You are on page 1of 18

Adaptive Wave Field Synthesis for Sound Field

Reproduction: Theory, Experiments, and


Future Perspectives*
PHILIPPE-AUBERT GAUTHIER, AES Student Member, AND ALAIN BERRY
(philippeaubertgauthier@hotmail.com) (Alain.Berry@USherbrooke.ca)
Groupe dAcoustique, Universite de Sherbrooke, Sherbrooke, Quebec, Canada, J1K 2R1
Wave field synthesis (WFS) is a sound field reproduction technology that assumes that the
reproduction environment is anechoic. A real reproduction space thus reduces the objective
accuracy of WFS. Adaptive wave field synthesis (AWFS) is defined as a combination of WFS
and active compensation. With AWFS the reproduction errors are minimized along with a
departure penalty from the WFS solution. Analysis based on the singular value decomposition
connects WFS, active compensation, and Ambisonics. The decomposition allows the prac-
tical implementation of AWFS based on independent radiation mode control. Results of
experiments in different rooms support the theoretical propositions and show the efficiency
of AWFS for sound field reproduction.
0 INTRODUCTION
Sound field reproduction is one of the possible techno-
logical approaches for spatial sound. It focuses on the
simulation of the physical stimulus of spatial hearing,
which is the sound field in which an audience or a listener
is immersed in a natural hearing situation. Sound field
simulation, by opposition to perception simulation (level-
difference stereophony, time-difference stereophony, bin-
aural audio technologies), which is based on illusion [1],
mainly relies on the recreation of a physical acoustic quan-
tity, such as acoustic pressure [2][5], intensity field [6],
and direct or diffuse fields [7][9]. Sound field reproduc-
tion for spatial audio relies on the fundamental hypothesis
that if the sound field is reproduced accurately, in objec-
tive terms, the listener will be exposed to appropriate bin-
aural and monaural cuesinteraural time differences, in-
teraural intensity difference, and spectral cues caused by
external ear filtering as a function of the incident direction
of the sound wave [10]. The reproduction of spatial cues
would then create a spatial impression corresponding to
the target that must be reproduced.
The best known sound field reproduction techniques
address the interior problem of sound field reproduction.
The objective is then to reproduce or synthesize a given
sound field inside a volume or a listening region sur-
rounded by a reproduction source array. Typical tech-
niques are wave field synthesis (WFS) [2], [11] and Am-
bisonics [12], [13], [5]. The WFS approach derives from a
simple source formulation of the KirchhoffHelmholtz in-
tegral and the Ambisonics approach derives from a mode-
matching approach [14]. Theoretical connections between
WFS and Ambisonics are already known [15]. Some ap-
proaches for sound field reproduction are derived from
active noise control methods and often include a cost func-
tion for the minimization of reproduction errors [16][20].
This paper presents the adaptive wave field synthesis
(AWFS) method, which is a combination of WFS and
active noise control techniques.
*Presented at the 123rd Convention of the Audio Engineering
Society, New York, 2007 October 58; revised 2007 November 26.
Editors Note: This Journal paper is a fully reviewed submission, which was awarded the Student Technical Papers Award at
the 123rd Convention of the Audio Engineering Society in New York, 2007 October 58. The student, Philippe-Aubert
Gauthier, presented the paper and was honored at the Convention. The Editor hopes this will encourage future student
submissions. The criteria for the award can be found at http://www.aes.org/events/124/authors.
PAPERS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1107
Recent research on WFS and sound field reproduction
addresses different problems, such as spatial aliasing,
sound reinforcement, room effects, and panel actuators
[21][24]. One recent research topic is the effect of system
response or room response on the objective performance
of the sound field reproduced by WFS [25][27]. Since
WFS development assumes that the reproduction space is
anechoic, the real room response introduces a severe dis-
crepancy between the target sound field and the sound
field reproduced by WFS. This has led several researchers
to work on room compensation for WFS [4], [5], [26]
[29]. AWFS is proposed as a method combining WFS and
active room compensation based on the minimization of
reproduction errors. The amount of compensation is con-
trolled by a penalization parameter, which limits the de-
parture from the classical WFS solution [5].
Typical applications of sound field reproduction are
music reproduction and virtual acoustics. In these cases
the goal is to reproduce the complete sound field of a
virtual scene made of several virtual sources (such as
plane waves, spherical waves) so that the perceived rela-
tive movement of the virtual sources is consistent with the
listeners movement in the listening area. Although the
most obvious applications of sound field reproduction are
in the field of audio (music reproduction, sound reinforce-
ment, virtual reality, videoconferencing [30][32]), sev-
eral applications of sound field reproduction are concerned
with experimental acoustics and psychoacoustics [33][35].
This is an interesting and rich research avenue for spa-
tial sound and sound field reproduction. Future research
perspectives for AWFSas will be discussed in Section
3are related to sound field reproduction for sound en-
vironment reproduction and synthesis, which present very
specific constraints and technological problems.
This paper presents AWFS and is divided into three
main parts. In the first part WFS is reviewed and AWFS is
accordingly defined and analyzed on the basis of the sin-
gular value decomposition (SVD) of the electroacoustical
plant matrix (from the loudspeakers to the error micro-
phones). In the second part AWFS is described and evalu-
ated from an experimental point of view. A 24-loud-
speaker system was built and tested in three different
rooms: hemianechoic chamber, laboratory space, and re-
verberation chamber. In the third part, a review of the
future research perspectives for sound field reproduction
and sound environment reproduction is presented in light
of the knowledge gained from the definition of AWFS and
the experiments.
1 ADAPTIVE WAVE FIELD SYNTHESIS
1.1 Wave Field Synthesis
WFS originates from proposals by Berkhout in the late
1980s and early 1990s [2], [11], [36]. WFS is developed as
follows. From the simple source formulation of the Kirch-
hoffHelmholtz integral theorem [37], WFS operators are
defined for a given simple virtual source (also called the
primary source, typically creating spherical or plane waves
in a horizontal plane, the listening plane), typically fed by
a monophonic signal, as a filter set for a loudspeaker array
(also called secondary source array), which reproduces the
acoustic field of the virtual source (which is the target, or
virtual, sound field).
The problem is usually limited to the horizontal plane
with a finite number of discrete secondary sources using
appropriate simplifications of the integral formulation
[11]. With WFS systems it is known that plane and spheri-
cal waves can be reproduced approximately in a broad
planar listening area surrounded by the loudspeaker array.
The benefit of current WFS prototypes is their effec-
tiveness in transmitting a spatial impression (in terms of
localization) over a broad area surrounded by the loud-
speakers. One WFS drawback is related to the definition of
the WFS operators, which implies that the reproduction
room response or the electroacoustical system limits [25]
are not considered in the process, except in specific studies
[25][27]. The typical WFS system is consequently based
on an open-loop architecture assuming a free field as the
reproduction space. Active room compensation or system
equalization for WFS is an active research topic for ob-
jective sound field reproduction in a real situation [4],
[16], [19], [20], [25][29], [38]. AWFS falls within that
topic.
1.1.1 WFS Operators
AWFS is partly based on the WFS solution for a given
virtual sound field. The WFS operators are defined as
follows [11]:
Q
WFS
x
l
l
, = Aj

jk
2
cos
e
jkr
0
r
0

r
ref
r
ref
+ r
0

l
(1)
q
WFS
x
l
l
, =
4Q
WFS
x
l
l
,
j
(2)
where A() is the primary-source monopole amplitude,
[rad] is the angle between the primary source and the
normal to the reproduction line at the secondary-source
position x
l
(l)
, r
0
| x
o
x
l
(l)
| is the distance [m] between
the primary source (in x
o
) and the secondary source (in
x
l
(l)
), and r
(ref)
| x
(ref)
x
l
(l)
| is the distance [m] between
the secondary source and the reference line along the line
r
0
. The position vector x
(ref)
corresponds to the point,
aligned with the primary source and the secondary source
l, which belongs to the reference line. These geometric
parameters are illustrated in Fig. 1. In Eq. (1)
l
[m] is the
secondary-source separation (
l
| x
l
(l)
x
l+1
(l)
| ), k is the
wave number, and j 1. Eq. (1) expresses the WFS
monopole source amplitudes Q
WFS
(x
l
(l)
, ) to produce the
virtual sound field p
(virt)
(x, ) while Eq. (2) transforms
WFS monopole amplitudes into source strength ampli-
tudes q
WFS
(x
l
(l)
, ) [39].
A secondary source is active if | | < /2 [rad]. (See
Fig. 1 for a definition of .) An active secondary source
implies that Q
WFS
(x
l
(l)
, ) 0). Other secondary sources,
with | | /2, are not active (artificially forced to Q
WF
-
S(x
l
(l)
, 0). This highlights the fact that a WFS solution
GAUTHIER AND BERRY PAPERS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1108
implies that only a part of the reproduction source array is
active for a given primary source. A spatial weighting
window is used to reduce the WFS strength amplitudes
progressively. This is applied to the secondary sources
closer to the end of the active part of the secondary source
array. Such spatial windowing limits the finite aperture
effects of the secondary source array. Here the spatial
weighting window is a rectangular window with half-
Hanning windows at its ends. The half-Hanning window
extends over 0.5 m [11].
The stationary phase approximation [37] in the devel-
opment of the WFS operators also leads to the definition
of a reference line [11] on which the amplitude of the
reproduction error should theoretically be zero in compari-
son with points outside the reference line, again for the
free field. Indeed, the stationary phase approximation in
WFS implies that the reproduced sound field would be
reproduced perfectly only on a line in front of the second-
ary source array. For WFS this line is chosen arbitrarily,
and this corresponds to the WFS reference line. This line
is shown in Fig. 1. In this paper the secondary source array
is placed on a circle, and the reference line is also defined
as a circle, centered on the primary source position x
0
and
passing through the secondary source array center. The
choice of a given reference line definition is not relevant to
the AWFS fundamental idea, but it remains important for
the WFS definition (for the reproducibility of numerical
and experimental results) since the reference line partici-
pates in the definition of the WFS operators of Eq. (1).
Practical WFS implementation implies discrete repro-
duction sources on a finite line. The WFS reproduced
sound field in free space is then
p
rep
x, =

l=1
L
Q
WFS
x
l
l
,
e
jkr
r
(3)
with r | x x
l
(l)
| .
1.2 Adaptive Wave Field Synthesis
AWFS is introduced as an optimal control approach
with the cost function
J
AWFS
= e
H
e + Q Q
WFS

H
Q Q
WFS
(4)
which must be minimized, with being a real positive
scalar. The superscript H denotes Hermitian transposition.
Eq. (4) introduces the M-component reproduction error
column vector e(x
m
(m)
, ) p
(virt)
(x
m
(m)
, ) p
(rep)
(x
m
(m)
,
) p
(virt)
(x
m
(m)
, ) G()Q(x
l
(l)
, ), where p
(virt)
is the
complex column vector of the target sound pressure field
at the M sensor locations in x
m
(m)
, and G() is an M by L
matrix of the plant complex response (from the L repro-
duction source monopole amplitudes to the M acoustic
pressures at the error sensor array). The first term in Eq.
(4) represents the summation of the quadratic reproduction
errors with respect to the target sound field, observed at
the error sensor locations. The first objective is then to
reduce these reproduction errors caused possibly by room
response or system limitations. The second term in Eq. (4)
is weighted by the regularization parameter and repre-
sents a quadratic departure penalty of the solution Q with
respect to the fixed WFS solution Q
WFS
. The penalization
for any departure from the WFS solution is motivated by
simple facts: the WFS direct sound field approaches the
target sound field and the WFS solution accordingly pro-
vides correct sound localization. More simply, AWFS im-
plies a reproduction error minimization along with a favor
for the WFS solution. For active sound field reproduction
using few error sensors (as reported herein), the WFS
solution introduces supplementary spatial information,
which would not otherwise be accessible using only re-
production error minimization at a few sets of error sen-
sors. This is an advantage of AWFS.
This formulation of the problem, using a classical in-
verse problem with Tikhonov regularization with a priori
Fig. 1. Definition of WFS operators. Virtual source is located in x
0
; reproduction source l is located in x
l
(l)
(x
l
(l)
, . . . , x
l
(l)
, . . . , x
L
(l)
).
x
(ref)
point that belongs to reference line; xany field point, x
m
(m)
mth sensor position (x
1
(m)
, . . . , x
m
(m)
, . . . , x
M
(m)
).
PAPERS ADAPTIVE WAVE FIELD SYNTHESIS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1109
solution (WFS) [40], is the principal originality of AWFS.
Also note that the a priori solution could be of any other
type, such as Ambisonics or multichannel stereophony. In
that case, although it could not be called AWFS, the re-
maining theoretical developments would apply (Q
WFS
be-
ing replaced by any solution Q
priori
). This could be a sub-
ject of future research.
J
AWFS
is a quadratic function of Q. Although the repro-
duction system p
(rep)
(x
m
(m)
, ) G()Q(x
l
(l)
, ) is under-
determined (assuming M < L, as for the remainder of the
paper) [41], the WFS departure penalty term, weighted by
, ensures that there will be a unique minimum of Eq. (4).
That is, the Hessian matrix ([G
H
G + I]) of the quadratic
Eq. (4) is positive definite, provided that > 0. The op-
timum source vector minimizes the cost function of Eq. (4)
[41],
Q
AWFS
= G
H
G + I
1
G
H
p
virt
+ Q
WFS
. (5)
When , the solution reduces to the WFS operators
Q
WFS
; when 0, the solution is the optimal control
solution, which minimizes the reproduction error in a
least-mean-square (LMS) sense. These two situations are
the limiting cases of AWFS [5].
1.2.1 Simple AWFS Examples in Free Field
Some preliminary examples of WFS and AWFS simu-
lations in free field are now introduced for illustrative
purposes. They illustrate the very general behavior of
AWFS. AWFS as an active sound field reproduction and
active room compensation method will be discussed later
on the basis of experimental results.
The theoretical configuration, which also corresponds to
the experimental system, is shown in Fig. 2. The system
includes 24 secondary or reproduction sources grouped in
a circular array and four error sensors for which the re-
production errors should be minimized according to the
AWFS cost function of Eq. (4). This configuration is cho-
sen to illustrate and study the AWFS behavior. For the
practical experiments, a linear array of eight monitoring
sensors is used to evaluate the reproduced sound field. For
the theoretical simulations and examples, a set of eight
concentric circular arrays, including a total of 96 monitor-
ing sensors in the circular region circumscribed by the
black circular solid line shown in Fig. 2, is used.
Simulation results are reported in Figs. 3 and 4. The
results are quantified by the performance criterion E
LS
,
E
LS
=
e
H
x
m
m
, ex
m
m
,
p
virt
H
x
m
m
, p
virt
x
m
m
,
(6)
Fig. 2. AWFS system configuration including reproduction sources, virtual source, error sensors, monitoring sensors, and typical virtual
sound field at 220 Hz. Solid black circular lineregion covered by 96 monitor sensors used for simulation only.
Fig. 3. Performance criterion E
LS
evaluated at error sensors as a
function of frequency. WFS; AWFS using a single
general penalization parameter 22.5 [Eq. (5)]; AWFS
based on independent radiation mode control (SVD,
1
22.5,

2

3
0.75,
4
0.0125).
GAUTHIER AND BERRY PAPERS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1110
which represents the mean normalized quadratic reproduc-
tion errors at the error sensors. When E
LS
reaches 0, the
reproduction error is null. When E
LS
is 1, the reproduction
error is as large as if the reproduced sound field were
perfect silence. This scalar criterion can also be evaluated
for the 96 monitor sensors; x
m
(m)
is then replaced by the
monitor sensor positions. The reproduction error e is com-
puted for the optimal AWFS solution Q
AWFS
or for the
WFS solution Q
WFS
.
In Fig. 3 the performance criterion for sound field re-
production E
LS
is shown for the error sensors. Since this
illustrative case is computed in the free-field situation,
WFS already performs well (E
LS
K 1) since some of the
hypotheses implied in the development of the WFS solu-
tion are respected. However, the WFS solution provides an
E
LS
greater than for AWFS using a single penalization
parameter as in Eq. (5). (The penalization parameter was
set to 22.5.) This proves that the reproduction error is
effectively reduced by AWFS at the error sensor array.
Typical WFS reproduction errors in the free field are
caused by the inherent WFS assumptions [11] and by the
finite size of the active reproduction source array in com-
parison with the wavelength. Since the configuration used
for these simulations is relatively small, this type of error
is dominant for the low frequencies. The third curve
(heavy line) will be discussed in Section 1.2.2.
To illustrate the effect of AWFS outside the control
region defined by the error sensor positions, Fig. 4 shows
the reproduction error minimization at the 96 monitor sen-
sors. (See the region covered in Fig. 2.) Clearly, the re-
production error minimization at the error sensors effec-
tively reduces the WFS reproduction errors outside this
control region. This illustrates the effectiveness of AWFS
in compensating for reproduction errors in a region larger
than the control region covered by the error sensors. This
issue is further developed in the following section. The
third curve will be discussed in Section 1.2.2.
The corresponding real parts of reproduced sound fields
at 220 Hz are shown in Figs. 5 and 6 for the target sound
field of Fig. 2. In these figures several results and infor-
mations are included. The secondary source array is shown
as a heavy circle. The reproduced sound field is shown as
a gray surface. The wavefronts of the target sound pressure
field are shown as dashed contour lines for p
(virt)
(x, )
0. The contour lines of the local normalized performance
criterion E
LS
are also superimposed on the reproduced
sound fields.
As shown in these figures, WFS effectively performs
sound field reproduction in the free field, but the repro-
duction errors (visible through the local E
LS
contour lines)
are smaller for the AWFS reproduced sound field. This
Fig. 4. Performance criterion E
LS
evaluated at 96 monitor sensors
as a function of frequency. WFS; AWFS using a single
general penalization parameter 22.5 [Eq. (5)]; AWFS based
on independent radiation mode control (SVD,
1
22.5,
2

3
0.75,
4
0.0125).
Fig. 5. Real part of reproduced sound field at 220 Hz by WFS in
free field.
Fig. 6. Real part of reproduced sound field at 220 Hz by AWFS,
using a single penalization parameter , in free field.
PAPERS ADAPTIVE WAVE FIELD SYNTHESIS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1111
illustrates the fact that AWFS can effectively reduce the
reproduction errors inherent in WFS. Room compensation
by AWFS is considered from experimental results in Sec-
tion 2.
1.2.2 Singular Value Decomposition of the
Plant Matrix
The singular value decomposition of the plant transfer
matrix is [42]
G
ML
= U
MM

ML
V
LL
H
(7)
where the subscripts indicate matrix dimensions. U is the
matrix whose columns are the left singular vectors u
i
of G,
V is the matrix whose columns are the right singular vec-
tors v
i
of G, and is the rectangular matrix of G real
positive singular values
i
,
=

1
0 0
0
2
0 0

0 0
M
0

ML
(8)
where
i
>
i+1
. The rank r of G (r M < L) is equal to
the number of nonzero singular values. In Eq. (8) it was
assumed that r M for convenience of notation.
The matrices U and V are unitary matrices (U
H
U
UU
H
V
H
V VV
H
I) since the left and right singular
vectors are orthonormal vectors. SVD has well-known
properties [40], [42]. The span of the left singular vectors
u
i
constitutes the range of the matrix G and the span of the
last L r right singular vectors v
i
is the null space of G
(Gv
i
0 for i > r and i L) [42], [43]. In order to
explicitly relate left and right singular vectors to physical
parameters, right singular vectors v
i
and left singular vec-
tors u
i
will hereafter be identified as source modes and
pressure modes, respectively. Each source and pressure
mode pair with the corresponding singular value is here
defined as a radiation mode. This is an AWFS-specific
terminology.
Eq. (7) produces this transformed radiation equation
p
rep
= Q

(9)
with the transformed monopole amplitudes Q and acoustic
pressures p,
p
virt
= U
H
p
virt
p
rep
= U
H
p
rep
(10)
Q

= V
H
Q.
Since is a pseudodiagonal rectangular matrix (the ele-
ments are nonzero on the main diagonal only), the radia-
tion problem of Eq. (9) is a set of independent acoustical
radiations between source modes v
i
and pressure modes u
i
.
The singular values
i
then represent the coupling, or gain,
between the corresponding source mode and pressure
mode.
Theoretical examples of source modes and pressure
modes in the free field are shown in Figs. 7 and 8, where
a simple free-field radiation model was used for the sec-
ondary source system [see Eq. (3)].
The source modes at the secondary source array corre-
spond to harmonic functions. These harmonic functions
typically give simple multipole patterns as pressure modes
at the error sensor array. According to the pressure modes
Fig. 7. First four source modes at 220 Hz in free field.
Fig. 8. First four pressure modes at 220 Hz in free field. sensor
position; positive real part; negative real part. Symbol di-
ameter illustrates magnitude of corresponding value.
equivalent computed free-field directivity.
GAUTHIER AND BERRY PAPERS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1112
shown in Fig. 8, these multipole patterns are finite-
difference approximations of pressure, perpendicular pres-
sure gradients, and second-order mixed spatial derivatives
at the error sensor array.
The harmonic sound fields produced by each of these
radiation modes are shown in Fig. 9. Clearly, the radiation
modes correspond to finite approximations of pressure,
pressure gradients, and second-order mixed spatial deriva-
tives at the error sensor array, which is located at the
center of the reproduction source array. It is then expected
that the higher order radiation modes, which correspond to
higher order spatial derivatives, will play a significant role
in determining the size of the region where the reproduc-
tion error would be reduced in comparison with WFS.
Taking the AWFS solution of Eq. (5) and using the
SVD of G, the AWFS solution can be expanded over the
source modes v
i
, Q
AWFS

L
ii
Q

i
v
i
, as follows:
Q

i
=

i
2
+
i
p
i
virt
+

i

i
2
+
i
v
i
H
Q
WFS
i r M
v
i
H
Q
WFS

i
0, i r,
i L
0,

i
= 0, i r,
i L
(11)
where the independent contributions of v
i
in Q
AWFS
were
now individually regularized with
i
. AWFS thus proceeds
to control and regularize each radiation mode indepen-
dently. The AWFS solution of Eq. (4) implicitly assumed
an identical regularization parameter for all source modes,

i
for all i. Note that Eq. (11) was slightly modified
in comparison to Eq. (5) for the particular case
i
0. Eq.
(11) is forced to the pseudoinverse definition by the addi-
tion of the third case. The AWFS solution [Eq. (11)] sug-
gests that the underlying mechanism is independent radia-
tion mode control. This mechanism can be used for
practical AWFS implementation.
Examples of AWFS based on independent radiation
mode control are included in Figs. 3 and 4, where the
performance criterion E
LS
is represented by dashed lines
and was computed with different penalization parameters
for each of the radiation modes. Clearly the independent
radiation mode control implementation yields a better per-
formancethe size of the effective reproduction region is
expected to be larger.
Since the singular values
i
decrease with the mode
orders i, the higher order radiation mode penalization pa-
rameters
i
were also reduced in amplitude. This is nec-
essary since the effect of penalization for each of the ra-
diation modes depends on the relative magnitude of the
square of the singular value and the corresponding penal-
ization parameter [see Eq. (11)]. As the higher order ra-
diation modes typically correspond to higher order spatial
derivatives, it is important not to overpenalize these modes
as they might enlarge the effective control region. (This
was the case for AWFS using a single penalization param-
eter, which is fully equivalent to Eq. (11) using
i
.)
This is supported by Fig. 10, which shows the reproduced
sound field by AWFS based on independent radiation
mode control. In this regard, future research on AWFS
using more error sensors in a compact sensor array would
be interesting since it is probable that the total number of
higher order spatial derivatives included in the pressure
Fig. 9. (a)(d) Sound field radiated by first four source modes. (e)
Sound field radiated by null space, representing sum of null
space radiation modes. zero-pressure contours of sound
pressure fields.
PAPERS ADAPTIVE WAVE FIELD SYNTHESIS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1113
mode basis would increase, as would the effective control
region.
The AWFS realization based on independent radiation
mode control (Fig. 10) implies a larger effective reproduc-
tion region (characterized by a low local E
LS
) than AWFS
based on a single penalization parameter (Fig. 6). This
stems from the ability to penalize the radiation modes
individually and from the contribution of the higher order
radiation modes.
1.2.3 AWFS, SVD, and Ambisonics
The SVD of the plant matrix as an analysis tool for
AWFS introduces a link between WFS, AWFS, and Am-
bisonics. Theoretical connections between WFS and Am-
bisonics are already known on the basis of theoretical
orthogonal functions of the wave equation in cylindrical
coordinatescylindrical harmonics [15]. The SVD of the
plant matrix for AWFS, however, introduces a new con-
nection on the basis of orthogonal functions derived from
numerical plant decoupling.
As the pressure modes (see Fig. 8) approach cylindrical
harmonics, or Ambisonics components [5] at the error
sensor array, it is possible to understand that AWFS op-
erates as a closed-loop implementation of Ambisonics,
with a penalty for the departure from the WFS solution
since the pressure modes, which approach Ambisonics
components, are individually controlled on the basis of
least-mean-square minimization, as for Ambisonics,
which typically implies open-loop (that is, in a theoretical
a priori step) minimization of the errors for these compo-
nents. Accordingly these very specific shapes of the pres-
sure modes at the sensor array allow for a conceptual
connection between Ambisonics and AWFS using the
configuration presented. Further considerations and ex-
amples are given [5].
1.3 Signal Processing for Adaptive Wave
Field Synthesis
For the sake of brevity, this section describes the im-
plied signal processing for AWFS as defined in the pre-
vious sections. Signal processing for AWFS is the topic of
a paper accepted for publication by the Journal of Acous-
tical Society of America.
1.3.1 Modification of the Filtered-Reference LMS
Algorithm (FXLMS)
In a classical adaptive minimization of Eq. (4), a cen-
tralized multichannel adaptive algorithm would be imple-
mented using a feedforward architecture [44]. This is il-
lustrated in Fig. 11. A set of z-transformed adaptive filters
w(z) are adapted to reduce the quadratic summation of the
M reproduction errors e
M
(n), where n is the sample index.
Fig. 10. Real part of reproduced sound field at 220 Hz by AWFS
based on independent radiation mode control.
Fig. 11. AWFS using modified multichannel adaptive algorithm. Capitalized subscripts denote signal dimensions.
GAUTHIER AND BERRY PAPERS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1114
The discrete filters w(z) are fed by a reference signal x(n)
and produce the L command signals y
L
(n) for the L sec-
ondary sources. Here G(z) is the only physical system. All
other operations are achieved in a discrete-time signal-
processing domain. This includes G

(z), which is an iden-


tified approximation of G(z). The target sound field defi-
nition at the error sensors is obtained by passing the
reference signal x(n) through the target pressure modeler
a(z). The WFS solution departure penalty [see Eq. (4)] is
automatically taken into account by the LMS adaptation,
which is based on a gradient algorithm of the discrete-time
representation of the AWFS cost function of Eq. (4). Us-
ing M error sensors and L reproduction sources, this will
be a multiple-inputmultiple-output (MIMO) L M sys-
tem. The computational burden involved in the block dia-
gram of Fig. 11 becomes prohibitive as L, M, and the
adaptive filter order increase [44], [45]. Moreover, this
may slow down the gradient LMS algorithm convergence
as L and M increase [45]. This type of algorithm implies a
general convergence coefficient and a unique general
penalization parameter [46] [see Eq. (5)].
1.3.2. Independent Radiation Mode Control
Based on Plant Singular Value Decomposition
As noted in Eq. (11), each source mode contribution Q

i
in Q
AWFS
is solely defined by the corresponding p
i
(virt)
,
i
,

i
, and v
i
H
Q
WFS
, which is the projection of the WFS
operator Q
WFS
on the ith source mode v
i
. It is possible to
exploit this independence of the radiation modes for prac-
tical implementation of AWFS.
This leads to another approach, illustrated as a block
diagram in Fig. 12. G(z) is again the only physical system.
With this diagram we seek to produce a set of M (with
M L) independent SISO (single-input-single-output)
adaptive systems w
i
(z). This type of algorithm then implies
an independent convergence coefficient
i
and an inde-
pendent penalization parameter
i
for each radiation mode.
This is an advantage of AWFS based on independent ra-
diation mode controlindependent fine-tuning of the con-
vergence and penalization properties of each radiation
mode. This is not the case with the modified multichannel
FXLMS algorithm for AWFS. Here each independent
adaptive system generates the signal sent to each source
mode v
i
for i r M. This type of independent control
using SVD was investigated by Bai and Elliott [45] for
loudspeaker-to-ear crosstalk cancellation, but without an a
priori solution like Q
WFS
. This independent radiation
mode control of AWFS is also a modification of the prin-
cipal-component LMS (PC-LMS) algorithm for active
noise control of tonal disturbance [46][48]. The PC-LMS
is based on SVD, but for AWFS the SVD decoupling must
operate over the entire frequency range of interest [5],
[45].
This implementation requires the additional MIMO syn-
thesis and MIMO analysis filters
(s)
G(z) and
(a)
G(z), re-
spectively, which must be generated in a prior identifica-
tion stage. In this case an identification of the frequency
response functions (FRFs) G

ml
() between each secondary
source l and each error sensor m must be achieved. The
SVD of the corresponding transfer matrix for each fre-
quency provides the source modes v
i
() and the corre-
sponding pressure modes u
i
(). The inverse Fourier trans-
form of v
i
()/
i
(), i r M, and u
i
() generates the
z-transformed discrete-time impulse response (IR) matri-
ces
(s)
G(z) and
(a)
G(z), respectively. The matrix
(s)
G(z)
generates the inputs of the individual reproduction sources
from the source mode inputs with a delay of
SVD
samples
[45]. Note that division by
i
() implies plant whitening
along uncoupling. (Plant whitening is optional; results pre-
sented in this paper only include plant decoupling.) Con-
versely, the matrix
(a)
G(z) is used to generate the pressure
mode errors from the individual sensor errors, again with
Fig. 12. Block diagram of AWFS least-mean-square adaptive digital signal processing implementation based on independent radiation
mode control. Subscripts denote signal dimensions.
PAPERS ADAPTIVE WAVE FIELD SYNTHESIS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1115
a delay of
SVD
samples [5], [45]. The second synthesis
filter matrix
(s)
G(z) is created from the null-space source
modes to produce the fixed part of the AWFS solution [see
the last two cases of Eq. (11)]. The delay block (Fig. 12)
of 2
SVD
samples then represents the uncoupled plant
model. This type of realization was more extensively de-
scribed in [45] for loudspeaker-to-ear cross talk cancella-
tion. Also, because of the WFS departure penalty in the
AWFS definition, an additional set of fixed filters produce
the outputs for the i > r (and i L) WFS fixed source
mode contributions q
i
, which belong to the null space.
This introduction of an a priori solution is the principal
difference compared to the work of Bai and Elliott [45].
1.3.3 Synthesis and Analysis Filters: Radiation
Mode Reordering
The general procedure just described involves supple-
mentary practical considerations before being usable for
the creation of the synthesis and analysis filtersit re-
quires further operations.
Numerical SVD algorithms typically arrange the left
and right singular vectors in V() and U() according to
decreasing singular values. Then, since the plant SVD is
computed for each frequency independently, a radiation
mode reordering algorithm is needed to smooth the fre-
quency response functions of the uncoupling filters. This
smoothing typically yields shorter impulse responses for
the corresponding filters in the time domain.
For the AWFS system the reordering algorithm simply
performs a two-step smoothing by 1) rearranging the col-
umns in V() and U() so that source modes vary
smoothly as a function of frequency, and 2) verifying and
correcting any sign changes in V() as the frequency in-
creases. This proved to increase greatly the quality (com-
pactness in the time domain) of the synthesis and analysis
filters. (More details and examples will be given in the
paper on signal processing for AWFS.) Note that such
considerations for left and right singular vector reordering
was not included in Bai and Elliotts original work on
broad-band plant decoupling [45].
2. ADAPTIVE WAVE FIELD SYNTHESIS:
EXPERIMENTAL RESULTS
This section summarizes the experimental results ob-
tained for AWFS in three different rooms: hemianechoic
chamber, laboratory space, and reverberation chamber
with damping material. These rooms cover a wide range of
typical acoustical responses, from nearly free field to
nearly diffuse field. A limited set of results are presented
here to validate the AWFS concept. The AWFS system is
shown in Fig. 13 in the laboratory space. The loudspeaker
and microphone setup corresponds to the configuration
shown in Fig. 2.
2.1 Hemianechoic Chamber
Sound fields reproduced by WFS and AWFS in a hemi-
anechoic chamber are presented to demonstrate the inter-
est of AWFS in a very simple acoustical environment. The
WFS reproduced impulse responses (IRs) at the monitor-
ing array are shown in Fig. 14. These reproduced sound
fields were measured using swept sines as excitation sig-
nals. Frequency response functions (FRFs) were first com-
puted by averaging over 200 recordings in the frequency
domain. Inverse Fourier transforms of the resulting aver-
Fig. 13. Experimental AWFS setup in laboratory space.
Fig. 14. Reproduced (heavy gray lines) and virtual (thin black lines) impulse responses at monitoring sensor array (shown in Fig. 2)
for WFS with system in hemianechoic chamber.
GAUTHIER AND BERRY PAPERS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1116
age FRFs were used to obtain 256-coefficient impulse re-
sponses. The bandwidth is limited to 600 Hz, which is just
below the WFS aliasing frequency for this setup (634 Hz
assuming at least two reproduction sources per smallest
wavelength).
Clearly, the target sound field is approached by the
WFS direct sound field. However, there are some discrep-
ancies between this reproduced direct sound field and the
target sound field, presumably caused by the loudspeaker
responses and the WFS approximations. After the direct
sound field impinges on the monitoring sensor array, the
floor reflection crosses this array. In this figure the late
reflection corresponds to the low-frequency echo of the
hemianechoic chamber, which is not anechoic below ap-
proximately 100 Hz. All these differences are potentially
compensated by AWFS.
Using the modified FXLMS algorithm and the indepen-
dent radiation mode control algorithm for AWFS, the re-
produced sound fields in the hemianechoic chamber are
shown in Fig. 15 for the same virtual source position.
Clearly, the two implementations of AWFS compensate
both for 1) the direct sound field coloration and 2) the
room response. This illustrates the interest of AWFS as an
active sound field reproduction method.
Although hardly visible on the figure, the AWFS algo-
rithm based on independent radiation mode control per-
forms better than the AWFS algorithm based on the modi-
fication of the classical FXLMS algorithm. This is shown
in Fig. 16, where the normalized energies of the reproduc-
tion errors are computed for the reproduced sound fields of
Figs. 14 and 15 and for a supplementary virtual source
position. The normalized energies of the errors are com-
puted from the differences between virtual and reproduced
IRs shown in Figs. 15 and others, measured for different
virtual source positions. The normalized energies are com-
puted as the sums of the quadratic error signals (differ-
ences between virtual IRs and reproduced IRs in the time
domain) over the length of the IRs normalized by the total
quadratic sum of the virtual IRs, divided by the number of
monitoring microphones. The normalization is thus
achieved by division by the mean virtual IR energies at the
monitoring microphones.
According to the results shown in Fig. 16, the AWFS
algorithms reduce on average the reproduction errors as
compared to WFS by controlling the reproduction error at
the four error sensors (two of which are monitoring sen-
sors number 4 and 5). As suggested earlier in this paper,
AWFS based on independent radiation mode control ef-
fectively provides a larger reproduction region since the
higher order modes are included in the controller in that
case. Note that the size of the effective control region is
also blurred by the fact that this type of representation
includes all the frequencies at once. AWFS modified by
FXLMS provides a significant reproduction error reduc-
tion at the two central monitoring microphones, but the
reproduction errors are not so small for the other monitor-
ing microphones. AWFS performs better than WFS for all
the virtual source positions reported; active room compen-
sation by AWFS is effective and proven. The convergence
coefficients and
i
, the penalization parameters and
i
,
Fig. 15. Reproduced (heavy gray lines) and virtual (thin black
lines) impulse responses at monitoring sensor array (shown in
Fig. 2) for AWFS with system in hemianechoic chamber. (a)
FXLMS algorithm with 20. (b) Independent radiation mode
control.
Fig. 16. Normalized energies of error signals at each monitoring
microphone for two virtual source positions in hemianechoic
chamber. Position #1x
0
[0, 4, 0], position #2x
0
[ 4, 0,
0]. WFS errors; errors of AWFS by FXLMS; errors of
AWFS by independent radiation mode control.
PAPERS ADAPTIVE WAVE FIELD SYNTHESIS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1117
and the number of control coefficients are summarized in
Table 1 for all the results reported, including those for the
laboratory space and the reverberation chamber.
2.2 Laboratory Space and
Reverberation Chamber
Since AWFS is proposed as an active room compensa-
tion method for sound field reproduction, several ex-
amples of room compensation in more hostile environ-
ments are summarized. To validate some theoretical
propositions, Fig. 17 presents the pressure modes at the
error sensor array at 220 Hz in the reverberant chamber.
These are the pressure modes u
i
, as was done for the
theoretical case shown in Fig. 8. Although they differ from
those obtained in the theoretical case, the pressure modes
effectively yield finite approximations of pressure, pres-
sure gradients, and second-order mixed spatial derivatives
of pressure at the error sensor array. This proves experi-
mentally that AWFS applied to the specific loudspeaker
and sensor layout presented herein can be understood as a
closed-loop implementation of Ambisonics since the pres-
sure modes approach the typical low-order Ambisonics
components (further details are given in [5]).
IRs reproduced by WFS in laboratory space are shown
in Fig. 18. As for the WFS experiment in the hemi-
anechoic chamber, the reproduced direct sound field ap-
proaches the target sound field. The late part of the IRs
contains, as expected for the room shown in Fig. 13, far
more energy than that obtained in the hemianechoic space.
This increases the discrepancy between WFS and the tar-
get sound field and can potentially be compensated by
AWFS.
Reproduced sound fields by AWFS illustrated in Fig. 19
show that this discrepancy can be reduced drastically. A
long part of the tail of the reproduced IRs caused by the
room response is compensated. The reproduced sound
fields by AWFS show less effect of the room than WFS,
and even the reproduced direct sound field approaches
more precisely the target direct sound field.
As shown earlier for the reproduced sound field in
hemianechoic space (Fig. 16), Fig. 20 shows the normal-
ized energies of the reproduction errors at the eight moni-
toring sensors (see Fig. 2) for two virtual source positions,
one of which was used to produce the sound field shown
in Figs. 18 and 19.
According to the results shown in Fig. 20, the AWFS
algorithms reduce on average the reproduction errors
when compared to WFS by controlling the reproduction
error at the four error sensors. As for the hemianechoic
space, AWFS based on independent radiation mode con-
trol effectively provides a larger reproduction region in
comparison with AWFS based on the modification of the
FXLMS algorithm. AWFS performs better than WFS for
all the virtual source positions reported. Active room com-
pensation by AWFS is effective even in a more hostile
acoustic environment than the hemianechoic chamber.
Further WFS and AWFS experiments were carried out
in a reverberation chamber. Several sheets of damping
material were placed on the floor and in the corners of the
room to reduce the reverberation time. Otherwise the ma-
jor source of acoustical absorption was the system opera-
tor, and this typically caused excessively long system IRs.
The addition of these absorbent panels reduced the length
and level of the room reverberation, although these re-
mained above the reverberation length and level obtained
in the laboratory space.
Results are summarized in Fig. 21, which shows the
normalized energies of the reproduction error at the moni-
toring sensors for four different virtual source positions.
Fig. 17. Measured pressure modes at 220 Hz in reverberant
chamber. sensor position; positive real part; negative real
part; + positive imaginary part; negative imaginary part. Sym-
bol diameter illustrates magnitude of corresponding value.
equivalent computed free-field directivity.
Table 1. Number of coefficients (for AWFS filters and reproduced sound field), convergence coefficients, and penalization
parameters for different rooms and algorithms.
Rooms and Algorithms Coefficients or
i
or
i
Hemianechoic, modified FXLMS 256 0.00001 20
Hemianechoic, independent radiation mode control 256 0.0001 0.0004 2 0.2
0.0004 0.002 0.2 0.1
Laboratory, modified FXLMS 512 0.000002 30
Laboratory, independent radiation mode control 512 0.0000075 0.00003 2 0.2
0.00003 0.00015 0.2 0.1
Reverberant, modified FXLMS 1024 0.0000005 50
Reverberant, independent radiation mode control 1024 0.0000125 0.00005 4 0.4
0.00005 0.00025 0.4 0.04
GAUTHIER AND BERRY PAPERS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1118
Clearly, the normalized energies of the reproduction errors
for WFS are greater than those obtained for the hemi-
anechoic chamber and the laboratory space. Once again,
AWFS effectively reduces the reproduction errors. AWFS
performs better than WFS for progressive sound field re-
production (spherical wave from the virtual source in the
horizontal plane). This proves that AWFS can effectively
reduce the effect of the room response on reproduced
sound fields by WFS even in a hostile environment. More-
over, the AWFS algorithm based on independent radiation
mode control performs better than AWFS based on the
modified FXLMS algorithm. This concludes the experi-
mental validation of the AWFS concept for active sound
field reproduction in rooms.
2.2.1 Effect of the Higher Order
Radiation Modes
In Section 1.2.2 it was claimed that the higher order
radiation modes are of relevance in increasing the effec-
tive control region. This was verified experimentally in the
laboratory space. For the dedicated AWFS experiment the
WFS operators were forced to zero, Q
WFS
0, and the
independent radiation mode controllers were sequentially
added in the solution. The WFS solution is forced to zero
to illustrate the effects of the individual radiation modes
one to four. The test was made for harmonic sound field
reproduction by AWFS at 400 Hz.
The results of these experiments are shown in Fig. 22,
where the normalized criterion E
LS
E[e(n)
T
e(n)/p
(virt)
T
(n)
p
(virt)
(n)] at the monitor sensors is represented. E[] rep-
resents mathematical expectation.
Clearly, when AWFS based on independent radiation
mode control includes only one radiation mode, the results
correspond to AWFS by FXLMS. When the number of
higher order modes included in AWFS based on independent
radiation mode control increases, the size of the effective
reproduction region increases from a quarter wavelength
(one radiation mode) to half a wavelength (four radiation
modes). This supports the previous observations regarding
the importance of the higher order radiation modes to en-
large the effective active sound field reproduction region.
3. SOUND FIELD REPRODUCTION FOR SOUND
ENVIRONMENT REPRODUCTION
Independent of the chosen spatial sound technology,
one of the very promising applications for spatial sound,
Fig. 18. Reproduced (heavy gray lines) and virtual (thin black lines) impulse responses at monitoring sensor array (shown in Fig. 2)
for WFS with system in laboratory space.
Fig. 19. Reproduced (heavy gray lines) and virtual (thin black
lines) impulse responses at monitoring sensor array (shown in
Fig. 2) for AWFS with system in laboratory space. (a) FXLMS
algorithm with 20. (b) Independent radiation mode control.
PAPERS ADAPTIVE WAVE FIELD SYNTHESIS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1119
including sound field reproduction, is sound environment
reproduction for psychoacoustics or acoustics experi-
ments. Sound field reproduction systems or spatial sound
technologies are interesting because they simplify the ex-
perimental process and they do not imply in situ subjective
measurements. They also simplify the process of AB
comparative listening, and they allow virtual interactive
use of the system. Sound environment reproduction is used
in the automotive sector for the prediction of sound in
virtual automobiles. Examples are using multichannel
acoustic simulation and binaural transfer path analysis
[49], [50]. Similar work has been conducted in the aero-
space sector to create virtual environments for aircraft
community noise [51], [52], as well as for aircraft cabin
noise [53] and cockpits [54], and to evaluate the sound
quality of aircraft interior noise using modeling tools [55].
Other recent examples of sound field reproduction for
sound environment reproduction within the aerospace con-
text are given in [34], [35], one of which is concerned with
sound reproduction in a helicopter mock-up.
Most of this work has been using auralization and bin-
aural reproduction techniques to reproduce acoustic
stimuli at ear locations [56], [57]. However, these tech-
niques are limited to small reproduction zones, the listen-
ers ears. This limits the application of these techniques in
mock-up situations where it is expected that the auditor
navigates in a visual installation. Research on interactive
sound field reproduction of aircraft cabin and cockpit
noise proves to be an interesting future research avenue for
the spatial audio community as it includes very specific
problems and constraints, which can lead to significant
new knowledge and technologies. One of the great chal-
lenges is to provide and define an integrated solution that
includes sound field characterization (both subjectively
and objectively), mock-up characterization, hierarchical
classification of the sound field elements, and a combina-
tion plus an adaptation of several known spatial sound
technologies to fit the sound field element classification
(that is, select or create appropriate techniques for differ-
ent types of field components such as modal response,
diffuse field, localized sources, direct sound fields, aero-
Fig. 20. Normalized energies of error signals at each monitoring
microphone for two virtual source positions in laboratory space.
Position #1x
0
[0, 4, 0]; position #2x
0
[ 4, 0, 0].
WFS errors; errors of AWFS by FXLMS; errors of AWFS
by independent radiation mode control.
Fig. 21. Normalized energies of error signals at each monitoring
microphone for four virtual source positions in reverberant
chamber. Position #1x
0
[0, 4, 0]; position #2x
0
[ 4, 0,
0]; position #3x
0
[0, 1.5, 0]; position #4x
0
[ 1.19,
0.91, 0]. WFS errors; errors of AWFS by FXLMS;
errors of AWFS by independent radiation mode control.
Fig. 22. Normalized E
LS
criterion at monitoring sensors using
harmonic algorithms at 400 Hz after convergence in laboratory
space using AWFS with Q
WFS
0. Wave length and some
corresponding fractions are included.
GAUTHIER AND BERRY PAPERS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1120
acoustic noise sources, and stationary versus transient
sounds). Moreover, this constitutes an interdisciplinary
challenge, gathering acousticians, psychoacousticians, en-
gineers, and vehicle users. As an example, the project
might include an ecological validation of the proposed
reproduction system [58] to validate the proposed sound
environment reproduction process.
Our future research in sound field reproduction will be
dedicated to sound environment reproduction in vehicle
mock-ups. This goal, which differs somewhat from the
objectives and ideas presented in this paper on AWFS, is
discussed briefly in this section.
To highlight the differences, consider, as an example,
the extreme experimental AWFS case where active sound
field reproduction was tested in the reverberation chamber.
In that situation the objective was to reproduce locally,
inside the reproduction source array, a progressive sound
field, like in an anechoic chamber. In that situation the
difficulty arises from the acoustical response differences
between the virtual space and the reproduction space.
Transforming a part of a reverberant chamber into an an-
echoic space using active sound field reproduction is a
considerable challenge. Although such an experiment was
achieved to evaluate the performance limitations of AWFS
in hostile acoustic situations, it is not practically interest-
ing. Indeed, a reverberation chamber does not correspond
to any common multichannel listening room for spatial
sound.
Typically, classical WFS systems are installed in nor-
mal listening rooms with an acoustical response more at-
tenuated than in the reverberation chamber. This type of
reproduction spaces for spatial sound implies a specific set
of technical problems for the chosen technology. The sys-
tem must adapt to various a priori unknown reproduction
spaces with more or less reverberant responses and various
differences between rooms (as the listening rooms are ex-
pected to be well designed for multichannel listening).
That being said, for spatial sound application (music re-
production, virtual reality display) the method must be
general enough to tolerate various environments and differ-
ent loudspeaker and microphone arrays without a problem.
On that very specific point, sound field reproduction in
vehicle mock-ups that imitate faithfully the visual aspects
of the original vehicles implies a reduced acoustical dif-
ference between the virtual space and the reproduction
space. If the mock-up is a convenient visual and spatial
representation of the real vehicle (including geometry, ma-
terials, furniture), it is expected that the acoustical re-
sponse of the mock-ups enclosed volume will be nearly
similar to that of the real vehicle. This poses a very spe-
cific set of problems, constraints, and simplifications in
comparison with the general aim of progressive sound
field reproduction in rooms as reported herein. In the case
of sound field reproduction in mock-ups, the method can
be very case-specific since it strongly depends on the situ-
ation, and the system is only designed for a given vehicle.
Accordingly, it is possible to imagine that a fine and
case-specific sound field component hierarchical classifi-
cation according to physical and perceptive sound field
characteristics can be applied to that problem to select the
most efficient methods and needed precision or fidelity for
each of the retained sound field or sound environment
components. This is the topic of our future research in
spatial sound reproduction for sound environment repro-
duction in vehicle mock-ups. We then expect to elaborate
a comprehensive and general method for sound field char-
acterization and reproduction in mock-ups from our expe-
rience with a given vehicle.
Within that context, AWFS might be used to compen-
sate for the discrepancies between the real vehicle re-
sponse and the mock-up response and for the electroacous-
tical system response. Moreover, further, yet very simple,
modifications of the AWFS definition can enhance its con-
tribution within such a research project. It is possible to
generalize the cost function of Eq. (4) using any a priori
solution. This might be a subject of further research.
4 CONCLUSION
In this paper WFS was first reviewed from a general
description. One of the important aspects that was recalled
for WFS is the reproduction error introduced by the room
response, which was also illustrated by measurements in
three different spaceshemianechoic chamber, laboratory
space, and reverberant chamber.
To compensate for reproduction errors caused by the
listening room response, AWFS was introduced as a re-
production error minimization combined with a penalty
for any departure from the WFS solution. This is the major
originality of AWFS. Note that AWFS is not limited to the
configuration used in this paper. Testing the cost function
of Eq. (4) for different loudspeaker and error sensor setups
could be the topic of further research on AWFS. Usage of
the WFS solution as a favored solution in AWFS is simply
motivated by the fact that WFS, although potentially suf-
fering from room coloration, provides relatively accurate
sound localization over 360 degrees for a large listening
region delimited by a reproduction source array. It was
shown that AWFS can then reduce the reproduction errors
without being very different from the WFS solution, de-
pending on the penalization parameters ( or
i
for AWFS
based on independent radiation mode control). Analysis of
AWFS using SVD showed the underlying mechanism of
AWFSindependent radiation mode control. It also in-
troduced a unified vision of WFS, Ambisonics, AWFS,
and control using plant SVD.
Two algorithms for AWFS were then introduced
modified FXLMS and independent radiation mode control
on the basis of plant decoupling using SVD. One of the
advantages of AWFS based on independent radiation
mode control is the ability to fine-tune the convergence
properties of each radiation mode controller and its con-
tribution in the total solution. As the higher order radiation
modes imply higher order spatial derivatives at the error
sensor array, this causes an enlargement of the effective
reproduction region where the reproduction error is re-
duced. Further research on AWFS could be dedicated to
the development of more efficient adaptive algorithms.
PAPERS ADAPTIVE WAVE FIELD SYNTHESIS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1121
Experimental investigations of both WFS and AWFS in
several rooms showed how AWFS can effectively reduce
both the room response and the electroacoustical system
coloration. This demonstrates the effectiveness of AWFS
to compensate for room responses, the original objective
behind this project. Moreover it was shown that improved
sound field synthesis in a reverberant environment is
possible.
Experimental comparison between the two algorithms
supports the fact that AWFS based on independent radia-
tion mode control enlarges the effective control region by
allowing the contribution of the higher order radiation
modes to be more present in the resulting solution. Further
research on the practical implementation and algorithms of
AWFS based on independent radiation mode control (in-
cluding plant SVD) could be interesting. More details on
AWFS can be found in the first authors thesis [59].
5 ACKNOWLEDGMENT
This work was supported by NSERC (Natural Sciences
and Engineering Research Council of Canada), NATEQ
(Fond Quebecois de la Recherche sur la Nature et les
Technologies), VRQ (Valorisation Recherche Quebec),
and the Universite de Sherbrooke. This research has been
conducted in collaboration with CIRMMT (Centre for In-
terdisciplinary Research in Music, Media, and Technol-
ogy, McGill University). The authors acknowledge the
contribution of Emmanuel Corratge who contributed to the
construction of the harmonic AWFS system and to the
harmonic experiments.
6 REFERENCES
[1] W. B. Snow, Basic Principles of Stereophonic
Sound, J. SMPTE, vol. 61, pp. 567589 (1953).
[2] A. J. Berkhout, D. de Vries, and P. Vogel, Acoustic
Control by Wave Field Synthesis, J. Acoust. Soc. Am.,
vol. 93, pp. 27642778 (1993).
[3] A. J. Berkhout, M. M. Boone, and D. de Vries,
Generation of Sound Fields Using Wave Field Synthe-
sisAn Overview, in Proc. Active 95 (1995).
[4] P. A. Gauthier, A. Berry, and W. Woszczyk,
Sound-Field Reproduction In-Room Using Optimal Con-
trol Techniques: Simulations in the Frequency Domain,
J. Acoust. Soc. Am., vol. 117, pp. 662678 (2005).
[5] P. A. Gauthier and A. Berry, Adaptive Wave Field
Synthesis with Independent Radiation Mode Control for
Active Sound Field Reproduction: Theory, J. Acoust.
Soc. Am., vol. 119, pp. 27212737 (2006).
[6] J. W. Choi and Y. H. Kim, Manipulation of Sound
Intensity within a Selected Region Using Multiple
Sources, J. Acoust. Soc. Am., vol. 116, pp. 843852 (2004).
[7] I. Veit and H. Sander, Production of Spatially Lim-
ited Diffuse Sound Field in an Anechoic Room, J. Au-
dio Eng. Soc., vol. 35, pp. 138143 (1987 Mar.).
[8] J. Merimaa and V. Pulkki, Spatial Impulse Re-
sponse Rendering I: Analysis and Synthesis, J. Audio
Eng. Soc., vol. 53, pp. 11151127 (2005 Dec.).
[9] V. Pulkki and J. Merimaa, Spatial Impulse Re-
sponse Rendering II: Reproduction of Diffuse Sound and
Listening Tests, J. Audio Eng. Soc., vol. 54, pp. 320
(2006 Jan./Feb.).
[10] C. I. Cheng and G. H. Wakefield, Introduction to
Head-Related Transfer Functions (HRTFs): Representa-
tions of HRTFs in Time, Frequency, and Space, pre-
sented at the 107th Convention of the Audio Engineering
Society, J. Audio Eng. Soc. (Abstracts), vol. 47, p. 100
(1999 Nov.), preprint 5026.
[11] E. N. G. Verheijen, Sound Reproduction by Wave
Field Synthesis, Ph.D. thesis, Delft University of Tech-
nology, Delft, The Netherlands (1997).
[12] M. A. Gerzon, Ambisonics in Multichannel
Broadcasting and Video, J. Audio Eng. Soc., vol. 33, pp.
859871 (1985 Nov.).
[13] D. G. Malham and A. Myatt, 3-D Sound Spatial-
ization Using Ambisonic Techniques, Computer Music
J., vol. 19, pp. 5870 (1995).
[14] M. A. Poletti, Three-Dimensional Surround
Sound Systems Based on Spherical Harmonics, J. Audio
Eng. Soc., vol. 53, pp. 10041025 (2005 Nov.).
[15] R. Nicol and M. Emerit, 3D-Sound Reproduction
over an Extensive Listening Area: AHybrid Method Derived
from Holophony and Ambisonic, in Proc. AES 16th Int.
Conf. (Rovaniemi, Finland, 1999 April 1012), pp. 436453.
[16] S. J. Elliott and P. A. Nelson, Multiple-Point
Equalization in Room Using Adaptive Digital Filters, J.
Audio Eng. Soc., vol. 37, pp. 899907 (1989 Nov.).
[17] S. Ise, A Principle of Sound Field Control Based
on the KirchhoffHelmholtz Integral Equation and the
Theory of Inverse System, Acta Acustica, vol. 85, pp.
7587 (1999).
[18] J. Garas, Adaptive 3D Sound Systems, Tech-
nische Universiteit Eindhoven, Eindhoven, The Nether-
lands (1999).
[19] M. Bouchard and S. Quednau, Multichannel Re-
cursive-Least-Squares Algorithms and Fast-Transversal-
Filter Algorithms for Active Noise Control and Sound
Reproduction Systems, IEEE Trans. Speech Audio Pro-
cess., vol. 8, pp. 606618 (2000).
[20] A. O. Santillan, Spatially Extended Sound Equal-
ization in Rectangular Rooms, J. Acoust. Soc. Am., vol.
110, pp. 19891997 (2001).
[21] E. Corteel, On the Use of Irregularly Spaced
Loudspeaker Arrays for Wave Field Synthesis, Potential
Impact on Spatial Aliasing Frequency, in Proc. 9th Int.
Conf. on Digital Audio Effects (DAFX) (2006).
[22] S. Spors and R. Rabenstein, Spatial Aliasing Ar-
tifacts Produced by Linear and Circular Loudspeaker Ar-
rays Used for Wave Field Synthesis, presented at the
120th Convention of the Audio Engineering Society, J.
Audio Eng. Soc. (Abstracts), vol. 54, p. 689 (2006 July/
Aug.), convention paper 6711.
[23] B. Klehs and T. Sporer, Wave Field Synthesis in
the Real World: Part 1In the Living Room, presented at
the 114th Convention of the Audio Engineering Society, J.
Audio Eng. Soc. (Abstracts), vol. 51, p. 408 (2003 May),
convention paper 5727.
GAUTHIER AND BERRY PAPERS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1122
[24] T. Sporer and B. Klehs, Wave Field Synthesis in
the Real World: Part 2In the Movie Theater, presented
at the 116th Convention of the Audio Engineering Society,
J. Audio Eng. Soc. (Abstracts), vol. 52, p. 796 (2004 July/
Aug.), convention paper 6055.
[25] E. Corteel, Equalization in an Extended Area Us-
ing Multichannel Inversion and Wave Field Synthesis, J.
Audio Eng. Soc., vol. 54, pp. 11401161 (2006 Dec.).
[26] S. Spors, A. Kuntz, and R. Rabenstein, An Ap-
proach to Listening Room Compensation with Wave Field
Synthesis, presented at the AES 24th International Con-
ference, Banff, Canada (2003).
[27] S. Spors, M. Renk, and R. Rabenstein, Limiting
Effects of Active Room Compensation Using Wave Field
Synthesis, presented at the 118th Convention of the Au-
dio Engineering Society, J. Audio Eng. Soc. (Abstracts),
vol. 53, p. 681 (2005 July/Aug.), convention paper 6400.
[28] T. Betlehem and T. S. Abhayapala, Theory and
Design of Sound Field Reproduction in Reverberant
Rooms, J. Acoust. Soc. Am., vol. 117, pp. 21002111
(2005).
[29] L. Fuster, J. J. Lopez, A. Gonzalez, and P. D. Zuc-
carello, Room Compensation Using Multichannel Inverse
Filters for Wavefield Synthesis Systems, presented at the
118th Convention of the Audio Engineering Society, J.
Audio Eng. Soc. (Abstracts), vol. 53, p. 686 (2005 July/
Aug.), convention paper 6401.
[30] D. de Vries, A. J. Reijnen, and M. A. Schonewille,
The Wave-Field Synthesis Concept Applied to Genera-
tion of Reflections and Reverberation, presented at the
96th Convention of the Audio Engineering Society, J. Au-
dio Eng. Soc. (Abstracts), vol. 42, p. 396 (1994 May),
preprint 3813.
[31] D. de Vries, E. W. Start, and V. G. Valstar, The
Wave-Field Synthesis Concept Applied to Sound Rein-
forcement: Restrictions and Solutions, presented at the
96th Convention of the Audio Engineering Society, J. Au-
dio Eng. Soc. (Abstracts), vol. 42, p. 396 (1994 May),
preprint 3812.
[32] H. Buchner, S. Spors, and W. Kellermann, Full-
Duplex Systems for Sound Field Recording and Auraliza-
tion Based on Wave Field Synthesis, presented at the
116th Convention of the Audio Engineering Society, J.
Audio Eng. Soc. (Abstracts), vol. 52, p. 812 (2004 July/
Aug.), convention paper 6120.
[33] T. Bravo and S. J. Elliott, Variability of Low Fre-
quency Sound Transmission Measurements, J. Acoust.
Soc. Am., vol. 115, pp. 29862997 (2004).
[34] N. Epain, E. Friot, and G. Rabau, Indoor Sonic
Boom Reproduction Using ANC, in Proc. Active 2004
(2004).
[35] M. Keller, A. Roure, and F. Marrot, Acoustic
Field Reproduction for Psychoacoustic Experiments: Ap-
plication to Aircraft Interior Noise, Proc. Active 2006
(2006).
[36] E. W. Stewart, D. de Vries, and A. J. Berkhout,
Wave Field Synthesis Operators for Bent Line Arrays in
a 3D Space, Acustica/Acta Acustica, vol. 85, pp. 883892
(1999).
[37] E. G. Williams, Fourier AcousticsSound Radia-
tion and Nearfield Acoustical Holography (Academic
Press, London, 1999).
[38] A. Asano and D. C. Swanson, Sound Equalization
in Enclosures Using Modal Reconstruction, J. Acoust.
Soc. Am., vol. 98, pp. 20622069 (2002).
[39] A. D. Pierce, Acoustics: An Introduction to Its
Physical Principles and Applications (Acoustical Society
of America, Woodbury, NY, 1991).
[40] C. Hansen, Rank-Deficient and Discrete Ill-Posed
Problems (SIAM, Philadelphia, PA, 1998).
[41] P. A. Nelson and S. J. Elliott, Active Control of
Sound (Academic Press, London, 1992).
[42] G. H. Golub and C. F. van Loan, Matrix Compu-
tations (Johns Hopkins University Press, Baltimore, MD,
1996).
[43] P. Lancaster and M. Tismenetsky, The Theory of
Matrices (Academic Press, Orlando, FL, 1985).
[44] S. Elliott, Signal Processing for Active Control
(Academic Press, London, 2001).
[45] M. R. Bai and S. J. Elliott, Preconditioning Mul-
tichannel Adaptive Filtering Algorithms Using EVD- and
SVD-Based Signal Prewhitening and System Decou-
pling, J. Sound Vib., vol. 270, pp. 639655 (2004).
[46] S. J. Elliott, Optimal Controllers and Adaptive
Controllers for Multichannel Feedforward Control of Sto-
chastic Disturbances, IEEE Trans. Signal Process., vol.
48, pp. 10531060 (2000).
[47] R. H. Cabell and C. R. Fuller, A Principal Com-
ponent Algorithm for Feedforward Active Noise and Vi-
bration Control, J. Sound Vib., vol. 227, pp. 159181
(1999).
[48] R. H. Cabell, D. Palumbo, and J. Vipperman, A
Principal Component Feedforward Algorithm for Active
Noise Control: Flight Test Results, IEEE Trans. Contr.
Sys. Technol., vol. 9, pp. 7683 (2001).
[49] J. Perisse, F. Magand, M. Henry, S. Chouteau, and
N. Leclerc, SD-Lab: An Interactive Sound Design Simu-
lator for Vehicle Interior Noise, Acta Acustica, vol. 89,
suppl., p. S90 (2003).
[50] K. Genuit and W. R. Bray, Prediction of Sound
and Vibration in a Virtual Automobile, J. Sound Vib., vol.
36, no. 7, pp. 1219 (2002).
[51] J. Chevret, R. Maier, J. M. Nogues, J. Perisse, and
C. Thirard, Towards Virtual Sound for Aircraft Simula-
tion, Acoust. Bull., vol. 31, no 1, pp. 2527 (2006).
[52] S. A. Rizzi and B. M. Sullivan, Synthesis of Vir-
tual Environments for Aircraft Community Noise Im-
pact Studies, in Collection of Technical Papers11th
AIAA/CEAS Aeroacoustics Conf., vol. 4, pp. 22912306
(2005).
[53] R. Weber, I. Baumann, N. Freese, R. Kruse, and V.
Mellert, Psychoacoustic Analysis of Sound in the Cabin
of Passenger Aircrafts, Acta Acustica, vol. 89, suppl., p.
S10 (2003).
[54] Y. He, A. Djordjevich, and L. Fuyuan, Sound
Rendering and Its Application in Virtual Cockpit, in
Proc. Int. Conf. on Signal Processing, vol. 2, pp. 1412
1415 (1998).
PAPERS ADAPTIVE WAVE FIELD SYNTHESIS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1123
[55] A. Vecchio, T. Polito, K. Janssens, and H. van der
Auwearaer, Real-Time Sound Quality Evaluation of Air-
craft Interior Noise, Acta Acustica, vol. 89, suppl., p. S5
(2003).
[56] M. Kleiner, B. I. Dalenback, and P. Svensson, Au-
ralizationAn Overview, J. Audio Eng. Soc., vol. 41, pp.
861875 (1993 Nov.).
[57] L. Savioja, J. Huopaniemi, T. Lokki, and R.
Vaanenen, Creating Interactive Virtual Acoustic Envi-
ronments, J. Audio Eng. Soc., vol. 47, pp. 675705 (1999
Sept.).
[58] C. Guastavino, B. F. G. Katz, J. D. Polack, D. J.
Levitin, and D. Dubois, Ecological Validity of Sound-
scape Reproduction, Acta Acustica/Acustica, vol. 91, pp.
333341 (2005).
[59] P. A. Gauthier, Synthe`se adaptative de champs
sonores, Ph.D. thesis, Universite de Sherbrooke, Sher-
brooke, Quebec, Canada (2007).
THE AUTHORS
P.-A. Gauthier A. Berry
Phillipe-Aubert Gauthier was born in 1976. From the
age of 14, for nearly 8 years, he played electric bass guitar
while learning jazz, improvisation, slapping, tapping, and
so on. He received a B.Ing. degree in mechanical engi-
neering from Universite Laval, Quebec City, Canada, in
2000 and an M.Sc. degree in 2003. In 2007 he received a
Ph.D. degree from Sherbrooke University, Quebec,
Canada. His thesis was dedicated to adaptive wave field
synthesis. In 2006 he received the Leonard de Vinci medal
from the Engineering Faculty of Sherbrooke University.
He is interested in spatial sound, multichannel sound re-
production, audio signal processing, room acoustics, loud-
speaker and microphone arrays, spatial hearing, psycho-
acoustics, and active control. At present he is involved as
a postdoctoral student in an industrial research project
(GAUS, CIRMMT) dedicated to sound environment
reproduction.
Dr. Gauthier is also an active sound artist mostly in-
volved in electroacoustic music, electronic music, genera-
tive music, sound installation, performance, and writing.
Several of his artistic projects were collaborations with
Tanya St-Pierre. He received a grant from Sherbrooke
City in 2006 for the realization of a sound installation for
the Pure-Data Convention 2007 in Montreal, Canada. His
artistic production has been presented in Canada, the
United States, and Europe, including at the Fine-Art Mu-
seum of Sherbrooke and at the Spark Festival (Weisman
Art Museum, Minneapolis). Since the beginning of his
graduate studies in sound and vibration, his artistic and
scientific activities have been linked through a multidis-
ciplinary approach.
G
Alain Berry graduated in aerospace engineering from
ENSICA, France, in 1985. He received M.Sc. and Ph.D.
degrees in acoustics from the Universite de Sherbrooke,
Que., Canada, in 1988 and 1991, respectively. After a post-
doctoral fellowship at INSA-Lyon, France, he was appointed
professor in the Department of Mechanical Engineering,
Universite de Sherbrooke, in 1992. He carries out his re-
search within GAUS (Groupe dAcoustique de lUniversite
de Sherbrooke), a research group focused on engineering
acoustics and vibrations. His research interests include struc-
tural acoustics, active noise and vibration control, vibration
and acoustic transducers, and source localization techniques.
Dr. Berry is the author of over 50 peer-reviewed pub-
lications in scientific journals. In recent years he has been
involved in a number of multi-institutional research projects
on acoustic comfort issues, mainly in the automotive and
aerospace sectors. Since 2007 he has been holding a senior
industrial chair in aviation acoustics in collaboration with
Bombardier Aerospace, Pratt & Whitney Canada, and Bell
Helicopters. He is an associate member of CIRMMT
(Centre for Interdisciplinary Research in Music Media and
Technology) centered at McGill University in Montreal.
Since 2002 he has been collaborating with CIRMMT re-
searchers on sound field reproduction techniques and the
development of novel loudspeaker concepts.
GAUTHIER AND BERRY PAPERS
J. Audio Eng. Soc., Vol. 55, No. 12, 2007 December 1124

You might also like