Regularized LMS methods for baseline wandering removal in wearable ECG devices

The acquisition of electrocardiogram (ECG) signals by means of light and reduced size devices can be usefully exploited in several health-care applications, e.g., in remote monitoring of patients. ECG signals, however, are affected by several artifacts due to noise and other disturbances. One of the major ECG degradation is represented by the baseline wandering (BW), a slowly varying change of the signal trend. Several BW removal algorithms have been proposed into the literature, even though their complexity often hinders their implementation into wearable devices characterized by limited computational and memory resources. In this study, we formalize the BW removal problem as a mean-square-error regression with an ℓ1 or ℓ2 penalty function and propose low-complexity least mean squares (LMS) solutions that comply with a wearable device implementation.