An adaptive kalman filter for removing baseline wandering in ECG

Baseline wandering interference misleads ECG annotators from accurate identification of the ECG features. Previous work that deals with baseline wandering removal requires the identification of the QRS complex or other ECG features prior to baseline removal. This paper proposes an adaptive Kalman filter for the real time removal of baseline wandering using a polynomial approximation independent of the signal characteristics. A state space model is used with an adaptive Kalman filter to estimate the state variables, including the baseline wandering approximation from the previous values of the original ECG signal. This approach is applied to the (PTB) Diagnostic ECG Database and to a ECG signal disturbed by white noise and a second order baseline wandering. The results show accurate and improved baseline wandering estimation and removal as compared to moving averaging and cubic spline techniques.