Optimized Identity Authentication via Channel State Information for Two-Factor User Verification in Information Systems
Chuangeng Tian, Fanjia Li, Xiaomeng Liu, Juanjuan Li · Sensors · 2025
Traditional user authentication mechanisms in information systems, such as passwords and biometrics, remain vulnerable to forgery, theft, and privacy breaches. To address these limitations, this study proposes a two-factor authentication framework that integrates Channel State Information (CSI) with conventional methods to enhance security and reliability. The proposed approach leverages unique CSI variations induced by user-specific keystroke dynamics to extract discriminative biometric features. A robust signal processing pipeline is implemented, combining Hampel filtering, Butterworth low-pass filtering, and wavelet transform threshold denoising to eliminate noise and outliers from raw CSI data. Feature extraction is further optimized through a dual-threshold moving window detection algorithm for precise activity segmentation, a subcarrier selection method to filter redundant or unstable channels, and principal component analysis (PCA) to reduce feature dimensionality while retaining 90% of critical information. For classification, a kernel support vector machine (SVM) model is trained using a randomized hyperparameter search algorithm. The SVM classifies the CSI feature patterns obtained from user-specific keystroke dynamics, which are processed by Hampel filtering, Butterworth low-pass filtering, wavelet transform threshold denoising, a dual-threshold moving window detection algorithm, a subcarrier selection method, and PCA, to achieve optimal performance. The experimental results show that the user recognition accuracy of this algorithm is 2-3% better than current algorithms.