Sunteți pe pagina 1din 20

Audio Engineering Society

Convention Paper
Presented at the 131st Convention
2011 October 2023 New York, USA
This paper was peer-reviewed as a complete manuscript for presentation at this Convention. Additional papers may be obtained
by sending request and remittance to Audio Engineering Society, 60 East 42
nd
Street, New York, New York 10165-2520, USA; also
see www.aes.org. All rights reserved. Reproduction of this paper, or any portion thereof, is not permitted without direct permission
from the Journal of the Audio Engineering Society.
Beamforming regularization, scaling
matrices and inverse problems for sound
eld extrapolation and characterization:
Part II Experiments
Philippe-Aubert Gauthier
1,2
,

Eric Chambatte
1,2
, C edric Camier
1,2
, Yann Pasco
1,2
, and Alain Berry
1,2
1
Groupe dAcoustique de lUniversit e de Sherbrooke, Univ. de Sherbrooke, Qu ebec, J1K 2R1 Canada
2
Centre for Interdisciplinary Research in Music, Media and Technology, McGill Univ., Montr eal, Qu ebec, H3A 1E3 Canada
Correspondence should be addressed to Philippe-Aubert Gauthier
(Philippe-Aubert.Gauthier@USherbrooke.ca)
ABSTRACT
Sound eld extrapolation (SFE) is aimed at the prediction of a sound eld in an extrapolation region using
microphone array. For sound environment reproduction purposes, sound eld characterization (SFC) aims
at a more generic or parametric description of a measured or extrapolated sound eld using dierent physical
or subjective metrics. In this paper, experiments with a recently-developed SFE method (Part I Theory)
are reported in a rst instance. The method is based on an inverse problem formulation combined with a
recently proposed regularization approach: a beamforming matrix in the discrete smoothing norm of the cost
function. In a second instance, the results obtained from the SFE method are applied to SFC as presented in
Part I. The SFC classication method is veried in two environments that recreate ideal or complex sound
elds. In light of the presented results and discussion, it is argued that SFE and SFC proposed methods are
eective.
1. INTRODUCTION
For spatial sound reproduction technologies based on
physical simulation such as Wave Field Synthesis (WFS)
[1, 2], the underlying hypothesis is that the immersion
of a listener in a physical reconstruction of a target
sound eld will lead to an appropriate sound percep-
tion over a large listening area. In this area, the local-
ization cues (interaural level difference, interaural time
difference and spectral modications) are naturally pro-
Gauthier et al. Sound eld extrapolation and characterization II
duced from the interaction of the listeners body and ex-
ternal ears with the recreated sound eld. To reproduce
or recreate a real sound eld or real sound environment,
WFS and other physical reproduction techniques require
a complete physical description of the target sound eld.
Sound eld extrapolation using microphone array tech-
nologies is appropriate for this purpose. In this paper, a
sound eld extrapolation and characterization method-
ology is presented (Part I) and experimentally tested
(Part II). This work is part of a larger project which in-
volves the entire sound eld reproduction of an airplane
cabin in a full-scale mock-up. The objective of the re-
ported theory and experiments is to get preliminary in-
sights on the efciency and validity of the sound eld ex-
trapolation (SFE) and sound eld characterization (SFC)
methods in practical situation. A complete introduction
to the purposes of this work is presented in Part I The-
ory.
This paper completes the theoretical results of its com-
panion paper: it presents experimental validations in an
hemi-anechoic roomand in a reverberation chamber. For
a detailed description of the SFE and SFC methods, the
reader is redirected to Part I Theory. In the companion
paper, it was concluded that the proposed SFE method
was efcient for theoretical sound elds without mea-
surement noise. In this part, the SFE method is tested in
more realistic situations with real sound elds and micro-
phones, including unknown measurement noise and im-
precision and difference between microphones. Fromthe
SFC developments and results in the companion paper, it
was concluded that the energy vector magnitude and the
directional diffusion were the scalar metrics which are
the most able to distinguish archetypical sound elds:
source in free eld, standing wave and diffuse sound
eld. In this second part, the SFC and classication
method is tested in real non-ideal situations with imper-
fect sound elds that approach archetypical sound elds.
As it will be shown in this paper, the SFE and SFC meth-
ods work well in practical situations.
1.1. Paper structure
Section 2 formulates the objectives of the experiments
and the selected methodology to experimentally investi-
gate the SFE and SFC methods. The measurement sys-
tem, microphone arrays, hardware and setups are de-
scribed in Sec. 3. The results for the hemi-anechoic
room and reverberation chamber are presented in Sec. 4.
The comparative results of eld-type classications and
scores are reported in Sec. 5. The companion paper
(Part I Theory) presents further details on the SFE and
SFC methods and realizations. Full-page gures are at
the end of the paper.
2. OBJECTIVES AND METHODOLOGY
The aim of this paper is twofold: 1) illustrate the ef-
ciency of the SFE method and 2) investigate the capa-
bility of the SFE and SFC methods to deal with various
archetypical and arbitrary sound elds.
In a rst instance, to illustrate the efciency of the pro-
posed method for SFE and sound source localization, a
sound eld was measured using a microphone array in a
hemi-anechoic room. The sound eld was created us-
ing an omnidirectional source. The array was moved
in the vicinity of the original source position and the
sound eld was again measured in that extrapolation re-
gion. Subsequent multichannel SFE was applied to the
rst measurement data to predict the sound eld in the
extrapolation region. To illustrate the efciency of the
method we summarize the comparison between the mea-
sured sound pressure eld in the extrapolation region and
the extrapolated sound pressure eld in that region. The
presented result was already published in Ref. [3] but is
reproduced here to illustrate the approach. Additional
experimental verications of the SFE method are also re-
ported in Ref. [3].
In a second instance, to evaluate the capability of the SFC
method to distinguish between different types of sound
eld, (namely a source in free eld, a diffuse sound eld
and a standing wave), we made several sound eld mea-
surements in various known environments to emulate
these archetypical sound elds. Field-type classication
and eld-type scores as presented in Part I Theory are
then evaluated and discussed on the basis of the a priori
known information about the environments.
3. MICROPHONE ARRAYS AND EXPERIMEN-
TAL SETUPS
The microphone arrays are made of custom-made mi-
crophones using 6-mm electret capsules, cylindrical alu-
minum chassis and RCA connectors. The microphones
are shown in Fig. 1(a). We assembled 96 of these micro-
phones. The sensitivity of each microphone is calibrated
at 1 kHz. The microphones are not phase calibrated since
it was veried that intermicrophonic phase mismatches
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 2 of 20
Gauthier et al. Sound eld extrapolation and characterization II
are smaller than the equivalent phase error caused by
the uncertainty in microphone positioning in the array
(including uncertainty caused by array manipulation, as-
sembly, etc.). The custom-made preampliers are based
on eight-channel circuits shown in Fig. 1(b). Each PCB
circuit is equipped with DB-9 connectors to simplify the
connection of 8-microphone groups. The preamplier
circuits are grouped to form24-channel preamplier rack
with power supply and toroidal transformers. One of the
24-channel preamplier racks is shown in Fig. 1(c). Four
racks were built in our laboratory facility to form a 96-
channel preamplier rack.
The complete measurement setup is schematically rep-
resented in Fig. 2(a). The preamplier outputs are digi-
tized using four MOTU
TM
24IO sound cards connected
to a computer through Audiowire
TM
cables.
The two microphone array congurations are made of 96
or 80 custom-made microphones arranged in a double-
layer (12.25 cm vertically spaced) rectangular array and
aligned on a horizontal grid with a spacing of 12.25 cm.
These two congurations are shown in Fig. 2. The
array width and length (from capsule to capsule) are
0.993 m and 1.3475 m, respectively, and the array height
is 12.25 cm. The microphone arrays used in the hemi-
anechoic room and in the reverberation chamber are
shown in Fig. 3.
The virtual source distribution used in the direct and in-
verse problems is a spherical distribution of 642 plane
waves, the direction cosines of which are shown in Fig. 4.
4. EXPERIMENTAL RESULTS: SOUND FIELD
EXTRAPOLATION AND SOUND FIELD CHAR-
ACTERIZATION
In this section, several SFE and SFC experiments are
reported to validate the two methods in two scenarios:
hemi-anechoic room and reverberation chamber. The di-
rect comparison of SFE results with directly measured
sound eld in hemi-anechoic room for rigorous vali-
dation of the SFE methods based on irregular micro-
phone array, inverse problems and beamforming regular-
ization matrix was specically investigated and reported
in Ref. [3]. These investigations showed that the SFE
method was effective for the prediction of the sound eld
in an extended area. One result from Ref. [3] is repro-
duced here. Therefore, in this paper, the validation is
mostly devoted to the SFC method as reported in the
(a) Electret microphones with casing including 6-mm
capsules
(b) 8-channel preampliers circuits
(c) 24-channel preamplier rack
Fig. 1: Pictures of the custom-built electret microphones
and 24-channel preamplier racks (design by Y. Pasco).
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 3 of 20
Gauthier et al. Sound eld extrapolation and characterization II
Microphone array
Preamplifiers
Sound cards
Computer
(a)
x
2
x
1
x
3
(b)
x
2
x
1
x
3
(c)
Fig. 2: (a): Data acquisition chain. Schematic represen-
tation of the two microphone array congurations ((b):
96-channel and (c): 80-channel versions). The micro-
phone capsules are shown as black dots. The grey mesh
corresponds to the positions in the x
1
x
2
plane of the mi-
crophones.
companion paper. Furthermore, we investigate the capa-
bility of the sound eld classication method to distin-
guish between the archetypical cases of a sound source
in nearly free-eld conditions, a room mode forced re-
sponse and a diffuse sound eld.
4.1. Descriptions of the tested environments
The Groupe dAcoustique de lUniversit e de Sherbrooke
(a)
(b)
Fig. 3: Pictures of the microphone arrays in hemi-
anechoic room (a) and reverberation chamber (b). In the
hemi-anechoic room, the sound source is an omnidirec-
tional source shown in the foreground. In the reverbera-
tion chamber, the sound source is a public address loud-
speaker shown in the top-left corner of the picture.
hemi-anechoic room dimensions are, approximately,
6.55 m6.25 m3.05 m, which gives a total volume of
125 m
3
. The walls and ceiling are truly anechoic above
100 Hz. A picture of a part of the hemi-anechoic room
where the experiment took place is shown in Fig. 3(a).
One notes that only a fraction of the room was used and
that mobile absorbing walls were surrounding the setup
(on the left). The oor is made of thick concrete.
The reverberation chamber volume V is 142 m
3
. The di-
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 4 of 20
Gauthier et al. Sound eld extrapolation and characterization II
0.5
0
0.5
1
0.5
0
0.5
1
1
0.5
0
0.5
1
cos(
l
)cos(
l
)
sin(
l
)cos(
l
)
s
i
n
(

l
)
Fig. 4: Spherical distribution of L = 642 incoming plane
waves. Each propagating direction
l
and
l
is shown as
a black dot on the sphere with the corresponding direc-
tion cosines.
mensions are 7.5 m 6.2 m 3.05 m. Since its volume
is larger than 50 m
3
, the room is judged as able to pro-
duce a diffuse sound eld. The room reverberation dis-
tance [12] was directly evaluated fromthe average spatial
decay curve from an omnidirectional sound source. We
recall that the reverberation distance or radius is the dis-
tance from an omnidirectional sound source where the
direct sound energy density is equal to the diffuse and
homogeneous energy density. At distances larger than
the reverberation distance, the reverberation is stronger
that the direct sound eld. The reverberation distance
oscillates between 0.46 m and 0.66 m depending on the
computation method (from the direct and diffuse sound
eld asymptotic expressions or the 3 dB gain above the
homogeneous sound pressure level). For the reported ex-
periment, the distance between the sound source and the
microphone array was much larger than the reverberation
distance. Therefore, we assume that the microphone ar-
ray is solely exposed to reected sounds. A picture of
the room is shown in Fig. 3(b).
To illustrate the modal behavior of the reverberation
chamber, a frequency response function (FRF) was mea-
sured from the sound source to the rst microphone of
the array. This FRF is shown in Fig. 5(a). Reported test
frequencies (54.5 and 600 Hz) are marked in this FRF.
The corresponding impulse response and Schroeder en-
ergy decay curve [13] are shown in Fig. 5(b). According
10
2
10
3
60
50
40
30
20
10
0
10
Freq. [Hz]
G
a
i
n

[
d
B

r
e
f

1
]
(a)
0 1 2 3 4 5
100
80
60
40
20
0
Time [s]
I
m
p
u
l
s
e

r
e
s
p
o
n
s
e

[
d
B

r
e
f

1
]


Instantaneous IR
Schroeder Curve
T20 prediction
EDT prediction
60 dB
(b)
Fig. 5: (a): Measured frequency response function
(FRF), (b): instantaneous impulse response (IR) and
Schroeder curve in the reverberation chamber. The
test frequencies (54.5 Hz and 600 Hz) are identied as
vertical lines in the FRF. Linear approximation of the
Schroeder curve used for the computation of the early
decay time [s] and reverberation time [s] are also shown
as dashed lines over the IR.
to this Schroeder curve, the estimated early decay
time (EDT) is 4.3280 s and the reverberation time T
60
(60 dB fall-off) evaluated from a linear t of the rst
20 dB fall-off is 6.4346 s. Accordingly, the Schroeder
cutoff frequency f
Sch
2000(T
60
/V)
1/2
[14] is approxi-
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 5 of 20
Gauthier et al. Sound eld extrapolation and characterization II
5 4 3 2 1 0 1 2
2
1.5
1
0.5
0
0.5
1
1.5
2
x
1
[m]
x
2

[
m
]
a) b)
Fig. 6: Top view of the experimental conguration in
hemi-anechoic room. The sound source is shown as
a large black circle. The acoustic center of the sound
source is shown as a small white dot. Two microphone
array positions are labeled a) and b).
mately 426 Hz, i.e. below the 600 Hz test frequency. At
600 Hz the modal density N( f ) = 1/2V/
2
(2 f )
2
/c
3
(modes per unit angular frequency in rad/s) is 2.5336
(or 15.9191 modes per frequency in Hz) and the aver-
age spacing
f
= 1/(2N( f )) in hertz between room
resonance frequencies is 0.0628 Hz [14]. According
to the Schroeder frequency f
Sch
we expect that the ex-
cited sound eld at 600 Hz should be diffuse. How-
ever, the relatively low theoretical modal density N( f ) at
600Hz (which is independent of the reverberation time
and low mode damping) suggests that only three or less
modes might strongly dominate the room response at
600 Hz. Therefore, given the theoretical requisites for
a diffuse sound eld, the sound eld is not exactly dif-
fuse at 600 Hz. This partly explains some of the SFC
results reported in the forthcoming.
4.2. SFE and SFC in hemi-anechoic room
This section presents the SFE, computed metrics and
sound eld classication results for the hemi-anechoic
room described in Sec. 4.1 with the setup described in
Sec. 3. For this experiment, the sound source was pro-
ducing a 600 Hz continuous tone.
The geometrical arrangement of the sound source and
microphone array is shown in Fig. 6.
To demonstrate the efciency of the SFE method, a di-
rect comparison of the extrapolated sound eld and the
measured sound eld in a monitoring area is shown in
Fig. 7. Clearly the extrapolated sound eld approaches
(a) Measured sound eld p(x) in the monitoring area
(b) Extrapolated sound eld p(x) in the monitoring area
(c) Extrapolation error |e(x)| =| p(x) p(x)| in the moni-
toring area
Fig. 7: SFE validation in hemi-anechoic room by direct
comparison of the measured p(x) and extrapolated p(x)
sound elds in a monitoring region a) different from the
measurement region b) (see Fig. 6) (from Ref. [3]). The
real and imaginary parts of the sound elds are shown on
left and right, respectively.
the measured sound eld in the monitoring area. As ex-
pected, the extrapolation error increases with distance
from the measurement microphone array. This com-
pletes the proof of efciency. Further details are given
in Ref. [3].
In the remainder of this section, the measuring micro-
phone array was in position a) (Fig. 6). Once the complex
sound eld p(x
m
) was measured at the microphone ar-
ray, SFE was achieved using the inverse problemmethod
combined with the beamforming regularization matrix as
described in Part I Theory. The obtained solution is
either described as a plane wave amplitude distribution
q
BF

(
l
,
l
) or by the corresponding directional pressure
p
l
and the directional energy density E
l
. These two met-
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 6 of 20
Gauthier et al. Sound eld extrapolation and characterization II
rics, introduced in the companion paper, are presented in
Fig. 8. The regularization parameter was set to 0.5.
Clearly both the directional pressure and energy density
are useful to localize the sound source and the ground
reection. Superimposed on these two quantities, the
energy vector direction is also shown. As reported in
the companion paper for the theoretical test case of two
sound sources in free-eld, the energy vector gives an av-
erage direction between the two localized sound sources.
This is consistent with the energy vector shown in these
two gures.
One should also note that the direct sound, beside be-
ing apparently softer than the ground reection (see
Fig. 8(a)), involves a wider spread in the spherical plane
wave distribution. The latter effect compensates for the
former: i.e. a larger but softer lobe in the plane wave dis-
tribution might be as loud as a smaller but stronger lobe.
Therefore, for the reported experiment, we cannot state
the ground reection is stronger than the direct sound
solely from the maximum values of the observed lobes.
The SFE results in the horizontal plane (x
1
x
2
) and the
vertical plane (x
1
x
3
) are shown in Figs. 21(a) and 21(b),
respectively. The wave fronts curvature observed for
SFE in the horizontal plane (Fig. 21(a)) approaches that
of concentric spheres centered at the exact sound source
position. One also notes that SFE is smooth over space,
i.e. it does not diverge with distance from the micro-
phone array. This is an important feature of SFE us-
ing the beamforming regularization matrix in the inverse
problem. Indeed, although it is not expected that SFE is
efcient at very large distance from the array, it still pro-
vides a smooth sound eld. This is a great advantage for
subsequent sound eld reproduction: no extreme sound
pressure level variations are expected even beyond the
effective SFE area. The SFE results in the vertical plane
(Fig. 21(b)), reveal the direct sound in the source and
array plane and the ground reection that travels along
positive x
1
and x
3
. By comparing the direct sound in-
cidence to the reection incidence while taking account
the ground plane, one notes that the detected reection
incident angle is consistent with the presented congu-
ration. This validates the capability of the SFE method
to effectively localize sound source positions in directive
sound elds such as those created by several sources in
nearly free-eld conditions.
As explained in the theoretical companion paper, sev-
eral sound eld metrics such as sound intensity I(x),
direction-of-arrival (DOA) n
DOA
(x) = I(x)/||I(x)||
2
,

x
1
x
3
Directional pressure p
l
(
l
,
l
), color: dB ref 1
x
2

Ground reflection
Direct sound
10
8
6
4
2
0
(a)

x
1
x
3
Directional energy density E
l
(
l
,
l
), color: dB ref 1
x
2

Ground reflection
Direct sound
10
8
6
4
2
0
(b)
Fig. 8: Spherical plots of (a): the normalized directional
pressure |p
l
(
l
,
l
)|/||p
l
(
l
,
l
)||

(linear (radius) and


dB ref 1 (color) scale), (b): the normalized directional
energy density E
l
/||E
l
||

(linear (radius) and dB ref 1


(color) scales) and the energy vector E (shown as a large
arrow) in hemi-anechoic room at 600 Hz for the extrap-
olated sound eld shown in Fig. 21(a).
energy density E(x) and diffuseness (x) [9] are de-
tailed eld metrics: they provide a detailed view of the
measured sound eld. These four eld metrics were
computed from the inverse problem solution for SFE
region in the horizontal and vertical planes. They are
shown in Figs. 9 and 10. From the intensity elds, and
especially from the DOA elds, we see that the energy
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 7 of 20
Gauthier et al. Sound eld extrapolation and characterization II
3 2.5 2 1.5 1 0.5 0 0.5
2
1.5
1
0.5
0
0.5
1
1.5
2
I(x)
N
= 0.00028135
I(x) [W/m
2
], Max(|I(x)|) = 0.002545 W/m
2
x
1
[m]
x
2

[
m
]
(a)
3 2.5 2 1.5 1 0.5 0 0.5
2
1.5
1
0.5
0
0.5
1
1.5
2
n
DOA
(x)
x
1
[m]
n
DOA

N
x
2

[
m
]
(b)
Fig. 9: (a): Intensity eld I(x) [W/m
2
], average intensity
vector I(x)
N
[W/m
2
], (b): DOA n
DOA
(x) and average
DOAn
DOA

N
(N =625) for the extrapolated sound eld
shown in Fig. 21(a). The microphone array is shown in
light grey. Average vectors are shown as a large arrows.
ow and DOA are aligned with the original sound source
over an effective SFE area that is larger than the micro-
3 2.5 2 1.5 1 0.5 0 0.5
2
1.5
1
0.5
0
0.5
1
1.5
2
I(x)
N
= 0.00020287
I(x) [W/m
2
], Max(|I(x)|) = 0.002545 W/m
2
x
1
[m]
x
3

[
m
]
(a)
3 2.5 2 1.5 1 0.5 0 0.5
2
1.5
1
0.5
0
0.5
1
1.5
2
n
DOA
(x)
x
1
[m]
n
DOA

N
x
3

[
m
]
(b)
Fig. 10: (a): Intensity eld I(x) [W/m
2
], average in-
tensity vector I(x)
N
[W/m
2
], (b): DOA n
DOA
(x) and
average DOA n
DOA

N
(N = 625) for the extrapolated
sound eld shown in Fig. 21(b). The microphone array
is shown in light grey. The average vectors are centered
at the microphone array and shown as a large arrows.
The source plane x
3
= 0 m is shown as black dotted line
and the ground plane x
3
1.2 m is shown as a black
dashed line.
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 8 of 20
Gauthier et al. Sound eld extrapolation and characterization II
phone array. This is especially true for positions which
are aligned with the sound source and the array (along x
1
in this specic case). From the evaluated intensity and
DOA elds in the vertical plane, the sound beam caused
by the ground reection is easily noticeable.
The energy density E(x) in the horizontal plane
(Fig. 11(a)) shows that most of the energy is concentrated
around the true sound source position with some spread-
ing along the propagative direction (namely, x
1
), indi-
cating that the sound pressure should be stronger around
the sound source. As expected, the diffuseness (x) and
average diffuseness (x)
N
approach zero in most of
the SFE area. We recall that the diffuseness is bounded:
0 (x) 1. In some very localized areas, the dif-
fuseness approaches unity, a result which is not expected
for a nearly free-eld situation. This happens in specic
regions where the intensity is very low (see Figs. 9(a)
and 10(a)). This should be understood as local behav-
ior caused by the interference of the direct and reected
sound. As reported in the companion paper, this exam-
ple also highlights the fact that diffuseness strongly de-
pends on position. Therefore, diffuseness might not be
the most appropriate parameter to globally quantify the
global diffuseness of a sound eld, at least in the fre-
quency domain.
In Part I of this paper, several global eld metrics were
identied: energy density deviation
N
, directivity fac-
tor Q, directivity index DI, directional diffusion d, en-
ergy vector E magnitude, azimuth and elevation. These
scalar or global eld metrics are reported in Tab. 1 for the
600 Hz sound eld in the hemi-anechoic room. Among
these global eld metrics, two were identied as classi-
cation metrics, i.e. metrics that are able to distinguish
between archetypical sound elds. These are the en-
ergy vector magnitude R
E
and the directional diffusion
d. They are emphasized in Tab. 1. In Part I of this paper,
we shown how these two metrics are able to distinguish
free-eld situations from diffuse elds or even standing
wave patterns.
Fromthese two classication metrics, a set of sound eld
type scores were proposed in the companion paper. The
free-eld score is given by
S
ff
= R
E
. (1)
The modal or standing-wave score is given by
S
m
= (1 R
E
)(1 d/60)
2
, (2)
(a)
(b)
Fig. 11: (a): Energy density E(x) [J/m
3
] 10
6
, average
energy E(x)
N
[J/m
3
], (b): diffuseness (x) and aver-
age diffuseness (x)
N
(N =4096) for the extrapolated
sound eld shown in Fig. 21(a).
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 9 of 20
Gauthier et al. Sound eld extrapolation and characterization II
Metrics Values
Deviation of E(x) (
N
) 79.1337 %
Directional diffusion (d) 18.2335 %
Directivity factor (Q) 0.2595
Directivity index (DI) -5.8588 dB ref 1
Energy vector azimuth (
E
) -3.1310 rad
Energy vector elevation (
E
) -0.3615 rad
Energy vector radius (R
E
) 0.7454
Free-eld score (S
ff
) 0.7454
Diffuse-eld score (S
d
) 0.2311
Modal score (S
m
) 0.0235
Table 1: Scalar metrics for the sound eld measured
in hemi-anechoic room at 600 Hz (see Figs. 21(a) and
21(b)) with N = 4096.
and the diffuse-eld score is given by
S
d
= (1 R
E
)(d/60)
2
. (3)
These scores are bounded between zero and unity. The
sound eld type scores for the experiment in the hemi-
anechoic room are reported in Tab. 1. The largest score
is S
ff
, which validates the SFC and scoring method pro-
posed in the companion paper. It is normal that S
ff
does
not approach unity in this real situation. This is caused
by the ground reection that decreases the energy vec-
tor magnitude since the incoming sounds are distributed
over two main incidences. Interestingly, the free-eld
score S
ff
= 0.7454 approaches much more closely the
free-eld score of the theoretical test case with two sound
sources in free-eld S
ff
= 0.7387 (see Part I Theory)
than the free-eld score of the theoretical test case with
a single sound source in free-eld S
ff
= 0.9174 (see
Part I Theory). This is consistent with the presence
of the ground reection which acts as a secondary sound
source.
4.3. SFE and SFC in reverberation chamber
For the experiments in the reverberation chamber, two
types of sound eld are targeted: a single standing wave
pattern below the Schroeder frequency [12] and a diffuse
sound eld above the Schroeder frequency. To excite a
single standing wave pattern, several frequency response
functions of the room were identied to localize a strong
response peak in the low frequency range (see Fig. 5(a)).
For the diffuse sound eld excitation, the same frequency
that was originally used for the hemi-anechoic case is
used: 600 Hz since it is assumed that this is above the
Schroeder frequency. Note that the excitation at 54.5 Hz
exactly falls on one of the sharp room resonances in the
low frequency range. According to theoretical prediction
of natural frequencies of a rigid cavity (the size of which
is equal to the actual reverberation chamber), a resonance
is expected at 55 Hz. This is an axial mode along x
2
with
a corresponding modal index of 2 that should create two
nodal planes in the room.
4.3.1. Room mode excitation
This section presents the SFE and SFC results for the
excitation of the reverberation chamber by a continuous
54.5 Hz tone. The inverse problem solution was com-
puted for =5. The resulting directional pressure p
l
and
energy density E
l
are shown in Fig. 12. Clearly, an axial
mode along x
2
is detected. Indeed, the sound eld is de-
scribed as two facing plane wave lobes. It should be kept
in mind that a low frequency such as 54.5 Hz implies a
much longer wavelength than at 600 Hz. Therefore, the
relative size of the microphone array to the acoustical
wavelength makes the array much less accurate in terms
of angular resolution. This explains that the two facing
plane waves appear as large lobes in the resulting plane
wave distribution.
The extrapolated sound eld is shown in both the hori-
zontal and vertical planes in Fig. 22. To facilitate the ob-
servation of the standing wave pattern, the room bound-
aries, the expected nodal planes location and actual nodal
planes of the extrapolated sound eld are all shown in
this gure. For a rigid rectangular room as described in
Sec. 4.1, one of the theoretically predicted resonant fre-
quencies is approximately 55 Hz, this mode is an axial
mode along x
2
with modal index 2. The extrapolated
sound eld approaches this theoretical result. Informal
listening tests in the reverberation chamber excited at
54.5 Hz also conrmed the presence of the illustrated
nodal planes. Also note that the extrapolated mode is
approximately independent of x
1
and x
3
, in comparison
with its strong dependency on x
2
in a region delimited by
the room boundaries.
The intensity I(x) and DOA n
DOA
(x) elds for this test
case are shown in Fig. 13. The energy density eld E(x)
and diffuseness elds (x) are shown in Figs. 14(a) and
14(b). Just as it was observed for the theoretical test case
of a standing wave pattern (see Part I Theory of this
paper), the diffuseness approaches unity over the entire
SFE region. This indicates that diffuseness might not be
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 10 of 20
Gauthier et al. Sound eld extrapolation and characterization II

x
1
x
3
Directional pressure p
l
(
l
,
l
), color: dB ref 1
x
2

10
8
6
4
2
0
(a)

x
1
x
3
Directional energy density E
l
(
l
,
l
), color: dB ref 1
x
2

10
8
6
4
2
0
(b)
Fig. 12: Spherical plots of (a): the normalized direc-
tional pressure |p
l
(
l
,
l
)|/||p
l
(
l
,
l
)||

(linear (radius)
and dB ref 1 (color) scale), (b): the normalized direc-
tional energy density E
l
/||E
l
||

(linear (radius) and dB


ref 1 (color) scales) and the energy vector E (shown as a
large arrow) in reverberation chamber at 54.5 Hz for the
extrapolated sound eld shown in Fig. 22(a).
3 2 1 0 1 2 3
3
2
1
0
1
2
3
I(x) [W/m
2
], Max(|I(x)|) = 0.0090782 W/m
2
x
1
[m]
I(x)
N
= 0.0015504
x
2

[
m
]
(a)
3 2 1 0 1 2 3
3
2
1
0
1
2
3
n
DOA

N
n
DOA
(x)
x
1
[m]
x
2

[
m
]
(b)
Fig. 13: (a): Intensity eld I(x) [W/m
2
], average inten-
sity vector I(x)
N
[W/m
2
], (b): DOA n
DOA
(x) and aver-
age DOA n
DOA

N
(N = 625) for the extrapolated sound
eld shown in Fig. 22(a). The microphone array is shown
in light grey. Average vectors are shown as a large arrow.
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 11 of 20
Gauthier et al. Sound eld extrapolation and characterization II
(a)
(b)
Fig. 14: (a): Energy density E(x) [J/m
3
] 10
5
, average
energy E(x)
N
[J/m
3
], (b): diffuseness (x) and aver-
age diffuseness (x)
N
(N =4096) for the extrapolated
sound eld shown in Fig. 22(a).
Metrics Values
Deviation of E(x) (
N
) 38.9008 %
Directional diffusion (d) 34.1550 %
Directivity factor (Q) 0.0792
Directivity index (DI) -11.0147 dB ref 1
Energy vector azimuth (
E
) -1.1541 rad
Energy vector elevation (
E
) -0.5297 rad
Energy vector radius (R
E
) 0.0792
Free-eld score (S
ff
) 0.0792
Diffuse-eld score (S
d
) 0.2984
Modal score (S
m
) 0.6224
Table 2: Scalar metrics for the modal sound eld
measured in reverberation chamber at 54.5 Hz (see
Fig. 22(a)) with N = 4096.
most appropriate metric to distinguish a standing wave
pattern from a diffuse sound eld. This is discussed in
much detail in the companion paper.
The scalar metrics computed from the inverse problem
solution are reported in Tab. 2 for the case of the room
mode excitation. As theoretically introduced in the com-
panion paper, a standing wave typically involves a low
energy vector radius R
E
since various facing plane waves
tend to cancel out each other in the computation of the
energy vector E. This is observed for the reported exper-
imental case: R
E
= 0.0792. Since the directional energy
is polarized for two facing directions, the directional dif-
fusion d should be low. For the reported experiment, the
directional diffusion d is relatively low: d = 34.155 %.
For the theoretical test case of a standing wave reported
in Part I of this paper, the directional diffusion was much
lower: d = 12.3839 %. This difference is explained
by the fact that working at very low frequency to trig-
ger a single room mode will inevitably involve larger
lobes in the plane wave description of the sound eld
(see Figs. 12(a) and 12(b)). This spherical spreading of
the inverse problem solution with decreasing frequency
leads to an overestimation of the directional diffusion.
Despite these limitations, the sound-eld type scores, as
presented in Eqs. (1) to (3), are able to detect the modal
character of the measured sound eld. This partially con-
rms the validity of the proposed sound eld classica-
tion and scoring method. Further details about the depen-
dency of the directional diffusion on frequency are given
in forthcoming sections.
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 12 of 20
Gauthier et al. Sound eld extrapolation and characterization II
4.3.2. Diffuse sound eld excitation
This section presents the SFE and SFC results for the
excitation of the reverberation chamber by a continuous
broadband noise. The SFE and SFC were computed at
600 Hz from these measurements. The inverse problem
solution was computed for = 1. The resulting direc-
tional pressure p
l
and energy density E
l
are shown in
Fig. 15. According to the directional energy density, it
is clear that sound energy comes from many different
directions, hence suggesting that the sound eld is the
combination of sound waves propagating in many direc-
tions. However, strong principal directions are observed,
suggesting that the eld is not ideally diffuse.
The corresponding extrapolated sound eld is shown in
both the horizontal and vertical planes in Fig. 23. Again,
the obtained SFE shows a spatial distribution approach-
ing a diffuse sound eld. However, this seems to be only
the case around the microphone array, i.e. in the valid
SFE area.
The intensity I(x) and DOA n
DOA
(x) elds for this test
case are reported in Fig. 16 for the horizontal plane. The
energy density eld E(x) and diffuseness elds (x) are
shown in Figs. 17(a) and 17(b). As noticed for the theo-
retical test case of a diffuse sound eld in Part I Theory,
the diffuseness varies strongly with spatial position and
oscillates between zero and unity. The average diffuse-
ness (x)
N
= 0.43518 suggests a moderately diffuse
sound eld, as for the studied theoretical case in Part I.
The detailed eld metrics suggested that the measured
sound eld is partly diffuse. This is also in accordance
with known room metrics such as the Schroeder fre-
quency and modal density (see Sec. 4.1). More generic
information can be obtained from the scalar metrics.
They are computed from the inverse problem solution
and are reported in Tab. 3 for the case of the diffuse
sound eld created by white noise and post-processed at
600 Hz. A diffuse sound eld typically have a low en-
ergy vector radius R
E
since various plane waves tend to
cancel out each other in the computation of the energy
vector E. This is observed for the reported experimen-
tal case: R
E
= 0.142. For a diffuse sound eld, the di-
rectional diffusion was expected to be higher than 40 %.
However, for the reported experiment, the directional dif-
fusion d is slightly lower than for the theoretical case:
d = 49.5443 % versus d = 53.9748 % for the theoreti-
cal test case reported in Part I. Such a slight difference
might be caused by: 1) the fact that the measured sound

x
1
x
2
x
3
Directional pressure p
l
(
l
,
l
), color: dB ref 1

10
8
6
4
2
0
(a)

x
1
x
3
Directional energy density E
l
(
l
,
l
), color: dB ref 1
x
2

10
8
6
4
2
0
(b)
Fig. 15: Spherical plots of the (a): normalized direc-
tional pressure |p
l
(
l
,
l
)|/||p
l
(
l
,
l
)||

(linear (radius)
and dB ref 1 (color) scale), the (b): normalized direc-
tional energy density E
l
/||E
l
||

(linear (radius) and dB


ref 1 (color) scales) and the energy vector E (shown as a
large arrow) in reverberation chamber at 600 Hz for the
extrapolated sound eld shown in Fig. 23(a).
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 13 of 20
Gauthier et al. Sound eld extrapolation and characterization II
2 1 0 1 2
2
1.5
1
0.5
0
0.5
1
1.5
2
I(x) [W/m
2
], Max(|I(x)|) = 1.6614e007 W/m
2
x
1
[m]
x
2

[
m
]
(a)
2 1 0 1 2
2
1.5
1
0.5
0
0.5
1
1.5
2
n
DOA
(x)
x
1
[m]
n
DOA

N
x
2

[
m
]
(b)
Fig. 16: (a): Intensity eld I(x) [W/m
2
], average in-
tensity vector I(x)
N
[W/m
2
], (b): DOA n
DOA
(x) and
average DOA n
DOA

N
(N = 625) for the extrapolated
sound eld shown in Fig. 23(a). The microphone array
is shown in light grey. Average vectors are shown as a
large arrows.
(a)
(b)
Fig. 17: (a): Energy density E(x) [J/m
3
] 10
9
, average
energy E(x)
N
[J/m
3
], (b): diffuseness (x) and aver-
age diffuseness (x)
N
(N =4096) for the extrapolated
sound eld shown in Fig. 23(a).
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 14 of 20
Gauthier et al. Sound eld extrapolation and characterization II
Metrics Values
Deviation of E(x) (
N
) 81.8899 %
Directional diffusion (d) 49.5443 %
Directivity factor (Q) 0.1285
Directivity index (DI) -8.9101 dB ref 1
Energy vector azimuth (
E
) 1.9031 rad
Energy vector elevation (
E
) 0.7498 rad
Energy vector radius (R
E
) 0.1420
Free-eld score (S
ff
) 0.1420
Diffuse-eld score (S
d
) 0.5850
Modal score (S
m
) 0.2730
Table 3: Scalar metrics for the sound eld measured
in reverberant chamber above Schroeder frequency (see
Fig. 23(a)) with N = 4096.
eld is not diffuse in the most strict theoretical sense or
2) the fact that the array and spatial processing involves
a limited or approximated view of the entire sound eld.
In light of the room properties reported in Sec. 4.1, this
difference is easily justied and tolerated.
Despite these limitations, the sound-eld type scores
shown in Tab. 3, as introduced in Eqs. (1) to (3), are again
able to detect the diffuse character of the measured sound
eld, i.e. the highest score is S
d
. Again, this conrms
the validity of the proposed sound eld classication and
scoring method. In the next section, the dependency of
the directional diffusion on frequency is illustrated. This
explains some of the observation reported in this section.
5. COMPARATIVE RESULTS OF THE FIELD-
TYPE CLASSIFICATION AND SCORES
In this section, the SFC results are compared for the three
reported experiments. The scalar metrics that have been
recognized as potential eld-type identiers are shown in
Fig. 18 in the d/100R
E
plane. For comparison purpose,
the theoretical results presented in Part I are superim-
posed in the background of the gure. It is obvious that
the three points that correspond to the three eld types
fall in the proper regions identied as white and grey
rectangles. These regions were dened in Part I. The
points that correspond to the measured standing wave
and the diffuse sound eld are less separated than for the
theoretical results. As mentioned earlier, this difference
is created by the fact that the comparison is made at two
different frequencies (54.5 and 600 Hz) and the behav-
Fig. 18: Field-type classication in terms of energy vec-
tor magnitude R
E
and directional diffusion d/100 for the
reported experiments. The classication regions, as de-
ned in Part I, are shown as white and grey rectangles.
The theoretical results obtained in Part I are shown as
pale gray lines and markers in the background. The
experimental hemi-anechoic case at 600 Hz is shown
as a black-lled circle. The experimental reverberation
chamber case at 54.5 Hz is shown as black-lled square.
The experimental reverberation chamber case at 600 Hz
is shown as a black-lled triangle. The theoretical case
of an axial mode at 600 Hz is shown as black-lled dia-
mond.
ior of the microphone array and presented SFE method
change with frequency. To highlight this blurring behav-
ior in the low frequency range, a new theoretical case
is introduced. This case corresponds to an axial mode
along x
2
at 600 Hz and the sound pressure p(x
m
) at the
array was scaled to the same level that the measured
standing wave. The inverse problem was then solved us-
ing the same regularization parameter than for the true
measured standing wave. The obtained directional en-
ergy density is shown in Fig. 19. By comparison with the
true standing wave at 54.5 Hz (Fig. 12(b)), the two fac-
ing lobes are much sharper and the directional diffusion
is much lower for this high-frequency mode. The cor-
responding point in the d/100R
E
plane is also shown
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 15 of 20
Gauthier et al. Sound eld extrapolation and characterization II

x
1
x
3
Directional energy density E
l
(
l
,
l
), color: dB ref 1
x
2

10
8
6
4
2
0
Fig. 19: Spherical plot of the directional energy density
E
l
(linear (radius) and dB ref 1 (color) scales) for a theo-
retical axial mode at 600 Hz.
ff m d
0
0.5
1
S
f
f
,
S
m
,
S
d
Hemianechoic
ff m d
0
0.5
1
Rev. at 54.5 Hz
ff m d
0
0.5
1
Rev. at 600 Hz
Fig. 20: Field-type scores S
ff
, S
d
and S
m
for the three
reported experiments.
in Fig. 18. Clearly, the distinction between this new
point and the measured diffuse sound eld is much more
marked. This conrms the fact that the SFC method pre-
sented in Part I is valid and that the distinction between
standing waves and diffuse sound eld is much more dif-
cult in the low end of the spectrum.
The eld-type scores S
ff
, S
m
and S
d
as dened in Eqs. (1),
(2) and (3) are shown in Fig. 20. The highest of the three
scores is identied in black. It is clear the identied max-
imum values of the scores all t the corresponding ex-
perimental cases. This validates the scoring method as
proposed from theoretical observations in Part I.
6. CONCLUSION
The aim of this paper was to present the experimental
validation of the theoretical SFE and SFC methods pro-
posed in Part I Theory. Experimental validation re-
sults in an hemi-anechoic room and in a reverberation
chamber were reported. Spatial extrapolation was vali-
dated in hemi-anechoic conditions by direct comparison
of a measured sound eld and extrapolated sound eld
in an extrapolation region which was different from the
measurement region. This comparison highlighted the
effectiveness of the SFE method based on inverse prob-
lem with the regularization matrix in terms of sound eld
prediction in an extrapolation area. The SFC method was
rst evaluated by means of comparison between the var-
ious eld and scalar metrics for thee archetypical sound
elds: source in nearly free-eld, low-frequency stand-
ing wave in reverberation chamber and diffuse sound in
reverberation chamber. The classication method based
on the two identied classication metrics, namely the
energy vector magnitude and the directional diffusion,
was effective in the distinction of the three eld types.
The eld-type scoring method was also efcient in terms
of automatic sound-eld type detection. The three test
cases were identied correctly.
Although the SFC and classication methods correctly
identied the sound elds, the distinction between the
standing wave at 54.5 Hz and the diffuse sound eld
at 600 Hz was the most difcult to achieve with cer-
tainty. As mentioned in the paper, this may be caused
by the fact that the comparison was not achieved for the
same frequencies. In all cases, it highlights the fact that
the distinction of a standing wave from a diffuse sound
eld solely based on the directional diffusion might need
some renements. One of the possible improvements
might be in the renement or modication of the direc-
tional diffusion metric. Indeed, the directional diffusion
is a normalized measure of the variance of the directional
energy density that does not take into account the neigh-
boring relationships between the l-th propagation direc-
tions. A measure of the spatial smoothness of the direc-
tional energy might be a promising candidate to rene
the distinction between these two types of sound eld.
Pattern detection or spherical autocorrelation might also
be an interesting research avenue for standing wave de-
tection.
The general conclusions and perspectives of future work
regarding the proposed SFE and SFC method are re-
ported in Part I of this paper.
7. ACKNOWLEDGEMENTS
This is a CRIAQ (Consortium for Research and Innova-
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 16 of 20
Gauthier et al. Sound eld extrapolation and characterization II
tion in Aerospace in Qu ebec) research project supported
by NSERC (Natural Sciences and Engineering Research
Council of Canada), Bombardier and CAE Electronics.
8. REFERENCES
[1] A.J. Berkhout, D. de vries, P. Vogel, Acoustic con-
trol by wave eld synthesis, Journal of the Acousti-
cal Society of America 93 (1993) 27642778.
[2] P.-A. Gauthier, A. Berry, W. Woszczyk, Sound-
eld reproduction in-room using optimal control
techniques: Simulations in the frequency domain,
Journal of the Acoustical Society of America 117
(2005) 662678.
[3] P.-A. Gauthier,

E. Chambatte, C. Camier, Y. Pasco,
A. Berry, Derivation of Ambisonics signals and
plane wave description of measured sound eld us-
ing irregular microphone arrays and inverse prob-
lem theory, Ambisonics Symposium 2011, Lexing-
ton, Kentucky, June 2011.
[4] C. Hansen, Rank-decient and discrete ill-posed
problems, SIAM, Philadelphia, 1998.
[5] P.-A. Gauthier, C. Camier, Y. Pasco, A. Berry,

E. Chambatte, Beamforming regularization matrix


and inverse problems applied to sound eld mea-
surement and extrapolation using microphone ar-
ray, accepted for publication, Journal of Sound and
Vibration (2011).
[6] E. Hulsebos, D. de Vries, E. Bourdillat, Improved
Microphone Array Congurations for Auralization
of Sound Fields by Wave-Field Synthesis, Journal
of the Audio Engineering Society 50 (2002) 779
790.
[7] C. Verron, G. Pallone, M. Aramaki, R. Kronland-
Martinet, Controlling a spatialized environmental-
sound synthesizer, IEEE Workshop on Applications
of Signal Processing to Audio and Acoustics, New
Paltz, NY, October 2009.
[8] C. Verron, M. Aramaki, R. Kronland-Martinet, G.
Pallone, A 3-D immersive synthesizer for envi-
ronmental sounds, IEEE Transactions on Audio,
Speech, and Language Processing 18 (2010) 1550
1561.
[9] J. Merimaa, V. Pulkki, Spatial impulse response
rendering I: Analysis and synthesis, Journal of the
Audio Engineering Society 53 (2005) 11151127.
[10] V. Pulkki, J. Merimaa, Spatial impulse response
rendering II: Reproduction of diffuse sound and lis-
tening tests Journal of the Audio Engineering Soci-
ety 54 (2006) 320.
[11] V. Pulkki, Spatial sound reproduction with direc-
tional audio coding, Journal of the Audio Engineer-
ing Society 55 (2007) 503516.
[12] H. Kuttruff, Room acoustics, Spon Press, New
York, 2000.
[13] J. Jouhaneau, Acoustique des salles et sonorisation,
Lavoisier Tec&Doc, 1997.
[14] A.D. Pierce, Acoustics: An introduction to its phys-
ical principles and applications, Acoustical Soci-
ety of America, Woodbury, 1991.
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 17 of 20
Gauthier et al. Sound eld extrapolation and characterization II
(a) SFE in the horizontal plane
(b) SFE in the vertical plane
Fig. 21: Real and imaginary parts (left and right, respectively) of the extrapolated sound eld p(x) in hemi-anechoic
room at 600 Hz using the inverse problem solution with the beamforming regularization matrix. The microphones are
shown as small black dots. The sound source position is shown as a black and white circle. For illustration purpose,
the anechoic boundaries are shown as dashed lines for SFE in the horizontal plane. The source plane x
3
= 0 m is
shown as black dotted line and the ground plane x
3
1.2 m is shown as a black dashed line for SFE in the vertical
plane.
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 18 of 20
Gauthier et al. Sound eld extrapolation and characterization II
(a) SFE in the horizontal plane
(b) SFE in the vertical plane
Fig. 22: Real and imaginary parts (left and right, respectively) of the extrapolated sound eld p(x) in reverberation
chamber at 54.5 Hz using the inverse problem solution with the beamforming regularization matrix. The microphones
are shown as small black dots. The sound source position is shown as a black and white circle. For illustration purpose,
the room boundaries are shown as thick black lines. The theoretical nodal planes for the excited axial mode are shown
as dashed lines. Measured nodal planes ([p(x)] = 0 and [p(x)] = 0) are shown as thin black lines.
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 19 of 20
Gauthier et al. Sound eld extrapolation and characterization II
(a) SFE in the horizontal plane
(b) SFE in the vertical plane
Fig. 23: Real and imaginary parts (left and right, respectively) of the extrapolated sound eld p(x) in reverberation
chamber at 600 Hz using the inverse problem solution with the beamforming regularization matrix. The microphones
are shown as small black dots. For illustration purpose, the room boundaries are shown as thick black lines.
AES 131
st
Convention, New York, USA, 2011 October 2023
Page 20 of 20

S-ar putea să vă placă și