An efficient method for L-filter design

For a given or estimated noise type, the determination of optimal L-filter coefficients is computationally intensive and extremely time consuming. Thus, even though the L-filters obtained are very successful in restoring signals corrupted by nonGaussian noise, their on-line design in real-time applications is not practical. In this paper, a new methodology for the design of L-filters is presented in which the L-filter coefficients are obtained by approximating the covariance matrix of the ordered samples through the use of Taylor expansion. The paper also presents several examples that compare the proposed approach to the classical approach. The simulations show the value of the proposed approach.