the Software Defined Radio peripheral USRP and portable oscilloscope,this paper implements an integrated electromagnetic interference signal detection hardware platform and studies the signal modulation recognition algorithm based on deep learning to identify the modulation type of the radiated interference signal.Finally,the hardware platform and the software algorithm are integrated to realize the electromagnetic interference detection and identification system.The main research contents and innovations of this paper are as follows:(1)Based on USRP and portable oscilloscope,a small integrated electromagnetic interference detection unit is realized.Using GNU Radio and USRP to design and develop a detection and acquisition device for radiated interference signals.And through the multi-unit expansion method of cooperative scanning,the frequency domain monitoring bandwidth is increased,we realized the large-bandwidth real-time spectrum monitoring function with low cost.This provides new ideas for the development of intelligent and highly dynamic new electromagnetic interference detection equipment and systems.(2)Research on wireless signal modulation recognition method based on deep learning.A ResNet residual network model is designed,and the self-attention mechanism is introduced to improve the long-term timing-dependent information learning ability of the model.The data augmentation methods of interpolation,extraction and noise addition,meanwhile,are used to expand the sample space of the data set to enhance the robustness and generalization ability of the model.Finally,the 25 kinds of modulation signals are identified.Experimental results show that when SNR?3dB the average accuracy rate is over 96%.(3)Based on the deep learning multi-task method,a multi-task network model for both signal modulation recognition and signal-to-noise ratio prediction is designed.By using the backbone neural network with shared parameters to learn the common characteristics between tasks,the two tasks are jointly optimized and mutually promoted.Finally,when SNR?5dB,the average classification accuracy is over 98%,and the mean absolute error of the signal-to-noise ratio prediction is less than 0.56dB as a whole.The results show that,the multi-task learning network could well accomplish tasks modulation classification and signal-to-noise ratio prediction.