Wi-Crack: a smartphone keystroke recognition system based on multi-dimensional information

Abstract With the popularity of mobile payment, password protection has become more and more important. Channel state information (CSI) has recently been used to crack passwords in public WiFi environments. The feasibility of password cracking lies in the fact that different keystrokes lead to different finger movement directions and distances, resulting in unique interference to WiFi signal transmission. The unique interference can be recorded by CSI and used for keystroke inference. In this paper, we propose Wi-Crack, a keystroke recognition system for numerical keypad input on smartphones. Two computers equipped with commercial off-the-shelf WiFi NIC comprise the system, with one serving as the transmitter and the other as the receiver. Previous keystroke recognition systems only used the amplitude of CSI for keystroke recognition. Different from them, Wi-Crack combines the amplitude, phase, amplitude difference and phase difference of CSI for keystroke recognition. The use of multi-dimensional information has enabled Wi-Crack to improve keystroke recognition. Experimental results show that Wi-Crack improves the accuracy of keystroke segmentation to consistently above 90%. It also improves the keystroke recognition accuracy on DTW-KNN, SVM, 1D-CNN, and LSTM with over 90% in the best case.