Comprehensive Overview of the Identity Authentication System Based on Phase Space Reconstruction and Time Series Prediction of Pulse Wave Signals
In recent years, the rapid development of internet and mobile devices has led to an increasing risk of identity theft. As a critical tool for protecting personal information and preventing unauthorized access, identity recognition technology has garnered significant attention. Traditional methods, such as passwords or PINs, are vulnerable to attacks, while biometric technologies offer higher convenience and security by leveraging unique physiological or behavioral characteristics. Among various biometric features, photoplethysmography (PPG) signals have emerged as a promising candidate due to their inherent advantages, including liveness detection, security, low cost, and ease of acquisition.
PPG signals, which reflect blood volume changes in microvascular tissues, are influenced by cardiac activity and vascular dynamics. These signals exhibit nonlinear and chaotic characteristics, making them suitable for identity authentication when analyzed using advanced computational techniques. Existing research on PPG-based biometrics primarily focuses on time-domain or frequency-domain feature extraction, often overlooking the rich dynamical properties embedded in the signals. This study addresses this gap by proposing a novel identity authentication framework that combines phase space reconstruction (PSR) and long short-term memory (LSTM) networks to exploit the nonlinear dynamics of PPG signals.
Background and Motivation
Biometric systems based on fingerprints, facial recognition, iris scans, and voiceprints have been widely adopted. However, these methods face challenges such as spoofing attacks, environmental sensitivity, and high implementation costs. PPG signals, in contrast, provide a non-invasive and cost-effective alternative. Early studies demonstrated the feasibility of PPG for biometric identification by extracting features such as pulse wave amplitude, duration, and derivative characteristics. However, these approaches often rely on handcrafted features, limiting their robustness and generalization.
Recent advancements in deep learning have enabled more sophisticated analysis of PPG signals. For instance, convolutional neural networks (CNNs) and recurrent neural networks (RNNs) have been employed to classify PPG waveforms. Despite their success, these methods do not fully exploit the nonlinear dynamical properties of PPG signals, which are intrinsic to cardiovascular physiology. Recognizing this limitation, the present work introduces a novel framework that leverages PSR to reconstruct the underlying dynamical system of PPG signals and employs LSTM networks for time series prediction.
System Design and Methodology
The proposed identity authentication system consists of two main phases: enrollment and verification. During enrollment, users register their PPG signals, which undergo preprocessing and PSR to determine optimal embedding dimensions and time delays. These parameters, along with a personalized LSTM model, are stored in a database. During verification, the system captures a user’s PPG signal, reconstructs its phase space using the stored parameters, and evaluates the prediction accuracy of the corresponding LSTM model. If the prediction error falls below a predefined threshold, the user is authenticated.
A key innovation of this approach is the use of PSR to uncover the hidden dynamical structure of PPG signals. According to Takens’ embedding theorem, a one-dimensional time series can be reconstructed into a higher-dimensional phase space that preserves the topological properties of the original system. The selection of delay time and embedding dimension is critical for accurate reconstruction. This study employs mutual information to determine the delay time and the false nearest neighbors (FNN) method to estimate the embedding dimension.
To validate the chaotic nature of PPG signals, the Grassberger-Procaccia (G-P) algorithm is used to compute the correlation dimension. Results confirm that PPG signals exhibit saturated correlation dimensions, indicating their chaotic behavior. This finding justifies the use of nonlinear dynamical analysis for PPG-based biometrics.
Experimental Results and Analysis
The performance of the proposed system is evaluated using both collected PPG data and public datasets. Four subjects participate in the experiments, with each providing 50 seconds of PPG recordings at a sampling rate of 125 Hz. The first 40 seconds of data are used for model training, while the remaining 10 seconds are reserved for testing.
Phase space reconstruction reveals distinct attractor trajectories for different individuals, demonstrating the uniqueness of PPG dynamics. The LSTM models, trained on these reconstructed sequences, achieve high prediction accuracy for their respective subjects while producing larger errors for imposters. A support vector machine (SVM)-based thresholding mechanism is applied to distinguish genuine users from imposters, yielding a 100% authentication accuracy on a test set of 100 PPG samples from 87 subjects.
The system’s robustness is further assessed by examining its generalization to remote PPG (rPPG) signals, which are acquired non-contact using video cameras. Although rPPG signals are noisier than contact-based PPG, preliminary experiments show that they retain discriminative dynamical features. Variance analysis confirms significant differences in nonlinear parameters (e.g., permutation entropy, fuzzy entropy) between individuals, suggesting the potential of rPPG for biometric applications.
Discussion and Future Directions
The success of this study highlights the importance of nonlinear dynamical features in PPG-based identity authentication. By combining PSR with LSTM networks, the proposed framework achieves superior performance compared to traditional feature-based methods. The system’s ability to generalize to rPPG signals opens new possibilities for non-contact biometrics, particularly in scenarios where hygiene or convenience is a priority.
Future research directions include expanding the dataset to include diverse demographics, investigating the system’s performance under dynamic conditions (e.g., motion artifacts), and exploring hybrid biometric systems that integrate PPG with other modalities. Additionally, advancements in deep learning, such as attention mechanisms, could further enhance the model’s predictive capabilities.
Conclusion
This study presents a groundbreaking approach to identity authentication by leveraging the nonlinear dynamical properties of PPG signals. The integration of phase space reconstruction and LSTM-based prediction provides a robust and scalable solution for biometric systems. Experimental results demonstrate the method’s high accuracy and potential for real-world deployment. As remote sensing technologies evolve, the proposed framework could pave the way for next-generation non-contact biometric systems.
doi.org/10.19734/j.issn.1001-3695.2024.08.0307
Was this helpful?
0 / 0