Active Control of Noise by Wave Field Synthesis 5092 (D - 3)

Michele Zanolin (l), Prof.Paolo Podini (2), Stefano De Stabile (3),

Paolo Vezzoni (3) and Angelo Farina (4)
(1) Massachusetts Institute of Technology,Cambridge, MA
(2) Physics Department, University of Parma, Italy
(3) Acoustic consultant, Gamma Engineering, Reggio Emilia, Italy
(4) Industrial Engineering Department, Area delle scienze, 43100 Parma ,Italy

Presented at IAU D 101

the 108th Convention
2000 February 19-22
Paris, France
This preprint has been reproduced from the author’s advance
manuscript, without editing, corrections or consideration by the
Review Board. The AES takes no responsibility for the
contents.

Additional preprints may be obtained by sending request and

remittance to the Audio Engineering Society, 60 East 42nd St.,
New York, New York 10165-2520, USA.

All rights reserved. Reproduction of this preprint, or any portion

thereof, is not permitted without direct permission from the
Journal of the Audio Engineering Society.

AN AUDIO ENGINEERING SOCIETY PREPRINT

ZANOLIN ET AL. PREPRINT 5092 ACTIVE CONTROL OF NOISE

Active control of noise by Wave Field Synthesis

M.Zanolin
M I T , 77 Mass.Ave, Cambridge, MA, USA
P. Podini
Physics Dept., UniversitA di Parma, Campus 43100, PR, Italy
A. Farina
Industrial Engineering Dept., Campus 43100, PR, Italy
S.De Stabile, PYezzoni
Gamma Engineering SPA, RE, Italy

Abstract
In this paper we present the results of a research about active control
of noise. We have built a prototype to lower the direct field of a known
source using a linear array of 40 loudspeakers controlled by delay lines
and a digital signal processor. The inspiring theoretical tool is the Wave
Field Synthesis (WFS), an holographic approach to sound control that
was proposed by Berkhaut et al. in 1988 at the TU of Delft. We have
obtained an average reduction of 3 db on a area of 16 squared meters for
a frequency of 1000 Hz.

1 Introduction
Sound reduction by means of destructive interference is a well assessed
methodology but, due to the physical nature of the sound field, it is very
difficult to achieve sensible results over wide areas. On the other side the
holographic approach of the Wave Field Synthesis to the acoustic field
control is interesting because it is a tool to control the direct field of a
given source on a volume. The basic idea of our research was to verify if
the basic concepts of the WFS can be employed in order to build up a
phase cancellation that, despite to what happens with other approaches,
can be obtained over a wide area. Physically speaking we wanted to re-
produce a wave front that has the same spatial properties of the one of a
known source, but opposite in phase so to obtain destructive interference
between the two.

AES 108th CONVENTION, PARIS, 2000 FEBRUARY 19-22 1

ZANOLIN ET AL. PREPRINT 5092 ACTIVE CONTROL OF NOISE

In term of impulse response of the test source, this first approach is de-
voted to reduce the energy contained in the first lo-15 milliseconds. Such
formulation of the problem should be in principle able to overcome the
major difficulties that are usually encountered in traditional noise cancel-
lation approaches, when a phase cancellation is obtained by means of a
reduced number of loudspeakers placed in certain points distant from the
source and usually near the receiving point where the sound reduction is
required. In such a case it is well known that phase cancellation can be
only a local solution, and to a certain noise reduction in a certain area
corresponds a noise increase somewhere else.

We used a system devoted to control the sound on a plane (2 dimen-

tional approximation). The array we built following the approach of the
WFS has the nice property that, in the plane, irradiate very poorly out-
side of the angle where we expected the best reproduction.

Line of secondary sources

Area of best reproduction

Primary Source

Secondary
iource

Figure 1: Two dimensional view of the geometrical setup

We made some preliminary estimations of the area where it is possible

to lower the direct field of a known omnidirectional monopolar source (in
the experiments we used a 8” loudspeaker). These consideration plus the
way we decided to implement the anti-spatial aliasing filter suggested us
the geometrical caracteristics of the experimental set up.
The experiment has been performed inside an industrial “capannone”
(manufacturing floor) of approximate dimensions: 50 X 20 X 8 (LxWxH).

AES 108th CONVENTION, PARIS, 2000 FEBRUARY 19-22 2

ZANOLIN ET AL PREPRINT 5092 ACTIVE CONTROL OF NOISE

AMPLL
_ INVERTER

Figure 2: Block diagram of the setup

Inside that environment there were several reflecting surfaces and the re-
verberation time was about 2.5 set at 1 kHz; in this way we were able to
conduct our test in a situation that can be considered as good replica of
a real industrial environment.

The linear array was made up of 40 4” loudspeaker units. The dist-

nce between adjacent centers of emission was 12 cm so to reach a lenght
of about 5 meters. Array, test source, and receivers, was positioned at
2.5 meters of quota and about 5 meters from the ceiling. The Syquist
frequency regarding spatial aliasing of wave fronts that propagate in all
the directions is about 1300 Hz. The distance between the source and the
array was 2 meters and the area of the optimal reduction starts about 2
meters in front the array, it is large 4 meters and long at least 4. The set-
up has been prepared to work optimally in the octave of 1000 Hertz. The
invertion of the phase in the signal of the test source has been obtained
by an inverter in the the amplifier.

Each loudspeaker has tested, measuring its impulse and polar response
by means of an MLSSA analyser and an array of 255 microphone’s posi-
tions (see for the relative technique [3]); in this way we were able to select
the loudspeakers that showed to have the most similar behaviour in the
polar responses and in the temporal shape of the axial impulse response.

AES 108th CONVENTION, PARIS, 2000 FEBRUARY 19-22 3

ZANOLIN ET AL. PREPRINT 5092 ACTIVE CONTROL OF NOISE

For figure nuber are depicted the axial impulse response of three loud-
speakers. It can be noted that the shape in the first millisecond (where
the biggest part of the energy is) is very similar.

0.25 _........_..
_ . .._....
_...... . _ _ .._.. ... ..... .~ _ .. ..
0.20

0.15

0.10

0.05

0.00

-0.05

-0.10 . . .

-0.15

-0.20

-0.25
:
auf0
5.0 6.0 7.0 8.0

Time - met

Figure 3: -4sial impulse responses of three loudspeakers mesured at the distance

of one meter. -4 good homogeneity can be noted in the first millisecond.

Loudspeakers showing the less uniform characteristics were placed at

the outer border of the array where the tapering i
that we applied reduced their importance in the reconstructed sound
field (we’ll discuss this topic in the description of the experimental set-
up). -41~0 the WFS itself, as we will see in the following, require that the
loudspeakers far from the primary source (at the end of the array in our
case) became less important (by means of a geometrical gain).

The use of a linear array introduce an error in the amplitude of the wave
front due to the fact that the real field decay in amplitude like the inverse
of the distance from the source, but the field reproduced by the array
goes like one on the square root of the distance. So we decided to set the
amplitude gain to obtain a correct amplitude at a certain distance from
the array (in the middle of the area that we considered of “best reproduc-
tion”). Going very close to the array it is obvious that the performance
of the system degrades, because a receiver is able to recognize that the
sound comes from the loudspeakers of the array and not from the primary
source. This approach, and the possible upgrades devoted to manage the
early reflections, is two dimensional, in the sense that all the contributions

‘IVe turned off the last 5 loudspeakers in the border following the formula
(1 + cos(i7;/6))/2,i is the outer loudspearker index. (i=1,..,5).

AES 108th CONVENTION, PARIS, 2000 FEBRUARY 19-22 4

ZANOLIN ET AL. PREPRINT 5092 ACTIVE CONTROL OF NOISE

coming from the celing and the floor are neglected. Often, in places like
a theatre where the listeners absorb part of the energy contained in the
sound that travels in the vertical direction, it is not a too bad aproxi-
mation. On the other side the fact that an array of monopolar sources
irradiates uniformly in all the direction equivalent under a rotation along
the array, can improve the amount of these contribution. To solve this
problem is under study the use of oval loudspeakers that have an higher
directivity in the vertical axis than in the orizontal one.

2 Theory
What we are going to analize here is a two dimensional aproximation
of the Kirchhoff-Helmhotz integral. The reason for this aproximation is
that that the goal of our research is to understand how the direct field
can be contrlled, around a fixed quota where the heads of the listeners
are supposed to stand (for exemple in a factory or in a theatre). Let
us consider the the K-H integral (see for example [l]) in the case where
the primary source is a monopole. In the formalism that we choosed
v is the primary source and S(w) the time Fourier transform of his
amplitude2. The secondary sources are dispaced on a straight line in front
of the primary source. Under the far field conditions (see for example [4]
at pg.20 and the following):

kr, kAr > 1; (I)

it is possible to obtain the so called 24 D Rayleigh Integral for the time

Fourier tranform of the acoustic field, on a straignt line of receivers posi-
tions in front of the line of the secondary sources:

where g is given by:

g(R,n) = (1+ z”;I ;;p2.

The geometrical quantities (4.t7LCincluded) are indicated in the figure
[4]. The integrand of (2) can be written as the product of an amplitude
(that we will call Q( 2,~)) and the monopolar propagator between the
position of the secondary source and the receiver. Q is the product of
different factors and it is worth to explain what they will became in the
prototype:

fi : it will became the digital filter. The effect on the filter due to
the discretization of the secondary source distribution is a moltiplication
for a band pass filter (different from every source) in the case we want to

we follow the usual convention: S(w) = s-‘,” exp (-iwt)s(t)dt .

AES 108th CONVENTION, PARIS, 2000 FEBRUARY 19-22

ZANOLIN ET AL PREPRINT 5092 ACTIVE CONTROL OF NOISE

A
x (integration variable)

Angle
of incidence

Figure 4: Geometry for the 21i2 Rayleigh integral

drive frequencies higher than the spatial Nyquist frequency (V = c/2Ax

where c is the speed of sound and Ax is the spacing between the loud-
speakers.

S(w) : spectrum of the monopolar test source (in our prototype is a

delta around 1000 Hz.).

co~(&~,)/sqrt(r): This will become a geometrical gain implemented by

an analogical (in this first prototype) board.

ezp(-ilcr) : geometrical delay controlled by bucket brigade delay lines.

g(R, TL) : For this term we write a more detailed digression.

Due to g the equation (2) has an unpleasant aspect: the amplitude (note:
not the phase!) of the secondary sources depends on the position of the
receiver R. This is a problem because we would like that the amplitude de-
pends only on the position of the source and the geometrical caracteristics
of the system (position and shape of the array and area where we want to
reproduce as correctly as possible the field). A solution that works can be
found, because at this stage, our goal is not to cancel exactly the direct
field, but estimate the area where it is possible to lower enought the direct
sound level. From here to the following we will concentrate our attention
in a reduction of 8 dB.3. To this goal let us note that if we sum to a given
primary field (PI) a secondary one (Pz) that is in counter-phase and with

30ur first prothotype works only on the direct field and we do not aim, at this stage to
obtain an higher reduction.

AES 108th CONVENTION, PARIS, 2000 FEBRUARY 19-22 6

ZANOLIN ET AL PREPRINT 5092 ACTIVE CONTROL OF NOISE

a different amplitude

+ r>
f’l = Acos(qb),~2 = BCOS(~$ (4)

PI + P2 = A(1 - A/B)cos(~) (5)

we can observe that the control of the amplitude, to obtain a good re-
duction, is not very critical. In fact from the expression of the lowering
respect the primary one:

AL = Lpi - Lpl+pz = 10 * Log((1 - A/B)2) (6)

it is easy to see that to lower of 8 dB a given field (using another field

exactly in counter phase) is allowed an error in amplitude up to 40 %.

0 iw MO 300 400 500 600 7w 800 900 ,ooo

Figure 5: g plotted in function of x = ~~~f~

From figure (5),where we plotted g in the form (2) in function of

x = :izt”, , is possible to note that if we set g for a line of receivers
that stands 4 meters away from the array (x = 2) the condition of not
exeding the error of 40 per cent in amplitude is respected about from 1,5
meter of distance from the array.

Spatial Aliasing

Spatial aliasing (SA) is one of the problems that has to be faced in prac-
tical realizations of a system that controls the spatial properties of the
acoustic field. To explain intuitively the effect of SA, let us observe that
the distance between adjacent loudspeakers (AX) can be used to define

AES 108th CONVENTION, PARIS, 2000 FEBRUARY 19-22 7

ZANOLIN ET AL. PREPRINT 5092 ACTIVE CONTROL OF NOISE

a scale of frequencies ~~~~~~~~~ = c/2Az). SA is an error inducted by

the array that emerges when we try to reproduce wave-pakets contain&
high frequencies (compared with UNyguist), that move trasversal to the
line of the loudspeakers. The Nyquist condition, that can be derived sub-
stituing a line distribution of sauces with an infinite array of equal-spaced
omnidirectional punctual sauces

means that if the spatial Fourier tranform of the amplitude of the monopo-
lar distribution of sources Q(k,) 4 has no component with ]k,] 2 r/Ax
the field is correctly reproduced by the array.

How many sources in the array?

Another issue that we whould like to mention regards the number of

sources. In fact if only few sources are used there are limitation in the
reproduction of the spatial properites of the field (see for example [8]).

These limitations have a different nature respect the case of the spatial
aliasing. SA is an undesired effect caused by the implementation of the
continuous line of secondary sources with a “discrete” number of loud-
speakers. As explained above the limitations are on wave-packets that
propagate tranversally to the array. The limitation that emerges when we
use an array composed by few loudspeakers ,on the other side, regards
more the lenght of the array and concern the reproduction of wave-packets
that move perpendicular to the array. For example we can’t reproduce
the spatial properties of a plane wave with an omnidirectional poin source
(one secondary sauce case).

In general if you want to reproduce a plane wave and you position a

receiver at a distance equal or larger than the diameter of the array the
wave fronts starts to became curve (and not flat). Physically means that
if we want to reproduce wave fields with no components for k, # 0 we ob-
tain a field that have these components. Let us consider, for example, an
array of 2M + 1 loudspeakers equally spaced (Ax). the spatial transform
of the amplitude of the secondary sources (Q(k,: t)), has to be convolved
with the functional 5

A(k,) = 1 + 5 cos(k,nAx); (8)

n=O

This term is given by the Fourier tranform of a sum of an finite num-

ber of delta located at the positions of the loudspeakers (see for example
[5]). W e ch ose d a p au. number of secondary sources because in the limit

4Q(kz) is the spatial Fourier transform of Q(z,z,w) respect z.

5A(k,) in gene ral is a functional (and not function) in the sense that performing the limit
M ++ 00, It has a sense only inside the integral.

AES 108th CONVENTION, PARIS, 2000 FEBRUARY 19-22 8

ZANOLIN ET AL. PREPRINT 5092 ACTIVE CONTROL OF NOISE

M ++ 0 the array goes in the “one source” case. to derive (6) we took
the origin of the x axis on the central secondary source. When M c) oo
the functional A(&) converges towards a sum of infinite Dirac’s deltas
translated of a multiple of the quantity 27r/Ax that is the double of the
spatial Nyquist frequency. In this limit the spatial spectrum of Q(lc,, t)
became simply periodized, by the convolution with A(Jc,) and the stan-
dard ideas on the anti-aliasing filtering lead to a correct implementation
of the field.

-5 ,

- x/Ax T/AX

Figure 6:A(&) plotted in the Nyquist region for 5,9,39 loudspeakers

In real cases M is finite and so A(/c=) is not a sum of Dirac’s delta. In

figure(l) we have plotted A(L,) for tre different numbers of loudspeakers.
It is clear that if we use a low number of loudspeakers, the convolution be-
tween Q(L, t), and A(&) changes the shape of Q(lc,, t) inside the Nyquist
zone, (even in the case Q(lc,, t) = 0 for l&l > r/Ax). We can imagine
the curves in the figure as the result of a convolution of a sine (the Fourier
transform of a squared window that has the lenght of the array), with the
sum of Dirac’s deltas (of the infinite array). In our case (40 loudspeakers)
can be proved (we have work in progress on it) that this effect is not the
main source of error 6 .

Matlab simulations

Some matlab simulation have been performed to get a qualitative insight

6 These arguments are extensively covered in the litterature of radar and sonar processing.

AES 108th CONVENTION, PARIS, 2000 FEBRUARY 19-22 9

ZANOLIN ET AL. PREPRINT 5092 ACTIVE CONTROL OF NOISE

on the performance of the array The sources of the array have been con-
sidered monopolar with a delta impulse response. We have considered a
primary source emitting a sine containing the frequencies of the octave
of 1000 hz. We have calculated the field on a grid of points in front the
array. the grid is large 8 meters and long 4. the pass of the grid is 0.5
cm. In fig 7 is depicted a configuration at a fixed time of the sum of the
two field. In the area that we considered of best reproduction the primary
it is possible to note the partial cancellation of the primary fie1d.A good
property is that utside is not amplified (due to the directivity of the ar-
ray). If we take one slice of the configuration in the central zone (x=800)
and plot the two fields we can obtain the figure 8 that suggest an upper
limit in the reduction of 20 dB for this range of frequencies.

Figure 7: Matlab simulation:this is a map, at a fixed time of the sum of the

primary and secondary field. The primary one is a sine containing frequencies
in the octave of 1000 Hz. The area (located two meters from the array) is 8
meters wide and 4 meters long. The grid of the figure has a step of 0.5 cm. In
the central area the two fields smoth each other, and in the sides the level the
primary field is not highered.

3 Experimental Setup
Some remark can be made about the experimental set up. The goal is to
show how some practical problem have been faced and how the perfor-
mance of the system can be improved.

- Delay lines (Bucklet brigade, model MN 3004) controlled by a two phase

AES 108th CONVENTION, PARIS, 2000 FEBRUARY 19-22 10

ZANOLIN ET AL PREPRINT 5092 ACTIVE CONTROL OF NOI:

0.02 -

-0.02’
0 100 zoo 300 4GQ 500 600 700 800 30 D

Figure 8: A(&) Primary and secondary field plotted on a line perpendicular to

the array from a distance of 2 meters until 6 meters away.

clock generator (MN3101) have been used in order to set the geometrical
delay required for the signals to be driven to the loudspeakers of the array.
The clock generators have shown an intrinsic instability in mantaining
the frequency so the delay of the signal have probably became randomized
and the reproduction has probably suffer of it. Is important that we have
obtain a positive result even in the presence of this weakness. Future pro-
totypes will be developed including the geometrical contribution to delay
and gains inside the work of the digital signal processor.

-Loudspeakers interact between eachother. For example we installed three

loudspeaker in the wood structure of the array and mesured the axial im-
pulse responses of each loudspeaker mantaining the other two turned off
(see fig (9)). Th e energy contained in each IR was bigger than the en-
ergy of a loudspeaker alone. The speaker with the other two turned off
positioned in the two lateral sides, had an impuse response containig an
energy bigger than the other two. This is an effect of the fact that the
copuling depends on the distance. From the figure (9) can be noted that
for wavelenght short respect the spacing:

the copuling vanish. A tecnique to analize in a quantitative way the

spatial properties of the effective sources is discussed in [3]. What we
expected was that installing the more omogeneous sources in the centre
of the array they interact giving omogeneus “effective” sources.

AES 108th CONVENTION, PARIS, 2000 FEBRUARY 19-22

ZANOLIN ET AL. PREPRINT 5092 ACTIVE CONTROL OF NOISE

Transfer Function flaynitude - dB voltsn~olts

-60 .O

roob .o
log Frequency - Hz

Figure 9: Frequency content of the impulse responses of three adjacent loud-

speakers in the array. Every mesurment of the impulse response is performed
taking one of them turned on an the others turned off. The most energetic one
is the central loudspeaker.

was not like an array of monopolar sources, but probably more close to
a dipolar one . Closing the back of the array should lead to a more
“monopolar” emission (fig 10, B). The reproduction has probably suffer
also of this fact, especially for wide angle respect the line that connect the
source with the center of the array.

4 Experimental results

The experiment have been performed giving to the test source and the
the electronic system that pilot the array the same signal. The gain of
the test source has been adjusted to to give the the same pressure level
in the in the center of the receivers area. Unfortunately the first test
source we used was able to give a 90 dB level when the signal of the
array was 91.8 dB. This is another point were the performance of the
can be improved. The result of the mesurments of SPL performed by a
BruelKjaer 2236 Integrating Sound Level Meter, are depicted in the figure
num. The imprortant aspect is that the reduction is on a wide area as we
desired.

AES 108th CONVENTION, PARIS, 2000 FEBRUARY 19-22 12

ZANOLIN ET AL PREPRINT 5092 ACTIVE CONTROL OF NOISE

Figure 10: Qualitative description of the expected modification in the polar

patterns of the speakers of the array, closing the back of the array.

L 4m
+O

1
0 2m
0 E
P
test source

2m 1
-0 0
2 5

Figure 11: position of the experimental mesurments

5 Conclusions
We have analized some necessary issue to understand how to cancel the di-
rect field of a known source.We have also analize some limitations (most of
them easy solvable) of our first prototype. The fact that the experimental
result were positive (even in the presence of these limitations) is an inter-
esting point. An aspect that we whould to remark is that the reduction
has been obtain on a wide area. The main goal of this research is to de-
sign a low-cost device to be employed in industrial environments in order

AES 108th CONVENTION, PARIS, 2000 FEBRUARY 19-22 13

ZANOLIN ET AL. PREPRINT 5092 ACTIVE CONTROL OF NOISE

POSITION PRIMARY LEV. PiRRAY LEV. SUM LOWERING (dB.)

90.1 91.8 85.9 -4.23

2 91.6 93.2 90.1 -1.14

91.4 93.1 90.4 -1

4 88.4 90.1 86 -2.46

5 88.3 90.2 85.8 -2.5

Figure 12: Mesured data

to obtain sensible noise reduction over large portion of space, but other
practical implementations are currently under investigation too. Research
to control early reflections and wide band signal are on progress.

References
PI -Bleistein,N., Mathematical Methods for wave phenomena, Academic
Press (1984)

PI -Berkhaut “A holographic approach to acoustic control”

J. Audio Eng. Sot. 36 (12) p.977.(1988)

PI -A.Farina,M.Zanolin “Mesurment of the acoustic diffusion coefficient

by a Wave Field Synthesis tecnique” ,108 AES convention,Febbr
2000, Paris.

14 -Berkhaut, De Vrie and Vogel, “Acoustic control by wave field Syn-

thesis”
J. Acoust.Soc.Am. 93(5) p.2764. (1993)

PI E.Start, Direct sound enhancment by Wave Field Synthesis,Ph.D.

thesis TU Delft,(1997)

PI -W.A.Kupermann, T.Akal, D.R.Jackson “Phase conjugation in the

ocean:
Experimental demonstration of an acoustic time-reversal mirror”
J.Acous.Soc.

P-1-M.Boone, Edwin N. Verheijen, G.Jansen “Virtual Reality by

Sound Reproduction Based on Wave Field Synthesis” J.Audio
Eng.Soc.(1996)

PI -O.Kirkeby, F.Nelson et al. “local sound field reproduction using dig-

ital signal processing” J Acous.Soc.Am. 100(3),pp 1584-1593,(1996)

AES 108th CONVENTION, PARIS, 2000 FEBRUARY 19-22 14