Performance analysis of a new implementation for the frequency-domain LMS adaptive filter

Abstract

The author analyzes the performance of a novel implementation of the frequency-domain LMS (least mean square) adaptive filter. This implementation is based on using the LMS adaptive algorithm to implement the DFT (discrete Fourier transform) part of the frequency-domain LMS adaptive filter. Numerical issues are also discussed. This is significant in this implementation because the performance of the frequency-domain LMS adaptive filter is dependent on the accuracy of the DFT computation obtained by using the LMS algorithm. The effects of inaccurate transform (DFT) computation of the performance of the transform-domain LMS adaptive filter are analyzed. The author also used a linear filter model to explain some of the observed nonideal effects. This indicated that, for large filter lengths, the accuracy of the DFTs computed diminishes.<>