Publication detail

Nonlinear Stochastic Gradient Algorithm with Variable Step-Size

Vladimir Malenovsky, Ing. Radek Zezula, Ing. Ivan Koula, Ing.

Original Title

Nonlinear Stochastic Gradient Algorithm with Variable Step-Size

Type

conference paper

Language

English

Original Abstract

This letter proposes a new algorithm, which uses an optimal step-size (OSS) weight-adjustment scheme. This strategy leads to better convergence rate and misadjustment in environments with sudden change of parameters and for colored input data. The computational complexity is comparable with the well-known RLS. The performance of the novel approach is verified by simulations under system identification scenario and compared with the peroformance of the NLMS and RLS algorithms. Experimental results for car-interior echo cancelation are presented and a discussion is provided for improving the performance using exponentially averaged gradient vector.

Keywords

adaptive filter, echo cancelling, handsfree, noise suppression, gradient vector

Authors

Vladimir Malenovsky, Ing. Radek Zezula, Ing. Ivan Koula, Ing.

RIV year

2005

Released

1. 9. 2005

Publisher

VUT Brno

Location

Brno

ISBN

80-214-2972-0

Book

Proceedings of the Intl. Conference TSP 2005

Edition

28

Edition number

1

Pages from

103

Pages to

108

Pages count

6

BibTex

@inproceedings{BUT15145,
  author="Vladimír {Malenovský} and Ivan {Koula} and Radek {Zezula}",
  title="Nonlinear Stochastic Gradient Algorithm with Variable Step-Size",
  booktitle="Proceedings of the Intl. Conference TSP 2005",
  year="2005",
  series="28",
  volume="28",
  number="1",
  pages="6",
  publisher="VUT Brno",
  address="Brno",
  isbn="80-214-2972-0"
}