PAPERS

Design of a Highly Directional Endfire

Loudspeaker Array*

MARINUS M. BOONE, AES Fellow

([email protected])

Laboratory of Acoustical Imaging and Sound Control,

Delft University of Technology, 2600 GA, Delft, The Netherlands

AND

WAN-HO CHO AND JEONG-GUON IH, AES Member

([email protected]) ([email protected])

Center for Noise and Vibration Control, Department of Mechanical Engineering,

Korean Advanced Institute of Science and Technology, Daejeon 305-701, Korea

The applicability of a loudspeaker line array, of which the main directivity is in the length
direction of the array, is discussed. Hence this acoustic array aims at the endfire beamforming
direction, resulting in a “spotlight” of sound in a preferred direction. Optimized beamforming
techniques are employed, which have been developed earlier for the reciprocal problem of
directional microphone arrays. Effects of the design parameters of the loudspeaker array system
on its performance have been investigated. It is shown that the stability factor of the optimized
beamformer can be a useful parameter to control the directional characteristics of the array. In
addition the effect of mutual interaction between individual loudspeaker elements in the array
system has been considered to reduce the difference between designed and actual performance
of the array system. A prototype constant-beamwidth array system has been tested by simula-
tion and measurement and the results support findings in a parametric analysis.

0 INTRODUCTION A lot of research work on the characteristics of transducer

array systems has already been conducted in the field of
Directional loudspeaker systems have been studied in- antennas, which are for electromagnetic use, and for micro-
tensively by many researchers because of their useful phone systems. Various spatial filtering methods have been
application, such as a column array that addresses sound devised from this previous work, and the method is in
information in the plane of the listeners’ ears. For a single general called beamforming [1]. For microphone arrays,
loudspeaker unit the directional characteristics depend on directivity control has been one of the major research
the Helmholtz number, which is related to the size of the topics, in particular on the design of a constant beamwidth
radiating membrane and the wavelength. When multiple- over a broad frequency range [2], [3]. Some array designs
loudspeaker units are concerned, forming a so-called have been applied to hearing aids [4], [5] to obtain a highly
loudspeaker array, the directional characteristics depend directive characteristic. The representative methods to ob-
on the placement of the loudspeaker units within the array tain highly directive beam patterns can be summarized by
and on the filtering of the audio signals that are provided three methods—the delay and sum technique [1], the gradi-
to the loudspeakers. ent method (such as Jacobi arrays) [6], and optimal beam-
forming [7], [8]. Among these, the optimal beamforming
method is known to deliver a relatively high directivity as
* This work was presented in part at the 122nd Convention of compared to other methods [9], [10]. The solution for opti-
the Audio Engineering Society, Vienna, Austria, 2007 May 5–8. mal beamforming was suggested in the middle of the 20th
Manuscript received 2008 February 22; revised 2009 February century. However, at that time it was only considered to be
11 and April 6. of academic interest because of noise problems associated

J. Audio Eng. Soc., Vol. 57, No. 5, 2009 May 309

BOONE ET AL. PAPERS

with the available equipment [11], but also because the of the ith transducer, and Gn, n ¼ 1, 2, . . ., N, denotes
implementation of the required filters was not possible with the directional factor of each transducer. In the case of
the analog equipment at that time. A constrained solution to an endfire array the propagation vector W is described
solve the noise problem was suggested by Gilbert and by [10]
Morgan [12], and with the advent of modern digital signal 2 ! 3
processing equipment, this technique has been applied to 1 ð; ; !Þej c z1 cos 
6 2 ð; ; !Þej c z2 cos  7
!

many practical situations. One of these applications is the 6 7

Wð; ; !Þ ¼ 6 .. 7 (3)
optimized beamforming that has been implemented in the 4 . 5
!
hearing glasses [13]. These are highly directive hearing aids N ð; ; !Þej c zN cos 
mounted in the arms of a pair of spectacles, with four
microphones at each side. where zi is the longitudinal position of the ith transducer.
Theoretically, based on the reciprocity principle of the A broadside array concentrates the acoustic information
acoustic field, the theory for a microphone array system onto a two-dimensional plane. On the other hand, an end-
can also be applied to a loudspeaker array system. In this fire array focuses the sound on a one-dimensional line.
study the theory to optimize the directivity of microphone For that reason an endfire array is more suitable for the
arrays is reviewed and then applied to a loudspeaker array objective of this study than a broadside array to obtain a
system. Moreover, a modification process has been highly directive beam pattern. The geometry of an endfire
adapted to overcome the problems induced by the differ- array is shown in Fig. 2.
ence between microphones and loudspeakers. The major To compare the acoustic performances of array sys-
concern in this investigation is the optimization of the tems, many evaluation parameters have been suggested,
design parameters of the optimal beamformer. These are of which the directivity factor is the most important. For
directivity index and noise sensitivity for microphone loudspeaker systems the directivity factor is defined by
arrays and directivity index and power index for loud- the ratio of the acoustic intensity at a far-field point in
speaker arrays. In these optimizations the stability factor
that will be introduced in Section 1.2 will play an impor-
tant role.

1 BASIC THEORY
1.1 Basics of the Array System
Fig. 1 shows the configurations of typical microphone
and loudspeaker arrays. Here Fn(o), (n ¼ 1, 2, . . ., N),
denotes the filters that control the input and output, and
which are connected to each acoustic transducer (micro-
phone or loudspeaker). They can be written in vector
form as

Fð!Þ ¼ ½F1 ð!Þ F2 ð!Þ  FN ð!ÞT (1)

where o is the angular frequency and the superscript T
denotes the transpose of the vector. In general, the acous-
tic transducers are distributed in an orderly manner in a
real implementation. Most representative configurations
are broadside arrays, in which the transducers are aligned
perpendicular to the direction of sound propagation, and
endfire arrays, in which the transducers are aligned along
the direction of sound propagation. For a broadside loud-
speaker array the vector of transfer functions to a far-field
immission point consists of the directivities of the indi-
vidual transducers and their relative propagation delays,
as described by the vector equation [10]
2 ! 3
1 ð; ; !Þej c x1 sin  cos 
6 2 ð; ; !Þej c x2 sin  cos  7
!
6 7
Wð; ; !Þ ¼ 6 .. 7: (2)
4 . 5
!
N ð; ; !Þej c xN sin  cos 
Here y is the azimuthal angle, f the zenithal angle (see Fig. 1. Typical basic configurations of acoustic array systems.
Fig. 2), c the speed of sound, xi the transverse position (a) Microphone array. (b) Loudspeaker array.

310 J. Audio Eng. Soc., Vol. 57, No. 5, 2009 May

PAPERS HIGHLY DIRECTIONAL LOUDSPEAKER ARRAY

a preferred direction and the intensity obtained at the

same point with a monopole source that radiates the FH ð!ÞFð!Þ
ð!Þ ¼ : (7)
same acoustic power as the loudspeaker system [14]. This FH ð!ÞW  ð!ÞW T ð!ÞFð!Þ
measure implies how much available acoustic power is
concentrated in the preferred direction by the designed The noise sensitivity C(o) is the reciprocal of the white
system as compared to the total radiated power. In matrix noise gain [8]. It is not only a measure for the sensitivity
notation this definition is expressed by [10] of the microphone array to internal microphone noise
but also a measure of the robustness of the array to
max; fFH ð!ÞW  ð;;!ÞW H ð;;!ÞFð!Þg errors in signal processing and deviations of the transdu-
Qð!Þ ¼ (4)
FH ð!ÞSTzz ð!ÞFð!Þ cers from their ideal gains (in amplitude and phase).
Eq. (7) is also a useful performance measure for loud-
where* denotes the conjugate operator, H denotes the speaker arrays. Here the denominator relates to the
Hermitian transpose, and Szz(o) is defined by its elements, sound intensity in the target direction, whereas the nom-
Szz ð!Þ¼ ½Smn  inator, which is the sum of the squared filter values, is
proportional to the electric input power of the loud-
1 R 2R  (5)
 speaker array. Hence we designate C(o) for loudspeaker
¼ W m ð;;!ÞWn ð;;!Þsin  d d
4 0 0 arrays the power factor (PF) or, when expressed on a dB
scale, the power index (PI).
Here the subscripts m and n denote the indexes of the
With the notation described in this section the di-
loudspeaker elements.
rectivity pattern of the discrete array system can be
For microphone array systems the same expressions
expressed as
can be used. Here the directivity factor expresses the ratio
between the sensitivity in a main direction and the sensi- ð; ; !Þ ¼ FT ð!ÞWð; ; !Þ: (8)
tivity in a diffuse or isotropic sound field. The matrix
Szz(o) is commonly called the coherence matrix of the
array. For an endfire array the elements of the coherence
matrix can be written as sinc functions [10], [11], 1.2 Optimal Beamformer
As a starting point let us take a look at a simple
sin½kðzm  zn Þ
Smn ¼ : (6) delay-and-sum microphone array beamformer in endfire
kðzm  zn Þ configuration. In this case the transducer elements are
Here zm and zn are the positions of the transducer elements only filtered with delays that compensate for the travel
and k denotes the wavenumber. Usually the directivity times of a wave from the target direction to the different
index (DI), which is the logarithmic value in dB of the microphones. The consequence is that this gives a high
directivity factor, is used. output from the target direction because all contributions
Another important evaluation parameter for micro- of the desired wave are completely in phase. This results
phone arrays is the noise sensitivity (NS). This quantity in a high ratio between the target response and the
shows the amplification ratio of uncorrelated noise, so- internal noise response. Hence the noise sensitivity of
called internal noise, to the signal and is given by [10] such a system is low. The directivity of the array is a
result of the fact that from directions other than the
target direction the wavefield contributions will add up
more or less destructively. However, at low frequencies
the phase differences between those nontarget waves are
small and hence the directivity index is small at low
frequencies.
The optimal beamformer tries to solve this problem by
not maximizing the output of the array in the target direc-
tion, but by maximizing the ratio between the response
from the target direction and the average uncorrelated
response from all other directions.
This optimization problem of an array system can
be defined by the following minimization expression
[8], [10]:

min FH ð!ÞSTzz ð!ÞFð!Þ subject to FT ð!ÞWð!Þ ¼ 1: (9)

Fð!Þ

The real-valued cost function derived by the Lagrange

method is given by
Fig. 2. Configuration of endfire loudspeaker array and direc- min JðFÞ ¼ FH STzz F þ ðFT W  1Þþ  ðFT W   1Þ (10)
tional definition of radiated sound field. Fð!Þ

J. Audio Eng. Soc., Vol. 57, No. 5, 2009 May 311

BOONE ET AL. PAPERS

where m is a Lagrange multiplier. The optimal value of To solve this problem, Gilbert and Morgan [12] sug-
F is obtained at the stationary point where the differen- gested a method adding a specific value b to the diagonal
tial of J(F) is zero. The optimal solution Fopt is esti- of the coherence matrix. Cox et al. [8] suggested a
mated as generalized form of the optimization problem. To restrict
the noise sensitivity or power index, an additional con-
rF JðFÞ ¼ Szz Fopt þ W ¼ 0 (11a) straint is required, such as
or
FH ð!ÞFð!Þ ¼ 2 max : (14)

FTopt ¼  W H S1
zz : (11b)
Using the Lagrange multiplier b, the real-valued cost
Substituting the constraint into Eq. (11b), the Lagrange function can be written as [9],[10]
multiplier is expressed as
1 min JðFÞ ¼ FH ðSzz þ IÞT F þ ðFT W  1Þ
 ¼ H 1 : (12) Fð!Þ
(15)
W Szz W
þ  ðFT W   1Þ  2
Then the optimal solution of Eq. (9) is given by [15]
where I is an identity matrix of size N. The solution of
W H ð!ÞS1
zz ð!Þ Eq. (15) is obtained as [9], [10]
FTopt ð!Þ ¼ H 1
: (13)
W ð!ÞSzz ð!ÞWð!Þ
W H ð!Þ½Szz þ ð!ÞI 1
The optimized beamformer aims at maximum directivity FTopt ð!Þ ¼ : (16)
and, at the same time, frequency-independent sensitivity W H ð!Þ½Szz þ ð!ÞI 1 Wð!Þ
in the target direction. This gives a principle problem at
low frequencies, because the magnitudes of the individual The role of b is to obtain a behavior of the array that is
transducer filters will increase considerably. between the maximized directivity result of Eq. (13) and
This characteristic can be demonstrated with a simple the delay-and-sum result. We will call the coefficient
example. Fig. 3 shows the filter coefficients estimated by b(o) the stability factor, because it controls the robustness
the optimal beamforming method with N ¼ 2, d ¼ 0.15 m of the array.
as a function of frequency. We see that the magnitudes of
the filter coefficients increase drastically in the low-fre-
quency range. Also notice an opposite phase relationship 2 OPTIMAL DESIGN OF AN ARRAY SYSTEM
between both filters. Due to this characteristic, the self- 2.1 Design Parameters
noise of the microphones is amplified greatly. It also
The directional characteristics of the loudspeaker array
means that the system should have a high precision to
system depend on several design parameters—the number
obtain the desired target response. For loudspeaker arrays
of transducers, their spacing and distribution pattern, the
this means that high input signals are needed and also that
directional characteristics of the individual loudspeakers,
the output control must be very precise for a correct re-
and the applied beamforming filters. For the optimal
sponse in the target direction. Therefore in the past the
beamformer the filter shape of the array system is deter-
optimal beamformer could not be implemented easily by
mined by Eq. (16). Therefore the parameter to be opti-
means of conventional analog circuits.
mized is the stability factor b(o).
To investigate the effect of each design parameter, a
parametric study was conducted with Eqs. (4) and (7).
Each loudspeaker element is assumed to be a monopole
and the effects of reflection and scattering are ignored.
With uniform spacing d and the same number of transdu-
cers, it is observed that the same directional characteris-
tics apply if we normalize the frequencies according to
the high-frequency limit (in Hz) given by
c
fh ¼ : (17)
2d
Fig. 4 shows the directivity index and power index of an
array with four equally spaced loudspeakers. The stability
factor b was set at 0.01. Also, using the delay-and-sum
beamformer, it was compared with the array system
having the same configuration. The filter derived by the
delay-and-sum beamformer is given by [16]
Fig. 3. Filter coefficients estimated using optimal beamforming
method with N ¼ 2. Both filters have the same magnitude. — W  ð!Þ
Fds ð!Þ ¼ : (18)
phase of first filter; – – phase of second filter. W H ð!ÞWð!Þ
312 J. Audio Eng. Soc., Vol. 57, No. 5, 2009 May
PAPERS HIGHLY DIRECTIONAL LOUDSPEAKER ARRAY

For the case of an endfire array, Eq. (18) can be rewrit- lower than fh. The frequency with the maximum DI value
ten as also increases, but it remains below fh. The power index
shows a decreasing tendency with the increase in fre-
1h iT
quency, reaching a minimum value at f ¼ fh. These results
Fds;endfire ð!Þ ¼ 1 ejkðz2 z1 Þ . . . ejkðzN z1 Þ : (19)
N are in agreement with the aforementioned theory.
Fig. 6 shows the changes in directional characteristics
The result shows that the directivity index is higher for with varying stability factors (b ¼ 100–106). Here the
the optimal beamformer than for the delay-and-sum number of equally spaced transducers was eight. With
beamformer in the frequency range below f/fh  0.9. This increasing b the DI and PI values decrease up to the
is consistent with previous work, which mentioned that frequency of DImax. At frequencies higher than fh DI and
the optimal beamformer converges to the delay-and-sum PI are no longer controllable by b.
beamformer at fh [10], [11].
The number of loudspeaker elements determines the
maximum value of the directivity index. For an endfire 2.2 Design Procedure
array system, the maximum directivity index (in dB) is Fig. 7 illustrates the design steps of the optimal beam-
determined by [17] former for a loudspeaker array system. First the target
values of the evaluation parameters DI and PI and the
DImax ¼ 20 log N (20) high-frequency limit should be selected. Considering these
target values, the basic design parameters, i.e. the number
where N denotes the number of transducers. Fig. 5 depicts of transducers and spacing, are determined using Eqs. (17)
the results of a parametric study varying the number of and (20). The second step is the selection of b as suggested
transducers (N ¼ 4–8) with b ¼ 0.01. The increase in DI to solve the power index problem of the equipment.
follows the increase of N over the whole frequency range However, it can also be applied to control the directional

Fig. 4. (a) Directivity index and (b) power index of array with Fig. 5. Changes in (a) directivity index and (b) power index
four uniformly spaced loudspeakers. — optimal beamformer; for various numbers of loudspeakers. — N ¼ 4; – – N ¼ 6; • • •
– – delay-and-sum beamformer. N ¼ 8.

J. Audio Eng. Soc., Vol. 57, No. 5, 2009 May 313

BOONE ET AL. PAPERS

characteristics of the array system without changing the

array configuration. Consequently this parameter can be
determined after selecting the other parameters. Finally
the optimal filter is derived by Eq. (16).
The optimal value of b, considering both DI and PI,
cannot be obtained directly. For this reason several itera-
tive methods were suggested to obtain the optimal value
[10]. It is thought that, as an alternative method, the
plot of PI versus DI can be used in the selection of b.
Consider the array system with N ¼ 8, which was used
in the previous section. The range of b was varied from
106 to 100. Fig. 8 shows the DI–PI plots for various
normalized frequencies (f/fh ¼ 0.08–0.75). It can be

Fig. 6. Change in (a) directivity index and (b) power index

for various stability factors b. — b ¼ 106; – – b ¼ 105; • • •
b ¼ 104; –  – b ¼ 103; –  – b ¼ 102, - - - b ¼ 101, 
b ¼ 100. Fig. 7. Design steps for optimal beamformer.

Fig. 8. Directivity index versus power index for array system with N ¼ 8 for different values of b and f / fh.

314 J. Audio Eng. Soc., Vol. 57, No. 5, 2009 May

PAPERS HIGHLY DIRECTIONAL LOUDSPEAKER ARRAY

seen that, with the increase of b, both DI and PI decrease assumption of omnidirectionality. In the case of a loud-
for a certain frequency or, conversely, the range of varia- speaker array, however, the size of the loudspeakers is
tions of DI and PI decreases with an increase in fre- much larger than that of the microphones to obtain
quency for a fixed value of b. This fact is related to the enough output power. Therefore not only the directivity
result of the previous section, namely, that the directional of the single transducer itself related to its own geometry
characteristics are no longer controllable at frequen- should be considered, but also the scattering from the
cies higher than fh. If the target performance of the array other transducers.
system is given by a specific range of DI and PI, the Usually the scattering effect is considered as being
value of the stability factor can be selected in such a induced by an incident field. The total field is described
DI–PI plot. by a summation of incident and scattered sound fields.
The directional pattern of the individual unit can be
obtained by summing the direct field from the transducer
3 CONSIDERATIONS FOR A PRACTICAL itself and the scattering field induced by the other units.
SYSTEM The analytical solution for the scattered field can be found
under some special conditions [18]. However, the direc-
3.1 General Directivity Pattern
tional pattern of a general array system is not easy to
In the foregoing sections the effect of reflection and obtain theoretically, because the scattering field of each
scattering induced by the loudspeaker enclosures has been loudspeaker also becomes the incident field to the other
ignored, that is, Gn ¼ 1, n ¼ 1, 2, . . ., N. In the case of a loudspeakers, recursively. Therefore numerical methods
microphone array system, the size of the transducers is or measurements would be useful to obtain the resultant
usually sufficiently small compared to the smallest wave- sound field.
length of interest. However, for loudspeaker arrays it is
recommended that the loudspeaker units be of a larger
size to obtain a sufficiently large radiation power. In prac- 3.3 Modified Design Procedure
tice the loudspeaker is not a simple source, but its radia- Fig. 9 explains the modified design procedure for the
tion pattern is a function of frequency or the Helmholtz design of an optimal beamformer. Considering the acous-
number, which is the ratio of diaphragm size to wave- tical facts as outlined in the previous sections, an addi-
length. Therefore the Gn value in Eq. (3) cannot be con- tional step to obtain the directional pattern of each
sidered unity, and the coherence function of the array transducer in the array system is required. In this step the
system described by Eq. (6) may not be valid. This fact directional pattern must contain the directional pattern of
will affect the design of the optimal filters, especially in
the selection of the stability factor to obtain the target
directivity index value, because Eq. (4) is a function of
the coherence matrix Szz (o).
Two methods are applicable to solve this problem.
The first is to return to Eq. (5) as a more general
definition of the coherence matrix. The other method is
using the general definition of directivity, which is
given for a source with only radiation dependence as a
function of y,
R
Qð!Þ ¼ 2= 0 2 ðÞ sin  d (21)

where it is assumed that G2 (ytarget) ¼ 1 in the target

direction.
Eq. (6) can be used as one of the evaluation parameters,
but it can also be employed in the derivation of the filter
coefficient, as in Eqs. (13) and (16). Therefore it is
expected that a direct calculation using Eq. (5) to design
the optimal filter can yield a further optimized solution
than the use of Eq. (6).

3.2 Mutual Interaction between Transducers

Scattering effects induced by objects increase accord-
ing to the size of the objects and decrease according to the
distance compared to the wavelength [18]. In the case of a
microphone system, the size of the microphones is usually Fig. 9. Modified design steps for optimal beamformer consider-
small enough to ignore the scattering effect; hence the ing directivity pattern and mutual interaction between loud-
equations described in Section 2 can be used with the speaker elements.

J. Audio Eng. Soc., Vol. 57, No. 5, 2009 May 315

BOONE ET AL. PAPERS

each transducer as well as the scattering effect induced by 4.3 Example 2—Constant-Beamwidth Array
every other transducer. Fig. 10 depicts the geometrical As a second example we considered a constant-
definition of the observation plane to obtain the direc- beamwidth array (CBA) system. The simplest concept to
tional pattern of the total sound field for each transducer design a CBA is using the different array sets, as com-
in the array system. Values of Gn, n ¼ 1, 2, . . ., N, can be puted for different values of the Helmholtz number kd.
obtained by either calculation or measurement of the With this method, however, we need redundant acoustic
sound pressure over the selected plane. When the direc- devices. In a specific array system one can say that the
tional pattern of the first unit was calculated, only the same value of DI means the same beamwidth. Therefore
diaphragm center of the first loudspeaker was activated the CBA system can be designed by selecting the fre-
and the other units were inactive. quency-dependent factor b(o) that gives a constant DI
over the entire target frequency range.
The same array system as described in the previous
4 NUMERICAL SIMULATION
example was used for the test. Values of DI and PI of this
4.1 Modeling of the Loudspeaker Array System system as a function of b are as shown in Fig. 8. The
As a design example, a loudspeaker array system was target frequency range was 0.1–1 kHz and the target value
chosen that consisted of eight loudspeakers at regular of DI was 12 dB, which is the highest value in Fig. 8 for
intervals of 0.15 m. The size of each loudspeaker box the condition of PI < 20 dB. To satisfy these conditions,
was 0.11 (W)  0.16 (H)  0.13 (D) m, and the diam- the b values along the DI line of 12 dB were selected from
eter of the loudspeaker diaphragm was 0.075 m. The Fig. 8.
boundary element method (BEM) was used to calculate Figs. 15 and 16 show the directivity index and pattern
the directional pattern of each transducer in the given of the designed loudspeaker system. One can observe
array configuration. Each loudspeaker was modeled by that every case satisfies the target value within 2 dB.
106 linear triangular elements, as illustrated in Fig. 11. As in the previous test example, the filter designed
The characteristic length of the elements was 0.057 m, considering the directional pattern of each loudspeaker
which gives a high-frequency limit of 1 kHz based on in the array system, as shown in Fig. 15(b), is very
the l/6 criterion (fh of the array system was 1.1 kHz). much like the result obtained using the simple source
All elements except the center of the diaphragm were assumption.
modeled as rigid boundaries. In order to obtain the di- Another factor that should be considered is the power
rectional pattern of each loudspeaker in the array sys- index of the designed filters. Although it is possible to
tem, the calculation was carried out by activating the obtain a constant directivity index as a function of
units one by one as part of the complete system. The frequency, this may lead to high power index values
observation plane was selected as a circle, as shown in with consequently a low array output and also the pos-
Fig. 10, which centered at the active node of each loud- sibility of large deviations from the desired pattern in
speaker with a wavelength of the lowest frequency of real situations that do not strictly follow the model
interest as the radius (3.43 m).

4.2 Example 1—Maximization of the Directivity

Index
The target performance of the array system was the
maximization of the directivity index within a range of
the power index that does not exceed 20 dB. Optimal
filters were calculated by two methods. In the first
method we assumed that every loudspeaker unit would
behave as a monopole and the scattering effect was
ignored. In the second method the directional pattern of
each unit and the effect of scattering were taken into
account in the design procedure. The acoustical perfor-
mance of the designed filters was tested by numerical
simulation using BEM.
Fig. 12 shows the power index of the designed filters
calculated by Eq. (7), and Fig. 13 shows the directivity
index of the designed loudspeaker system using the opti-
mal beamforming method to maximize the directivity
index. One can observe that, compared to the result using
the simple source assumption, values of the directivity
index considering the scattering effect are very similar to
the predicted values. This fact is also confirmed in the Fig. 10. Observation planes to obtain directional pattern of total
directional pattern, as shown in Fig. 14. sound field for each loudspeaker element in system.

316 J. Audio Eng. Soc., Vol. 57, No. 5, 2009 May

PAPERS HIGHLY DIRECTIONAL LOUDSPEAKER ARRAY

Fig. 11. Boundary element model. (a) One loudspeaker. (b) Entire loudspeaker array.

Fig. 12. Power index of filters. ▪ filter designed under simple source assumptions; ○ filter designed considering directional pattern of
each loudspeaker in system.

J. Audio Eng. Soc., Vol. 57, No. 5, 2009 May 317

BOONE ET AL. PAPERS

Fig. 13. Directivity index of loudspeaker system designed using optimal beamforming method to maximize directivity index. ▪
Eq. (16); ○ BEM. (a) Filter designed under simple source assumptions. (b) Filter designed considering directional pattern of each
loudspeaker in system.

Fig. 14. Comparison of directivity patterns of loudspeaker system designed using optimal beamformer to maximize directivity index.
(a) Predicted by Eq. (16) using filter designed under simple source assumptions. (b) Predicted by BEM using filter designed under
simple source assumptions. (c) Predicted by Eq. (16) using filter designed considering directional pattern of each loudspeaker in
system, (d) Predicted by BEM using filter designed considering directional pattern of each loudspeaker in system.

318 J. Audio Eng. Soc., Vol. 57, No. 5, 2009 May

PAPERS HIGHLY DIRECTIONAL LOUDSPEAKER ARRAY

Fig. 15. Comparison of directivity patterns of loudspeaker system designed using optimal beamforming method to obtain constant
beamwidth. ▪ Eq. (16); ○ BEM. (a) Filter designed under simple source assumptions. (b) Filter designed considering directional
pattern of each loudspeaker in system.

Fig. 16. Comparison of directivity patterns of loudspeaker system designed using optimal beamformer to obtain constant beamwidth.
(a) Predicted by Eq. (16) using filter designed under simple source assumptions. (b) Predicted by BEM using filter designed under
simple source assumptions. (c) Predicted by Eq. (16) using filter designed considering directional pattern of each loudspeaker in
system, (d) Predicted by BEM using filter designed considering directional pattern of each loudspeaker in system.

J. Audio Eng. Soc., Vol. 57, No. 5, 2009 May 319

BOONE ET AL. PAPERS

used. Fig. 17 shows an example of such a case. It is recorded signal. A swept-sine signal was adopted in the
observed that the difference between predicted and test, and the filter designed for the optimal beamformer
simulated results increases at the points of high power was applied to the swept signal. For multichannel sound
index (400 Hz). From this result one can conclude that
it is helpful to maintain a low power index for precise
predictions.

5 MEASUREMENTS

5.1 System Configuration

To see the performance of the designed filters in a real
situation, measurements were conducted in an anechoic
condition. Fig. 18 is a schematic of the loudspeaker array
and measurement system setup. Fig. 19 shows an endfire
array system. The loudspeaker sizes and all the mea-
surements involved were the same as in Section 4. The
loudspeaker array was mounted on a turntable. In the
measurement system the distance between the micro-
phone and the frontal surface of the loudspeaker array

was 2.5 m when the angle was 0 . It is noted that the
geometric center of the array system was also the rotating
center of the turntable. Because of the limited distance
between loudspeaker array and measurement microphone
some near-field effects may occur. One effect is that in
the endfire orientation, level differences occur because of
distance changes. For the outer loudspeakers of the array
this results in a level change at the microphone location of
 1.5 dB. Another effect occurs in broadside orientation.
Here the outer loudspeakers produce signals with in-
correct phases at the microphone location of about 5 at
100 Hz and 50 at 1000 Hz.
Measurements were conducted in steps of 10 from 0 to
180 , and symmetrical data are assumed in the other half-
circle. The sound pressure was measured using a sound
level meter (B&K 2239) and the signal was transferred to
the control computer. Fig. 17. Effect of power index. ▪ PI limited to 20 dB; ○ PI
Signal processing was done using the MATLAB soft- limited to 50 dB. (a) Applied power index. (b) Resulting direc-
ware for both signal generation and analysis of the tivity index.

Fig. 18. Schematic diagram of loudspeaker array and measurement system setup.

320 J. Audio Eng. Soc., Vol. 57, No. 5, 2009 May

PAPERS HIGHLY DIRECTIONAL LOUDSPEAKER ARRAY

playback the filtered signal was sent to an ADAT audio This may be caused by the near-field effects in the mea-
interface (RME ADI-648) using a sound card (RME surements, as mentioned in Section 5.1.
HDSP MADI), and then this signal was amplified by a
multichannel amplifier (Sonic Emotion M3S Amp).
In all measurements two different filter design methods 5.3 Test 2—Constant-Beamwidth Array
were used as discussed in the previous section—the filter In a second example the filter of the constant-
design under simple source assumptions and the filter beamwidth array, which was introduced in Section 4.3,
design including the directional characteristics of each was implemented. The target value of the directivity in-
loudspeaker for the total radiation field. dex was chosen to be 12 dB. Fig. 22 compares the target
values and the results of two designed filters. The filter
designed considering the directional characteristics of
5.2 Test—Constant b
the total radiation field of each loudspeaker shows values
As a first example we considered the acoustic perfor- that are higher and closer to the target than the filter
mance of the optimal beamformer with constant b for all using a uniform pattern. However, it is noted that the
frequencies. The b value was set to 0.01, which is usually designed filter has a still higher sound level in the off-axis
a satisfactory value for optimized beamformers. A com- directions than expected from theoretical prediction.
parison of the directivity index of the designed and Fig. 23 shows the measured directional patterns of the
measured loudspeaker system is given in Fig. 20. Similar array system, which supports the aforementioned findings
to the foregoing simulation results, the filters designed with regard to the acoustic performance of the designed
considering the directional characteristics of the loudspea- filters.
kers were closer to the predicted values than the filters
using the uniform pattern. Moreover a far higher directiv-
ity index value was obtained by the former filter than the
latter, which did not consider loudspeaker directivity.
Fig. 21 compares simulated and measured directional
patterns of the designed loudspeaker system. In the
measured result the beamwidth of the main lobe is nearly
the same or somewhat narrower in the low-frequency
range, but with higher sidelobes, than the predicted value.

Fig. 20. Comparison of directivity patterns of loudspeaker sys-

tem designed using optimal beamforming method with b ¼ 0.01.
▪ Eq.(16), ○ measured. (a) Filter designed under simple source
assumptions. (b) Filter designed considering directional pattern
Fig. 19. Endfire array system. of each loudspeaker in system.

J. Audio Eng. Soc., Vol. 57, No. 5, 2009 May 321

BOONE ET AL. PAPERS

6 CONCLUSIONS
In this study the basic theory of an endfire loudspeaker
array system was investigated with a parametric study.
The design parameters included the number of loud-
speaker units, their spacing, the array length, and the effect
of the stability factor of the optimal beamformer. Two
types of filter design procedures were tested—one method
was to design a filter under a simple source assumption
and another was to design a filter by additionally consider-
ing the directional pattern of each loudspeaker in the array
system. Measurements on an array of eight loudspeakers
revealed that inclusion of the directivity patterns of the
loudspeakers resulted in an increase in the directivity
by 2–3 dB as compared with a design based on a mono-
Fig. 22. Comparison of the directivity patterns of loudspeaker
pole assumption. Moreover it was observed that the system designed using optimal beamforming method to obtain con-
difference between expected and actual results of the stant beamwidth with DI ¼ 12 dB. ▪ target value; ○ simple source
directivity index also changed from 3–6 dB to 1–3 dB assumptions, △ realistic source model with loudspeaker directivity.

Fig. 21. Comparison of directivity patterns of loudspeaker system designed using optimal beamformer with b ¼ 0.01. (a) Predicted by
Eq. (16) using filter designed under simple source assumptions. (b) Measured result using filter designed under simple source
assumptions. (c) Predicted by Eq. (16) using filter designed considering directional pattern of each loudspeaker in system.
(d) Measured result using filter designed considering directional pattern of each loudspeaker in system.

322 J. Audio Eng. Soc., Vol. 57, No. 5, 2009 May

PAPERS HIGHLY DIRECTIONAL LOUDSPEAKER ARRAY

when considering a realistic radiation condition without [3] D. B. Ward, R. A. Kennedy, and R. C. Williamson,
too much effort. “Theory and Design of Broadband Sensor Arrays with
Frequency Invariant Far-Field Beam Patterns,” J. Acoust.
Soc. Am., vol. 97, pp. 1023–1034 (1995).
7 ACKNOWLEDGMENT [4] W. Soede, A. J. Berkhout, and F. A. Bilsen,
“Development of a Directional Hearing Instrument Based
This work was supported in part by the Korea Research
on Array Technology,” J. Acoust. Soc. Am., vol. 94, pp.
Foundation under grant KRF-2006-612-D00004, the BK21
785–798 (1993).
Project, and NRL (M10500000112-05J0000-11210).
[5] W. Soede, F. A. Bilsen, and A. J. Berkhout, “As-
sessment of a Directional Microphone Array for Hearing-
8 REFERENCES Impaired Listeners,” J. Acoust. Soc. Am., vol. 94, pp. 799–
808 (1993).
[1] B. D. van Veen and K. M. Buckley, “Beamforming: [6] D. E. Weston, “Jacobi Sensor Arrangement for
A Versatile Approach to Spatial Filtering,” IEEE Acoust., Maximum Array Directivity,” J. Acoust. Soc. Am., vol.
Speech, Signal Process. Mag., vol. 5, pp. 4–24 (1988 Apr.). 80, pp. 1170–1181 (1986).
[2] J. L. Flanagan, D. A. Berkley, G. W. Elko, and [7] H. Cox, R. M. Zeskind, and T. Koou, “Practical
M. M. Sondhi, “Autodirective Microphone Systems,” Supergain,” IEEE Trans. Acoust., Speech, Signal Pro-
Acoustica, vol. 73, pp. 58–71 (1991). cess., vol. ASSP-34, pp. 393–398 (1986).

Fig. 23. Comparison of directivity patterns of the loudspeaker system designed using optimal beamformer to obtain constant
beamwidth with DI ¼ 12 dB. (a) Predicted by Eq. (16) using filter designed under simple source assumptions. (b) Measured result
using filter designed under simple source assumptions. (c) Predicted by Eq. (16) using filter designed considering directional pattern
of each loudspeaker in system, (d) Measured result using filter designed considering directional pattern of each loudspeaker in system.

J. Audio Eng. Soc., Vol. 57, No. 5, 2009 May 323

BOONE ET AL. PAPERS

[8] H. Cox, R. M. Zeskind, and M. M. Owen, “Robust sented of the 120th Convention of the Audio Engineering
Adaptive Beamforming,” IEEE Trans. Acoust., Society, J. Audio Eng. Soc. (Abstracts), vol. 54, p. 728
Speech, Signal Process., vol. ASSP-35, pp. 1365–1376 (2006 July/Aug.), convention paper 2869.
(1987). [14] L. E. Kinsler, A. R. Frey, A. B. Coppens, and J. V.
[9] J. M. Kates and M. R. Weiss, “A Comparison of Sanders, Fundamentals of Acoustics (Wiley, New York,
Hearing-Aid Array-Processing Techniques,” J. Acoust. 2000), chap. 7.
Soc. Am., vol. 99, pp. 3138–3148 (1996). [15] D. A. Brandwood, “A Complex Gradient Operator
[10] I. Merks, “Binaural Application of Microphone and Its Application in Adaptive Array Theory,” IEE Proc.,
Arrays for Improved Speech Intelligibility in a Noisy pts. F and H, vol. 130, pp. 11–16 (1983).
Environment,” Ph.D. thesis, Technical University of [16] A. J. Berkhout, “Pushing the Limits of Seismic
Delft, Delft, The Netherlands (2000). Imaging. Part 1: Prestack Migration in Terms of Double
[11] M. Brandstein and D. Ward, Microphone Arrays. Dynamic Focusing,” Geophysics, vol. 62, pp. 937–954
(Springer, New York, 2001), chap. 2. (1997).
[12] E. N. Gilbert and S. P. Morgan, “Optimum [17] R. L. Pritchard, “Maximum Directivity Index of a
Design of Directive Antenna Arrays Subject to Ran- Linear Point Array,” J. Acoust. Soc. Am., vol. 26, pp.
dom Variations,” Bell Sys. Tech. J., vol. 34, pp. 637– 1034–1039 (1954).
663 (1955). [18] E. G. Williams, Fourier Acoustics—Sound Radia-
[13] M. M. Boone, “Directivity Measurements on a tion and Nearfield Acoustical Holography (Academic
Highly Directive Hearing Aid: The Hearing Glasses,” pre- Press, London, 1999), chap. 6.

THE AUTHORS

M. M. Boone W.-H. Cho J.-G. IH

Marinus M. Boone was born in 1947. He received Wan-Ho Cho was born in Seoul, Korea, in 1980.
an M.Sc. degree in subjective acoustics, with the He received a B.S. degree in mechanical engineer-
topic of loudness perception, from the Delft Univer- ing and M.S. and Ph.D. degrees in acoustics from
sity of Technology, Delft, The Netherlands. the Korean Advanced Institute of Science and
Since 1974 he has been with the Laboratory of Technology (KAIST) in 2002, 2004, and 2008, re-
Acoustical Imaging and Sound Control at Delft Uni- spectively. He is now a postdoctoral fellow in the
versity of Technology, where he received a Ph.D. de- Acoustics Laboratory at KAIST.
gree on the development of a 32-channel microphone During his doctoral studies he joined the Labora-
system for directional outdoor noise measurements. tory of Acoustical Imaging and Sound Countrol at
Later on he designed several microphone systems, Delft University of Technology, Delft, The Nether-
based on the same technology, for the measurement of lands, as a guest student researcher from 2006 to
traffic and aircraft noise as well as microphone arrays 2007, doing research on the directional loudspeaker
for directional hearing aids, leading to the design of the array system. His doctoral thesis was about sound
so-called hearing glasses. At present his interests are in field control based on acoustical holography using
audio transducers and multichannel sound reproduc- loudspeaker array systems. His research interests are
tion for applications with wave field synthesis. in the area of audio engineering, especially sound
Dr. Boone is a fellow of the Audio Engineering field control and reproduction by multichannel au-
Society, and from 1996 to 2006 he was secretary of dio systems. He is also involved in product sound
the Netherlands Section of the AES. quality (PSQ) and noise control.

324 J. Audio Eng. Soc., Vol. 57, No. 5, 2009 May

PAPERS HIGHLY DIRECTIONAL LOUDSPEAKER ARRAY

Jeong-Guon Ih received a B.S. degree in mechani- with the Daewoo Motor Company (now GM Daewoo
cal engineering in 1979 from the Seoul National Uni- Auto Company) in Inchon, Korea, from 1979 to 1990.
versity and M.S. and Ph.D. degrees in Acoustics/ From 1985 to 1990 he was in charge of the Noise,
mechanical engineering in 1981 and 1985 from Vibration and Harshness (NVH) Group in the Tech
KAIST. He was with Loughborough University (UK) Center. In 1987 he spent a year at ISVR, Southampton
in 1999, with Seikei University (Japan) in 2005, University (UK) as a postdoctoral researcher.
and with Canterbury University (New Zealand) in Dr. Ih received domestic and international awards,
2006 as a visiting professor, either doing research or including academic awards from the Acoustical So-
lecturing. ciety of Korea (ASK) and the Korean Society for
He joined the Department of Mechanical Engi- Noise and Vibration Engineering (KSNVE). He was
neering of the Korean Advanced Institute of Science Secretary General of Inter-Noise 2003 in Jeju,
and Technology (KAIST) in Daejeon, Korea, as an Korea. Currently he is a vice president of ASK and
assistant professor in 1990, teaching courses in formerly he served as editor in chief of the Journal of
acoustics and vibrations, and is currently a professor the Acoustical Society of Korea. He is a member of
there. Before joining the faculty at KAIST he was the editorial board of Applied Acoustics.

J. Audio Eng. Soc., Vol. 57, No. 5, 2009 May 325