Adaptive dual augmented extended Kalman filtering of ECG signals

چکیده مقاله

In the previous investigation, we demonstrated that the augmented extended Kalman filter (AEKF) can effectively address the limitations of the conventional extended Kalman filter (EKF) in denoising electrocardiogram (ECG) signals contaminated with nonstationary noise. The AEKF employs a novel measurement model designed to accommodate non-Gaussian, non-stationary additive disturbances. However, its performance relies on careful tuning of certain hyperparameters. In this study, we introduce an advanced version of the AEKF, referred to as the adaptive dual augmented extended Kalman filter (ADAEKF), which features an automated mechanism for optimizing hyperparameter settings. Furthermore, we use a methodology to adaptively adjust state and measurement noise covariance matrices, leading to substantial improvements in ADAEKF's performance. Unlike the original AEKF, which assumed nonstationary noise in ECG signals followed an autoregressive order one (AR1), our proposal allows for the possibility of second-order (AR2) or third-order (AR3) processes. We develop alternative measurement models based on these hypotheses. Notably, the ADAEKF does not require prior knowledge of the specific noise type and can effectively handle white Gaussian processes as well. Our work represents the first effort to integrate and assess the dual EKF framework within the domain of ECG signal processing. Extensive evaluations conducted on the MIT-BIH Normal Sinus Rhythm Database (NSRDB) using three distinct types of contaminants - white Gaussian noise, synthetic pink noise, and real muscle artifact noise obtained from the Physionet noise stress database - demonstrate the superiority of the proposed ADAEKF over traditional EKF and AEKF approaches in both stationary and non-stationary environments. Furthermore, the proposed method was benchmarked against other nonlinear Kalman filter-based frameworks, including the ensemble Kalman filter (EnKF), unscented Kalman filter (UKF), and the recently introduced adaptive cubature Kalman filter (ACKF) for ECG denoising. The comparative results revealed that the proposed ADAEKF outperformed UKF, ACKF, and EnKF in nonstationary environments at low input SNRs. Additionally, the proposed ADAEKF improved the overall performance of EKF-based frameworks compared to the state-of-the-art marginalized particle extended Kalman filter (MP-EKF), while offering a simpler and more computationally efficient solution.