Recognition Of Electromagnetic Leakage Signals Of RF Based On Machine Learning

Today,with the increasing development of informatization,networks and electronic devices are everywhere,so that information exchange occurs all the time.At the same time,information security issues are increasingly valued.Information security threatens both hardware and software.On the hardware side,there is a device cloning method to forge the hardware and threaten information security.On the software side,ensuring information security mainly depends on the encryption operation of the encryption chip.During the operation of the encryption chip,the electromagnetic leakage caused by the electromagnetic change and the current change is used as a side-channel attack to threaten the information security.This article uses electromagnetic leakage and RF-DNA to classify individual chips and decrypt encrypted information,which is of great significance for evaluating information security and improving information security protection.This article mainly has the following innovations:First,for the problem of excessive number of single-model classifications of chip that is the same type,based on deep learning,the same type of radiation source classification scheme that prioritizes batch classification and then chip individual classification is designed to reduce the number of single-model classifications required,and the final optimal accuracy rate reaches 99.9%.And,in the office environment,experiments have verified that the classification model has good anti-interference ability.Second,based on the deep random forest method and the Hamming weight model,a key recovery method for side channel analysis is proposed,and combined with the encrypted information,a deep random forest model suitable for the classification is trained.The optimal model is achieved by optimizing the number of forests,the number of trees in a single forest,and the size of the density window.Simulation experiments show that the model has better accuracy than the random forest model.Third,a ResNet10-RF model based on the combination of residual neural network and random forest is proposed.Aiming at the electromagnetic leakage characteristics of the reclassified encrypted information according to Hamming weight,the final classifier of the residual neural network is modified,and the combination of deep learning and machine learning is used.The combination of deep learning self-extracting feature characteristics and random forest is used to improve key recovery.Accuracy,combined with transfer learning methods,improves training speed.