A Robust Method Research About Keystroke Eavesdropping Based On Acoustic Signals

With the development of smartphones and various smart devices,high-precision sensors are playing an increasingly important role in people’s lives.While these sensors enrich the life of users,they can also easily become a breakthrough for privacy leakage.Currently,attacking user privacy through acoustic signals collected by smart devices has become one of the most common side channel attacks chosen by attackers,and there are still many challenges and problems in this area of research,such as lack of robustness to various instabilities and low accuracy in the face of fine-grained classification tasks.Such problems reduce the threat of this side-channel attack to user privacy,while detaching it from the actual attack environment.In response to these problems,this thesis proposes two side-channel attack methods based on acoustic signals,which respectively eavesdrop on keyboard input from unfamiliar victims and input environments,revealing the serious threat to user privacy and the need for preventing privacy leakage.The main contents and contributions of this dissertation are as follows:1.A keystroke side-channel attack scheme robust to unfamiliar victims is proposed.Aiming at the possibility of the unknown victim in reality,this thesis proposes an algorithm that is robust to the force,speed,gesture of typing,and ambient noise.This thesis analyzes the acoustic characteristics of keystrokes,extracts fine-grained and robust features from acoustic signals,and uses Support Vector Machine(SVM)as a classifi cation model to make key prediction,which is independent of unstable factors affected by the victim.In addition,this thesis enhances the robustness of the algorithm to the environment and the anti-noise performance through the key detection algorithm and denoising.Based on the test results under different experimental scenarios,the proposed scheme is effective in predicting keystrokes and significantly improve the performance of sidechannel attack.2.A keystroke side-channel attack scheme robust to unfamiliar environments is proposed.In practical scenarios,the attackers of side-channel attacks may not have any prior knowledge,and the input environment also faces frequent changes,that is,the attackers do not have real information about the victim and the input environment.Starting from this actual situation,this thesis proposes an environment estimation algorithm without real training samples,including keyboard type recognition,microphone height estimation,microphone coordinate estimation and angle correction,to convert unknown environments into known ones.Subsequently,this thesis utilizes the steps of environment simulation,small training sample collection and model training with small samples to predict keystrokes,so as to eavesdrop keystrokes from unknown environments.Through testing and comparative experiments under different conditions,this thesis verifies the robustness,accuracy and usability of the proposed scheme.3.This thesis uses a single commercial smartphone to implement the proposed scheme and conduct the corresponding experiments.This shows that the attack method still poses a huge threat to privacy even in the absence of real training samples,lack of understanding of the attack object,and extremely simple attack equipment.In conclusion,this thesis indicates the necessity for users to be more vigilant and strengthen prevention against such kind of attack.