Content uploaded by Woon-Seng Gan
Author content
All content in this area was uploaded by Woon-Seng Gan
Content may be subject to copyright.
Active Noise Control: Open Problems and Challenges
Sen M. Kuo and Kevin Kuo
Department of Electrical Engineering
Northern Illinois University
DeKalb, Illinois 60115, USA
Woon Seng Gan
School of Electrical and Electronic Engineering
Nanyang Technological University
Singapore
Abstract—This paper introduces active noise control (ANC)
techniques focusing on challenges in developing practical
applications and open problems for research from the signal
processing perspective. We propose advanced strategies to
further improve ANC performance that include virtual sensing,
residual noise masking, and active sound quality control. We
also present several challenges for real-world applications such
as canceling high-frequency and impulse-like noises, and
reducing noises generated from moving sources. Finally, ANC
applications in consumer products and healthcare devices are
used to demonstrate some potential add-on functions for
promoting cost-effective ANC products.
I. INTRODUCTION
Active noise control systems [1-6] cancel the undesired noise
by generating a secondary noise of equal amplitude and
opposite phase of the primary noise for the cancellation of
both noises based on the principle of superposition. The ANC
system is very effective for attenuating low-frequency noises
where passive methods are expensive and ineffective. Most
practical ANC systems use adaptive filters to automatically
track variations of unknown noise characteristics and
environment. The most commonly used adaptive filter is the
finite impulse response (FIR) filter with the filtered-X least-
mean-square (FXLMS) algorithm [7, 8]. The commercially
available digital signal processors [9] simplified real-time
experimentation of ANC algorithms, thus promoting
development and application of ANC systems.
This paper briefly reviews basic ANC algorithms and
proposes some open problems for further research from signal
processing perspectives. We also present advanced techniques
such as virtual sensing, residual noise masking, and active
sound quality control to improve system performance for
commercialization. In addition, we introduce some add-on
functions to increase the overall value of ANC systems and
make them more attractive and cost-effective for real-world
applications.
II. ANC SYSTEMS AND OPEN PROBLEMS
The basic single-channel ANC system using the FXLMS
algorithm is illustrated in Fig. 1 [2], where the reference
signal x(n) is picked up by a reference sensor. The reference
signal is processed by the adaptive filter W(z) to generate the
canceling signal y(n) driven by a secondary loudspeaker. The
error sensor is used to monitor the ANC performance by
sensing the residual noise e(n). The use of the adaptive
algorithms for ANC systems is necessary to compensate for
the secondary path S(z) [2].
P(z)
Σ
W(z) S(z)
LMS
d(n) x(n)
y
(
n
)
y′(n)
e(n)
+
-
Fig. 1 Basic single-channel ANC system using the
FXLMS algorithm
In this section, we use basic ANC algorithms to present some
open problems for further research from adaptive signal
processing viewpoints.
A. Broadband Feedforward Systems
As shown in Fig. 1, the secondary signal y(n) is computed as
[2]
)()()( nnny Txw=, (1)
where
[]
w() () () ()nwnwn wn
L
T
=−01 1
... and
x() [() ( ) ( )]nxnxn xnL T
=−−+ ... 11
are the coefficient and
signal vectors of W(z), respectively, and L is the filter length.
The FXLMS algorithm updates the coefficient vector as
wwx( ) () ()()nnnen+= + ′
1
μ
, (2)
where μ is the step size (or convergence factor) that
determines the convergence speed,
′=∗xx() () ()nsn n
(3)
is the filtered signal vector and ()sn is the impulse response
of the secondary-path estimation filter, ()Sz.
The most popular filter structure used for ANC systems is
the FIR filter given in (1). It can be replaced by infinite
impulse response filters, lattice filters, transform-domain
filters, subband filters, etc. [2]. The convergence rate of the
978-1-4244-6878-2/10/$26.00 ©2010 IEEE
164
FXLMS algorithm given in (2) can be improved by using
variable step sizes, recursive least square algorithms, affine
projection algorithms, Kalman algorithms, etc. In addition,
nonlinear adaptive filters with associated algorithms can be
used to reduce nonlinear effects caused by different factors. It
is still an area of open research to develop more advanced
adaptive algorithms in order to further improve the
performance of ANC systems.
The FXLMS algorithm expressed in (2) and (3) is very
tolerant of errors made in the estimation of S(z) by the filter
()Sz. Within the limit of slow adaptation, the algorithm will
converge with nearly 9
0
o of phase error between ()Sz and
S(z). Therefore, off-line modeling can be used to estimate S(z)
during the initial training stage for most ANC applications,
however, on-line modeling may be required for some
applications that involves fast changing environments. The
detailed procedure for off-line modeling and some on-line
secondary-path modeling algorithms are summarized in [2].
The development of robust and efficient on-line secondary-
path modeling algorithms still deserves further research. In
particular, development of on-line modeling techniques
without using additive noise is critical.
The acoustic ANC system shown in Fig. 1 uses a reference
microphone to pick up the reference noise. Unfortunately, the
anti-noise generated by the secondary loudspeaker also
radiates upstream to the reference microphone, resulting in a
undesired acoustic feedback that corrupts the reference signal
x(n). One solution for eliminating acoustic feedback is to use a
feedback neutralization filter [2]. Since the primary noise is
highly correlated with the anti-noise, the on-line adaptation of
the feedback neutralization filter must be inhibited when the
ANC system is in operation. Therefore, analysis on the effects
of acoustic feedback and the development of effective on-line
feedback-path modeling techniques still remains open for
further research [11].
B. Narrowband Feedforward Systems
A narrowband ANC system uses an internally synthesized
reference signal x(n). Two types of reference signals are
commonly used in narrowband ANC systems: (1) an impulse
train with a period equal to the inverse of the fundamental
frequency of the periodic noise [12], and (2) sinusoidal
signals that have the same frequencies as the corresponding
harmonic components [13].
A digital recursive quadratic oscillator generates two
orthogonal cosine and sine components. These two signals
are separately weighted and then summed to produce the
canceling signal y(n). In practical applications, periodic noise
usually contains multiple narrowband components. This
requires higher-order adaptive filters that can be implemented
by using direct, parallel, direct/parallel, or cascade forms [2].
Analysis of narrowband ANC systems is usually based on
a single-frequency case. The analysis and optimization of
widely-used parallel and other forms for multiple-frequency
cases is open for research. Also, analysis of using single error
signal e(n) to update multiple adaptive filters deserves further
study and improvement. In addition, a narrowband ANC
system assumes the reference signal x(n) has the same
frequency as the primary noise d(n) at the error sensor
location. In many practical applications, the reference
sinusoidal frequencies used by the adaptive filters may be
different than the actual frequencies of primary noise. This
frequency mismatch will degrade the performance of ANC
systems, and these effects have been analyzed recently [14];
however, effective solutions remain to be developed for
practical applications.
C. Adaptive Feedback ANC
A single-channel adaptive feedback ANC system synthesizes
(or regenerates) the reference signal x(n) based on the
adaptive filter output y(n) with the secondary-path model
()Sz and the error signal e(n). Thus, this technique is also
known as internal model control [15]. We can estimate the
primary noise d(n) exactly if () ()Sz Sz≈, and use this
estimated primary noise as the reference signal x(n). This
adaptive feedback ANC algorithm has many advantages and
can be applied in ANC headsets and other industrial
applications [16].
There are many open problems for further study of
adaptive feedback ANC systems. For example, what kinds of
noises can be effectively reduced by the adaptive feedback
ANC, the effects of secondary-path modeling error on the
accuracy of synthesized primary signal x(n) and the
performance of the ANC system, and the stability and
convergence rate of the systems.
III. CHALLENGES FOR IMPROVING ANC PERFORMANCE
Fig. 2 Experimental setup of the snore ANC system
In this section, we use the snore ANC system installed on a
traditional headboard [17] as an example for presenting
several challenges to further improve performance of ANC
systems for practical applications. The snore ANC for the bed
partner is shown in Fig. 2. Two secondary loudspeakers and
two error microphones are mounted on the headboard. A
model of a human torso called the KEMAR (Knowles
Electronics Mannequin for Acoustics Research) is used as the
bed partner. Two microphones inside the ear cavities of the
KEMAR are used to evaluate the performance at the ears of
165
the bed partner, which mimics real performance perceived by
the human.
The spectral plots of signals picked up by the error
microphones and the microphones inside the ears of the
KEMAR were obtained using an HP dynamic signal analyzer.
These plots show the spectra of noises before (ANC OFF)
and after (ANC ON) the operation of active snore
cancellation.
Fig. 3 Spectra of signals at the left error microphone
Figure 3 shows the spectra of snore signals picked up by
the left error microphone before (original snore) and after
(residual snore) the operation of ANC system. The average
noise reduction is about 10-20 dB. Figure 4 shows the spectra
of signals sensed by the microphone inside the left ear of the
KEMAR. The average snore noise reduction is about 5-10
dB. These results clearly show that better ANC performance
can be achieved by placing the error microphones close to the
ears of KEMAR, especially at the high-frequency region that
has a smaller quiet zone.
Fig. 4 Spectra of snore picked up by the microphone inside
the left ear of KEMAR before and after ANC
A. Virtual Sensing Techniques
As shown in Figs. 3 and 4, the optimum locations of error
microphones in snore ANC system are close to the listener’s
ears for maximum noise reduction because the “quiet zones”
are centered at the error microphones. Therefore, the
challenge is to create quiet zones at the locations of virtual
sensors (i.e., ears of the bed partner) by placing physical error
sensors on the headboard as shown in Fig. 2.
1) Class 1 - requiring preliminary identification
This class of algorithms [18, 19] requires system models or
transfer functions to process the data obtained from physical
sensors. During the training stage, sensors are deployed at the
physical locations permanently and virtual locations
temporarily. The sensing data is used to derive the transfer
functions needed for virtual sensing methods. After the
training stage, the sensors at the virtual locations are
removed.
The system performance of this class of algorithms
depends heavily on the accuracy of system models obtained
during the off-line training stage, and is very sensitive to the
locations of the physical sensors. In addition, it may need to
repeat the off-line training process for the changing
environment during the operation of ANC. Another limitation
is that the off-line training is inapplicable (or difficult to
realize) in some real-world applications.
2) Class 2-not requiring offline identification
The second class of algorithms does not need the off-line
training stage. For example, Moreau [20] developed the
stochastically optimal tonal diffuse field virtual sensing
method for a spatially-fixed target zone, and extended it to
moving targets. A contribution of this method is that it
doesn't require off-line training stage to obtain system models
(or transfer functions), thus it can be more efficiently and
effectively applied to practical systems.
In many practical ANC applications, especially in
consumer electronics and medical instruments, it is desired to
shift quiet zone(s) away from the locations of the physical
error sensors to the desired virtual locations for optimum
performance. Therefore, development of effective virtual
sensing techniques for ANC systems is a very important and
challenging work, and it deserves further research and
development.
B. RESIDUAL NOISE MASKING
ANC systems do not completely reduce primary noise due to
many physical limitations. In some applications, the residual
noise can be masked using suitable masking signals such as
music or nature sounds. Properly selected audio sound can
create a very soothing atmosphere to help a person relax and
sleep. This is one of the key reasons in exploring the use of
audio for masking residual noise in snore ANC systems. In
addition, the same audio can be used as a training signal for
off-line secondary-path modeling, and extended to form an
on-line secondary-path modeling technique. The use of audio
is explored in the following areas:
1) Residual noise masking with soothing audio using
principles of psychoacoustics.
2) Off-line modeling of secondary paths.
166
3) On-line secondary-path modeling for the FXLMS
algorithm to update ANC filter.
A psychoacoustic processor is designed and integrated
with the ANC systems to mask the residual snore noise at a
volume as low as possible [21]. This system involves playing
modified audio simultaneously with anti-noise to mask the
residual noise. Figure 5 shows the integrated audio/ANC
system performing both audio interference cancellation and
on-line modeling of S(z) [22]. The comfort audio wa(n) is
added with the canceling signal y(n) and the mixed signal
u(n) is output to drive the secondary loudspeaker. Thus, the
signal e(n) picked up by the error microphone contains both
the true error signal and the desired audio component. The
audio component is subtracted to obtain the true error signal
)(ne′ to update the adaptive filter W(z).
For consumer devices like snore ANC systems, it is
preferred to use an enjoyable piece of music or audio as the
lossless training signal in modeling performance. This
motivates the use of nature sounds like raining, running
streams, etc. for modeling the secondary paths.
d(n)
+
)(
ˆzS
LMS
W(z)
)(
ˆzS
LMS
-
+
+
+
w
a
(n)
y(n) x(n) u(n)
-
1
)(ne′
e(n)
Acoustic Domai
n
S(z)
copy
+
+
Fig. 5 Integrated audio/ANC system that provides both audio
cancellation and on-line secondary-path modeling
In order to mask the residual noise seamlessly, the audio
signal )(nwa must be chosen and processed by the
psychoacoustic processor based on the characteristics of the
residual noise. In addition, it is necessary to control the
volume of the audio signal such that it matches the level of
true residual noise. Therefore, more research is needed on
psychoacoustics and human perception of sounds.
C. ACTIVE SOUND QUALITY CONTROL
In some applications, it is desirable to retain a low-level
residual noise with a desired spectral shape or changed noise
signature. Active sound quality control (ASQC), which
changes amplitudes of noise components with predetermined
values, is a useful and important extension of ANC. The
system for individually controlling each harmonic of a
periodic noise is called the narrowband active noise
equalization [23]. The narrowband ASQC algorithm can be
implemented and analyzed in frequency domain [24], and
extended to control broadband noises [25]. The block
diagram of the broadband ASQC system is illustrated in Fig.
6. The broadband ASQC algorithm uses a shaping filter C(z)
to control the residual noise spectrum. The filter is designed
such that |C(z)|2 is the desired shape of the residual noise
spectrum.
P
(
z
)
Σ
W
(
z
)
d(n) x(n)
y(n) y′(n)
+-
Ŝ
(
z
)
LMS
e(n)
C
(
z
)
Ŝ
(
z
)
S
(
z
)
Σ
Σ
+
-
+
-
e′(n)
Fig. 6 Block diagram of the broadband ASQC system
Active sound quality control inherits the problems of ANC
systems. These problems include passband disturbance due to
the uncorrelated interference at frequencies of the magnitude
response of secondary path with high gains, and slow
convergence due to the eigenvalue spread of the input
autocorrelation matrix determined by the secondary path.
Further research is needed to design shaping filter C(z), to
analyze transient behaviors of the algorithm, and the modeling
error of secondary-path estimates on the performance of
system.
D. Some Special Noises
Active noise control systems are mainly developed for
reducing low-frequency noises and stationary noises
generated by engines, compressors, fans, and etc. This section
introduces some challenging cases.
1) High Frequency Noises
Many new developments of ANC applications in consumer
electronics and medical instruments involve reducing high-
frequency noises in three-dimensional space. The first
example is the development of privacy-phone handsets that
picks up near-end speech, transmits it, and at the same time
generates the out-of-phase speech to cancel the original
speech in space, thus allowing private and quiet voice
communication in public areas. Another potential application
is the development of ANC systems for dental drills, where
the frequency range of the dominant noise varies from 2000
to 6000 Hz for most of drills, depending on the rotation speed
of the bearing [26].
Combination of both active and passive noise
control techniques in small spaces such as headsets is the
current solution for reducing noise with high and wide
frequency range. With advances in acoustic theory and
167
control techniques, innovative ANC systems can be extended
to higher frequencies.
2) Impulse-Like Noises
There are many practical ANC applications involve impulse-
like noises. For example, soldiers need ANC helmets or
headphones for protection against extremely high-level
impulse-like noises from bomb explosions, gun shots, and
etc. The real challenge is that the ANC headphones must
selectively cancel harmful noises only, while still pass spatial
and environmental information to the user. As for the second
example, there are many loud impulse-like noises generated
by stamping machines in industrial manufacturing plants.
Fig. 6 ANC setup based on the GE Giraffe® incubator
In hospitals, there are a lot of life-saving equipment such
as breathing and IV pumps that generate impulse-like noises.
For example, infant incubators shown in Fig. 6 are used in
neonatal intensive care units (NICU) to increase the survival
of premature and ill infants. The application of ANC for
reducing incubator noise in NICU was developed in [27]. The
recorded NICU noise has lot of warning sounds and IV pump
sounds, which are impulse-like noises. The development of
the nonlinear filtered-X least mean M-estimate algorithm for
reducing impulse-like noise in incubators was reported in
[27]. Further performance analysis, real-time experiments,
and development of more effective ANC algorithms for
impulse-like noises is needed.
3) Dynamic Active Noise Control
In recent years, traffic noise coming from streets, highways,
railways, and airports has been of increasing concern. There
is some research on improving the performance of noise
barriers by using ANC systems, but the noise sources in these
studies are assumed to be fixed. In practice, the positions of
noise sources in real-world transportation applications are
time varying, and it is necessary to study and develop
dynamic ANC systems for moving noise sources relative to
the ANC installation.
Little research has been done in active control of noise
from moving sources [28, 29]. These publications show that
the optimal noise controller for moving noise sources will
vary when the positions of the primary noise sources change.
However, the relationships among the noise reduction
performance of ANC systems, the changed system transfer
functions caused by the speed and direction of the moving
sources, and the corresponding frequency shifting due to the
Doppler effect have not yet been investigated. Therefore, there
is need to research these three critical issues: (1) Time-varying
primary paths, (2) the Doppler effects, and (3) large-scale and
fast-convergence control algorithms.
IV. ADD-ON FUNCTIONS
One of the reasons for slow adoption of ANC is due to the
higher initial cost as compared with available passive
methods. To overcome this economic factor, some ANC
systems can be more cost effective and attractive by adding
additional functions on the products without increasing the
overall system cost. This goal can be achieved by sharing the
hardware resources such as amplifiers, loudspeakers,
microphones, and etc.
The first example is the integrated audio/ANC system for
infant incubators as shown in Fig. 6. It should be noted that
the womb has a rich sound environment with sound mainly
coming from the heart and blood vessels of the mother [30].
These sounds help the infant’s neurological development and
improve their understanding of rhythm and melody.
Therefore, it is strongly desired to decrease the NICU noise
and at the same time provide the infant with the ambience of
the mother’s womb. An audio-integration algorithm [29]
introduces a healthy intrauterine sound with the ANC system
to soothe the infants. The integrated system uses the same
power amplifier and the loudspeakers inside the incubator
(see Fig. 6), thus the overall system cost remains the same.
The second example of an add-on function is to use the
microphones inside the incubator to detect and recognize the
infant cry in order to alert and inform the nurse or caretaker
of the most likely reason behind the cry [31]. This algorithm
utilizes signal boundary detection and linear predictive
coding (LPC) coefficients in order to analyze and extract
features from infant cry instances. Consistent reference
signals for three separate cry pathologies (hunger, wet diaper,
and a need for attention) were decomposed to generate
training vectors for cry recognition. Qualitative matching
was defined on the basis of similarity between unknown cry
LPC coefficients to the weighted coefficients of each of the
three training vectors. Figure 7 shows that different cry
signals have varying LPC coefficients. This suggests that
LPC can be used to differentiate cries. Further research is
needed to evaluate different features and sound recognitions
techniques.
Another example of add-on functions to ANC systems is to
integrate audio entertainment and hands-free phone
communications features into the snore ANC system as
shown in Fig. 3. In this case, the loudspeakers on the
headboard and the associated amplifiers are available for
audio entertainment purpose. In addition, with the
microphones on the headboard along with the loudspeakers
and associated amplifiers, a hands-free speakerphone can be
developed as an add-on feature to the snore ANC system.
168
This additional feature involves the development of acoustic
echo cancellation techniques.
Fig. 7 LPC coefficient sets of attention, diaper and
hungry cries
V. CONCLUSIONS
This paper presented open problems and challenges of
ANC systems from the signal processing viewpoint. Several
open research and development problems were given. Three
important techniques: virtual sensing, residual noise masking,
and active sound quality control, were proposed to further
improve performance of ANC systems. We also presented
some potential but difficult ANC applications for high-
frequency and impulse-like noises, and traffic noise with
moving noise sources. Finally, we suggested add-on functions
for designing more attractive and cost-effective ANC systems.
REFERENCES
[1] P. A. Nelson and S. J. Elliott, Active Control of Sound, San Diego:
Academic Press, 1992.
[2] S. M. Kuo and D. R. Morgan, Active Noise Control Systems - Algorithms
and DSP Implementations, New York: Wiley, 1996.
[3] C. R. Fuller, S. J. Elliott, and P. A. Nelson, Active Control of Vibration,
San Diego: Academic Press, 1996.
[4] C. H. Hansen and S. D. Snyder, Active Control of Noise and Vibration,
London: E&FN Spon, 1997.
[5] S. J. Elliott, Signal Processing for Active Control, San Diego: Academic
Press, 2001.
[6] S. M. Kuo and D. R Morgan, “Active noise control: A tutorial review,”
Proc. IEEE, vol. 87, pp. 943-973, June 1999.
[7] B. Widrow and S. D. Stearns, Adaptive Signal Processing, Englewood
Cliffs, NJ: Prentice-Hall, 1985.
[8] D. R. Morgan, “An analysis of multiple correlation cancellation loops
with a filter in the auxiliary path,” IEEE Trans. Acoust., Speech, Signal
Processing, vol. ASSP-28, pp. 454-467, Aug. 1980.
[9] S. M. Kuo and W. S. Gan, Digital Signal Processors: Architectures,
Implementations, and Applications, Upper Saddle River, NJ: Prentice
Hall, 2005.
[10] L. J. Eriksson, M. C. Allie, and R. A. Greiner, “The selection and
application of an IIR adaptive filter for use in active sound attenuation,”
IEEE Trans. Acoust., Speech, Signal Processing, vol. ASSP-35, pp.
433-437, Apr. 1987.
[11] S. M. Kuo, “Active noise control system and method for on-line
feedback path modeling,” US Patents No. 6,418,227, July 9, 2002.
[12] S. J. Elliott and P. Darlington, “Adaptive cancellation of periodic,
synchronously sampled interference,” IEEE Trans. Acoust., Speech,
Signal Processing, vol. ASSP-33, pp. 715-717, June 1985.
[13] B. Widrow, J. R. Glover, J. M. McCool, J. Kaunitz, C. S. Williams, R.
H. Hern, J. R. Zeidler, E. Dong, and R. C. Goodlin, “Adaptive noise
canceling: principles and applications,” Proc. IEEE, vol. 63, pp. 1692-
1716, Dec. 1975.
[14] H.-J. Jeon, T.-G. Chang, and S. M. Kuo, “Analysis of frequency
mismatch in narrowband active noise control,” IEEE Trans. Audio,
Speech, and Language Processing, to be published.
[15] S. J. Elliott and T. J. Sutton, “Performance of feedforward and feedback
systems for active control,” IEEE Trans. Speech Audio Processing, vol.
4, pp. 214-223, May 1996.
[16] S. M. Kuo, S. Mitra, and W. S. Gan, “Active noise control system for
headphone applications,” IEEE Trans. on Control Systems Technology,
vol. 14, no. 2, March 2006, pp. 331-335.
[17] Sen M. Kuo, Sreeram R. Chakravarthy, Rakesh Gireddy, and Abhijit
Gupta, “Experiment of Active Snore Noise Control Systems,” Noise
Control Engineering Journal, vol. 56, no. 1, Jan-Feb., 2008, pp. 16-24.
[18] G. Bonito, S. J. Elliott, and C. C. Boucher, “Generation of zones of quiet
using a virtual microphone arrangement,” J. Acoust. Soc. Am., vol. 101,
no. 6, pp. 3498-3516, June 1997.
[19] B. Cazzolato, “An adaptive LMS virtual microphone,” Proc. Active
2002, pp. 105-116, July 2002.
[20] D. Moreau, B. Cazzolato, B. Zander, and C. Peterson, “A review of
virtual sensing algorithms for active noise control,” Algorithms, vol. 1,
no. 2, 2008, pp. 69-99.
[21] S. Singaraju and S. M. Kuo, “Noise masking using psychoacoustics,” in
Proc. Active 2006, Sept. 2006.
[22] W. S. Gan and S. M. Kuo, “An integrated audio and active noise control
headsets,” IEEE Trans. Consumer Electronics, vol. 48, no. 2, May
2002, pp 242-247.
[23] S. M. Kuo and M. J. Ji, “Development and analysis of an adaptive noise
equalizer,” IEEE Trans. Speech Audio Processing, vol. 3, pp. 217-222,
May 1995.
[24] S. M. Kuo, M. Tahernezhadi, and L. Ji, “Frequency-domain periodic
active noise control and equalization,” IEEE Trans. on Speech and
Audio Processing, vol. 5, no. 4, pp. 348-358, July 1997.
[25] S. M. Kuo and Y. Yang, “Broadband adaptive noise equalizer,” IEEE
Signal Processing Letters, vol. 3, no. 8, pp. 234-235, August 1996.
[26] E. Kaymak, M. Atherton, K. Kotter, and B. Millar, “Active Control at
High Frequencies,” in Proc. Int. Congress Sound and Vibration, 2006.
[27] Priya Thanigainayagam, Sen M. Kuo, and Ravi Yenduri, “Nonlinear
Active Noise Control for Infant Incubators in Neo-Natal Intensive Care
Units,” in Proc. IEEE ICASSP, April 2007, vol. I, pp.109-112.
[28] V. Martin, “Active control of moving sound source radiation: numerical
modeling in the space-frequency and space-time domains,” Journal of
Sound and vibration, vol. 228, pp. 477-509, 1999.
[29] J. N. Guo and J. Pan, “Active control of moving noise source: effects of
off-axis source position,” Journal of Sound and Vibration, vol. 215, pp.
457-475, 2000.
[30] J. M. Standley, “A Meta-Analysis of the Efficacy of Music Therapy on
Pretern Infants,” Journal of Pediatric Nursing, vol. 17, pp. 107-113,
2002.
[31] K. Kuo, “Feature Extraction and Recognition of Infant Cries,” 2010
IEEE Int. Conf. on Electro/Information Technology, Normal, Illinois.
169
