KeySign: WiFi-Based Authentication Using Keystroke Signatures

Authentication plays an important role in human-computer interaction. Personal Identification Number (PIN) is a commonly used method in authentication. However, PINs can be easily cracked, posing a threat to system security. Similar to fingerprints, irises, and faces, a user's keystroke signature is biologically unique and can be used as the authentication feature. In this paper, we propose the KeySign system, which records the user's keystroke through Channel State Information (CSI) in WiFi. The system consists of a transmitter and two receivers. We designed a four-branch Convolutional Neural Network (CNN) to extract feature from the amplitude and phase dimensions of the dual-receiver CSI. The network finally generates a vector that records user's keystroke feature. Finally, with the designed Class Density User Detection (CDUD) algorithm, we can distinguish between legal and illegal users while further identifying the specific identity of the legal user. We collected keystroke data from 8 users on 5 different sets of PINs. The experimental results show that for the distinction between legal and illegal users, KeySign has an average accuracy of 90.83%, a false acceptance rate of 7.08%, and a false rejection rate of 13.33%. For the identification of legal users, the average accuracy is up to 97.22%.