Computational noise effects on adaptive filter algorithms (Q795794)

This is a study of the propagation of errors in the adaptive least mean squares (LMS) algorithm for recursively estimating the weighting function coefficients of finite impulse response filters. Both noise induced by the assumed statistical nature of the processed data and finite word length arithmetic roundoff noise are included in the analysis. Propagation of the covariance matrix of the (normalized) estimation errors is studied under certain assumptions on the noise processes. These assumptions appear somewhat restrictive and as such this study should be seen as preliminary.