Expression for the estimation of time-averaged three-dimensional acoustic
energy density spectral density
Running title: Expression for 3D acoustic energy density
Ben S. Cazzolato�
and Justin Ghan†
School of Mechanical Engineering, The University of Adelaide, SA 5005, Australia.
PACS numbers: 43.58.-e, 43.60.Qv
�
Address all correspondence to this author. Phone: +61 8 8303 5449. Fax: +61 8 8303 4367; Electronic address:
[email protected]†Electronic address: [email protected]
1
Abstract
This paper builds on earlier work by the same authors to derive an expression for the time-
averaged three-dimensional acoustic energy density in the frequency domain using the auto-
and cross-spectral densities of multiple microphone elements. Expressions for the most com-
mon geometric arrangements are derived and validated numerically. It is shown that in the
presence of uncorrelated noise the various geometric configurations produce significantly dif-
ferent energy density estimates, unlike the perfectly correlated case when all configurations
yield almost identical energy density estimates.
2
I. INTRODUCTION
In several articles [1–5] an expression for the acoustic energy density spectral density along
a single axis using the two-microphone technique has been derived. This frequency domain
expression enables the calculation of acoustic energy density spectral density using the auto-
spectra and cross-spectrum of the two microphone signals. The advantage of this approach, as
opposed to the traditional approach of first calculating the mean pressure and particle velocity,
is that the energy density may be measured with a two-channel spectrum analyser with no
additional circuitry apart from the two microphone signals.
In this paper, the acoustic energy density estimate using spectral methods will be extended
to three dimensions. Several geometric configurations and spectral density estimation formula-
tions are discussed. Expressions involving the minimum number of terms necessary to estimate
the energy density spectral density are also derived to facilitate the calculation of the three-
dimensional energy density using a two-channel spectrum analyser. The resulting analytical
expressions are verified using a numerical simulation in the time domain. It is shown that under
certain conditions the different spectral density estimation formulations may vary significantly
from each other.
II. PHYSICAL 3D ENERGY DENSITY CONFIGURATIONS
Since the measurement of the three-dimensional acoustic energy density requires the same
set of measurements as the acoustic intensity, namely acoustic pressure and the three orthogonal
particle velocities, then any sensor capable of measuring one quantity can also be used to mea-
sure the other. There exists several different geometrical arrangements for which 3D acoustic
intensity and energy density may be measured[6–8]. This paper will loosely focus on the two
most common arrangements found in commercially available 3D sound intensity probes. These
are the four-microphone tetrahedral arrangement like those found in the Ono Sokki Tetra-phone
MI-6420 (Figure 1 a) and the six-microphone arrangement like those of the Brüel & Kjær Type
5356 Intensity Probe (Figure 1 b) or the GRAS Vector Intensity Probe (Figure 1 c).
A number of variants of the four-microphone sensor exist, the specific details of which are
discussed in detail in the following sections. A slight variant of the six-microphone sensor is the
3
seven-microphone sensor[9]. Both the six and seven-microphone sensors use three pairs of op-
posed closely spaced microphones to calculate the three orthogonal particle velocities. However
unlike its six-microphone sibling, which must estimate the pressure at the centre of the sensor
using a finite sum, the seven-microphone sensor uses the seventh microphone at its geometric
centre to directly measure the pressure. This is the most accurate of all the geometries[8] and
consequently has been used to accurately measure three-dimensional impulse response func-
tions for surround sound systems[9].
The bias errors, both inherent and systematic, for most of the three-dimensional geometric
arrangements have been documented by Cazzolato and Hansen [8], and by Parkins et al. [10]. It
should be noted that Parkins et al. [10] also investigated the errors arising from a six-microphone
arrangement with the microphones mounted on the surface of a rigid sphere (rather than the
open configurations shown in Figures 1 b&c). This mounted arrangement was employed in an
attempt to use the effects of diffraction caused by the sphere [11] to produce more favorable
bias conditions, and in doing so reduce the inherent errors. The effect of acoustic diffraction
around the rigid sphere caused the effective microphone spacing to be 50% greater than the
actual microphone spacing. When calculating the particle velocity estimate when using the
finite difference approximation, the effective microphone spacing should be used, rather that
the physical spacing.
III. ANALYTICAL DERIVATION
The analytical derivation in this paper closely follows the derivation by Ghan et al. [5]. The
instantaneous acoustic energy density, ED�t � , at a point is defined as the sum of the acoustic
potential energy density and the acoustic kinetic energy density at that point, given by [7]
ED�t ��� p2 � t �
2ρc2 �ρ � v � t ��� 2
2 � (1)
where p�t � is the instantaneous pressure and � v � t ��� is the magnitude of the total instantaneous
particle velocity, at that point, c is the speed of sound, and ρ is the mean density of the fluid. In
practice the total particle velocity is estimated using the three orthogonal terms (v�t ��� ivx
�t � �
4
jvy�t � � kvz
�t � ), therefore Equation (1) can be rewritten as
ED�t � � p2 � t �
2ρc2 � ∑i � x � y � z
ρv2i
�t �
2� (2)
For a three-dimensional sensor the pressure is typically estimated using an average of all the
microphone elements, thus for an sensor containing n microphones the pressure estimate is
p�t � � 1
n ∑ni � 1 pi
�t � � (3)
The measurement of the particle velocity can be obtained through a finite difference approx-
imation using two microphones located on one of the three orthogonal axes, thus for a micro-
phone separation distance of 2h, the particle velocity components are approximated by [7]
vi�t � � 1
2ρh
� t
� ∞ � pi1
�τ �� pi2
�τ �� dτ � (4)
In the following sub-sections frequency domain expressions are derived for the single-sided
time-averaged acoustic energy density estimate for a number of three-dimensional sensor ge-
ometries. The expressions are based on single-sided spectral densities so they may be directly
used with real-time spectrum analysers which usually calculate single-sided spectra.
A. Multichannel Expressions
In the case where a multi-channel spectrum analyser (such as a Brüel & Kjær Pulse system) is
available, it is possible to obtain exact frequency domain expressions in real time for any number
of three-dimensional energy density sensors. Expressions for the most common geometries are
derived in the following sections. It should be noted that the following estimates for energy
density can be calculated using a two-channel spectrum analyser by recording all auto and
cross-spectra, and then post processing the spectra in a software package such as Matlab or
Excel.
1. Cubic Four-microphone sensor with pressure averaging
The following analysis is for the four-microphone configuration as shown in Figure (2 a).
Three microphones (marked 2, 3 and 4), each forming one of the ends of the 3 orthogonal axes
5
(marked x, y and z respectively). This particular geometric configuration is known as a “cubic”
arrangement since the microphones lie in the corner of a cube, where the origin microphone
is in one corner of the cube and the three other microphones are located on the corresponding
vertices of the cube.
The specifics of this particular four-microphone sensor when used for energy density sens-
ing is discussed in detail by Cazzolato and Hansen [8]. This energy density sensor arrangement
has been widely applied to the characterisation of reverberant sound fields [12–20]. A fre-
quency domain expression for the sound intensity spectral density from this arrangement has
been derived[21], although to the authors knowledge, no such equivalent expressions exists for
energy density.
The pressure estimate is the mean of the pressures measured by all four microphones. The
particle velocities are calculated using Equation (4). The distance from the “origin” microphone
(microphone 1) to the other microphones was 2h. It should be noted that this particular sensor
does not have a unique acoustic centre [8, 21, 22]. This means that the positions in which
the particle velocities are estimated are not coincident with each other or the location of the
pressure estimate. The alternative four-microphone arrangements shown in Figures (2 b & c)
and discussed in more detail in Sub-sections III A 2 and III A 3 have a unique acoustic centre
which is located at the centre of gravity of the microphones.
The total time-averaged acoustic energy density is given by[5]
ED � limT � ∞
1T
E
� � T
0
�1
2ρc2 � p2 � t � �ρ2 � � v � t ��� 2 � dt � (5)
where the operator E is the expectation and T is the record length.
Using Parseval’s theorem [23] it can be shown that the three-dimensional double-sided time-
averaged acoustic energy density spectral density is,
ED�ω � � lim
T � ∞
1T
E
�1
2ρc2 � �P � ω � T � � 2 � ρ2 � �Vx
�ω � T ���
2 �ρ2 � �Vy
�ω � T ���
2 �ρ2 � �Vz
�ω � T ���
2 � (6)
where P�ω � T � and Vi
�ω � T � are the Fourier transforms of the pressure and velocity estimates.
6
From Equations (3) and (4),
P�ω � T �
� P1�ω � T � � P2
�ω � T � � P3
�ω � T � � P4
�ω � T �
4 � (7)
Vx�ω � T �
� 12ρh � P1
�ω � T � � P2
�ω � T �
jω � (8)
Vy�ω � T �
� 12ρh � P1
�ω � T � � P3
�ω � T �
jω � (9)
Vz�ω � T �
� 12ρh � P1
�ω � T � � P4
�ω � T �
jω� (10)
Equation (6) is valid for all three-dimensional sensor configurations, however the expressions
for the pressure and particle velocity differ for each sensor. Evaluating the expectation operator
in Equation (6) it can be shown that the single-sided time-averaged energy density spectral
density estimate for the four-microphone energy density sensor with pressure averaging is given
by
ED1-sided�ω � 4̄ � � 1
32ρc2 �3
8ρω2h2 � �G11 � �
� 132ρc2 �
18ρω2h2 � �
G22 � G33 � G44 �
� � 132ρc2 � 1
8ρω2h2 � �2Re � G12 � 2Re �G13 � 2Re � G14 � (11)
� � 132ρc2 � �
2Re � G23 � 2Re � G24 � 2Re � G34 �where Gxx is the single-sided auto-spectral density function for the pressure signal measured
by microphone x, and Re � Gxy is the real part of the single-sided cross-spectral density function
for microphone signals x and y. The subscript 4̄ represents the pressure estimate is obtained by
averaging all four microphone signals.
2. Regular Tetrahedral Four-microphone sensor with pressure averaging
Santos et al. [22] described a four-microphone arrangement in which the “acoustic centre”
is the same as the geometric centre of the four microphones (see Figure 2b). Their design
used microphones mounted on the corners of a regular tetrahedron. They combined pairs of
the microphone signals to estimate the pressure at the midpoint of each edge of the tetrahedron.
These six “virtual points” are then used to calculate the pressure gradient at the geometric centre
of the sensor.
7
This four-microphone sensor configuration uses the mean of all four microphones to provide
the pressure estimate. The double-sided time-averaged acoustic energy density spectral density
is given by Equation (6) where[22],
P�ω � T �
� P1�ω � T � � P2
�ω � T � � P3
�ω � T � � P4
�ω � T �
4 � (12)
Vx�ω � T �
� 1�
2ρh jω � � P1�ω � T � � P3
�ω � T �
2� P2
�ω � T � � P4
�ω � T �
2 � � (13)
Vy�ω � T �
� 1�
2ρh jω � � P1�ω � T � � P4
�ω � T �
2� P2
�ω � T � � P3
�ω � T �
2 � � (14)
Vz�ω � T �
� 1�
2ρh jω � � P1�ω � T � � P2
�ω � T �
2� P3
�ω � T � � P4
�ω � T �
2 � � (15)
It can be shown that the single-sided time-averaged energy density spectral density estimate
for the regular tetrahedron four-microphone energy density sensor with pressure averaging is
given by
ED1-sided�ω � Tetrahedron
� � 132ρc2 �
316ρω2h2 � �
G11 � G22 � G33 � G44 �
� � 132ρc2 � 1
16ρω2h2 � �2Re � G12 � 2Re � G13 � 2Re � G14 � (16)
�� 1
32ρc2 � 116ρω2h2 � �
2Re � G23 � 2Re � G24 � 2Re � G34 � �
3. Ono Sokki Four-microphone sensor with pressure averaging
The Ono Sokki four-microphone sensor [24–27] shown in Figure (1 a) also has the micro-
phones arranged in a regular tetrahedron. The weighting of the microphone signal differs from
that employed by Santos et al. [22] resulting in a different alignment of the three measurement
axes (see Figure 2c).
Using the approach employed by the Ono Sokki corporation[24–27] when processing the
data, the double-sided time-averaged acoustic energy density spectral density is given by Equa-
8
tion (6) where [26],
P�ω � T �
� P1�ω � T � � P2
�ω � T � � P3
�ω � T � � P4
�ω � T �
4 � (17)
Vx�ω � T �
� 12ρh jω � � P3
�ω � T � � P2
�ω � T � � � (18)
Vy�ω � T �
� 1
2�
3ρh jω � � 2P1�ω � T �� P2
�ω � T �� P3
�ω � T � � � (19)
Vz�ω � T �
� 1
2�
6ρh jω � � 3P4�ω � T �� P1
�ω � T �� P2
�ω � T �� P3
�ω � T � � � (20)
It can be shown that the single-sided time-averaged energy density spectral density estimate
for the regular tetrahedron four-microphone Ono Sokki energy density sensor with pressure
averaging is identical to that of the regular tetrahedral sensor given by Equation (16). This is to
be expected since the pressure and total velocity estimate is the same for both cases, only the
orientation with respect to the coordinate system has changed.
4. Six-microphone sensor with pressure averaging
Figure 2d shows an energy density sensor containing six microphones, with three pairs form-
ing three orthogonal axes. This sensor configuration uses the mean of all six microphones sig-
nals to provide the pressure estimate. The double-sided time-averaged acoustic energy density
spectral density is given by Equation (6) where,
P�ω � T �
� P1�ω � T � � P2
�ω � T � � P3
�ω � T � � P4
�ω � T � � P5
�ω � T � � P6
�ω � T �
6 � (21)
Vx�ω � T �
� 12ρh � P1
�ω � T � � P2
�ω � T �
jω � (22)
Vy�ω � T �
� 12ρh � P3
�ω � T � � P4
�ω � T �
jω � (23)
Vz�ω � T �
� 12ρh � P5
�ω � T � � P6
�ω � T �
jω� (24)
Therefore it can be shown that the single-sided time-averaged energy density spectral density
9
estimate for the six-microphone energy density sensor with pressure averaging is given by
ED1-sided�ω � 6̄ � � 1
72ρc2 �1
8ρω2h2 � �G11 � G22 � G33 � G44 � G55 � G66 �
� � 172ρc2 � 1
8ρω2h2 � �2Re � G12 � 2Re � G34 � 2Re � G56 � (25)
� � 172ρc2 �
�6
∑i � 3
�2Re � G1i � 2Re � G2i � �
6
∑i � 5
�2Re �G3i � 2Re �G4i ��� �
If the six microphones are mounted in a rigid sphere as described by Parkins et al. [10], then
the substitution h � 32r should be made, where r is radius of the sphere.
5. Seven-microphone sensor
With reference to the seven-microphone configuration in Figure 2(d), the additional “origin”
microphone (p0 � , located at the geometric centre of the six-microphone sensor, is used to mea-
sure the pressure for the sensor. By measuring the pressure directly rather than interpolating the
pressure measured by six microphones, the error in pressure associated with the finite-sum is
avoided.
The single-sided time-averaged energy density spectral density estimate for the seven-
microphone energy density sensor is given by
ED1-sided�ω � 7 � � 1
2ρc2 � �G00 �
� � 18ρω2h2 � �
G11 � G22 � G33 � G44 � G55 � G66 � (26)
� � 18ρω2h2 � �
2Re � G12 � 2Re � G34 � 2Re � G56 �
where G00 is the auto-spectral density function of the origin microphone signal. Note the sig-
nificant reduction in terms compared to the six-microphone sensor.
10
B. Reduced Order Expressions
Although it is desirable to calculate all the necessary auto- and cross-spectra simultaneously
on a multi-channel spectrum analyser, two-channel spectrum analysers (such as the HP 35665A)
may be used to measure the auto- and cross-spectral density estimates. The number of two-
channel measurements required is a function of the number of microphones and the particular
spectral density formulation. To calculate the energy density spectral density estimates using
Equations (11) and (16) requires 6 individual two-channel measurements and Equations (25)
and (26) would require 15 and 4 individual two-channel measurements respectively.
Clearly the large number of two-channel measurements for all but the seven-microphone
sensor would require a significant effort to obtain experimentally. It is therefore desirable to
have expressions requiring fewer measurements. The minimum number of separate measure-
ments is limited by the need for the cross-spectral density estimates used by the finite-difference
expression in the kinetic energy density estimate. In the following sections a number of ap-
proximations using less terms are derived to facilitate calculation using two-channel spectrum
analysers.
1. Cubic four-microphone sensor using origin microphone
An alternative formulation for the four-microphone sensor is to use the pressure at the “origin
microphone” of the cubic arrangement as the pressure estimate rather than the mean pressure
sensed by the four microphones as used previously. The particle velocities may be calculated
as before using Equations (8) to (10). This arrangement has been analysed by Cazzolato and
Hansen [8] and used experimentally for active noise control [28, 29].
For the case of the four-microphone energy density sensor with the pressure estimate given
by the microphone at the origin (p1), the single-sided time-averaged energy density spectral
11
density estimate is given by
ED1-sided�ω � 40
� � 12ρc2 �
38ρω2h2 � �
G11 �
� � 18ρω2h2 � �
G22 � G33 � G44 � (27)
� � 18ρω2h2 � � � 2Re � G12 � 2Re � G13 � 2Re � G14 � �
This formulation only requires three individual sets of measurements as compared to six for
Equations (11) and (16).
2. Six-microphone sensor - Reduced order
An approach commonly employed for three-dimensional sound intensity measurements [6,
30] is to use three one-dimensional probes. This approach was used by Parkins et al. [10] when
calculating the energy density for a six-microphone sensor. They used the mean of three two-
channel pressure estimates (obtained from the finite pressure sum of each of the three orthogonal
axes) to provide an overall pressure estimate. Thus, when using this approach the square of the
pressure magnitude in Equation (6) is given by
�P � ω � T ���2 � 1
3 ������P1�ω � T � � P2
�ω � T �
2 ����2
� ����P3�ω � T � � P4
�ω � T �
2 ����2
� ����P5�ω � T � � P6
�ω � T �
2 ����2 � �(28)
For this particular two-channel approximation, the single-sided time-averaged energy den-
sity spectral density estimate is given by
ED1-sided�ω � 63̄
� � 124ρc2 �
18ρω2h2 � �
G11 � G22 � G33 � G44 � G55 � G66 �
� � 124ρc2 � 1
8ρω2h2 � �2Re �G12 � 2Re � G34 � 2Re � G56 � � (29)
This formulation also only requires three individual sets of two-channel measurements com-
pared to 15 two-channel measurements needed when using Equation (25). This probably ex-
plains why this particular approach was used by Parkins et al. [10] when estimating the 3D
12
energy density using a two-channel spectrum analyser. It should also be noted that the expres-
sion derived by Parkins et al. [10] differs from Equation (29).
3. Cubic four-microphone sensor - Reduced order
The method used in Subsection III B 2 has been applied to a cubic four-microphone sensor
to derive an expression for the energy density using a reduced number of two-channel measure-
ments. Thus the square of the pressure magnitude in Equation (6) is given by
�P � ω � T � �2 � 1
3 � ����P1�ω � T � � P2
�ω � T �
2 ����2
� ����P1�ω � T � � P3
�ω � T �
2 ����2
� ����P1�ω � T � � P4
�ω � T �
2 ����2 �
(30)
For this four-microphone energy density sensor approximation, the single-sided time-
averaged energy density spectral density estimate is given by
ED1-sided�ω � 43̄
� � 18ρc2 �
38ρω2h2 � �
G11 �
� � 124ρc2 �
18ρω2h2 � �
G22 � G33 � G44 � (31)
� � 124ρc2 � 1
8ρω2h2 � �2Re �G12 � 2Re � G13 � 2Re �G14 � �
This formulation only requires three individual sets of measurements as compared to six
for Equations (11) and (16). It is interesting to note that Equation (31) contains the same
auto and cross-spectral density terms as Equation (27), however the weights of the individual
terms differ, therefore likely producing quite different results under certain conditions, which is
discussed further in Section III C.
The reason for the differing weights in the cubic four-microphone expressions given by
Equations (11), (27) and (31) is the location at where the pressure estimate is made. For Equa-
tion (11), the geometric centre of the four microphones is used, which is located a distance ofh � 3
2 from the origin microphone along a line normal to the plane intersecting the other three
points [28]. The pressure estimate using Equation (27) is obviously made at the origin micro-
phone. The estimate of the pressure using Equation (31) is made at a distance of h� 3
from the
13
origin microphone, which is half way between the origin microphone and the plane intersecting
the other three points.
Each component of the velocity is estimated halfway between the origin microphone and one
of the axis microphones. Given that the distance between the location of the pressure estimate
and velocity estimates using Equation (31) is slightly less that obtained by using Equation (11),
namely h � 2� 3
� 0 � 8165h compared to h � 32 � 0 � 8660h, it could be argued that the former is the
most accurate formulation. However, confirmation of this is beyond the scope of this paper.
C. Discussion
It was shown by Cazzolato and Hansen [8] that in the absence of uncorrelated noise and for
small non-dimensional microphone separations (2kh � 1) that there is very little difference in
the accuracy of the energy density estimate given by the configurations in Sub-sections III A 1,
III A 4, III A 5 and III B 1.
It is beyond the scope of this article to determine which of the expressions in Sections III A
and III B are the most accurate since accuracy is a function of the sound field, orientation and
also more importantly the coherence between the individual microphone elements. Poor co-
herence between microphone elements may be caused by any number of factors, such as poor
signal to noise ratios when there are low sound pressure levels [31] or when the FFT length is
insufficient resulting in resolution bias errors [23]. Even so, some general comments may be
made. If one takes the worst case scenario when there is no correlation between the microphone
elements, and assuming that the spectral densities at the microphones are equal (to Gpp), then
the following expressions can be obtained:
ED1-sided�ω � 4̄ � � 1
8ρc2 �3
4ρω2h2 � Gpp (32)
ED1-sided�ω � Tetrahedron � � 1
8ρc2 �3
4ρω2h2 � Gpp (33)
ED1-sided�ω � 6̄ � � 1
12ρc2 �3
4ρω2h2 � Gpp (34)
ED1-sided�ω � 7 � � 1
2ρc2 �3
4ρω2h2 � Gpp (35)
14
ED1-sided�ω � 40 � � 1
2ρc2 �3
4ρω2h2 � Gpp (36)
ED1-sided�ω � 43̄
� � 14ρc2 �
34ρω2h2 � Gpp (37)
ED1-sided�ω � 63̄
� � 14ρc2 �
34ρω2h2 � Gpp � (38)
Although the likelihood of the microphone signals being uncorrelated is quite unrealistic, it
can be seen that the estimates of the energy density spectral densities derived above can be quite
different when the microphone elements are poorly correlated.
IV. SIMULINK SIMULATION
Using the same technique employed by Ghan et al. [5] several Simulink models were created
to validate the three-dimensional energy density expressions derived above. The models were
used to obtain the time-averaged acoustic energy density estimate both by directly calculating
the average pressure and particle velocity, and also by using the expressions derived in Sections
III A and III B. Figure 3 shows the Simulink model for the four-microphone sensor without
pressure averaging. This particular model was chosen to illustrate the models as this is the
simplest of all the formulations.
A simulated acoustic environment was created with a monopole white noise point source
in free space. The four microphone pressure signals were then calculated using the following
expression for the transfer function from a monopole point source to a sensor at distance r:
G�s � � 1
re� r
c s
� (39)
where c � 343m � s� 1 is the speed of sound. These transfer functions were implemented using
a second order Padé approximation in series with a gain. A number of different sensor orien-
tations were trialled to ensure that the spectral density estimates held for any orientation. For
the results presented below the origin microphone was placed at distance r1 � 5 � 00m from the
monopole noise source and the remaining three microphones (separation distance, 2h � 50mm)
15
are r2 � 3&4 � 5 � 0289m from the source. The source produces continuous white noise of 1 � 0Pa
at 1m (94dBre20µPa). The density of air was ρ � 1 � 21kg � m� 3.
Figure 4 shows the sub-model used to calculate the energy density using the traditional
time-domain approach. The instantaneous pressure and particle velocities were estimated using
Equations (3) and (4) respectively (as shown in Figure 5). The velocity values were multiplied
by ρc so that all the outputs were dimensionally consistent. They were sampled at 1000Hz, and
a frequency spectrum of each was obtained by performing a 512 point FFT (see Figure 4). The
potential and kinetic energy auto-spectral densities were calculated, then summed and scaled
by the factor 12ρc2 to produce the time-domain based acoustic energy density spectral density
estimate (see Figure 4).
Figure 6 shows the implementation in the frequency domain. To compute an estimate di-
rectly in the frequency domain, a spectrum of each of the microphone pressure readings was
obtained by performing a 512 point FFT (see Figure 6). The auto- and cross-spectral densities
were then calculated, and Equation (27) was applied to produce the frequency-domain based
acoustic energy density spectral density estimate.
A comparison of the results is shown in Figure 7. The acoustic energy density spectral
densities obtained via each method of calculation are identical, verifying the derived Equation
(27). This same method of comparison was repeated for the expressions in Section III A. For
the two remaining reduced order frequency domain expressions shown in Equations (29) and
(31) it was not possible to obtain a time domain equivalent. Instead the results were compared
against the multi-channel equivalents shown in Equations (25) and (11) respectively. Although
the frequency domain spectral estimates were not exactly identical to the time domain esti-
mates, they were extremely close, with the difference being attributed to the differing weighted
contributions of the individual spectra and cross-spectra.
V. CONCLUSIONS
Several expressions for the time-averaged three-dimensional acoustic energy density esti-
mate have been derived using the auto- and cross-spectral densities between several closely
spaced microphones. These were validated numerically using a time domain simulation in
Simulink. The results obtained using these expressions were identical to the results using the
16
traditional method of using the weighted sum of the squared pressure and particle velocities. It
is shown that in the presence of uncorrelated noise the expressions may result in significantly
different spectral density estimates. Expressions containing a reduced number of terms were
also derived to facilitate the calculation of the three-dimensional energy density spectral den-
sity using two-channel spectrum analysers.
Acknowledgments
The authors gratefully acknowledge the financial support for this work provided by the Aus-
tralian Research Council.
17
[1] G. Elko. Frequency domain estimation of the complex acoustic intensity and acoustic energy den-
sity. PhD thesis, Pennsylvania State University, 1984.
[2] G.C. Steyer. Spectral methods for the estimation of acoustic intensity, energy density, and surface
velocity using a multimicrophone probe. PhD thesis, The Ohio State University, 1984.
[3] G.W. Elko. Simultaneous measurement of the complex acoustic intensity and the acoustic energy
density. Proceedings of the 2nd International Congress on Acoustic Intensity, pages 69–78, 1985.
[4] T.E. Vigran. Acoustic intensity - energy density ratio: An index for detecting deviations from ideal
field conditions. Journal of Sound and Vibration, 127(2):343–351, 1988.
[5] J. Ghan, B.S. Cazzolato, and S.D. Snyder. Expression for the estimation of time-averaged acoustic
energy density using the two-microphone method. Journal of the Acoustical Society of America,
113(5):2404–2407, 2003.
[6] G. Rasmussen. Measurement of vector fields. Proceedings of the 2nd International Congress on
Acoustic Intensity, pages 53–58, 1985.
[7] F. Fahy. Sound Intensity. E&FN Spon, London, 2nd edition, 1995.
[8] B.S. Cazzolato and C.H. Hansen. Errors arising from three-dimensional acoustic energy density
sensing in one-dimensional sound fields. Journal of Sound and Vibration, 236(3):375–400, 2000.
[9] A. Farina and L. Tronchin. 3D impulse response measurements on S. Maria del Fiore Church,
Florence, Italy. In Proceedings of ICA98 - International Conference on Acoustics, pages 26–30,
Seattle, Wahington State, 1998.
[10] J.W. Parkins, S.D. Sommerfeldt, and J. Tichy. Error analysis of a practical energy density sensor.
Journal of the Acoustical Society of America, 108(1):211–222, 2000.
[11] G.W. Elko. An acoustic vector-field probe with calculable obstacle bias. In Proceedings of Noise-
Con 91, pages 525–532, 1991.
[12] B. Morton. A system for the measurement of acoustic energy. Master’s thesis, The University of
Texas at Austin, 1976.
[13] B. Morton and E.L. Hixson. System for the measurement of acoustic energy density. Journal of
the Acoustical Society of America, 60(Sup 1):S59, 1976.
18
[14] M. Schumacher and E.L. Hixson. A transducer and processing system to measure total acoustic
energy density. Journal of the Acoustical Society of America, 74(S1):S62, 1983.
[15] M. Schumacher. A transducer and processing system for acoustic energy density. Master’s thesis,
The University of Texas at Austin, 1984.
[16] J.A. Moryl. A study of acoustic energy density in a reverberation room. Master’s thesis, The
University of Texas at Austin, 1987.
[17] J.A. Moryl and E.L. Hixson. A total acoustic energy density sensor with applications to energy
density measurement in a reverberation room. In Proceedings of Inter-Noise 87, pages 1195–1198,
1987.
[18] M.H.W. Budhiantho. Acoustic velocity related distributions. PhD thesis, The University of Texas
at Austin, 1997.
[19] M.H.W. Budhiantho and E.L. Hixson. Acoustic velocity related distributions. Report, Electroa-
coustics Research Laboratory, The University of Texas at Austin, 1997.
[20] M.H.W. Budhiantho and E.L. Hixson. A proposal to adopt Maxwell distribution as a measure of
acoustic field diffusness in a reverberation room. Journal of the Acoustical Society of America, 105
(2):935, 1998.
[21] J. Vandenhout, P. Sas, and R. Snoeys. Measurement, accuracy and interpretation of real and imagi-
nary intensity patterns in the near field of complex radiators. Proceedings of the 2nd International
Congress on Acoustic Intensity, pages 121–128, 1985.
[22] I.M.C. Santos, C.C. Rodrigues, and J.L. Bento Coelho. Measuring the three-dimensional acoustic
intensity vector with a four-microphone probe. Proceedings of Inter-Noise 89, pages 965–968,
1989.
[23] J.S. Bendat and A.G. Piersol. Random Data - Analysis and Measurement Procedures. John Wiley
& Sons, New York, 2nd edition, 1986.
[24] H. Suzuki, S. Oguro, and M. Anzai. A sensitivity compensation method for a three-dimensional
sound intensity probe. Proceedings of Inter-Noise 94, pages 1963–1966, 1994.
[25] K. Yamaguchi, K. Hori, T. Tananka, and M. Anzai. The development of instruments for the mea-
surement of sound intensity using 3-dimensional microphone probe. Proceedings on Inter-Noise
94, pages 1967–1970, 1994.
19
[26] M. Suzuki, H. Anzai, S. Oguro, and T. Ono. Performance evaluation of a three dimensional intensity
probe. Journal of the Acoustical Society of Japan (E), 16:233–238, 1995.
[27] H. Suzuki, S. Oguro, and T. Ono. A sensitivity correction method for a three-dimensional sound
intensity probe. Journal of the Acoustical Society of Japan (E), 21(5):259–265, 2000.
[28] B.S. Cazzolato. Sensing systems for active control of sound transmission into cavities. Ph.D.
Dissertation, The University of Adelaide, March 1999.
[29] B.S. Cazzolato and C.H. Hansen. Active control of enclosed sound fields using three-axis energy
density sensors: Rigid walled enclosures. International Journal of Acoustics and Vibration, 8(1):
39–51, March 2003.
[30] T. Loyau and J.-C. Pascal. Statistical errors in the estimation of the magnitude and direction of the
complex acoustic intensity vector. Journal of the Acoustical Society of America, 97(5):2942–2962,
1995.
[31] F. Jacobsen. Sound intensity measurement at low levels. Journal of Sound and Vibration, 166(2):
195–207, 1989.
20
List of Figures
1 Commercially available sound intensity sensors suitable for 3D energy density
measurement. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2 Multi-microphone three-axis energy density sensors. . . . . . . . . . . . . . . 23
3 Simulink model of a four-microphone energy density sensor. The “Time Do-
main Calculation” sub-model is shown in Figure 4, and the “Frequency Domain
Calculation” sub-model is shown in Figure 6. . . . . . . . . . . . . . . . . . 24
4 Simulink model for the calculation of the time-averaged acoustic energy density
estimate as the weighted sum of the auto-spectral densities of the pressure and
the three particle velocities. The block titled Energy Density Sensor (see Figure
5) calculates the time-domain quantities. . . . . . . . . . . . . . . . . . . . 25
5 Simulink model for the calculation of the pressure and the three particle veloc-
ities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
6 Simulink model for the calculation of the time-averaged acoustic energy density
estimate using the derived expression in terms of the auto- and cross-spectral
densities of the four pressure readings. . . . . . . . . . . . . . . . . . . . . 27
7 Time-averaged acoustic energy density estimates using the time domain method
and the frequency domain expression. . . . . . . . . . . . . . . . . . . . . . 28
21
(a) Ono Sokki Tetra-phone MI-6420.
Photo courtesy of Ono Sokki
(b) B&K Type 5356. Photo
courtesy of Brüel and Kjær
(c) GRAS Vector
Intensity Probe. Photo
courtesy of GRAS
Figure 1: Commercially available sound intensity sensors suitable for 3D energy density measurement.
22
1 2
4
3
x
y
z
2h
2h
2h
(a) Cubic four-microphone sensor
h
h
h
h
hh
h
h
1
2
3
4y
z
x
(b) Four-microphone regular tetrahedon
sensor. Small black circles represent the
virtual microphone locations.
2h
2h
2h
2h
3
2
1
4y
z
x
(c) Four-microphone Ono Sokki sensor
y
hh
h
h
h
h
x
z
12
3
45
6
0
(d) Six-microphone and
seven-microphone sensor
Figure 2: Multi-microphone three-axis energy density sensors.
23
Figure 3: Simulink model of a four-microphone energy density sensor. The “Time Domain Calculation”
sub-model is shown in Figure 4, and the “Frequency Domain Calculation” sub-model is shown in Figure
6.
24
Figure 4: Simulink model for the calculation of the time-averaged acoustic energy density estimate as the
weighted sum of the auto-spectral densities of the pressure and the three particle velocities. The block
titled Energy Density Sensor (see Figure 5) calculates the time-domain quantities.
25
Figure 5: Simulink model for the calculation of the pressure and the three particle velocities.
26
Figure 6: Simulink model for the calculation of the time-averaged acoustic energy density estimate using
the derived expression in terms of the auto- and cross-spectral densities of the four pressure readings.
27
0 50 100 150 200 250 300 350 400 450 500−85
−80
−75
−70
−65
−60
−55
−50
Frequency (Hz)
Aco
ustic
Ene
rgy
Den
sity
(dB
re
1 J/
m3 )
Time Domain MethodFrequency Domain Method
Figure 7: Time-averaged acoustic energy density estimates using the time domain method and the fre-
quency domain expression.
28