Analysis of the LMS algorithm with delayed coefficient update

Author

Ernst, Thomas ; Kaelin, August

Author_Institution

Inst. for Signal & Inf. Process., Swiss Federal Inst. of Technol., Zurich, Switzerland

Volume

2

fYear

1995

fDate

30 Apr-3 May 1995

Firstpage

1247

Abstract

There are many practical applications of the Least Mean Square (LMS) algorithm where a delay D in the error path is either necessary or unavoidable. We present a new and exact analysis of the Delayed LMS (DLMS) algorithm, which takes such a delay into account. The analysis leads to an upper bound on the allowable step-size that can be determined by finding the roots of a polynomial of order D+1. Computer simulations confirm the tightness of the bound

Keywords

Viterbi detection; adaptive signal processing; approximation theory; delays; discrete time systems; least mean squares methods; polynomials; LMS algorithm; Viterbi detection; adaptive signal processing; allowable step-size; computer simulation; critical step size; delayed LMS algorithm; delayed coefficient update; least mean square algorithm; polynomial roots; steady-state delay factor; steady-state excess mean squared error; upper bound tightness; Algorithm design and analysis; Computer errors; Computer simulation; Delay; Error correction; Least squares approximation; Polynomials; Stability; Upper bound; Viterbi algorithm;

fLanguage

English

Publisher

ieee

Conference_Titel

Circuits and Systems, 1995. ISCAS '95., 1995 IEEE International Symposium on

Conference_Location

Seattle, WA

Print_ISBN

0-7803-2570-2

Type

conf

DOI

10.1109/ISCAS.1995.520371

Filename

520371