Projective filtering based on L1-norm PC

The paper presents a modification of nonlinear state-space projections (NSSP) method. The proposed approach deals with the sub-space estimation problem. In the original NSSP method, the principal component analysis (PCA) is used for the subspace determination. The classical PCA uses L2-norm. It is well known that the L2-norm is sensitive to outliers. Thus, in this paper the L1-norm PCA is proposed a subspace determination. In numerical experiments an analytic signal and real ECG signals are processed with the proposed method. The signals are contaminated with Gaussian distributed noise with different signal to noise ratio (SNR). Obtained results confirm the usefulness of the proposed modification.