USING LINEAR PREDICTION TO MITIGATE END EFFECTS IN EMPIRICAL MODE DECOMPOSITION

It is well known that empirical mode decomposition can suffer from computational instabilities at the signal boundaries. These "end effects" cause two problems: 1) sifting termination issues, i.e. convergence and 2) estimation error, i.e. accuracy. In this paper, we propose to use linear prediction in conjunction with a previous method to address end effects, to further mitigate these problems. We compare the proposed mitigation to the existing method and provide simulations which demonstrate that the new approach improves intrinsic mode function estimation accuracy while significantly improving convergence.