An Unbiased Linear Artificial Neural Network with Normalized Adaptive Coefficients for Filtering Noisy ECG Signals

The electrocardiogram (ECG) is the most commonly used signal for diagnostic purposes in medicine. The adaptive filtering technique is suited for filtering ECG signals, which are inherently nonstationary. In this paper, we propose a novel neural-network-based adaptive filter to eliminate high-frequency random noise in ECG signals. We make use of a linear artificial neural network (ANN) with delayed values of the ECG time series as the filter inputs. The ANN does not contain a bias in its summation unit, and the coefficients are normalized. During the learning process, the normalized coefficients are used in the steepest-descent algorithm in order to achieve efficient online filtering of noisy ECG signals.