Research On Electromagnetic Space Anomaly Detection Method Based On Deep Learning

With the rapid development of communication technology,a large number of electronic devices are integrated into people's work and life,which brings new challenges to the safety of electromagnetic space side channel.In view of this application background,this paper studies the electromagnetic space anomaly detection model,determines whether the electromagnetic space is abnormal,classifies the abnormal conditions,and finds out the causes of electromagnetic space anomalies,so as to prevent and curb side channel attacks in time.Firstly,combined with the requirements of electromagnetic space information security,this paper summarizes the extremely complex factors causing electromagnetic space anomalies into two types: unintentional electromagnetic leakage that may lead to information leakage and malicious electromagnetic attack that may have been caused by side channel attack.The timefrequency characteristics of unintentional electromagnetic leakage are explained through modeling and calculation.This paper analyzes and studies the sampling method and time-frequency domain analysis technology,convolutional neural network and automatic encoder in deep learning,generation of countermeasure network and so on.Combined with the requirements of electromagnetic space information security,the factors causing electromagnetic space abnormalities are divided into two categories: one is unintentional electromagnetic leakage causing information leakage,and the other is malicious electromagnetic attack caused by side channel attack.Through modeling and calculation of the two factors,the time-frequency characteristics of electromagnetic leakage information are clarified.In this paper,the structure of coding module decoding module discrimination module is used to strengthen the learning ability of the model for the characteristics of frequency domain sequences without anomalies and the ability of data reconstruction.The reconstructed sequences of the model are significantly different from the sequences with anomalies.The test results show that the detection model based on the combination of variational automatic encoder and generation countermeasure network is effective in detecting abnormal data under-20 db signal-to-noise ratio in the frequency band of 2MHz,and the judgment accuracy of the actual collected abnormal data reaches 95%.Based on convolution neural network,this paper proposes a frequency domain based electromagnetic spatial anomaly classification model and a time-frequency domain based malicious electromagnetic attack anomaly classification model.The simulation results show that the electromagnetic spatial anomaly classification model has a good classification effect on the anomalies under-10 db signal-to-noise ratio and has a certain classification ability for the anomalies under-15 db signal-to-noise ratio in the frequency band of 2MHz;The malicious electromagnetic attack classification model can classify more than-8db modulation types.For real-time malicious electromagnetic attack signals,time-frequency domain classification is proposed,which is more effective than frequency domain classification.