Abstract
This paper improves the learning dictionary construction method for morphological component analysis (MCA) to separate the atrial and ventricular signals. The incoherence is added into the objective function to reduce the sparsity ratio between the atrial and ventricular dictionaries. By using the dictionaries, atrial and ventricular activities are separated from the location of the coefficients. We test the methods on both the synthetic and real atrial data. While extracting AFW from synthetic data, we use the Poisson relation as the measure. The result shows that we can obtain greater relation value using the method this paper presents than using the methods of ABS and PCA. We also conduct spectral analysis on AFW extracted from real atrial data.