A study of two adaptive filters in tandem

Author

Ho, K.C.

Author_Institution

Dept. of Electr. Eng., Missouri Univ., Columbia, MO, USA

Volume

48

Issue

6

fYear

2000

fDate

6/1/2000 12:00:00 AM

Firstpage

1626

Lastpage

1636

Abstract

The tandem of adaptive filters is common in practice. An example is the tandem of echo cancellers in telecommunication networks. This paper analyzes the convergence characteristics and tracking behavior of two adaptive filters in tandem, together with a comparison of its performance with a single adaptive filter. The adaptive algorithm considered is the LMS and the analysis is on mean-square. The coefficient errors correspond to noise, lag bias and lag variance are examined separately The theoretical results are corroborated with simulations. The study shows that the tandem of two adaptive filters decreases the convergence speed compared to a single adaptive filter. In addition, in steady state and when the step-size is small tandeming increases the coefficient variance due to noise by a factor of 2.5, the coefficient variance due to tracking lag by a factor of 1.5, but decreases the mean-square coefficient bias due to lag by a factor of 2

Keywords

Markov processes; adaptive filters; adaptive signal processing; convergence of numerical methods; echo suppression; filtering theory; least mean squares methods; telecommunication networks; telephony; tracking filters; LMS adaptive algorithm; coefficient errors; coefficient variance; convergence characteristics; convergence speed; echo cancellers; first-order Markov model; lag bias; lag variance; mean-square analysis; mean-square coefficient bias; noise; performance; simulations; step-size; tandem adaptive filters; telecommunication networks; telephone connection; tracking behavior; tracking lag; Adaptive filters; Circuits; Convergence; Delay; Echo cancellers; Least squares approximation; Noise cancellation; Speech; Telephony; Testing;

fLanguage

English

Journal_Title

Signal Processing, IEEE Transactions on

Publisher

ieee

ISSN

1053-587X

Type

jour

DOI

10.1109/78.845920

Filename

845920