Vous êtes sur la page 1sur 13

ACTA ACUSTICA UNITED WITH ACUSTICA

Vol. 93 (2007) 824 836


Objective Evaluation of Room Effects on Wave
Field Synthesis
Philippe-Aubert Gauthier, Alain Berry
Groupe dAcoustique de lUniversit de Sherbrooke, Universit de Sherbrooke, 2500 boul. de lUniversit, Sher-
brooke, Qubec, Canada, J1K 2R1. philippe_aubert_gauthier@hotmail.com
Summary
This technical paper reports the objective evaluation of sound eld reproduction using wave eld synthesis (WFS)
in a listening room. WFS is an open-loop technology for spatial audio and it assumes a free eld as the repro-
duction space. The main objective of this experiment was to understand how much, and how, WFS performance
is reduced in-room situation in comparison with free-eld situation. These undesirable eects are characterized
by the coloration of the frequency response functions (FRFs) and the presence of echoes and reverberation in
the reproduced impulse responses of the WFS system. This paper only addresses the objective performance (fre-
quency response functions and wavefront shape; not the perceptual appreciation) of sound eld reproduction. On
that matter, this technical paper thus validates and complements other objective evaluations previously published
and performed with various WFS systems in dierent listening rooms. A comparative review of spatial aliasing
frequency denitions is also discussed in the context of sound eld reproduction.
PACS no. 43.38.Md, 43.60.Tj, 43.38.Ar
1. Introduction
Sound eld reproduction has applications in multiple do-
mains. The most commonly known is spatial audio where
one is interested by the articial reproduction of the natu-
ral spatial character of hearing. In this context, sound eld
reproduction corresponds to a physical approach which
can be divided in two subclasses: interior and exterior
problems of sound eld reproduction. The wave eld syn-
thesis (WFS) system considered in this applied paper em-
phasizes on the interior problem, i.e. reproducing a sound
eld over an extended region surrounded by acoustical
sources. The exterior problem is dened by sound eld
reproduction around acoustical sources. For more details
on this functional classication or spatial audio, see refer-
ences [1, 2, 3].
Sound eld reproduction also nds applications in ac-
tive control of noise (canceling a sound eld is equivalent
to reproducing it with a sign dierence), panel transmis-
sion loss measurements [4] (low-frequency diuse eld
reproduction in reverberant chambers), electroacoustical
device measurements [5] (diuse sound eld reproduction
in anechoic spaces), experimental acoustics and psychoa-
coustics [2, 6] and potentially more.
Wave eld synthesis (WFS) is a specic method of
sound eld reproduction which has been introduced for
audio applications [7, 8, 9, 10, 11, 12, 13]. One of the
Received 16 February 2006, revised 15 November 2006,
accepted 13 June 2007.
WFS assumptions is that the reproduction space is ane-
choic [11]. In a practical utilization of WFS, the listening
room, or the reproduction space, is not anechoic and there-
fore reduces, in objective terms, the quality of the repro-
duced sound eld [2, 14, 15, 16, 17].
This paper reports the measurements of various sound
elds reproduced by a WFS system in a studio. This in-
cludes frequency responses, room eects, direct wave-
fronts and global room eects. The main objective was to
understand how, and by which dominating eects, WFS
reproduced sound elds dier from the virtual sound elds
which have to be reproduced. This technical paper thus
validates and complements other WFS evaluations previ-
ously published [14, 15, 16, 18] and performed with dier-
ent WFS systems in various listening rooms. The results
presented herein can thus serve as a basis for compari-
son and to enlarge the available examples of in-situ WFS
objective evaluation. As originally pointed by Boone and
Verheijen [17], objective measurements (like multi-trace
impulse responses used in this paper) for the evaluation
of reproduced sound eld allow for later comparison be-
tween dierent setups and published experimental reports.
The work reported in this technical paper is motivated by a
preliminary study on sound eld reproduction using adap-
tive control and WFS [2, 19, 20] to compensate for the
undesirable room eects on WFS.
A similar evaluation of WFS systems has been reported
by Kutschbach [18] but his work focused on the denition
of a measurement method for the verication of spatial
sound systems. His method was proposed to circumvent
two potential drawbacks: (1) the less precise sound eld
824 S. Hirzel Verlag EAA
Gauthier, Berry: Room effects on wave eld synthesis ACTA ACUSTICA UNITED WITH ACUSTICA
Vol. 93 (2007)
evaluation based on wave eld extrapolation techniques
(from a linear or circular array of microphones) and (2)
the time-consuming process of measurement over an en-
tire 2D microphone grid (which can include over than one
hundred measurement points) in the reproduction space.
Kutschbach performed measurements over an entire 2D
grid using 24 moving microphones using a 2-axis linear
stepper motor system which created powerful analysis and
visualization tools. In this paper, a more conventional and
simpler method was used by utilizing a xed linear micro-
phone array without eld extrapolation [17].
This work diers fromother published WFS evaluations
regarding the WFS system, the listening room and the re-
search intentions. In 1997, Start [14] focused on the objec-
tive and subjective evaluations of WFS systems for sound
enhancement (or sound reinforcement) in concert halls.
The tested systems were typically front-oriented (linear
or convexly bent arrays) and placed on a stage. In agree-
ment with the sound reinforcement intention, the physical
evaluation was based on the comparison between repro-
duced sound eld and the measured virtual sound eld ob-
tained when a real acoustic source was placed at the vir-
tual source position, on stage. These reported experiments
showed that WFS is able to reproduce the virtual source
direct sound eld in a large space such as concert halls
and in an anechoic chamber. Boone and Verheijen [17] re-
ported objective WFS evaluation methods for a rectangular
WFS system surrounding the listening region. The multi-
trace impulse responses showed that the virtual source di-
rect sound eld is eectively reproduced by WFS. In 2003,
Klehs and Sporer [15] published solely subjective evalu-
ations of a modied WFS system in a living room. The
modications were evoked for practical reasons. First, the
loudspeaker array did not encircle the listening region en-
tirely since large gaps were needed for two doors and the
loudspeaker were in proximity of the walls. Also, since
the number of channels was limited, the independent loud-
speaker spacing was dierent on the front (0.17 m spac-
ing) compared to the back and to the sides (0.17 m, but
two adjacent speakers reproduced the same signal). The
experiments took place in a small roomwith a oor surface
of approximately 25.4 m
2
. Various parameters were tested
and evaluated by the researchers, including the special and
a practical conguration, loudspeaker spacing, etc. From
these subjective evaluations, it was shown that WFS can
tolerate some practical compromises without considerable
sound quality degradation. Similar experiments were later
conducted in a movie theater [16] where the listening room
was larger (with a oor surface of approximately 96 m
2
)
and the tested WFS system was made of a frontal linear
loudspeaker array (roughly 6.6 m wide with a loudspeaker
spacing of 0.17 m). In agreement with previous evalua-
tions [15], it was shown that WFS can tolerate practical ap-
proximations without signicant sound quality reductions.
In this objective evaluation of room eect on WFS, the
WFS system completely encircles the listening area with
a uniform loudspeaker spacing and the listening room,
which is a small studio, has a oor surface of roughly
31.5 m
2
. This setup is comparable to the living room used
in Klehs subjective evaluations [15]. The objective eval-
uation presented herein complements the existing WFS
evaluations. As it will be shown, WFS eectively repro-
duces the direct sound eld of the virtual source, but the
room eects causes serious colorations and alteration on
the reproduced wave eld. Several subjective experiments
with this WFS system in the same listening room were
reported by Usher et al. [21]. The experiments discussed
in this technical paper thus complete Ushers experiments
with an objective perspective.
Room eect, as evaluated in this paper, is an impor-
tant issue for the practical development and the future
of WFS: it contributes to the understanding that listen-
ing rooms have noticeable eects on objective physical
parameters and on subjective perception (sound quality,
sound localization, etc.) for audio commercial WFS ap-
plications. It is then possible to determine whether or
not WFS needs specically acoustically designed listen-
ing rooms. In recent research activities on WFS, room
eect is also relevant in relation to room compensation
[2, 20, 22, 23, 24, 25, 26, 27, 28]. The usefulness of ac-
tive room compensation in comparison with passive de-
sign methods of WFS listening rooms is currently being
debated, and is still unresolved.
A general review of WFS is presented in section 2, the
experimental setup and procedure are described in sections
3 and 4 while the results are reported in section 5. A dis-
cussion summarizes the important observations on WFS
physical performance in room and adresses potential mod-
ications of WFS to improve sound eld reproduction.
2. Wave Field Synthesis (WFS)
WFS has been introduced by Berkhout in the late 80s
[7, 8, 9, 10, 11, 12]. The underlying theory comes from
the Huygens construction principle which states that a
given wave eld, produced by a primary source, at a given
time, can be reconstructed, at a later time, by replac-
ing a given wavefront by a continuous set of secondary
sources on the initial wavefront. The general WFS concept
is depicted in Figure 1. This reconstruction idea is math-
ematically expressed and generalized by the Kirchho-
Helmholtz integral from which the basics of WFS are de-
rived [7, 8, 9, 10, 11, 12]. Practically speaking, WFS uses
this integral formulation along with simplications to de-
ne inputs (as a function of both reproduction sources cor-
responding to secondary sources, and virtual source, cor-
responding to primary source, positions) to a loudspeaker
array. The virtual wave eld is dened by virtual sources
(spherical waves, plane waves, etc.) in a free-eld virtual
space (see Figure 4). In its common form, WFS is an open-
loop system which is theoretically valid for a free-eld re-
production space. Such an assumption is not applicable
to common listening environments such as studios, the-
aters or living rooms including a real audio system. Real
applications typically include reproduction errors caused
by the system limitations (coloration, nite size, etc.) [29]
825
ACTA ACUSTICA UNITED WITH ACUSTICA Gauthier, Berry: Room effects on wave eld synthesis
Vol. 93 (2007)
Figure 1. Symbol denition for the derivation of the WFS opera-
tors. The virtual source is located at x
o
. The reproduction source
l is located at x
l
. x
(ref)
describes points which belong to the
reference line. x describes any eld or measurement point. L
is the reproduction source line, the virtual source is on the left
of the source line and the reproduction space is on the right of
the source line. All sources and sensors are located on the x
1
x
2
plane.
and by the reproduction room [24]. However, from sub-
jective and perceptive arguments, this free-eld simplica-
tion can be partly justied [30]. This paper focuses on the
physical measurements of the reproduced sound eld in a
real reproduction space using a real system. As noted ear-
lier, objective (physically valid) reproduction and evalua-
tion are still fundamentally important to understand how to
increase the physical WFS reproduction quality in rooms
[2, 20].
2.1. Derivation of the WFS operators
A more detailed and technical description will be intro-
duced for the WFS denition seen in Figure 1. The pri-
mary source is located at x
o
, the secondary source (acting
like a monopole) l is at x
l
while the measurement micro-
phones are located at x. The virtual wave eld is dened
as a primary monopole pressure eld in the frequency do-
main: p( x, ) = A()e
jk| x x
o
|
/| x x
o
| where [rad/s] is
the radial frequency, A() [Pa m] is the monopole ampli-
tude and k [rad/m] is the wave number [31]. Note that the
time convention is e
jt
for the complex variables. The
WFS operators, expressed in terms of secondary source
monopole amplitudes, are then dened as follows [11]
Q
W FS
( x
l
, ) = A()j

jk
2
cos (1)

e
jkr
o

r
o

r
(ref)
/(r
(ref)
+ r
o
)
l
,
where [rad] is the angle between the primary source
and the normal to the reproduction line at the secondary
source position x
l
, r
o
= | x
o
x
l
| is the distance [m]
between the primary source (in x
o
) and the secondary
source (in x
l
) and r
(ref)
= | x
(ref)
x
l
| is the distance
[m] between the secondary source and the reference line
along the line r
o
. In equation (1),
l
[m] is the secondary
source (loudspeaker) separation (
l
= | x
l
x
l+1
|). Equa-
tion (1) expresses the WFS monopole source amplitude
Q
W FS
( x
l
, ) to reproduce p( x, ) as dened earlier. In
this paper, capital letters are typically used for monopole
source amplitude (A and Q). For a given primary source
position, equation (1) gives the monopole amplitudes for
all the secondary sources. However, not all the secondary
source needs to be active. In other words the secondary
source l is active (Q
W FS
( x
l
, ) = 0) if || < 90 degrees.
The reproduced sound pressure in space is denoted
p
(rep)
( x, ). For a total of Lsecondary sources in free eld,
one nds that
p
(rep)
( x, ) =
L

l=1
Q
W FS
( x
l
, )e
jkr
/r, (2)
where e
jkr
/r represents the acoustical radiation of a sec-
ondary source and r is the distance between the secondary
source x
l
and the eld point x so that r = | x x
l
|. As one
might expect, the sound radiation of secondary sources
will in reality be aected by factors such as loudspeaker
response, as well as directivity and room response. These
eects are the focus of this paper.
Note that the interest is in the reproduced impulse re-
sponses (IR) and frequency response functions (FRF)
from the primary source to the measurement points:
h( x, ) = p
(rep)
( x, )/A().
The theoretical reproduced IRs and FRFs units are then
[Pa / Pa m] [1/m]. Theoretically reproduced FRFs and
IRs in free eld were compared to measured FRFs and IRs
in room to separate the room eect from classical WFS
approximations.
2.2. Reference line
As shown in Figure 1, a reference line is needed for the
denition of the WFS operators in equation (1). The refer-
ence line corresponds to the positions where the reproduc-
tion error is zero (for theoretical free-eld situation), i.e.
where there is no magnitude and phase errors in the repro-
duced sound eld. Outside the reference line, magnitude
errors exist but phase errors are still zero. Several proposi-
tions for the choice of the reference line have been made
[11]: linear, circular or optimal [32]. Typically, the refer-
ence line passes through the secondary source array center.
The secondary source array center is dened by the axis
origin (see Figures 1 and 4). The linear reference line is
perpendicular to the line between the primary source and
the secondary source array center. An example of linear
reference line is shown in Figure 4. For the WFS simula-
tions used as a basis for comparison, a linear reference line
was assumed.
2.3. Spatial sampling and spatial aliasing
Spatial sampling of a continuous source distribution, as
introduced in equation (1), by a set of discrete secondary
sources can create spatial aliasing if the spatial sampling
826
Gauthier, Berry: Room effects on wave eld synthesis ACTA ACUSTICA UNITED WITH ACUSTICA
Vol. 93 (2007)
is not dense enough. For most WFS systems, spatial sam-
pling is limited by the loudspeaker size which dictates
the smallest secondary sources separation distance. Above
the spatial aliasing (SA) frequency, solely depending on
the secondary source array conguration and the primary
source position, WFS sound eld reproduction fails from
an objective perspective since SA artefacts appear. These
artefacts are supplementary waves which do not corre-
spond to the virtual wave eld produced by the primary
source. Although the physical validity of the WFS fails
above this SA frequency, the SA artefacts are rarely au-
dible so that WFS is typically used for the entire audible
bandwidth. In fact, the SA artefacts audibility is nearly in-
existant: that is the artefacts are not localized by human
hearing. They are however typically perceived as a space-
dependent coloration of the reproduced signals which does
not reduce the WFS major benets [33]. In all reported
cases, and on the basis of personal experiences, these ef-
fects do not reduce the spatial quality of the WFS experi-
ence. However, audibility of SA artefacts is still a subject
of debate.
A review of the main publications on SA will now in-
troduced to select a criterion for SA within the context of
this WFS objective performance evaluation. For the pur-
pose of this discussion, the virtual wave eld, created by
the primary source, is assumed to be a plane wave.
Although the existence of SA is only dened by the sec-
ondary and primary sources positions, it is possible to con-
sider the SA eects from two dierent perspectives: (1) as
a spatial sampling of the WFS driving function (see equa-
tion 1) so that the sound eld reproduction is achieved with
a discrete set of secondary source positions in x
l
with l = 1
to L [34, 35] or (2) as an irregular time-domain sampling
at a listener position [36]. In that latter case, the irregu-
lar time-domain sampling arises fromthe discrete irregular
time arrival of each of the secondary source contributions
as function of the listener position in relation to the sec-
ondary source array. The latter case is more similar to a
denition of a SA criterion for the audibility of SA arte-
facts.
Originally, and according to the rst perspective, the
WFS SA frequency criterion was strictly dened by the
most simple and most severe rule: WFS requires at least
two secondary sources per wavelength so that the SA
strictly do not exist below f
#1
SA
= c/(2
l
) [11]. This crite-
rion is strict because it is true for all primary source po-
sitions (or all virtual plane wave incidences). Not long
after, a less severe criterion was introduced: at least two
secondary sources per trace wavelength along the sec-
ondary source array [34, 35] for the smallest trace wave-
length along the array, so that this SA frequency crite-
rion for WFS is also a function of the propagation direc-
tion of the virtual plane wave in relation to the secondary
source array geometry. Note that this second proposition
is a strict space-domain application of the time-domain
Shannon criterion. In that case, one nds that, for plane
wave reproduction, f
#2
SA
= c/(2
l
sin
max
) where is
dened in Figure 1 and
max
represents the maximum an-
gle of the plane wave components of the virtual wave eld.
A problem arises if a plane wave impinges on the sec-
ondary source array with a perpendicular incidence (that
is =
max
= 0), one nds a SA frequency criterion
equal to innity, which is not realistic: SA occurs far be-
low f
#2
SA
= [33]. In accordance with more recent pa-
pers on SA and WFS [36, 33], this second SA frequency
criterion is too optimistic. Another recent paper on spa-
tial sampling of the WFS driving function by Spors [33]
proposes a SA frequency criterion that is more optimistic
than the rst criterion but more rigorous than the second
criterion. This criterion by Spors [33] is in fact a strict def-
inition of the spatial aliasing frequency above which SA
artefacts appear. The value of the SA frequency proposed
by Spors is f
#3
SA
= c/(
l
(1 + | sin()|). For a grazing in-
cidence plane wave, = /2, one recovers the most
severe criterion f
#3
SA
= c/(2
l
) = f
#1
SA
. For a perpen-
dicular incidence = 0, the SA frequency criterion is
f
#3
SA
= c/(
l
) = 2f
#1
SA
. Note that all three criteria are in-
dependent of the listeners position and they only depend
on the virtual sound eld and secondary sources arrange-
ment.
According to the second perspective (irregular time-
domain sampling at the listeners position), another deriva-
tion of the SA frequency criterion is suggested by Corteel
[36, 29]. This SA frequency criterion depends on the lis-
teners position. This new perspective does not necessar-
ily imply that SA does not exist below the corresponding
SA frequency. Rather, it suggests that SA artefacts are not
present at the listeners (or measurements) position. In-
terestingly, the SA frequency observed in Corteels results
[36] is in the range of the Spors predictions [33]. There-
fore, both rules seem in accordance with the exception that
one favors the sources perspective while the other favors
the listening positions perspective in the denition of a
SA criterion for WFS.
Several examples of harmonic plane wave reproduction
by WFS will now introduced to illustrate the SA artefacts
and validate the choice of the SA frequency criterion for a
given WFS system. Figure 2 shows the secondary source
array: a two-pieces linear array of 32 monopole sources in
a free eld. One part of the array is along x
1
and the other
is oblique. The virtual sound eld, produced by a primary
source, is a plane harmonic wavefront (in the x
1
x
2
plane)
propagating in the negative x
1
and x
2
directions (that is to-
ward the lower left corner of the gures). The secondary
sources are separated by 0.175 m. A spatial window (half-
Hanning) is applied to reduce the truncations eects [35]
caused by the nite size of the secondary source array. Ac-
cordingly, the four sources at each end of the nite array
are of reduced amplitudes to minimize these truncation ef-
fects. Also, since the complete reproduced sound eld is
the superposition of each secondary source, it is possible
to separate the contribution of the two linear segments of
the secondary source array, so that it is possible to observe
the SA contribution of each of these segments. Even if
the complete array is a piecewise nite-length linear array,
each part of the array is perfectly linear and will, in rela-
827
ACTA ACUSTICA UNITED WITH ACUSTICA Gauthier, Berry: Room effects on wave eld synthesis
Vol. 93 (2007)
Figure 2. Simulation of plane wave reproduction by WFS in free
eld at 900 Hz (left) and 1100 Hz (right). The secondary sources
are marked with white dots. The listening region is delineated by
the secondary source array and the white lines. The reproduced
sound elds are divided in two portions: sound elds from the
horizontal portion of the secondary source array [shown in (a)
and (d)] and sound elds from the oblique portion of the sec-
ondary source array [shown in (b) and (e)]. The complete sound
eld reproductions are shown in (c) [superposition of (a) and (b)]
and (f) [superposition of (d) and (e)].
tion with the corresponding angles, display dierent SA
frequencies consistent with the linear array theory of SA.
In Figure 2, the sound pressure amplitudes are arbitrarily
selected for illustration purposes.
For this array, with a secondary sources separation of
17.5 cm, the rst and severe aliasing criterion gives f
#1
SA
=
945.7 Hz assuming a sound speed of 331 m/s. As shown on
the left side of Figure 2 for 900 Hz, none of the two parts
of the array create SA so that the resulting wave eld (Fig-
ure 2c) is eectively a plane wavefront along negative x
1
and negative x
2
. In this case, WFS is physically eective.
At 1.1 kHz, SA starts to appear for the horizontal portion
of the array, as shown in Figure 2d. Note that the oblique
portion of the array (Figure 2e) does not create SA. This
is in perfect agreement with Spors [33] prediction of SA
frequency, where the SA frequency of the oblique part of
the array (with = 0) is f
#3
SA
= 1891.4 Hz. Typically, SA
artefacts appear as one or more additional beams of plane
wavefronts with a propagation direction dierent from the
virtual one. As shown in this gure, and as noted by Spors
[33], the width of the supplementary beams depends on
the aperture of the linear secondary source array. This also
dictates if the supplementary beams will reach the listeners
depending on their positions.
Figure 3. Simulation of plane wave reproduction by WFS in
free eld at 1300 Hz (left) and 2400 Hz (right). The secondary
sources are marked with white dots. The listening region is delin-
eated by the secondary source array and the white lines. The re-
produced sound elds are divided in two parts: sound elds from
the horizontal portion of the secondary source array [shown in
(a) and (d)] and sound elds from the oblique portion of the sec-
ondary source array [shown in (b) and (e)]. The complete sound
eld reproductions are shown in (c) [superposition of (a) and (b)]
and (f) [superposition of (d) and (e)].
Two other examples are given in Figure 3 for 1.3 kHz
and 2.4 kHz. At 1.3 kHz, only the horizontal portion of the
array creates SA. In comparison with the 1.1 kHz case, the
additional aliased beam introduces more energy in the lis-
tening region, as predicted by the SA analysis of Spors
[33]. WFS is then physically less eective since SA arte-
facts now contaminate the listening region. Note that the
oblique portion of the array does not create SA, as pre-
dicted by the value f
#3
SA
= 1891.4 Hz. The last example is
shown in Figure 3d to 3f for 2.4 kHz which is above all
the predicted SA frequencies except the second criterion
which predicts f
#2
SA
= with = 0 [34, 35]. Clearly
the two parts of the array create SA artefacts which ap-
pear as two additional undesirable beams of plane wave-
fronts for each portion of the array. This invalidates the
second criterion for the SA frequency but supports the
third one [33]. According to these free-eld simulation ex-
amples, SA starts to occur between f
#1
SA
= c/(2
l
) and
f
#3
SA
= c/(
l
(1 + | sin()|) with = 0.
At the beginning of this section, a second perspective
for the consideration of SA frequency criterion was de-
scribed [36] and is based on the listening positions. Al-
though the presented examples were not discussed in re-
lation to the listening position, it is worth noting that the
828
Gauthier, Berry: Room effects on wave eld synthesis ACTA ACUSTICA UNITED WITH ACUSTICA
Vol. 93 (2007)
examples show how, depending on the chosen perspective,
the SA phenomenon is interpreted: existence of additional
aliased beams of plane wavefronts or existence of addi-
tional aliased components (created by the SA beams) at
the listening positions.
In the case of WFS in a room, the relation between the
path direction of the additional aliased beams and the lis-
tening positions are less clear than for the reported free-
eld simulations [33]. Indeed, some aliased beams can
reach the listening region by reection or diraction from
surfaces and objects without any direct propagation. Ac-
cordingly, the SA frequency criterion which is used for the
following objective evaluation sticks to the most severe:
f
#1
SA
= c/(2
l
) below which, strictly no SA artefacts exist
for any primary source position, and above which some SA
artefacts might exist depending on the virtual sound eld
in relation to the secondary source array geometry. This
criterion, f
#1
SA
= c/(2
l
), can be described as the mini-
mal possible SA frequency for a given secondary source
array for any primary source type or position. Moreover,
any existing SA artefact would pollute the objective eval-
uation of room eects on WFS at the microphone array
since it would include room eects on WFS, which is the
main concern of this evaluation, and room eect on SA,
which is not addressed in this paper. Also, as any WFS
system (except with some modications like spatially l-
tered WFS [34]) might be used by various users to create a
plethora of dierent virtual sound elds, it is indeed risky
to state that the SA frequency criterion could be higher
than f
#1
SA
= c/(2
l
). For all of these reasons, it was de-
cided that it is best to adhere to the worst case scenario
for the denition of the SA frequency criterion. Note that
this denition corresponds to the classical spatial sampling
theorem from array theory [37].
3. Experimental setup
The experiments were performed with a WFS system
built by Fraunhofers Institute for Digital Media Technol-
ogy [38]. The system included 88 two-ways loudspeakers
mounted on 11 at units of 8 loudspeakers each. The loud-
speakers and microphones conguration are shown in Fig-
ure 4 and a photograph of the system is shown in Figure 5.
The secondary source array approximately forms a circle
in the horizontal plane (at a height of 1.22 m above the
oor) with a radius of about 2.2 m. The secondary source
array center is dened by the x
1
x
2
origin in Figure 4.
The WFS system [11, 38] is based on: (1) A reference
line dened by a line passing through the center of the
secondary source array and perpendicular to a line from
the primary source to the center of the secondary source
array (see Figure 4 for an example) and (2) a spatial win-
dow (half-Hanning) to progressively reduce the secondary
source amplitudes from the active secondary sources to
the non-active secondary sources [11]. The reproduction
room is located in the Redpath Hall (McGill University,
Montral, Canada) basement. The room is schematically
shown in Figure 6. Room partitions are made of 1.27 cm
Figure 4. The 88 loudspeakers of the WFS system (shown as
black squares) and the 8 microphones used in sound eld mea-
surements (shown as circles). O: The 6 dierent primary source
positions in the experiments, (a) to (c) being those described in
this paper. The reference line is shown as a dash-dot line for the
primary source (b).
Figure 5. Photograph of the WFS experiments including the front
loudspeaker array (four visible 8-loudspeaker units), the com-
puter interface and the microphone array.
(1/2 inch) plaster on brick walls and acoustical curtains
cover the whole surface of the walls (see Figures 5 and 6).
There is a suspended ceiling above which there is approxi-
mately 30 cm of compressed mineral wool-like material
for sound and thermal isolation while the concrete oor is
covered by a thin commercial carpet. The room dimen-
sions are 5.2 m6.05 m2 m (the 2 m height is the dis-
tance between the oor and the suspended ceiling), and
can be considered as a medium-sized listening room. The
background sound pressure level was estimated to be 47
[dB ref 2 10
5
] and the typical sound reproduction level
was estimated to be 78 [dB ref 2 10
5
] between 100 and
1000 Hz. The main noise sources were outdoor vehicles
and water pipes above the room.
Each 8-loudspeakers unit included ADAT optical input,
digital-analog converters and power ampliers while the
829
ACTA ACUSTICA UNITED WITH ACUSTICA Gauthier, Berry: Room effects on wave eld synthesis
Vol. 93 (2007)
Figure 6. Room geometry and relative WFS system position.
rendering system included four computers. The WFS sys-
tem also included a subwoofer channel for sound repro-
duction below the panel cut-on frequency. For these ex-
periments, the subwoofers were turned o and the analog
subwoofer channel (which is an unltered version of the
virtual source wave le) was used as the reference input
for FRFs and IRs identication. Since the loudspeakers
were separated by 17.5 cm, the WFS spatial aliasing fre-
quency was found to be 945.7 Hz assuming a sound speed
of 331 m/s and at least two reproduction (or secondary)
sources per wavelength to avoid spatial aliasing [11] (see
section 2.3). Therefore, the reference signal, also used to
feed the virtual source, was limited to 0 1 kHz (3 dB
cut-o point of a 12-order Butterworth lter). This fre-
quency limitation simply stems from the fact that we are
exclusively concerned with the eective WFS (below the
WFS spatial aliasing frequency) reproduction quality and
not with the entire audio-bandwidth quality.
Sound pressure was measured using a linear micro-
phone array (shown in Figures 4 and 5). The array in-
cluded 8 TMS microphones (model 130M01 with 130P10
preampliers) separated by 17.5 cm. For the sound eld
reproduction measurements, the linear array was placed
in the center of the loudspeaker array at the same el-
evation, i.e. 121.92 cm (48 inches), as the reproduction
sources (Figure 4). ICP conditioners (two 4-channels PCB
442B104) were used to store the microphones signals on a
DAT recorder (SONY PC216A) from which the data was
later exported and analysed. The microphones were cali-
brated for amplitude using a 1 kHz sound level calibrator.
4. Experimental and analysis procedure
The experimental setup and post-processing analysis pro-
cedure are both schematically shown in Figure 7. The
reference signal, also used to feed the primary source,
Figure 7. Schematic representation of the experimental setup and
the impulse response extraction.
was uncorrelated white noise low-pass ltered at 1 kHz to
avoid SA.
The microphone outputs and reference signal were
stored with a DAT recorder (a SONYPC216A, with a vari-
able sampling rate, set to sample data at 6 kHz and with
an anti-aliasing lter correspondingly adjusted to 2.5 kHz)
for later post-processing.
Before impulse response identication in the post-pro-
ssing operation, the measured pressures and the reference
signal were high-pass ltered above 100 Hz. This high-
pass ltering was used to remove uncorrelated measure-
ment noise (mainly coming from exterior vehicle trac)
due to the fact that the two-ways loudspeakers were not
eective at lower audible frequencies. Impulse responses,
between the reference signal and the measured pressures,
were then identied using an adaptive LMS algorithm
[39, 40]. The adaptive modeling proceeded on for approxi-
mately 6 minutes of data (2, 160, 000 samples). Validation
tests were performed and proved the validity of the result-
ing identications: these tests showed that the identica-
tions can predict a set of modeled pressures that matches
the measured pressures of the real system when using a
measured reference signal sequence, which has not been
utilized in the adaptive identication process [41]. The
adaptive LMS algorithm was used as an iterative identi-
cation method since this type of identication algorithm
is already included for on-line identication in the adap-
tive wave eld synthesis (AWFS) system [19]. Along with
other standard identication methods such as maximum-
length-sequence (MLS) or sweep sines, adaptive identi-
cation can typically converge towards very similar results
830
Gauthier, Berry: Room effects on wave eld synthesis ACTA ACUSTICA UNITED WITH ACUSTICA
Vol. 93 (2007)
obtained from the two others mentioned. This is a mat-
ter of convergence time, adaptation coecient, MLS se-
quence length, number of averages, etc.
Once the adaptive identication had converged, the re-
sulting impulse responses were low-pass ltered below
1 kHz. Thus, the frequency range of the system response
become strictly limited to a 1001000 Hz bandwidth. The
frequency response functions were obtained by Fourier
transformation from the identied impulse responses.
Since the identied system input is the analog refer-
ence signal [V] and the identied systemoutputs are sound
pressure [Pa], the impulse responses and frequency re-
sponses are expressed as sound pressure per volt [Pa/V].
All the following experimental results are based on the
aforementioned analysis procedure.
5. Experimental results
The experiments were performed for six primary source
positions. However, only three are presented here. These
positions (a), (b) and (c), shown in Figure 4, and were cho-
sen to create dierent incidence angles on the secondary
source and microphone arrays, and to vary the number of
active secondary sources (more active secondary sources
correspond to position c). The experiments focused on
frontal positions of the primary sources as this corresponds
to the most typical positions for primary sources.
The results are presented in two sections: the rst is ded-
icated to reproduced FRFs showing frequency coloration
by the room, the second shows the reproduced IRs to illus-
trate the room eects in terms of reections and wavefront
passages at the microphone array. As it will be shown, both
coloration and reections explain most of the discrepan-
cies between theoretical and experimental reproduced IRs
and FRFs by WFS.
5.1. Measured WFS frequency response functions
This section presents the FRFs between the reference sig-
nal and the microphones for three primary source posi-
tions (three dierent virtual wave elds). The objective is
to evaluate the room eects on the FRFs in comparison
with theoretical FRFs obtained from free-eld simulations
of WFS using the same conguration.
The rst reproduced sound eld is generated by a point
primary source located at x
1
= 0 m, x
2
= 4 m (position
(a) in Figure 4). Both the measurement [Pa/V] and the
simulation [1/m] gains are transformed in dB ref 1 gains.
The simulation gains are obtained with WFS simulations
(see reference [2] and section 2) in free eld and from
the division of the output sound pressures by the primary
monopole amplitude. As shown in Figure 8, the eight FRFs
measured by the microphone array display similar types
of responses. The various uctuations and dips that appear
above approximately 250 Hz are due to comb-ltering ef-
fects, destructive standing-wave interferences and possi-
bly nite aperture artefacts (diraction waves) produced
by the corners of the reproduction source array [11, 35].
Figure 8. Measured (thick line) and simulated (dashed line) WFS
FRFs gains [dB ref 1] for the primary source (a). Sensors #1 and
#8 are respectively the leftmost and rightmost sensors in Fig-
ure 4.
The corresponding FRFs dips are more signicant in the
frequency range above 250 Hz where the sound pressure
FRFs show a reduced spatial correlation, i.e. the FRFs
vary for each sensor. A reduced spatial correlation char-
acterizes a more diuse eld response and highlights the
destructive interference eect, which varies strongly as a
function of position. Below 250 Hz, the response seems to
be controlled by spatially correlated modal response. This
is mostly visible around 150 Hz where one possibly ob-
serves a strong room mode (or a group of modes, some-
times called a room formant [42]) resonant response. The
possible existence of damped standing waves (in the low
frequency limits) and comb-lter response (corresponding
to reection and diraction by objects and walls in the
higher frequency limits) suggests the need for WFS im-
provements in room situation (as already pointed by sev-
eral authors [2, 20, 28]). To support such observations,
Figure 8 also shows the comparison of the measured FRFs
with theoretical FRFs obtained from the free-eld WFS
simulations in the frequency domain for the same con-
guration. Since the measured and simulated FRFs units
are dierent, the free-eld simulation FRFs have been ad-
justed to t the measured data on average. Clearly, the
measured FRFs colorations are stronger than those of the
free-eld simulations, which are hardly visible on this g-
ure. The free-eld simulation colorations arise from vari-
ous eects including: nite aperture array (only a part of
the reproduction source array is active) and corner eects
831
ACTA ACUSTICA UNITED WITH ACUSTICA Gauthier, Berry: Room effects on wave eld synthesis
Vol. 93 (2007)
Figure 9. Gains [dB ref 1] of the measured (thick line) and sim-
ulated (dashed line) WFS FRFs for the primary source (b). Sen-
sors #1 and #8 are respectively the leftmost and rightmost sensors
in Figure 4.
[11, 35] (caused by piecewise linear secondary source ar-
rays). To illustrate this dierence and the soft coloration
of the free-eld FRFs, it is possible to evaluate the mean
(over the eight microphone positions) of the standard de-
viation of the FRFs gains between 100 and 1000 Hz for
both the experimental and theoretical cases. For the exper-
imental FRFs, the mean standard deviation is 5.7576 dB
(a variance of 33.1495) while for the theoretical free-eld
the mean standard evaluation is low as 0.3365 dB (a vari-
ance of 0.1132). This shows that the free-eld FRFs devi-
ation from the ideal at FRFs is small in comparison with
the room eect: room eects dominate the reproduction
errors. On that matter, one can see that any variation in
the WFS operators denition (approximations, position
of the reference line [32], spatial window [35], etc) would
not cause such large deviations as seen in the experimental
data.
Other experiments were conducted with ve dierent
primary source positions. Two of these measurements are
shown in Figures 9 and 10 for the primary source positions
(b) and (c) in Figure 4. Most of the comments presented
for the primary source position (a) apply for these two
other cases. Note that the dierence in the primary source
distances between (b) and (c) (see Figure 4) does not af-
fect the measured FRFs gains on average. This is simply
because the distance-dependent amplitude has not been
considered in these experiments (i.e. the loudness does not
Figure 10. Gains [dB ref 1] of the measured (thick line) and sim-
ulated (dashed line) WFS FRFs for the primary source (c). Sen-
sors #1 and #8 are respectively the leftmost and rightmost sensors
in Figure 4.
change with the distance of the primary source). The free-
eld WFS simulations include the distance-dependent am-
plitude, but have again been adjusted to t the measured
data on average. By comparing the FRFs for various pri-
mary source positions (Figures 8 to 10, primary monopole
source in positions (a), (b) and (c), respectively), one can
conclude that WFS FRFs departure from idealized ones is
mainly due to the electroacoustical system including the
loudspeakers, the furniture and the room response, and
much less to the primary source position or WFS specic
approximations.
This is supported by the fact that even if the FRFs
change with the primary source position, there is no clear
relation between the FRFs global trends and the primary
source position as it is for free-eld WFS simulations. That
is, in free-eld simulated situations, the WFS FRFs depar-
ture from the ideal primary source FRFs solely depends on
(1) frequency, (2) the primary source position in relation
with the secondary source array position and (3) the size of
the secondary source array. Some other eects like nite
loudspeaker array aperture and diraction from the cor-
ners of the secondary source array [11] can easily be lim-
ited using WFS modications such as spatial windowing.
Spatial windowing was used for both WFS simulations
and experiments. In all case, as shown by these three g-
ures, most of these free-eld colorations were dominated
by the prominent room response.
832
Gauthier, Berry: Room effects on wave eld synthesis ACTA ACUSTICA UNITED WITH ACUSTICA
Vol. 93 (2007)
Figure 11. Measured [Pa/V] (thick line) and simulated [1/m]
(dashed line) IRs for the primary source (a). Sensors #1 and #8
are respectively the leftmost and rightmost sensors in Figure 4.
Time of arrivals and amplitudes of the virtual eld are marked by
circles and they are connected by thick dotted line.
Again, by looking at the mean (along the microphone
positions) of the individual FRFs standard deviations (be-
tween 100 and 1000 Hz), it is possible to highlight the
insignicancy of the free-eld colorations in comparison
with the drastic room eect. For case (b), the experimen-
tal FRFs mean standard deviation is 5.3968 dB (variance
of 29.1259) while for the free-eld case, the FRFs mean
standard deviation is 0.3484 dB (variance of 0.1214). For
case (c), the experimental FRFs mean standard deviation
is 5.5827 dB (variance of 31.1669) while for the corre-
sponding free-eld case, the FRFs mean standard devia-
tion is 0.3502 dB (variance of 0.1227).
5.2. Measured WFS impulse responses
In this section, the multitrace IRs (impulse responses) will
be used to represent the directions of arrival and wavefront
curvatures of the reproduced sound eld that includes dis-
tinct sound reections.
Multitrace IRs are measured and analysed to evaluate
the geometry of the reproduced wavefronts. In this exper-
iment, the objective was to evaluate the room eect on
wavefront reconstruction by WFS. The IRs were obtained
from adaptive identication using the LMS algorithm with
a band-limited noise input reference as well as the micro-
phone outputs described in section 4.
The IRs are detailed in Figures 11 to 13 for the primary
source at positions (a), (b) and (c) respectively (see Fig-
ure 4 for the primary source positions). The arrival time
Figure 12. Measured [Pa/V] (thick line) and simulated [1/m]
(dashed line) IRs for the primary source (b). Sensors #1 and #8
are respectively the leftmost and rightmost sensors in Figure 4.
Time of arrivals and amplitudes of the virtual eld are marked by
circles and they are connected by thick dotted line.
and amplitude of the primary wavefronts are also shown
in these gures. This has once more been adjusted (global
amplitude and time delay) to t the measured data on av-
erage so that relative comparisons of amplitudes and de-
lays are possible. The free-eld WFS simulations are also
shown on these three gures in order to highlight the room
eects. The simulated WFS IRs were obtained by inverse
Discrete-Time Fourier Transform (DTFT) of the simulated
FRFs. Clearly, WFS produces a direct eld (the rst wave-
front that impinges the sensor array) which matches both
the free-eld WFS simulated reproduced eld and the su-
perimposed passage of the virtual eld. Here, this relation-
ship is noted in terms of relative delays and amplitudes
along the microphone array. It can thus be concluded that
direct eld reproduction by WFS is eective in rooms.
However, after the direct eld has reached the sensor ar-
ray, the reections on room walls are clearly visible, which
causes an important mismatch between the virtual wave
eld (or the free-eld WFS simulations) and the WFS re-
produced sound eld in a room.
Comparison of Figures 11 and 12 shows that the direc-
tion of the incident wave on the microphone array due to
the primary source angular position is properly achieved
by WFS. In Figure 11, all initial wavefronts arrive almost
simultaneously, corresponding to a normal incidence as
suggested by the relative positions of the primary source
(a) and the microphone array in Figure 4. By comparing
Figures 12 and 13, one can also observe that the change of
833
ACTA ACUSTICA UNITED WITH ACUSTICA Gauthier, Berry: Room effects on wave eld synthesis
Vol. 93 (2007)
Figure 13. Measured [Pa/V] (thick line) and simulated [1/m]
(dashed line) IRs for the primary source (c). Sensors #1 and #8
are respectively the leftmost and rightmost sensors in Figure 4.
Time of arrivals and amplitudes of the virtual eld are marked by
circles and they are connected by thick dotted line.
Figure 14. Measured IRs for the primary sources (a), (b) and (c).
Color scale on top; virtual positions (a), (b) and (c) in the middle;
time zoom for the virtual positions (b) and (c) in the bottom.
the virtual source distance (4 m to 12 m) is eectively re-
produced by a wavefront curvature which is larger in Fig-
ure 13 than in Figure 12. (This will be further explained
by Figure 14.)
In all the IR illustrations, one can see that the arrival
of the rst wavefront is preceded by growing (as time
increases) oscillations at 1 kHz. This is a signal process-
ing artefact which stems from the rectangular window l-
tering (low-pass at 1 kHz) of the IRs, in the frequency
domain, which creates a time-domain symmetrical band-
limited impulse (a sinc function with decaying oscillations
on both sides of the main impulse).
Figure 14 summarizes the results of Figures 11 to 13
in the time domain. The band-limited measured IRs are
plotted as a function of time and spatial position of the
microphone array for the three primary source positions.
The color scale contrast has been increased to enhance the
lower values of the IRs. The color scale [Pa/V] is shown at
the top of the gure. The bottom portion of the gure is a
time zoom around the arrival of the rst wavefront of the
IRs for primary positions (b) and (c). On this gure, the
arrivals of the virtual wave eld are also shown as dashed
lines for comparison purposes. This graphical representa-
tion, when compared with Figures 11 to 13, better illus-
trates the geometry of the reected wavefronts. The room
eects include strong reections (with wavefront curva-
tures similar to the virtual wave eld) and late diused
reverberation. The discrete early reection shapes, in rela-
tion to the virtual wave eld, are also visible for the repro-
duced sound elds shown in Figure 14.
According to these gures, WFS does not accomplish
objective (in physical terms) sound eld reproduction of
the virtual wave eld, except for the direct eld, which
approximately corresponds to the geometry of the virtual
wave eld created by the primary source.
6. Discussion
A physical interpretation of the experiments can be sum-
marized as follows. In terms of physical measurements,
the performance of WFS is aected by the presence of
the reproduction room which strongly colors FRFs and
introduces reections and reverberation in the IRs. Since
the virtual wave eld is generated by a primary monopole
source in a virtual free eld, the ideal FRFs have a at
frequency dependence and the corresponding IRs are sim-
ple band-limited impulses with a geometrical spreading in
space. The measurements clearly highlight the discrepan-
cies between this virtual eld and the reproduced FRFs
and IRs. On the other hand, the geometry of the direct
eld approaches the free-eld simulated WFS reproduc-
tion, which is itself similar to the virtual wave eld de-
nition. This includes wavefront curvature. The dierences
between the free-eld simulations and experiments high-
light potential technical improvements of WFS on a phys-
ical basis.
These results are in accordance with the WFS deni-
tion which relies on free-eld assumption for the repro-
duction space. According to the results presented in this
paper, correction of WFS response in room is needed to
increase the objective performance of sound eld repro-
duction with WFS. Since the WFS derivation from the
Kirchho-Helmholtz integral would be too dicult for
practical reproduction spaces (this would require an ac-
curate room model, leading to a very case-specic ap-
proach which would be unadaptable to adapt to varia-
834
Gauthier, Berry: Room effects on wave eld synthesis ACTA ACUSTICA UNITED WITH ACUSTICA
Vol. 93 (2007)
tions of the room characteristics in time), active com-
pensation using error sensors in the reproduction space
and adaptive signal processing is a promising research
area. This is the subject of current and recent researches
[2, 20, 22, 23, 24, 27, 43, 44, 45, 46, 47].
Although most of the previous sections described WFS
in terms of physically measurable quantities, the audio
applications of WFS address the human hearing system.
Therefore, a brief discussion relating to spatial hearing
perception and the measurements shown in this paper is
needed. This discussion mostly relates to the precedence
eect [30]. According to the precedence eect, human
sound localization in presence of a set of coherent wave-
fronts (in our case: the direct reproduced eld and the re-
ected and reverberated elds) uses the direction of ar-
rival of the rst wavefront - as long as the time separation
of the rst wavefront and the other coherent wavefronts is
less than the echo threshold time - to determine the local-
ization of the auditory event. This suggests that most of
the reected wavefronts in these experiments should not
inuence sound localization provided by the rst wave-
front (the direct eld which satisfactorily corresponds to
the virtual wavefront curvature) since the major WFS re-
ections (see Figure 14) appear before the echo threshold,
which is between 30 and 40ms for the two-channel stereo-
phonic conguration described by Blauert [30]. If this is
the case, most of the perceivable WFS objective perfor-
mance degradations caused by the room eect should be
the frequency-dependent colorations of the FRFs caused
by the rooms response and spatial localization should be
less inuenced. This should be veried by further exper-
imentation and suggests the need for frequency equaliza-
tion of WFS in rooms, specically at low frequency. On
this matter, one should note that the limited 1001000 Hz
bandwidth somehow limits the extent of the subjective ef-
fects interpretation.
7. Conclusion
In this technical paper, experiments on WFS sound eld
reproduction in rooms have been described as an objec-
tive evaluation of WFS performance in rooms. The results
have shown that WFS objective performance - described
with measured FRFs and IRs - is signicantly reduced in
comparison with free-eld WFS simulations and virtual
wave eld created by a primary source. These dierences,
entirely caused by the loudspeaker and room responses,
suggest the need for room compensation along with WFS
to increase the objective performance of the system. This
topic is the subject of current research using closed-loop
control and digital signal processing borrowed from ac-
tive noise control techniques [2, 20, 22, 23, 24, 27, 43,
44, 45, 46, 47]. The work presented in this technical paper
was a preliminary step towards what the researchers have
proposed as adaptive wave eld synthesis (AWFS) for
reproduction systems and room compensation with WFS
and active noise control (for more details, see reference
[20]). AWFS oers the possibility to control the amount of
room compensation. Current research activities on AWFS
have been devoted to signal processing and experimental
evaluations of AWFS versus WFS reproduced sound elds
in dierent reproduction rooms, hemi-anechoic chamber,
laboratory space and reverberant chamber. The results are
promising since AWFS eectively compensates the repro-
duction errors. This should be reported in upcoming re-
search papers. The results presented herein directly or in-
directly motivates further works on system limitations or
reproduction room compensation [24, 36]. The results also
revive the debate between room compensation or specic
room design for WFS. Aside from room compensation
based on dedicated signal processing and modication of
the classical WFS algorithms, one can imagine a reproduc-
tion room with considerable amounts of sound absorbing
material, that would induce reproduction errors to be dom-
inated by system limitations. This is an interesting simple
approach since typical WFS system are, and will be, used
in dedicated rooms. In such cases, direct sound eld equal-
ization for WFS [29] would be an ecient method to re-
duce the remaining WFS reproduction errors. The afore-
mentioned ideas are open issues for future development of
WFS.
Acknowledgements
This work has been supported by NSERC (Natural Sci-
ences and Engineering Research Council of Canada),
FQRNT (Fond Qubecois de la Recherche sur la Nature et
les Technologies), VRQ (Valorisation Recherche Qubec)
and Universit de Sherbrooke. The authors wish to ac-
knowledge the Institute for Digital Media Technology at
Fraunhofer in Ilmenau (Germany) for their technical sup-
port and for the lending of the WFS system to McGill Uni-
versity. This work has been conducted within CIRMMT
(Centre for Interdisciplinary Research in Music Media and
Technology, McGill University). The rst author wishes to
acknowledge John Usher from McGill University (CIR-
MMT) for his help and availability regarding the use of
the WFS system. The rst author wishes to acknowledge
Hugo Fourier for English language correction.
References
[1] F. Rumsey: Spatial audio. Focal, Oxford, 2001.
[2] P.-A. Gauthier, A. Berry, W. Woszczyk: Sound-eld repro-
duction in-room using optimal control techniques: Simula-
tions in the frequency domain. J. Acoust. Soc. Amer. 117
(2005) 662678.
[3] AES Sta writer: Multichannel audio systems and tech-
niques. Journal of the Audio Engineering Society 53
(2005) 329335.
[4] T. Bravo, S. J. Elliott: Variability of low frequency sound
transmission measurements. J. Acoust. Soc. Amer. 115
(2004) 29862997.
[5] I. Veit, H. Sander: Production of spatially limited diuse
sound eld in an anechoic room. Journal of the Audio En-
gineering Society 35 (1987) 138143.
[6] M. Keller, A. Roure, F. Marrot: Acoustic eld reproduction
for psychoacoustic experiments: application to aircraft in-
terior noise. Proceedings of Active 2006, 2006.
835
ACTA ACUSTICA UNITED WITH ACUSTICA Gauthier, Berry: Room effects on wave eld synthesis
Vol. 93 (2007)
[7] A. J. Berkhout: A holographic approach to acoustic control.
Journal of the Audio Engineering Society 36 (1988) 977
995.
[8] A. J. Berkhout, D. de Vries, P. Vogel: Wave front synthesis:
A new direction in electro-acoustics. Proceedings of the
93rd AES Convention, 1992.
[9] A. J. Berkhout, D. de Vries, P. Vogel: Acoustic control
by wave eld synthesis. J. Acoust. Soc. Amer. 93 (1993)
27642778.
[10] A. J. Berkhout, D. de Vries, J. J. Sonke: Array technology
for acoustic wave eld analysis in enclosures. J. Acoust.
Soc. Amer. 102 (1997) 27572770.
[11] E. N. G. Verheijen: Sound reproduction by wave eld syn-
thesis. Ph.D. thesis, Delft University of Technology, Delft,
1997.
[12] E. W. Stewart, D. de Vries, A. J. Berkhout: Wave eld syn-
thesis operators for bent line arrays in a 3D space. Acustica
united with Acta Acustica 85 (1999) 883892.
[13] G. Theile, H. Wittek: Wave eld synthesis: A promising
spatial audio rendering concept. Acoust. Sci. and Tech. 25
(2004) 393409.
[14] E. W. Start, M. S. Roovers, D. de Vries: In situ measure-
ments on a wave eld synthesis system for sound enhance-
ment. AES 102nd convention, 1997, Convention paper
4454.
[15] B. Klehs, T. Sporer: Wave eld synthesis in the real world:
Part 1 - In the living room. AES 114th convention, 2003,
Convention paper 5727.
[16] T. Sporer, B. Klehs: Wave eld synthesis in the real world:
Part 2 - In the movie theatre. AES 116th convention, 2004,
Convention paper 6055.
[17] M. M. Boone, E. N. G. Verheijen: Qualication of sound
generated by wave eld synthesis for audio reproduction.
AES 102nd convention, 1997, Convention paper 4457.
[18] H. Kutschback: Verication for spatial sound systems. Pro-
ceedings of the AES 24th international conference, Ban,
Canada, 2003.
[19] A. Berry, P.-A. Gauthier: Adaptive wave eld synthesis
with independent radiation mode control for active sound
eld reproduction. Proceedings of Active 2006, 2006.
[20] P.-A. Gauthier, A. Berry: Adaptive wave eld synthesis
with independent radiation mode control for active sound
eld reproduction: Theory. J. Acoust. Soc. Amer. 119
(2006) 27212737.
[21] J. Usher, W. Martens, W. Woszczyk: The inuence of the
presence of multiple sources on auditory spatial imagery
as indicated by a graphical response technique. Proceed-
ings of the 18th International Congress on Acoustics, Ky-
oto, Japan, 2004.
[22] A. Asano, D. C. Swanson: Sound equalization in enclo-
sures using modal reconstruction. J. Acoust. Soc. Amer.
98 (2002) 20622069.
[23] A. O. Santilla: Spatially extended sound equalization in
rectangular rooms. J. Acoust. Soc. Amer. 110 (2001) 1989
1997.
[24] S. Spors, A. Kuntz, R. Rabenstein: An approach to listening
room compensation with wave eld synthesis. Proceedings
of the AES 24th international conference, Ban, Canada,
2003.
[25] S. Spors, M. Renk, R. Rabenstein: Limiting eects of ac-
tive room compensation using wave eld synthesis. Pro-
ceedings of the 118th AES Convention, 2005.
[26] L. Fuster, J. J. Lpez, A. Gonzlez, P. D. Zuccarello: Room
compensation using multichannel inverse lters for wave
eld synthesis systems. Proceedings of the 118th AES Con-
vention, 2005.
[27] T. Betlehem, T. D. Abhayapala: Theory and design of
sound eld reproduction in reverberant rooms. J. Acoust.
Soc. Amer. 117 (2005) 21002111.
[28] S. Spors, H. Buchner, R. Rabenstein: Ecient active lis-
tening room compensation for wave eld synthesis. AES
116th convention, 2004, Convention paper 6119.
[29] E. Corteel: Equalization in an extended area using multi-
channel inversion and wave eld synthesis. Journal of the
Audio Engineering Society 54 (2006) 11401161.
[30] J. Blauert: Spatial hearing: The psychophysics of human
sound localization. MIT Press, 1999.
[31] A. D. Pierce: Acoustics - An introduction to its physical
principles and applications. Acoustical Society of America,
1991.
[32] J. J. Sonke, D. de Vries, J. Labeeuw: Variable acoustics
by wave eld synthesis: A closer look at amplitude eects.
AES 104th convention, 1998, Convention paper 4712.
[33] S. Spors, R. Rabenstein: Spatial aliasing artifacts produced
by linear and circular loudspeaker arrays used for wave
eld synthesis. AES 120th convention, 2006, Convention
paper.
[34] E. W. Start, V. G. Valstar, D. de Vries: Application of spa-
tial bandwidth reduction in wave eld synthesis. AES 98th
convention, 1995, Convention paper 3972.
[35] D. de Vries, E. W. Start, V. G. Valstar: The wave eld syn-
thesis concept applied to sound reinforcement: Restrictions
and solutions. AES 96th convention, 1994, Convention pa-
per 3812.
[36] E. Corteel: On the use of irregurlarly spaced loudspeaker
arrays for wave eld synthesis, potential impact on spa-
tial aliasing frequency. Proceedings of the 9th international
conference on digital audio eects (DAFX), 2006.
[37] L. J. Ziomek: Fundamentals of acoustic eld theory and
space-time signal processing. CRC, 1995.
[38] K. Brandenburg, S. Brix, R. Sporer: Wave eld synthesis:
New posibilities for large-scale immersive sound reinforce-
ment. Proceedings of ICA2004, Kyoto, Japan, 2004.
[39] B. Widrow, S. D. Steams: Adaptive signal processing.
Prentice-Hall, 1985.
[40] T. Sderstrm, P. Stioca: System identication. Prentice-
Hall, 1989.
[41] L. Ljung: System identication: Theory for the user.
Prentice-Hall, 1999.
[42] J. Jouhaneau: Acoustique des salles et sonorisation. Lavoi-
sier Tec&Doc, 1997.
[43] P. Nelson: Active control of acoustic elds and the repro-
duction of sound. Journal of Sound and Vibration 177
(1994) 447477.
[44] S. Ise: A principle of sound eld control based on the
Kirchho-Helmholtz integral equation and the theory of
inverse system. Acustica united with Acta Acustica 85
(1999) 7887.
[45] S. Takane, Y. Suzuki, T. Sone: A new method for global
sound eld reproduction based on Kirchhos integral
equation. Acustica united with Acta Acustica 85 (1999)
250257.
[46] P. A. Nelson: Active control for virtual acoustics. Active
2002, 2002.
[47] D. W. E. Schobben, R. M. Aarts: Personalized multi-
channel headphone sound reproduction based on active
noise control. Acta Acustica united with Acustica 91
(2005) 440449.
836

Vous aimerez peut-être aussi