Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
0
.
1 0
0
.
1
M
i
c
#
1
A c t # 9 , h ( t ) [ P a / U ]
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
0
.
2
5
0
.
1 0
0
.
1
M
i
c
#
1
4
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
0
.
2
5
0
.
1 0
0
.
1
M
i
c
#
8
0
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
0
.
2
5
5 0 5
x
1
0
3
A c t # 3 4 , h ( t ) [ P a / U ]
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
0
.
2
5
5 0 5
x
1
0
3
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
0
.
2
5
5 0 5
x
1
0
3
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
0
.
2
5
0
.
1 0
0
.
1
T
i
m
e
[
s
]
A c t # 4 1 , h ( t ) [ P a / U ]
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
0
.
2
5
0
.
1 0
0
.
1
T
i
m
e
[
s
]
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
0
.
2
5
0
.
1 0
0
.
1
T
i
m
e
[
s
]
Fig. 4: Examples of impulse responses from actuators
(columns) #9 (trim panel), #34 (oor shaker) and #41
(subwoofer) to microphones (rows) #1, #14 and #80.
sound eld at 4 of the microphones while the 36 trim-
panel shakers must reproduce the remainder of the audio
bandwidth over the 80 microphones. The crossover is
achieved using simple low-pass and high-pass 4th-order
Butterworth lters [17]. We use forward and reverse l-
tering to ensure that the phases of the crossover lters
are linear. This also gives a -6 dB gain at the cut-off fre-
quency so that there is no amplication at the crossover
frequency once the low- and high-frequency range sys-
tems are combined. Recently, we investigated the ef-
fect of crossover frequency in the aircraft mock-up [18].
From these studies, the crossover frequency is 140 Hz.
G
#
[k] G[k],
G[k]
+
M L M
p s
M
pe
j
e
j
r, r
E[k],
E[k]
e
Fig. 5: Block diagram of the sound eld reproduction
system. Signals are in the frequency domain. The thin
dashed block represents the equalized system. Signal
sizes are indicated by slash bars crossing the arrows.
4. MULTICHANNEL EQUALIZATION
The aim of multichannel equalization is simple: cre-
ate MIMO (Multiple-Input-Multiple-Output) lters that,
once placed before an actual MIMO electroacoustic sys-
tem, will make this system diagonal and at, i.e. the di-
agonal path should be white (and of unity gain) and the
cross-talk should cancelled. The signal processing archi-
tecture is represented in Fig. 5 as a block diagram in the
frequency domain in matrix form.
The experimental plant including the reproduction
sources, the vibroacoustic responses of the mock-up and
the microphone array is described for each frequency
bin k = 0, 1, ..., N 1 by a complex matrix G[k] C
ML
where M is the number of microphones, L the number
of potential reproduction sources for the frequency bin k
and N is the number of points in the frequency domain.
The measurement of G or an estimation of G will be de-
noted
G. Generally, we will use the hat to indicate that
the quantity is measured or estimated, i.e. it is not the
exact quantity. The multichannel equalization lter is
denoted by a complex matrix
G
#
[k] C
LM
. This l-
ter transforms, by matrix multiplication, the M measured
and digitized signals in the frequency domain p[k] C
M
into L digital transducer signals s[k] C
L
. This is written
as follows
s =
G
#
p. (1)
Ideally, the combination of the equalization lter with the
actual plant should lead to white and decoupled system
E[k] with a modeling delay of samples
E = G
G
#
Ie
j
, (2)
with 0 2, the normalized angular frequency, and
= 2k/N. Matrix E C
MM
will be called the equal-
ized system. The modeling delay is introduced to ensure
AES 52
ND
INTERNATIONAL CONFERENCE, Guildford, UK, 2013 September 24
Page 4 of 10
Gauthier et al. Sound eld reproduction in cabin mock-up
that the equalization lters
G
#
are causal in the time-
domain. Any deviation of the equalized system E (or
E)
from the identity matrix with additional modeling delay
phase shift e
j
will lead to reproduction errors, three
types of which exist: 1) channel coupling caused by off-
diagonal terms (E
mm
= 0, with m = m
), 2) non-at fre-
quency response function from recording microphone m
to error microphone m (|E
mm
| =1) and 3) phase errors in
the diagonal paths (E
mm
=e
j
).
In this paper, p C
M
is the measured target sound eld
at the microphone array and r[k] C
M
is the reproduced
sound eld at the microphone array with
r = E p = Gs. (3)
The measurement of the reproduced sound eld in the
cabin mock-up will be denoted r. An ideal reconstruc-
tion of the sound eld would lead to r =r = pe
j
. The
objective evaluations (Sec. 5) are solely based on actual
measurements of reproduced sound elds in the actual
mock-up.
4.1. Singular value decomposition
The singular value decomposition (SVD) of G[k] is given
by
G = U
H
V, (4)
where U[k] C
MM
= [u
1
, , u
M
] (U
H
U = UU
H
= I)
and V[k] C
LL
= [v
1
, , v
L
] (V
H
V = VV
H
= I) are
unitary matrices of singular vectors. Matrix [k] R
ML
is a diagonal matrix with the real, positive, singular
values
i
[k] in decreasing order on its main diagonal.
Tikhonov regularization and the proposed method di-
rectly interact with the singular value spectrum.
4.2. Tikhonov regularization
One of the simplest way to circumvent the limitation of
the pseudo-inverse approach [17] is to include a regular-
ization of the solution norm in a reproduction error min-
imization problem, in the least-mean-square sense. This
is known as Tikhonov regularization. The solution is de-
ned according to this minimization task
s = argmin{
Gs e
j
p
2
2
+
2
s
2
2
} (5)
where the amount of regularization is controlled by the
regularization parameter . As it was done in [17, 18],
we assume that is xed for all frequencies, the solution
of Eq. (5) is given by
s = (
G
H
G+
2
I)
1
G
H
e
j
p. (6)
Then, the equalization matrix is given by
G
#
= (
G
H
G+
2
I)
1
G
H
e
j
, (7)
and the resulting equalized system is
E = G(
G
H
G+
2
I)
1
G
H
e
j
. (8)
The role of the penalization parameter is to enhance the
main diagonal of the denominator matrix, hence prevent-
ing the instability of the inversion and reducing the solu-
tion 2-norm s
2
. If the identied system
G is equal to
the actual system G and that the regularization parameter
is zero ( = 0), then one obtains the ideal equalized sys-
tem E = G(G
H
G)
1
G
H
e
j
for the system under test.
However, the reader should keep in mind that for a rect-
angular MIMO system such as our with 80 microphones
and 41 actuators, it is not possible that the equalized sys-
tem become a perfect identity matrix with supplemen-
tal modeling delay since we work in the overdeterminate
case (M > L).
Using the SVD of
G, it is possible to rewrite Eq. (7):
G
#
=
rank(
G)
i=1
f
i
v
i
u
H
i
i
e
j
, (9)
with the lter factors f
i
given, for Tikhonv regularization,
by
f
i
=
2
i
2
i
+
2
. (10)
This formulation shows that for singular values
i
smaller than , the lter factors are less than one and
will lter out the corresponding singular components
( v
i
u
H
i
/
i
). For
i
larger than , the lter factors are one
and the corresponding singular components are not l-
tered.
4.3. Normalized Tikhonov regularization
In this section, we introduce a simple modication of the
multichannel equalization approach using least-mean-
square solution. In place of using an absolute regular-
ization parameter , we introduce a normalized regular-
ization parameter
dened by:
[k] =
1
[k], (11)
where
1
is the largest singular value from the SVD. Us-
ing this new denition, the lter factor used in Eq. (9) are
AES 52
ND
INTERNATIONAL CONFERENCE, Guildford, UK, 2013 September 24
Page 5 of 10
Gauthier et al. Sound eld reproduction in cabin mock-up
simply replaced by
f
i
=
2
i
2
i
+(
1
)
2
. (12)
This new lter factor is automatically adaptive in the
sense that the equivalent regularization parameter now
change with frequency. In other words, the new param-
eter
represents, as a fraction of the largest singular
value, the smallest of the
i
that will not be ltered in
Eq. (9). As an example, with
=0.01 any singular com-
ponents v
i
u
H
i
/
i
related to singular values smaller than
1% of the largest singular value
1
will be attenuated
or cancelled in the regularization process. This new ap-
proach corresponds to this minimization problem
s = argmin{
Gs e
j
p
2
2
+(
1
)
2
s
2
2
}. (13)
5. EXPERIMENTAL RESULTS
The experiments go as follow: Once the multichannel
reproduction system is experimetnally identied and that
target sound elds are stored in wave les, the reproduc-
tion signals are computed for the test cases. Then the re-
production signals and sent to the mock-up and the repro-
duced sound eld is measured with the microphone ar-
ray. Objective evaluation of each test case is then achieve
in a post-processing stage.
To explain more easily the results of modied mul-
tichannel equalization, let rst introduce the singular
value spectrum of the system matrices, i.e. the low-
frequency system with 5 inputs and 4 outputs and the
high-frequency system with 36 inputs and 80 outputs.
These singular value spectrum are shown in Figs. 6(a)
and 6(b). With increasing frequency the largest singu-
lar value decreases and the smallest increases, this cor-
responds to a better conditioning at higher frequencies.
The sudden drops of the singular values (Fig. 6) appear
at frequency limits of the swept sines used for the iden-
tication of the oor shakers (up to 1000 Hz), the sub-
woofer (up to 300 Hz) and trim panels (from 80 Hz).
Besides the singular value spectrum, Fig. 6 also includes
relevant lines. To exemplify the potential regularization
impact using the normalized regularization parameter
i
=0.01
1
4
0
H
z
1
0.1
1
0.01
1
4
(b) High frequency MIMO system: 3680
10
2
10
3
10
4
10
4
10
2
10
0
10
2
Frequency [Hz]
i
8
0
H
z
1
4
0
H
z
=0.1
1
0.1
1
0.01
1
36
Fig. 6: Singular value spectrum for the low frequency
and high frequency systems. Swept sine frequency limits
are shown as dashed vertical lines.
5.1. Objective evaluation metrics
Evaluation metrics are dened in the discrete frequency
domain k where signal spectra are represented by one-
sided modied periodograms obtained using Welchs
method and Hanning window of measured target and re-
produced sound elds. The averaged and normalized re-
production error is
E
LS
[k] =
r e
j
p
2
p[k]
2
=
e[k]
2
p[k]
2
, (14)
it represents, on the average, how well the sound eld is
reproduced over the microphone array. E
LS
= 0 means a
perfect reproduction. One should keep in mind that E
LS
is a relative reproduction error metric. As an example,
for a frequency-independent error 2-norm e[k]
2
and a
AES 52
ND
INTERNATIONAL CONFERENCE, Guildford, UK, 2013 September 24
Page 6 of 10
Gauthier et al. Sound eld reproduction in cabin mock-up
frequency-dependent target 2-norm p[k]
2
, one would
obtain a frequency-dependent E
LS
. To quantify the qual-
ity of the reproduced sound environment with respect to
timbre, the averaged magnitude spectrum error is dened
in dB ref 1 / Hz by
M
LS
[k] = 10log
10
| p[k]| 10log
10
| r[k]|
2
/M, (15)
it represents the reproduction error in terms of power
density function of the measured p[k] and reproduced
r[k] sound environments, i.e. it does not take into account
the spatial distribution of the phase but only the spatial
distribution of sound pressure amplitude. In Eq. (15),
absolute value | | and log
10
are elementwise. The source
signal 2-norm is given by
S
LS
[k] = s[k]
2
. (16)
5.2. Tikhonov regularization
Multichannel equalization with Tikhonov regularization
is used as a rst test case. From [18], the best regulariza-
tion parameters were identied to be 0.01 for the low-
frequency system and 0.1 for the high-frequency sys-
tem. The ideal crossover frequency was found to be
140 Hz. Figure 13 shows the obtained performance met-
rics E
LS
, M
LS
and S
LS
as function of frequency for this
test case. Generally speaking, the E
LS
is higher for the
lowest frequencies and it tends to increase with higher
frequency in the high-frequency range. Since from 1 kHz
to 10 kHz the typical target PSD falls of nearly 40 dB
(see Fig. 1), this reduction in p[k]
2
with increasing fre-
quency might explain the increase of relative reproduc-
tion error E
LS
. This general observation applies to the
following results. To illustrate the results, we also rely
on Welch periodograms of a selected monitoring micro-
phone recording of reproduced sound. The comparison
between the target and reproduction sounds at micro-
phone #14 is shown in Fig. 7. Clearly, the reproduced
PSD approaches the target PSD but it is not in exact cor-
respondence. As mentioned earlier, we remind that it is
impossible to obtain a perfect reproduction (in the objec-
tive sense) using more microphones than actuators.
5.3. Normalized Tikhonov regularization
Using the modied approach based on normalized
Tikhonov regularization (Sec. 4.3), several tests with
varying
were performed. Results for the high-
frequency range (above 140 Hz) are shown in Fig. 8.
We conduct separate studies for the two frequency ranges
10
2
10
3
10
4
Frequency [Hz]
P
o
w
e
r
/
F
r
e
q
.
[
d
B
r
e
f
1
/
H
z
]
Welch Power Spectral Density Estimate at Mic #14, ygrid: 10 dB
Target
Reproduced
Fig. 7: Example of Welch Power Spectral Density of
target and reproduced sounds at microphone #14 for the
case reported in Fig. 13 with Tikhonov regularization.
since non-linear distortions in one frequency range might
degrade the sound eld reproduction in the other range.
For a very small normalized regularization
= 0.001,
the regularization is not sufcient, the source norm S
LS
is large and squeaks and rattles are audible. This causes
a drastic increase in E
LS
and M
LS
. The most satisfying
normalized regularization parameter is
= 0.04. Note
that for the next gures, the worst case is shown as a
thick grey line, the best case as a thin black line and
the other cases as grey thin lines. From the compari-
son of 8(a) and (b), we note that the most satisfying E
LS
does not necessarily corresponds to the lowest M
LS
. The
PSD estimate at the monitoring microphone is shown in
Fig. 9. The correspondence is acceptable for the most
satisfying regularization amount obtained with
=0.04.
Similar measurements of reproduced sound elds were
done for the low-frequency range of the system, results
are shown in Fig. 10. In terms of reproduction error E
LS
minimization, it seems that the best results are obtained
with
= 0.004. Using these two most satisfying param-
eters, the full bandwidth results are shown in Figs. 11 and
12.
5.4. Experimental results comparison
In this section, we compare: 1) the reference test case
from [17] with the best results obtained with the Tikhonv
regularization and 2) the best results obtained with nor-
malized regularization parameter. The curves are shown
in Fig. 13. The results with normalized regularization
parameter show a lower reproduction error (E
LS
) and
a signicantly lower reproduction source 2-norm (S
LS
).
Therefore, the introduction of the normalized regulariza-
AES 52
ND
INTERNATIONAL CONFERENCE, Guildford, UK, 2013 September 24
Page 7 of 10
Gauthier et al. Sound eld reproduction in cabin mock-up
10
2
10
3
10
4
10
3
10
1
10
1
(a)
E
L
S
10
2
10
3
10
4
0
1
2
3
(b)
M
L
S
[
d
B
r
e
f
1
/
H
z
]
10
2
10
3
10
4
10
20
10
10
10
0
(c)
S
L
S
Frequency [Hz]
0.001
0.004
0.04
0.07
0.1
Fig. 8: (a) E
LS
, (b) M
LS
and (c) S
LS
for multichannel
equalization and various normalized regularization pa-
rameters
for the high frequencies. Best parameter in
terms of E
LS
is
= 0.04 and shown as black lines.
tion parameter
is benecial in the sense that a better
sound eld reproduction is achieved with much lower
10
2
10
3
10
4
Frequency [Hz]
P
o
w
e
r
/
F
r
e
q
.
[
d
B
r
e
f
1
/
H
z
]
Welch Power Spectral Density Estimate at Mic #14, ygrid: 10 dB
Target
0.001
0.004
0.07
0.04
0.1
Fig. 9: Example of Welch Power Spectral Density of
target and reproduced sounds at microphone #14 for the
case reported in Fig. 8 with various normalized regular-
ization parameters
for the high frequencies. Best pa-
rameter in terms of E
LS
is
= 0.04.
50 100 500
10
3
10
1
10
1
(a)
E
L
S
50 100 500
0
1
2
3
(b)
M
L
S
[
d
B
r
e
f
1
/
H
z
]
50 100 500
10
20
10
10
10
0
(c)
S
L
S
Frequency [Hz]
0.001
0.004
0.007
0.01
0.04
0.07
0.1
Fig. 10: (a) E
LS
, (b) M
LS
and (c) S
LS
for multichannel
equalization and various normalized regularization pa-
rameters
for the low frequencies. Best parameter in
terms of E
LS
is
= 0.004.
reproduction source signals: less squeaks and rattles
should emanate from the trim panels. This also automat-
ically introduces frequency-dependent regularization.
6. CONCLUSION
In this paper, we shown that it is possible to achieve
sound eld reproduction of real ight recordings in cabin
mock-up using multichannel equalization based on least-
mean-square approaches with microphone and vibration
transducer arrays. Results are satisfactory if one takes
into account several limitations and compromises: 1) the
use of an overdetermined problem (M > L) and regular-
ization to prevent large reproduction signals that could
lead to audible squeaks and rattles 2) the need for a sat-
isfactory global sound reproduction in the mock-up. In-
deed, for the second issue, using a smaller microphone
arrays with M = L could give very low reproduction er-
ror at the microphone array but with larger reproduction
errors outside the microphone array and possible reduced
AES 52
ND
INTERNATIONAL CONFERENCE, Guildford, UK, 2013 September 24
Page 8 of 10
Gauthier et al. Sound eld reproduction in cabin mock-up
10
2
10
3
10
4
10
3
10
1
10
1
(a)
E
L
S
10
2
10
3
10
4
0
1
2
3
(b)
M
L
S
[
d
B
r
e
f
1
/
H
z
]
10
2
10
3
10
4
10
20
10
10
10
0
(c)
S
L
S
Frequency [Hz]
Fig. 11: (a) E
LS
, (b) M
LS
and (c) S
LS
for multichan-
nel equalization with normalized regularization param-
eter using
= 0.004 and
= 0.04 for the low and high
frequency ranges, respectively.
robustness. The results are also concluding because they
were not done in ideal conditions with respect to robust-
ness. Indeed, system identication was performed hours
or even days before the actual measurements of repro-
10
2
10
3
10
4
Frequency [Hz]
P
o
w
e
r
/
F
r
e
q
.
[
d
B
r
e
f
1
/
H
z
]
Welch Power Spectral Density Estimate at Mic #14
Target
Reproduced
Fig. 12: Example of Welch Power Spectral Density of
target and reproduced sounds at microphone #14 for the
case reported in Fig. 11.
10
2
10
3
10
4
10
3
10
1
10
1
(a)
E
L
S
10
2
10
3
10
4
0
1
2
3
(b)
M
L
S
[
d
B
r
e
f
1
/
H
z
]
10
2
10
3
10
4
10
20
10
10
10
0
(c)
S
L
S
Frequency [Hz]
Reference test case: Tikhonov
Normalized regularization parameter
Fig. 13: Comparison of (a) E
LS
, (b) M
LS
and (c) S
LS
for multichannel equalization using reference test case
[17] with Tikhonov regularization and Tikhonov regu-
larization with best normalized regularization parameters
(from Fig. 11).
duced sound elds inside the mock-up.
Current and future works are oriented towards the im-
provement of the least-mean-square method. As it was
shown, the normalized regularization parameter can sim-
plify the selection of frequency-dependent regulariza-
tion, but the normalized regularization parameter still
difcult to select. To this end, we plan to work on further
modications of the method while introducing consid-
erations from the bounding property of condition num-
ber to approximately respect a maximum reproduction
source 2-norm. We also plan to work on new system
smoothing or modication algorithm before inversion to
enhance robustness, at the cost of increased reproduction
error. Example of work in this vein is [19].
Besides this, in a near future, the microphone array
recordings of the ight will be used within the context of
sound eld reproduction using commercial Wave Field
Synthesis system in a dedicated room. Standard Wave
Field Synthesis method will be adapted and the results
AES 52
ND
INTERNATIONAL CONFERENCE, Guildford, UK, 2013 September 24
Page 9 of 10
Gauthier et al. Sound eld reproduction in cabin mock-up
will be compared with sound eld reproduction evalua-
tion in the cabin mock-up.
7. ACKNOWLEDGMENTS
This work is part of a project involving: Consortium for
Research and Innovation in Aerospace in Qubec, Bom-
bardier Aronautique and CAE, supported by a Natural
Sciences and Engineering Research Council of Canada
grant.
8. REFERENCES
[1] C.A. Powell, J.M. Fields, Human response to air-
craft noise, in Aeroacoustics of ight vehicles,
Theory and practice, Volume 2: Noise control,
H.H. Hubbard Ed., Acoustical Society of America,
Woodbury, 1995.
[2] D.A. McCurdy and R. E. Grandle, Aircraft noise
synthesis system, NASA technical memorandum
89040, 1987.
[3] K. Janssens, A. Vecchio, H. Van der Auweraer,
Synthesis and sound quality evaluation of exterior
and interior aircraft noise, Aerospace Science and
Technology 12 (2008) 114-124.
[4] D. Berckmans, K. Janssens, H. Van der Auwer-
aer, P. Sas, and W. Desmet, Model-based synthe-
sis of aircraft noise to quantify human perception
of sound quality and annoyance, J. Sound Vib. 311
(2008) 1175-1195.
[5] H. Ploner-Bernard, A. Sontacchi, G. Lichteneg-
ger, S. Vssner, Sound-system design for a pro-
fessional full-ight simulator, presented at the 8th
International Conference on Digital Audio Effects
(DAFx05), 2005.
[6] S.T. Neely, J.B. Allen, Invertibility of room im-
pulse response, J. Acoust. Soc. Am. 66 (1979) 165
169.
[7] S.J. Elliott, P.A. Nelson, Multiple-point equaliza-
tion in a room using adaptive digital lters, J. Au-
dio Eng. Soc. 37 (1989) 899354.
[8] O. Kirkeby, F. Orduna, P.A. Nelson, H. Hamada,
Inverse ltering in sound reproduction, Measure-
ment + Control 26 (1993) 261266.
[9] F. Asanon, D.C. Swanson, Sound equalization in
enclosures using modal reconstruction, J. Acoust.
Soc. Am. 98 (1995) 20622069.
[10] P.A. Nelson, F. Orduna-Bustamante, H. Hamada,
Multichannel signal processing techniques in the
reproduction of sound, J. Audio Eng. Soc. 54
(1996) 973989.
[11] L. Fuster, A. Gonzlez, J.J. Lopez, P. Zuccarello,
Room compensation using multichannel inverse
lters for wave eld synthesis systems, Presented
at the 118th Audio Eng. Soc. Convention, 2005.
[12] E. Corteel, Equalization in an extended area using
multichannel inversion and wave eld synthesis, J.
Audio Eng. Soc. 54 (2006) 11401161.
[13] S. Bharitkar, C. Kyriakakis, Immersive audio signal
processing, Springer, New York, 2006.
[14] P. Basilio, J.J. Lpez, J. Escolano, L. Hrchens,
Multiactuator panels for wave eld synthesis:
Evolution and present developments, J. Audio
Eng. Soc. 58 (2010) 10451063.
[15] J. Nowak, M. Strau, Sound eld reproduction
analysis in a car based on microphone array mea-
surements, presented at the 48th Audio Eng.
Soc. International Conference: Automotive Audio,
2012.
[16] S. Cecchi, F. Bettarelli, F. Piazza, Multipoint
equalization for car audio systems, presented at
the 48th Audio Eng. Soc. International Conference:
Automotive Audio, 2012.
[17] P.-A. Gauthier, C. Camier, F.-A. Lebel, Y. Pasco, A.
Berry, Experiments of sound eld reproduction in-
side aircraft cabin mock-up, presented at the 133rd
Audio Eng. Soc. Convention, San Francisco, 2012.
[18] P.-A. Gauthier, C. Camier, O. Gauthier, Y. Pasco, A.
Berry, Aircraft sound environment reproduction:
Sound eld reproduction inside a cabin mock-up
using microphone and actuator arrays, Presented
at the 21st International Congress on Acoustics,
Montral, 2013.
[19] S. Cecchi, L. Palestini, P. Peretti, L Ramoli, F. Pi-
azza, Evaluation of a multipoint equalization sys-
tem based on impulse response prototype extrac-
tion, J. Audio Eng. Soc. 59 (2011) 110123.
AES 52
ND
INTERNATIONAL CONFERENCE, Guildford, UK, 2013 September 24
Page 10 of 10