Research On Attack Detection Based On Power Leakage Of Non-intrusive Side Channel

With the rapid development of computer technology,modern computing devices dynamically adjust their power consumption to match the requirement of the workload.As the computing equipment is performing different tasks,its power consumption information changes with time.Therefore,the power consumption of the device is often used as a typical side-channel signal to calculate the privacy detection and attack.By analyzing the power consumption of the computing device,the attacker can infer the current activities of the computing device.Existing methods of side-channel attack based on DC power are intrusive and require direct access to the internal components of the computing device.Compared with the former,the non-invasive side-channel attack based on AC power has better concealment,and the attacker can potentially observe the AC power side-channel signal and launch an attack,monitor the progress of the device in real-time,obtain the user's privacy,and even cause more disastrous consequences.Based on the AC power consumption,the feasibility of a non-invasive side-channel attack was studied.The main contents of our research were as follows:(1)We proposed a side-channel attack model based on AC power consumption analysis.In this model,the attacker only used limited AC power consumption to infer the performance of the application.At the same time,we designed three different attack scenarios to evaluate the feasibility and effectiveness of a side-channel attack.(2)Based on the correlation degree of power consumption,an offline attack method for applications was designed.This method mainly included data acquisition,data segmentation,feature extraction and application classification.A side-channel offline attack classifier based on power correlation is constructed,and the attack performance of the classifier is evaluated.The experimental results showed that the classification accuracy of applications on the target devices of different brands and systems can reach more than 92.6%.(3)Based on a convolutional neural network,a real-time attack method is designed.Using 2D convolutional neural network,the real-time attack performance of the side channel under different attack scenarios and different window lengths is evaluated,and the factors affecting the detection effect of the application are analyzed.The experimental results showed that under the three attack scenarios,the proposed algorithm can achieve pretty performance of online attack within 0.5 seconds of application startup.(4)To solve the problem that computing devices are vulnerable to malware attacks,a malware detection strategy based on behavior feature synthesis was proposed from the physical information.Aiming at the specific behavior of malware,we proposed a method to synthesize the behavior features of malware,which extracted a variety of behavior features from the power consumption to generate the behavior fingerprint of malware.Three common classification methods were selected to evaluate the performance of malware detection based on side-channel power analysis in a practical environment.The experimental results showed that the proposed system has an excellent defense ability against malware,the precision can reach more than 88.9%.