Volume 45, Number 6, 2004 599

Noise canceling in audio signal with adaptive filter

Drghiciu Nicolae, Reiz Romulus
University of Oradea,
3-5, Armatei Romane Str., Oradea, Romania, 410087


Abstract: - In this paper, an adaptive noise canceller will be presented and some useful
observations will be done with the canceling process over the audio signal

Key-words: Adaptive filter, input signal, output signal, noise.

1 Introduction

All the physical systems when they
operate, they produce physical noise,
which is a mixing of an infinitive number
of sinusoidal harmonics with different
frequencies. So, the initial signal
information is corrupted with this noise.
This complex signal, may be very noisy, so
much that the human ear or other system
which may follow it, cannot receive correct
the initial signal.
So, an algorithm has to be invented which
must be able to separate the physical noise
from the information signal and to output
the information signal without noise. This is
not possible, as there is not a perfect
system. So, this algorithm should have the
ability to reduce the noise level as much as
it can.
A good noise-canceling algorithm is the
algorithm, which is presented in this paper.
This algorithm is based on the Least Mean
Square algorithm. This system, operates
like an automatic control system with
feedback where the feedback performs an
adaptation to the filter coefficients,
otherwise it adjusts the filtering of the noise
in each sample.
Also, previous algorithms were proposed
from different peoples in the past years but
all of them or most of them could be
implemented on a good processor like a
DSP and only on it [2][3]. The algorithm
which is proposed on this paper can be
implemented and on a micro controller.



2 Problem formulation

A simple audio signal is represented by the
following mathematical equation
( ) ) * / * * 2 sin( ) ( 1 n fs fo pi n x =
Where is the frequency of the audio
signal, is the sampling frequency and
is the discrete time base.
0 f
fs
n



Fig.1 The simple audio signal without noise

When white noise is added to this signal
then it becomes
( ) ( ) ( ) ( ) n noise n fs fo pi n c + = * / * * 2 sin 2

So, the problem is to extract the useful
signal from this complex signal.

ACTA ELECTROTEHNICA 600


Fig.2 The audio signal with noise

A simple adaptive filter noise canceller is
presented in the following diagram.


Audio signal with noise
c( n) - u
k
Output - y
k



Noise
d(n)

error - e
k



Fig.3 Adaptive filter noise canceller

2.1 Brief adaptive filter theory

One of the most successful adaptive
algorithms is the LMS algorithm developed
by Widrow and his coworkers. Instead of
computing which is the optimal
solution of the wiener filter, the LMS
coefficients are adjusted from sample to
sample in such a way as to minimize the
MSE (mean square error).
opt
W

The LMS is based on the steepest descent
algorithm where the weight vector is
updated from sample to sample as in the
next equation.
( )
k k k
W W =
+

1
3
Where W
k
and are the weight or the
impulse response w (n) of the filter and the
true gradient vectors respectively, at the
sampling instant.
k

th
k
Controls the stability
and rate of convergence.

The steepest descent algorithm in the
previous equation still requires knowledge
of R and P , since
k
is obtained by the
next equation.
( ) 0 2 2 4 = + = = RW P
dW
dJ
k

The LMS algorithm is a practical method of
obtaining estimations of the filter weights
W
k
in real time without the matrix inversion
of the next equation or the
direct computation of the autocorrelation
and cross correlation. The Widrow Hopf
LMS algorithm for updating the weights
from sample to sample is given by
P R W
opt
1
=
( )
k k k k
e X W W 2 5
1
+ =
+
+
Where:
k
T
k k k
X W y e =
Clearly, the LMS algorithm above does not
require prior knowledge of the signal
statistics (that is the correlations R and P ),
but instead uses their instantaneous
estimations. The weights obtained by the
LMS algorithm are only estimations, but
these estimations improve gradually with
time as the weights are adjusted and the
filter learns the characteristics of the signal.
Eventually, the weights converge. The
condition of convergence is:
( )
max
1
0 6

> <
Where
max
is the maximum eigenvalue of
the input data covariance matrix In practice,
never reaches the theoretical optimum
(the Wiener solution), as the algorithm
iteration step can never be zero but it has a
discrete value.
k
W
Furthermore, the LMS algorithm can be
summarized in the next equation for the
implementation of the

filter.
( ) ( ) ( ) ( )

+
=
=
1
0
* 7
N
i
i
i k u k w k y
( ) ) ( ) ( ) ( ) 1 ( 8 e u h h + = +
In which: ( ) k y d e = ) ( ) (
Adaptive
Filter W
k+1
Volume 45, Number 6, 2004 601
In the above equation, denotes the
transpose of the vector . The desired
response
T
inv
h
) ( d can be found from the
adaptive filter scheme to be
( ) ) ( ) ( ) ( 9 g w h d
T
x
=
The notation ) ( u denotes a vector of
samples arranged as
( ) [
T
N u u u k u ) 1 ( . . .. . . . ) 1 ( ) ( ) ( 10 + = ]

Where is the order of the filter. N


3 Problem solution
The problem solution may be based on
standard adaptive techniques and especially
on the adaptive filter noise canceller.
This filter gets the physical noise and
filters it in such a way that the algebraic
addition with the complex signal produces a
new signal with a very low level of noise.
The schematic diagram of the adaptive filter
noise canceller was given above and it is
presented again in the following diagram.

Audio signal with noise
c( n) - u
k
Output - y
k



Noise
d(n)

error - e
k




Fig.4 The used adaptive noise canceller

As it can be observed, the adaptive filter
noise canceller is a system with feedback as
a system with automatic control. For this
reason, there is stability problem. The
problem arises when the error signals
amplitude increases in accordance with time
or when no convergence is obtained.
The mathematical algorithm, which is
used by the adaptive filter for the filters
coefficients update, is called LMS (Least
mean square) algorithm. This algorithm
tries to minimize the square of the error
signal so; the audio signal has been
reconstructed quite well.
The system output is the error signal. The
error signal is the algebraic subtraction
between the audio signal with noise and the
noise. When the LMS algorithm performs a
good convergence then the only error signal
is the audio signal. So, the error signal must
be the noise canceller output.
The LMS algorithm has also an initial
step variable , which is the step of the
convergence in each iteration. In the matlab
code where the adaptive filter noise
canceller is built, the NLMS algorithm
used. The only difference is that the step
variable , changes in each iteration with
the next equation:
( )
( )
2
) ( 11
k u +
=

where is a positive constant, usually less
than , and 2 is a small positive constant.
The symbol
2
) ( u is the energy of the
signal in a predefined buffer.


3.1 Implementation of the adaptive
in matlab and
waveforms
filter noise canceller
+
The adaptive noise canceller was
implemented in matlab as it offers the
opportunity for a quick and very
operation of the system.
professional way to study the whole
The matlab code, which was written, is
presented now.

function [x] = noice_canceller
clear all;
%**********************
%**** Time Base *******
%**********************
N = 2000;
n = 1:1:N;
%***********************
%**** System Vars
%***********************
fo = 500;
Fs = 44100;
db = 5;
%***********************
Adaptive
Filter W
k+1
ACTA ELECTROTEHNICA 602
%***** Signal Generation
%***********************
audio = sin(2*pi*fo/Fs*n); %+
0.7*sin(2*pi*5*fo/Fs*n));
%**********************
%**** Noise Generation
%*********************
%noise_plus_audio = awgn(audio,db);
noise = wgn(1,N,db)/2;
%noise = 0.1*randn(1,N);
for i=1:1:N
noise_plus_audio(i) = (audio(i) + noise(i)/2);
end
%**************************************
%**** Initialization of Noise Canceller
%**************************************
w0 = zeros(1,32); % Initial filter coefficients
%wi = zeros(1,255);
mu = 0.4; % LMS step size
S = initnlms(w0,mu,[],1);
%***************************************
%***** Apply Noise Canceller over signal
%***************************************
for i=1:1:N
[y,canceller_output(i),S] =
adaptnlms(noise(i),noise_plus_audio(i),S);
%***************************************
%***** Calculate FFT of Signals
%***************************************
ff_mix = abs(fft(noise_plus_audio,1024));
ff_output = abs(fft(canceller_output,1024));
%***************************************
%***** Calculate Error
%***************************************
error(i) = audio(i) - canceller_output(i);
%***********************
%*** Plots *************
%***********************
subplot(6,1,1);
plot(noise_plus_audio,'r');
axis([0 N -2 2]);
title('Signal and Noise');
subplot(6,1,2);
plot(audio,'g');
axis([0 N -2 2]);
title('Audio Signal');
subplot(6,1,3);
plot(canceller_output,'b');
title('Canceller Output');
axis([0 N -2 2]);
subplot(6,1,4);
plot(ff_mix,'r');
title('FFT of Mixed Signal');
subplot(6,1,5);
plot(ff_output,'r');
title(' FFT of Output');
subplot(6,1,6);
plot(error,'g');
axis([0 N -2 2]);
title('Error of FFT');
drawnow;
end

The output waveforms of the system are
plotted in the next figure.


Fig.5 Noise canceller output waveforms


4 Conclusion
To sum up, the adaptive noise canceller is
a very efficient and useful system in many
applications with sound, video etc. The only
disadvantage is that it needs digital signal
processor DSP for its operation.

References:
[1] Simon Haykin, Adaptive Filter Theory,
Prentice Hall 2002
[2] LMS adaptive-filtering
algorithm: Taylor, John T. and Qiuting
Huang: Electrical Filters. New York:
CRC Press, 1997 pp. 207-214.
[3] B. Widrow, J. R. Glover, Jr., J. M.
McCool, J. Kaunitz, C. S. Williams, R.
H. Hearn, J. R. Zeidler, E. Dong, Jr., and
R. C. Goodlin, "Adaptive noise
cancelling: Principles and applications,"
Proc. IEEE, vol. 63, pp. 1692--1716,
Dec. 1975.