Académique Documents
Professionnel Documents
Culture Documents
(8)
where a
i
is the amplitude of the i
th
order, f
i
is the
frequency of the i
th
order and the f
s
is the sampling rate.
In the conventional active noise control application, the
desired response is zero. However, in this study, for the
different purposes of enhancement, attenuation or
cancellation, the desired response d(n) can be
expressed as
N
i
s
E
i i
f nf d n d
1
) / 2 sin( ) (
(9
)
where d
i
is desired amplitude of the i
th
order,
E
i
f is the
frequency of the i
th
order, and f
s
is sampling rate. The
desired signal d(n) should include all frequencies that
are targeted for tuning in the control system.
Again in Figure 2, H(z) stands for the secondary path
transfer function that is the response from the control
speaker to the receiving microphone in the passenger
compartment. This transfer function can be estimated
from either a vehicle cabin model or experiment. Also,
( ) h z
(10)
From the results listed in Table 1, it is clearly that the
proposed FLMS algorithm will cost less computation
time than traditional time domain algorithm when the
block size is larger than 32, because those ratios are
less than 1. When the block size is equal to 32, the
computational costs of these two algorithms are almost
the same.
COMPUTER SIMULATION
STEP SIZE - Convergent rate is different for each
harmonic that is targeted for tuning control. For one
harmonic, the convergent rate depends on the step size
and the amplitude of frequency response function of the
secondary path. Since it is very difficult to change the
secondary path, adjusting step size is the best way to
achieve faster convergence and better performance.
The amplitude of frequency response function of
secondary path also varies with frequency. Thus, the
most suitable step size for each harmonic should be
different.
For the time domain algorithm, the step size is fixed for
all frequencies. Hence, the selected step size may be
optimal for only one harmonic, but not for another
harmonic. In fact, some of the other harmonics may
become divergent if the step size is not suitable. For
instance, in Figure 4, the dashed lines show the
simulation results when a fixed step size is used to
control the harmonics at 160Hz and 288Hz using the
time domain algorithm. The step size can not go any
larger in this case; otherwise, the harmonic at 288Hz will
diverge. It is seen that the harmonic at 160Hz (blue
dashed line) converge much slower than the harmonic at
288Hz (black dashed line).
x(n)
Speed Calculator
& Sine Wave
Generator
Tachometer
Signal
Powertrain
Noise
+
PT(n)
-
d(n)
+
-
y(n)
e(n)
Buffer
Buffer Buffer
Unbuffer IFFT
FFT
D(z)
( ) h z
FFT FLMS
W(z)
H(z)
FFT
For the frequency domain algorithm, the step size can
be adjusted for each frequency bin. In Figure 4, the solid
curves show the simulation results when the step size of
the harmonic at 160 Hz (blue solid line) is 10 times
larger than the step size of harmonic at 288Hz (black
solid line). It is seen that almost the same convergent
rates can be achieved for these two harmonics in Figure
4 and Figure 5. Even though the harmonic at 288Hz
(black dotted line) converge faster using the time domain
algorithm than that using the frequency domain
algorithm (black solid line), the total MSE (mean square
error) using time domain algorithm (red dotted line)
shows lower convergent rate than that of frequency
domain algorithm (red solid line). The reason is because
slower convergent rate of the harmonic at 160Hz using
time domain algorithm does affect the total system
performance. From the solid line, we can see that for the
first 700 samples, the frequency domain algorithm does
not show convergence. This is caused by the block
delay of the frequency domain algorithm. The amplitude
of frequency response function of the secondary path
used in this study is shown in Figure 3. It is clearly seen
that the amplitude response at 160Hz is much lower
than the amplitude response at 280Hz. This implies that,
to achieve similar convergent rate, a larger step size is
required at 160Hz.
100 120 140 160 180 200 220 240 260 280 300
40
45
50
Frequency (Hz)
M
a
g
n
i
t
u
d
e
(
d
B
)
Figure 3. Magnitude of the frequency response function
of the secondary path.
0 1000 2000 3000 4000 5000 6000 7000 8000
0
0.5
1
1.5
Samples
Two harmonics control, 160Hz and 288Hz
M
e
a
n
S
q
u
a
r
e
E
r
r
o
r
Figure 4. Mean square error (MSE) of two harmonics
control using frequency domain and time domain
algorithms. Solid line frequency domain algorithm;
dashed line --- time domain algorithm; red line or ---
total MSE of two harmonics; black line or --- MSE
of harmonic at 288Hz; blue line or --- MSE of
harmonic at 160Hz.
2200 2400 2600 2800 3000 3200 3400
0.16
0.18
0.2
0.22
0.24
0.26
0.28
0.3
0.32
Samples
Two harmonics control, 160Hz and 288Hz
M
e
a
n
S
q
u
a
r
e
E
r
r
o
r
Figure 5. Expanded view of two harmonics control
results using the frequency domain algorithm. Black
solid line MSE (mean square error) of harmonic at
288Hz; blue solid line MSE of harmonic at 160Hz.
POWERTRAIN NOISE CONTROL - The performance of
proposed active control system can be evaluated using
Matlab/Simulink. The primary powertrain disturbances
along with tachometer signal were recorded on an actual
vehicle when the engine crankshaft speed is set to about
3500 rpm. The reference signal is generated by a sine
wave generator including all frequencies targeted for
control. For demonstration purpose, the desired signal is
designed to reduce the response of the third order as
much as possible and to enhance the response of fourth
order to meet the desired amplitude. A transfer function
of the secondary path, which is from the driving signal of
the secondary speaker to the error sensor location, is
synthesized from a fast numerical model for vehicle
interior acoustics [12]. The secondary path was modeled
using a 300-tap finite impulse response filter
( ) h z
for all
simulations. The lengths of frequency domain adaptive
filter are N=128 and N=256. Thus, 256-Point FFT and
512-Point FFT are conducted in the relative simulations.
The spectral magnitude of primary disturbance signals
and pseudo-error signal are shown in Figure 6. In this
simulation, the last 8192 samples of pseudo-error signal
after the convergence is taken as steady-state signal.
The desired value of the 4
th
order is labeled by asterisk.
The black curve is the original response of powertrain
noise when the control is off. Both the blue curve and
red curve are the resultant responses when the control
is on. As shown in the plot, the reduction at the 3
rd
order
is 10dB when the block size is 128, and the reduction is
more than 16 dB when the block size is 256. This is
more reduction compared to only 10 dB reduction
applying the traditional time domain algorithm with the
same operating conditions [10]. The enhanced response
of the 4
th
order is very close to the desired response for
both simulations.
Expanded view as
shown in Figure 5.
120 140 160 180 200 220 240 260 280
Frequency (Hz)
R
e
l
a
t
i
v
e
a
m
p
l
i
t
u
d
e
(
d
B
)
Figure 6. Active noise control simulation results using
the proposed frequency domain active control system.
Black line primary powertrain noise; blue line
block size 128; red line block size 256; * desired
value.
However, the unintended overshoot is seen at
frequencies neighboring the controlled harmonics. There
are two ways to suppress the overshoot and to improve
the performance of the proposed system. Firstly, it is
seen that the overshoot of the red curve is less than that
of the blue curve in Figure 7. This is suggesting that the
overshoot can be suppressed as the block size
increases because the larger block size has smaller
frequency bin.
Secondly, one of the advantages of frequency domain
algorithm is that the step size can be adjusted
independently for each frequency bin. In Figure 7, the
black curve is still the baseline response of powertrain
noise. Both the blue curve and red curve are the sound
pressure responses after the controller is activated. The
step size of the blue curve is constant, while the step
size of the red curve is variable. As shown in Figure 7,
the overshoot is decreased significantly when the step
size at frequencies neighboring controlled response are
set to zero. This result is obtained when the block size is
128. Further refinement can be achieved using larger
block sizes, such as 256.
In traditional time domain algorithm, the amplitude of the
reference signal of each order is adjusted separately,
while tuning of the variable step size in the frequency
domain algorithm provides an alternative for the
proposed active control system. Furthermore, tuning
step size for each frequency bin according to the
secondary path transfer function can lead to faster
convergence and also avoid divergence in some cases.
120 140 160 180 200 220 240 260 280
R
e
l
a
t
i
v
e
a
m
p
l
i
t
u
d
e
(
d
B
)
Frequency (Hz)
Figure 7. Comparison between fixed step size and
variable step size for each frequency bin when step size
is 128. Black line primary powertrain noise; blue
line fixed step size; red line variable step size;
* desired value.
CONCLUSION
The proposed active control system for use to tune
powertrain noise was implemented and analyzed in
frequency domain. A case study is conducted in
MATLAB/Simulink, where the engine crank shaft speed
is constant. Simulations show that the proposed system
has better performance, such as more reduction at
targeted order, than conventional time domain algorithm.
Also, the overshoot problem can be suppressed by
using variable step size or by increasing block size.
Overall, we observed that this active control system
provides faster convergence, less computations, and
less overshoot while achieving the desired control
objectives. This feature is very desirable for use in
tuning interior sound quality, and is currently being
studied further for more variety of operating conditions.
REFERENCES
1. M. Harrison. Vehicle refinement: Controlling noise
and vibration in road vehicles. Butterworth-
Heinemann (2004).
2. S.M. Kuo and M.J. Ji, Development and analysis of
an adaptive noise equalizer, IEEE Transactions on
Speech and Audio Processing 3, pp. 217222
(1995).
3. S.M. Kuo, M. Tahernezhadi and L. Ji, Frequency-
domain periodic active noise control and
equalization, IEEE Transactions on Speech and
Audio Processing 5, pp. 348358 (1997).
4. S.M. Kuo, A. Gupta and S. Mallu, Development of
adaptive algorithm for active sound quality control,
Journal of Sound and Vibration 229, pp. 1221
(2007).
10 dB 10 dB
5. J. Ogue, T. Saito and Y. Hoshiko, A fast
convergence frequency-domain adaptive filter,
IEEE Transactions on Acoustics, Speech and Signal
Processing 31, pp. 13121314 (1983).
6. Y. Kajikawa, K. Ashitaka, and Y. Nomura,
Frequency domain active noise control system
using optimal step size, Electronics and
Communications in Japan, Part 3, Vol. 86, No. 7,
(2003).
7. E.R. Ferrara, "Frequency-domain Adaptive
Filtering, Chapter 6 of Adaptive Filters, edited by
Cowan and Grant, Prentice-Hall (1985).
8. A.V. Oppenheim, Signals and Systems, Prentice-
Hall, Inc, New Jersey (1983).
9. G. Clark, S.K. Mitra and S.R. Parker, Block
implementation of adaptive digital filters, IEEE
Transaction, Vol. CAS-28, No.6, pp. 584-592 (1981).
10. E.J. Sorosiak, et. al., An active noise control system
for tuning vehicle interior response, NoiseCon 08
(2008).
11. S.M. Kuo and D.R. Morgan, Active Noise Control
SystemsAlgorithms and DSP Implementations,
Wiley, New York (1996).
12. E.J. Sorosiak, et. al., A fast numerical model for
vehicle interior acoustics, NoiseCon 08 (2008).
CONTACT
Name: Teik C. Lim, PhD, PE, Fellows (ASME, SAE)
Position: Professor and Department Head
Director, Vibro-Acoustics and Sound Quality
Research Laboratory
Director, Hypoid Gear Mesh and Dynamic
Modeling Consortium
Director, UC Simulation Center
Address: Mechanical Engineering, Univ. of Cincinnati
598 Rhodes Hall, P.O. Box 210072
Cincinnati, OH 45221, USA.
Email: teik.lim@uc.edu