Modified filtered-reference / filtered-error LMS algorithm: Algorithm and stochastic model

This paper proposes a modified filtered-reference / filtered-error least-mean-square (MFxFeLMS) algorithm. In addition, a stochastic model for the first and second moments of the MFxFeLMS is derived. The proposed model is obtained without invoking the classic Independence Theory (IT), allowing for slow adaptation and Gaussian input signals. Numerical simulations confirm the very good agreement between the results obtained by the Monte Carlo (MC) method and from the proposed model for both white and colored Gaussian inputs.