Scribd
Upload
200 views

Convergence of LMS Algorithm

The document discusses the least mean square (LMS) algorithm. It introduces the LMS algorithm and explains that it was developed in 1960 as a simple stochastic gradient algorithm that does not require measuring correlation functions or matrix inversion. The LMS algorithm operates using a filtering process to generate an estimation error and an adaptive process to automatically adjust filter parameters based on this error via a feedback loop. The convergence behavior of the LMS algorithm depends on the eigenstructure of the correlation matrix of the filter tap inputs, with directional convergence occurring if the tap inputs are correlated.

Uploaded by

sivadanams
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
200 views

Convergence of LMS Algorithm

The document discusses the least mean square (LMS) algorithm. It introduces the LMS algorithm and explains that it was developed in 1960 as a simple stochastic gradient algorithm that does not require measuring correlation functions or matrix inversion. The LMS algorithm operates using a filtering process to generate an estimation error and an adaptive process to automatically adjust filter parameters based on this error via a feedback loop. The convergence behavior of the LMS algorithm depends on the eigenstructure of the correlation matrix of the filter tap inputs, with directional convergence occurring if the tap inputs are correlated.

Uploaded by

sivadanams
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

Convergence of

Least Mean-Square
Algorithm
Introduction to LMS Algorithm
The LMS algorithm was developed by Widrow
and Hoff in 1960.
The LMS algorithm is an important member of
the family of stochastic gradient algorithm.
LMS algorithm is a simple, does not require
measurement of the pertinent correlation
functions and also the matrix inversion.
Because of this, LMS algorithm became a
standard algorithm.
M.SUMALATHA
13KB1D3813
N.B.K.R.I.S.T 2
Structure and Operation of the LMS
Algorithm
The LMS algorithm is a Linear Adaptive Filtering Algorithm, consisting of two
basic processes:
1. A Filtering Process : Filtering Process involves two basic processes, namely:
(i) Computing the output of a linear filter in response to an input signal
(ii) Generating an estimation error by comparing this output with a desired
response
2. An Adoptive Process : Adoptive process involves the automatic adjustment of the
parameters of the filter in accordance with the estimation error.
The combination of these two processes working together constitutes to
feedback loop.
M.SUMALATHA
13KB1D3813
N.B.K.R.I.S.T 3
Feedback Loop
Fig-a: Feedback loop consists of a traversal filter, around which
the LMS algorithm is built.
This is responsible for performing the filtering sequence.
Fig-b: Consists of a mechanism for performing the adaptive
control process on the tap weights
of the transversal filter. Therefore,
it is known as adaptive
weight-control mechanisms.
M.SUMALATHA
13KB1D3813
N.B.K.R.I.S.T 4
Convergence of the LMS Algorithm
The Eigen structure of the correlation matrix R of a
traversal filters tap inputs has a profound impact on the
convergence behaviour of the LMS algorithm used to adapt
the filters tap weights.
If tap inputs are drawn from a white-noise process, they are
uncorrelated, the eigen valuespread (R) of the correlation matrix
is unity, LMS Algorithm is a non-directional convergence.
If the tap inputs are correlated and the eigen valuespread (R) is
large, the convergence of the LMS algorithm is directional in
nature.
M.SUMALATHA
13KB1D3813
N.B.K.R.I.S.T 5
Convergence of the LMS Algorithm Cont..
The directionality of convergence exhibited by the
LMS algorithm under non-white inputs manifests
itself in two ways:
1. The speed of convergence of the algorithm is faster in certain
directions in the algorithms weight space than in some other
directions.
2. Depending on the direction along which the convergence of the
algorithm takes place, it is possible for the convergence to be
accelerated by an increase in the eigenvalue spread (R).
These two aspects of the directionality of convergence are explained
with an example.
M.SUMALATHA
13KB1D3813
N.B.K.R.I.S.T 6
Convergence of the LMS Algorithm Cont..
M.SUMALATHA
13KB1D3813
N.B.K.R.I.S.T 7
Convergence of the LMS Algorithm Cont..
M.SUMALATHA
13KB1D3813
N.B.K.R.I.S.T 8
Thank You
&
Queries Please
M.SUMALATHA
13KB1D3813
N.B.K.R.I.S.T 9

You might also like