A wave theory of long LMS adaptive filters

H.J. Butterweck

Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademicpeer-review

1 Citation (Scopus)

Abstract

Long LMS filters of the tapped-delay line type are in widespread use, particularly in acoustic applications. For the limiting case of an infinite line length the behaviour of such filters is shown to be governed by remarkably simple laws. This is true for the steady state, where for small stepsizes the weight-error correlations become independent of the input signal, but also for the transient behaviour, where the spatial Fourier transform of the weight-error distribution decays exponentially. Moreover, a necessary and (probably) sufficient stability bound for the stepsize is derived. The 'wave theory' developed for the infinite line length also predicts the behaviour of rather short filters with sufficient accuracy, particularly for a moderately coloured input signal. No independence assumption is required and no assumption concerning the spectral distribution of the additive noise. Under steady-state conditions, the weight-error correlation between two line taps is solely determined by the noise autocorrelation, with the time delay replaced by the tap distance.
Original languageEnglish
Title of host publicationProc. Eusipco '2000, European Signal Processing Conference
EditorsM. Gabbouj, P. Kuosmanen
Pages2461-2464
Publication statusPublished - 31 Mar 2015
Event10th European Signal Processing Conference (EUSIPCO 2000) - Tampere, Finland
Duration: 5 Sep 20008 Sep 2000
Conference number: 10

Conference

Conference10th European Signal Processing Conference (EUSIPCO 2000)
Abbreviated titleEUSIPCO 2000
CountryFinland
CityTampere
Period5/09/008/09/00

Fingerprint Dive into the research topics of 'A wave theory of long LMS adaptive filters'. Together they form a unique fingerprint.

  • Cite this

    Butterweck, H. J. (2015). A wave theory of long LMS adaptive filters. In M. Gabbouj, & P. Kuosmanen (Eds.), Proc. Eusipco '2000, European Signal Processing Conference (pp. 2461-2464). [7075816]