You are on page 1of 5

[Dhanasekaran, 3(1): January, 2014] ISSN: 2277-9655

Impact Factor: 1.852


http: // www.ijesrt.com(C)International Journal of Engineering Sciences & Research Technology
[18-22]

IJESRT
INTERNATIONAL JOURNAL OF ENGINEERING SCIENCES & RESEARCH
TECHNOLOGY
Efficient Active Noise Cancellation for Decision Tree Technique of ANN
Dhanasekaran.B

Ph.d Scholar, Department. of ECE, JJT University, Rajasthan, India
dhanasekarme@gmail.com
Abstract
Efficient active noise cancellation (EANC) has wide application in next generation human machine
Interaction to Automobile Heating Ventilating and Air Conditioning (HVAC) devices. For EANC various
Algorithms can able to run and get good result in standard output and better performance meaning of Decision Tree
Technique is related Decision Tree non functional neurons (algorithm) from EANC has been proposed, the Wiener
filter based on Least mean Squared(Lms) algorithm family is most sought after solution of EANC.
This family includes LMS,NLMS,LPB,..member out algorithms and many more some of these non-linear
algorithms which provides better solution for non-linear noisy environment. The component of the(EANC) systems
like Microphone and loud speaker Exhibit non linear ties themselves. The non-linear transfer function create worst
situation. For example , Fxslms algorithms behaves well than the second order VfxLms algorithm is conditions of
non-minimum phase and more important in the mean square error. If by any means it can be decided that, which
particular algorithm will suit more to the problem, application, and give wonderful solution to solve the problem
efficiently . This is a task which is some sort of prediction of suitable solution to the problems. The Redial Basis
Function of Natural Networks (RBFNN) has been known to be suitable for non-linear function approximation. The
classical approach RBF implementation is to be fix the number of hidden neurons based on some property of the
input data , and estimate data , and estimate the weights connecting the hidden and output neuron using Linear least
square method. This removal process of ineffective hidden reason is called Decision Tree Technique. In this work,
neurons are algorithms to specified signal the result of least means square error is decided through this RBFNN
approach. So decision free becomes imperative for identification of nonlinear systems which changing dynamics
become failing the decision tree the network would result in neurons inactive hidden neuron being present on the
dynamics which caused their creation becomes existent.

Keywords: Active noise cancellation, Decision Tree technique , Artificial neural network, Least mean square,
Winner filter, VSSNDLMS algorithm,.

I. Introduction
Active Noise Cancellation (ANC) is a method
for reducing undesired noise in the system. ANC is
achieved by introducing a canceling antinoise wave
through secondary sources. Electronic system
interconnected by secondary sources through an using a
specific signal processing algorithm for the particular
cancellation scheme[1] [10]. This project is to build a
Noise-cancelling headphone by means of active noise
control. It involves using a microphone placed near the
ear and electronic circuitry which generates in the
microphone, an "antinoise" sound wave with the opposite
polarity of the sound wave arriving. This result in
pernicious interference, which is cancel out the noise
within the enclosed volume of the headphone. This
report will demonstrate the approaches that take on
tackling the noise cancellation effects, along with results
comparison. The Active noise cancellation (ANC)
has become a
effective mechanism for control of low frequency
acoustic noise. The main application areas of
ANC are exhaust , motor noise ,HVAC systems ,
mobile phone echo and vibration
cancellation.The wiener filter based Least means
squared (LMS)algorithm family is most sought
after solution of ANC .This famil y includes LMS
, Fx-LMS, VFx-LMS, FsLMS and many more .
Then there are Kal man filter algorithm which are
basically based on recursive least square
algorithm . Some of these are nonlinear
algorithm, which
provides better solution for non linear noisy environme
nt. Thecomponents of the ANC systems like
microphones and loudspeaker exhibit
nonlinearities themselves. The
nonlinear transfer function of primary and
secondary path may itself creates worse situation
[Dhanasekaran, 3(1): January, 2014] ISSN: 2277-9655
Impact Factor: 1.852

http: // www.ijesrt.com(C)International Journal of Engineering Sciences & Research Technology
[18-22]

.If there is only one reference microphone , one
loud speaker and one error microphone then
the situation is termed as Single channel case in
ANC. Single channel is not sufficient for complex
real time application , where noise i s normall y
of mul ti order nonl inear characteri stics
. Recentl y , done work suggests several
met hods for t he development of acti ve control
of noi se process for a si ngle channel case.
Multi channel approach model i s close to actual
condi tions of fi eld or area where noise removal
is required. Nonl i near Mul tichannel model wi t h
tri gonometric expansion gi ve better resul ts t han
vol tra seri es met hod [2]. The FsLMS
al gorit hm behave good t han second order
VFXLMS al gorit hm i n condi tions as t he
fol lowi ng. 1).First, The noise signal received by a
reference microphone is non-linear and predictable
(chaotic) ,while the secondary path transfer function of
an ANC system is nonminimum phase.

2) The primary path exhibits nonlinearity.
The It was Shown by Dr. Debi Das
that FsLMS out performs the FxLMS as well as VFXL
MS under the situations where the noise signal
received by a reference microphone is nonlinear
and chaotic, while the secondary
path transfer function has non minimum phase[11],[12].
So, in a particular noisy environment specific
algorithm may exhibit better performance than
otheralgorithm. So if by any mean it can be decid
ed that , which particular algorithm will suits most to th
e problem application , may give wonderful
solution to solve the problem efficiently. This is a
task which is some sort of a prediction of suitable
solution to problem.
The Neural Net work (NN) is suitable
for approximation problem. The radial basis Function
(RBF NN) have been known to be suitable for non
linear function approximation[3],[4].The
topology of RBF is a t wo layer net work with one
hidden layer and one output layer with the basis
function used for the hidden layer neurons being
usually Gaussian. The classical approach to RBF
implementation is to fix the number of hidden
neurons based on some property of the input data,
and estimate the weights connecting the hidden and
output neurons using linear least square method.
The disadvantages of this approach are that it
results in too many hidden neurons. If hidden
neuron ineffective which can be detected and
removed as
learning progresses then a more effective net wor
k topology can be realized. This removal
of ineffective hidden neurons is called Decision
Tree Technique[3]. So pruning becomes
imperative for identification of nonlinear systems
with changing dynamics because failing to
Decision tree the net work , in such case will
results in numerous inactive hidden neurons
being present as the dynamics which caused their
creation becomes nonexistent. This has been our
approach to combine Pruning Technique of
artificial neural network (ANN) to ANC
algorithms.


Figure 1 Flow diagram of proposed noise reduction
VSSNLMS algorithm.

II. System Model
Assuming that speech and noise are uncorrelated, the
noisy speech signal x(n) can be represented as
x(n) = s(n) + d(n), (1)
where s(n) is the clean speech signal and d(n) is the noise
signal. The signal is divided into the overlapped frames
by window and the short-time Fourier transform (STFT)
is applied to each frame. The time-frequency
representation for each frame is as follows. X (k, l) =
S(k, l) + D(k, l), where (k =1, 2,..., L) are the frequency
bin index and (l =1,2, ..., L) are the frame index. The
power spectrum of the noisy speech
2
) , ( l k X can be
represented as
) , ( ) , ( ) , (
2 2
l k D l k S l k X

+ (2)
Where
2
) , ( l k S the power spectrum of the clean
speech signal and
2
) , ( l k S is the power spectrum of
the noise signal. The proposed algorithm is summarized
[Dhanasekaran, 3(1): January, 2014] ISSN: 2277-9655
Impact Factor: 1.852

http: // www.ijesrt.com(C)International Journal of Engineering Sciences & Research Technology
[18-22]

in the block diagram shown in Fig. 1. It is consists of
main components: input signal ,adaptive noise,
VSSNLMS, and sigmoid function with a priori SNR.

III. Proposed Noise Estimation And
Reduction Algorithm
The experimental work has been done on
Matlab. The noise reduction algorithm is based on the
STD of the noisy power spectrum in a time and
frequency dependent manner as follows:


= =
= =
L
l
K
k
f l k X
L
k x l k X
K
l x
1
2
1
2
1 , ) , (
1
) ( , ) , (
1
) (
(3)
, ) ( ) , (
1
) (
2
1
2
(
(
(

|
|
|

\
|
=

=
K
k
t
t
l x l k X
K
l v (4)
, ) ( ) , (
1
) (
2
1
2
(
(
(

|
|
|

\
|
=

=
L
l
f
f
k x l k X
L
k v (5)
, ) (
1
, ) (
1
1
2
1
2

=

= =
K
K
f f
L
l
t t
k v
K
l v
L
(6)
,
) (
) ( ,
) (
) (
2 2

= =
f
t
f
t
t
l
k v
k
l v
l

(7)
where ) ( 1 l x is the average noisy power spectrum in the
frequency bin, ) (k x f is the average noisy power
spectrum for the frame index, and

2
t
and

2
f
are the
assumed estimate of noise power. (7) Gives the ratio of
the (STD) for the noisy power spectrum in the time-
frequency bin to its average. In the case of a region in
which a speech signal is strong, the STD ratio by (7) will
be high. The ratio is generally not high for a region
without a speech signal. Therefore, we can use the ratio
in (7) to determine speech presence or speech-absence in
the time-frequency bins [7]. Classification of speech-
presence and speech absence in frames using an adaptive
sigmoid function based on a posteriori SNR Our method
uses an adaptive algorithm with a sigmoid function to
track the threshold and control the trade-off between
speech distortion and residual noise:

(

+
=
)) ) ( .( 10 exp( 1
1
) (
t t
t
l
l

, (8)
Where ) (l
t
is the adaptive threshold using the sigmoid
function. We defined a control parameter
t..
This
threshold ) (l
t
is adaptive in the sense that it changes
depending on the control parameter
t
. The control
parameter
t
is derived from the linear function using the
a posteriori signal to noise ratio (SNR) in frame index
[9].

, ) ( .
off s t
l SNR + =
(9)


( )
,
2 , ) (
2 , ) , (
log . 10 ) (
2
2
(
(
(
(
(

|
|

\
|
=

k D norm
l k X norm
l SNR
(10)
Where

=

5
1
2 2
) , ( 5 1 ) (
l
l k X k D is the average of
the
2
) , ( l k X initial 5 frames during the period of the
first silence and norm is the Euclidean length of a vector.

, .
min max
SNR
s off
= (11)

,
min max
max min
SNR SNR
s

=


(12)

Where
s
is the slope of the
t
,
off
is the offset of the
t

. the more the a posteriori SNR increases, the more the
t

decreases. Simulation results show that an increase in the

t
parameter is good for noisy signals with a low SNR of
less than 5 dB, and that a decrease in
t
is good for noisy
signals with a relatively high SNR of greater than 15 dB.
We can thus control the trade-off between speech
distortion and residual noise in the frame index using
t.

the adaptive threshold using the sigmoid function allows
for a trade-off between speech distortion and residual
noise by controlling
t
[5],[6],[8] . If a speech signal is
present, the ) (l
t
calculated by (8) will be extremely
small (i.e., very close to 0). Otherwise, the value of
) (l
t
calculated by (8) will be approximately 1.

IV. Experimental Results And Discussion
This approach is based on decision tree
technique of ANN. This technique are using to
prune less effective neurons from net work and
[Dhanasekaran, 3(1): January, 2014] ISSN: 2277-9655
Impact Factor: 1.852

http: // www.ijesrt.com(C)International Journal of Engineering Sciences & Research Technology
[18-22]

finally running systems with successful onl y
neuron. In different conditions different neuron
may give better results. The survival
chances of neurons depend upon its algorithm
efficiency in given circumstances. The step size
factor was varied depending upon the optimization
required between complexity and amount of noise
cancellation.

Figure 2 VSSNDLMS algorithm samples and output

In Noisy conditions of case t wo
VSSNDLMS algorithm was most efficient.
Remaining all algorithm were decision tree, as
they were far away from standard threshold
value.

V. Conclusion
In this approach we found out those in particular
noisy conditions most suitable algorithms can be sort out
using our design Tree Method. This method can also be
applied to none cancellation, signal estimation and
vibration cancellation problems means of MSE Obtained
0.2469 deviation from them when calculated. It is
observed that VSSNDLMS is better than the other
unwanted application for EANC of SONG input or
speech for no. of iteration 2000. The Threshold criteria
result in 0.001 is the least values amongst the all the
applied algorithm.
Hence Pruning of all other algorithm include the
selection of that algorithm an Sum + Val neuron. Next
good performing Lms itself for 2000 Iteration the Excuse
performance of the Lms is due to Linear nature of the
song or speech input noise.
Here this destination have tried to reduce such
week signals by using basic adaptive approach But in
real time applications which involves thousand of such
week signals and knowing the fact but noise , at any cost,
cannot be cancelled at hundred percent, this approach
cannot do us a good favor and keep this research topic
active in future work.
The future ambition or task to apply this novel
concept for kalman and winer filters family algorithm
simultaneously , where an it is been tried to implement it
for only outer noise removal automobiles and noise
cancellation is head phones, mobile, noise performance
can be upgraded by finding proper step factor and
threshold Value . the algorithm level up gradation of
threshold and decision Tree Criteria may improve
performance.
Adaptive designed signal processing is rapidly
growing Branch of Digital signal Processing (DSP) used
has a great significance in the design of adaptive systems
. The various s signal processing applications demands
for reduction in fraud off between mid adjustment and
convergence rate, taking realization of algorithm into
account. These in slope of improvement in existing
algorithm which reduces complexity and fulfill and our
stringent conditions for stability. This thesis dealt with
trans versal FIR adaptive filter, this is only one method
of digital filtering and other techniques such inflict
impulse response (IIR) of lattice filtering may move to
be effective in an noise cancellation application. The real
time EANC system can be implemented successfully
using the TITmsc 6713 DSK.
Hence this system can only cancel those echos
that fair within the length of Adaptive filter (FIR). There
is lot, which can be done in future for improvement on
the methods for acoustic noise cancellation the fixed
digital signal processing and particularly adaptive
filtering is vast and further research and development in
this area can be result in some improvement on the
methods studying in this dissertation.

VI. References
[1] Dayang Zhou and Victor E. DeBrunner Efficient
Adaptive VolterraFilters For Active Nonlinear Noise
Control With A Linear Secondary Path in IEEE
2004.
[2] S. F. Boll, Suppression of acoustic noise in speech
using spectral subtraction, IEEE Trans. on
Acoustics, Speech, and Signal Processing, vol. 27,
no. 2, pp. 113-120, 1979.
[3] Lu.Yingwei and Narashiman Sundarjan
Performance Evaluation of a Sequential Minimal
Radial Baisi Function(RBF) Neural
Network Learning Algorithm in IEEE trans. On
Neural Networks, vol.9, No.2,March 21998.
[4] Lu Yingwei , N.Sundararajan, P.Saratchandran
Adaptive Nonliner Systems Identification Using
Minimal Radial Bsis Function Neural Networks in
IEEE 1996
[Dhanasekaran, 3(1): January, 2014] ISSN: 2277-9655
Impact Factor: 1.852

http: // www.ijesrt.com(C)International Journal of Engineering Sciences & Research Technology
[18-22]

[5] Y. Ephraim and D. Malah, Speech enhancement
using a minimum mean-square error short-time
spectral amplitude estimator, IEEE Trans. On
Acoustics, Speech, and Signal Processing, vol. 32,
no. 6, pp. 1109-1121, 1984.
[6] R. Sundarrajan and C. L. Philipos, A noise
estimation algorithm for highly non-stationary
environments, Speech Communication, vol. 48, pp.
220-231, 2006.
[7] S. J. Lee and S. H. Kim, Speech enhancement using
gain function of noisy power estimates and linear
regression, Proc. of IEEE/FBIT Int. Conf.
Frontiers in the Convergence of Bioscience and
Information Technologies, pp. 613-616, October
2007.
[8] Cohen, Speech enhancement using a noncausal a
priori SNR estimator, IEEE Signal Processing
Letters, vol. 11, no. 9, pp. 725-728, 2004.
[9] ITU-T, Subjective test methodology for evaluating
speech communication systems that include noise
suppression algorithm, ITU-T Recommendation, p.
835, 2003.
[10] S.M. Kuo and D.R.Morgan ,Active Noise Control
Algorithms and DSP Implementations. New York
:Wiley ,1996
[11] Debi Prasad Das ,Filtered s LMS Algorithm for
Multichannel Activeontrol Of Nonolinear Noise
Processes in IEEE Transcactions on SpeechAnd
Audio Processing , vol 14 ,no.5 Spetember 2006
[12] Debi Das and Ganpati Panda , Active Mitigation of
Nonlinear NoiseProcesses Using A Noval Filtered-s
LMS Algorithm in IEEETransactions of Speech
and Audio Processing, vol. 12,No..3,May 2004.