Noise assessment framework for optimizing ECG key generation

Bioelectrical signals such as electrocardiogram (ECG) have shown promise as biometrics, but their continuous nature and drastic acquisition variations make it difficult to deploy them for biometric-based key generation. In particular, it is nearly impossible to obtain raw ECG measurements from a large population under all possible test conditions. In this paper, we build upon our recent approach called IOMBA by combining it with a pre-assessment framework that uses synthetic ECGs to characterize the impact of different sources of noise on ECG-based keys. Our framework uses an auto-regressive (AR) model with three modulated sources of noise - baseline wander (BW), electromyography (EMG), and motion artifact (MA). The performance of the proposed framework is validated using normal ECG signals from popular ECG databases. Different feature extraction methods are applied for ECG key generation and the performance of each approach with each noise source is evaluated. The proposed framework can be used to optimize pre-processing approaches for low-cost applications.