Time-domain simulation of a guitar: Model and methoda)

Grégoire Derveauxb)
INRIA Roquencourt, Projet Ondes, 78153 Le Chesnay Cedex, France

Antoine Chaignec)
ENSTA, Chemin de la Hunière, 91761 Palaiseau Cedex, France
Patrick Joly and Eliane Bécache
INRIA Roquencourt, Projet Ondes, 78153 Le Chesnay Cedex, France

共Received 8 June 2003; revised 29 September 2003; accepted 3 October 2003兲

This paper presents a three-dimensional time-domain numerical model of the vibration and acoustic
radiation from a guitar. The model involves the transverse displacement of the string excited by a
force pulse, the flexural motion of the soundboard, and the sound radiation. A specific spectral
method is used for solving the Kirchhoff–Love’s dynamic top plate model for a damped,
heterogeneous orthotropic material. The air–plate interaction is solved with a fictitious domain
method, and a conservative scheme is used for the time discretization. Frequency analysis is
performed on the simulated sound pressure and plate velocity waveforms in order to evaluate
quantitatively the transfer of energy through the various components of the coupled system: from
the string to the soundboard and from the soundboard to the air. The effects of some structural
changes in soundboard thickness and cavity volume on the produced sounds are presented and
discussed. Simulations of the same guitar in three different cases are also performed: ‘‘in vacuo,’’ in
air with a perfectly rigid top plate, and in air with an elastic top plate. This allows comparisons
between structural, acoustic, and structural–acoustic modes of the instrument. Finally, attention is
paid to the evolution with time of the spatial pressure field. This shows, in particular, the complex
evolution of the directivity pattern in the near field of the instrument, especially during the attack.
© 2003 Acoustical Society of America. 关DOI: 10.1121/1.1629302兴
PACS numbers: 43.75.Gh, 43.40.Rj 关NHF兴 Pages: 3368 –3383

I. INTRODUCTION dimensional damped string model. The motion of the string

is assumed to be perpendicular to the top plate. The sound-
The scope of the reported study is the simulation of the
board itself is more sophisticated, in order to account for its
three-dimensional sound-pressure field of a guitar in the time
crucial role in the sound of a guitar. It is modeled as an
domain. In a guitar, the elastic waves in strings and sound-
board, and the acoustic waves in both external air and cavity, orthotropic heterogeneous damped Kirchhoff–Love plate,
are continuously evolving with time. Both fluid and struc- with a hole, and clamped at its boundaries. The other consti-
tural waves interact continuously so that the system com- tutive parts of the guitar 共back, sides, neck兲 are assumed to
posed by the instrument, the cavity, and the surrounding air be rigid. These latter assumptions could be eventually revis-
should be considered as a whole. This approach, which is ited in future studies. The plate radiates both inside the cav-
essential for a good understanding of the time-varying vi- ity and in the external free field. The cavity communicates
broacoustical phenomena involved in a guitar, has been made with the external field through the hole.
possible by the use of advanced numerical methods. In its present state, the purpose of the work is to show
The method used consists of modeling the vibratory and the feasibility and the main principles of a complete guitar
acoustical phenomena involved in the instrument from the modeling. No comparison with a particular instrument has
initial pluck to the radiation in a three-dimensional space. been made. The values of the various physical and geometri-
Since we wish to focus on the modeling of the soundboard cal parameters used in the simulations correspond to realistic
and on the fluid–structure interaction, the model used for the values extracted from the existing literature on the guitar.
other parts of the instrument is intentionally kept simple. The complexity of the system makes it necessary to use
Thus, an idealized plucking force is acting on a one- different numerical techniques for solving the problem. The
plate equation is solved with a specific spectral method. A
a兲
Preliminary results to the present study were first reported at: Forum Acus- fictitious domain method, similar to the one previously used
ticum, Séville, Spain 共Time-domain simulation of a guitar: Preliminary by Rhaouti et al. for timpani, is used for solving the fluid–
results, Forum Acusticum 2002, Special Issue of the Revista de Acustica,
structure interaction.1 Finally, a conservative scheme is used
Vol. 33兲 and 144th Acoustical Society of America Meeting, Cancun,
Mexico 关‘‘Time-domain simulation of a guitar,’’ J. Acoust. Soc. Am. 112, for the time discretization.
2409 共2002兲兴. The numerical aspects of the method were presented at the The paper starts with the presentation of the guitar
Second M.I.T. Conference on Computational Fluid and Solid Mechanics model, which is followed by the numerical formulation of
共June, 2003兲.
b兲
Present address: ENSTA-UMA, 32 Bd Victor, 75015 Paris, France. the problem. The fourth section is devoted to the presentation
c兲
Electronic mail: [email protected] of selected preliminary results whose aim is to show the

3368 J. Acoust. Soc. Am. 114 (6), Pt. 1, Dec. 2003 0001-4966/2003/114(6)/3368/16/$19.00 © 2003 Acoustical Society of America
 top plate of the instrument and ⌺ is the rest of the surface
共i.e., sides and back兲. The boundary of ␻ itself is divided into
two parts: ␥ 0 is the outer boundary of the top plate and ␥ f is
the inner boundary, along the hole. The surrounding air oc-
cupies the domain ⍀⫽R3 ⶿⌫. ⍀ corresponds to the internal
cavity and the external domain, which communicate via the
sound hole. The string of length l s is rigidly fixed to the neck
at a point denoted O, chosen as the origin of the coordinate
system 共see Fig. 1兲.

B. The string
The string is supposed to have uniform density ␳ s , uni-
form tension T, and no stiffness. In the following, only the
FIG. 1. Geometrical description of the guitar. vertical flexural motion of the soundboard will be consid-
ered. Therefore, it is also natural to consider only the vertical
main attributes of the method. The detailed presentation of polarization of the string. The motion of the string is thus
the application of the model to important features in the con- described by its transverse displacement 共perpendicular to
text of guitar acoustics, such as sound power and efficiency, the soundboard兲 denoted u s (x,t), x苸]0,l s 关 . Assuming that
is left for future studies. the amplitudes of vibration are sufficiently small, the string
With regard to previous work on guitars, one can sum- equation is a classical 1D damped wave equation
marize the originality of the present work as follows: First,
the cavity model and the air–structural coupling is more gen-
eral than in the boundary element method, developed by Ri-
␳s
⳵ 2u s
⳵t 冉
2 ⫺T 1⫹ ␩ s
⳵ ⳵ 2u s
⳵t ⳵x2 冊
⫹ ␳ sR s
⳵us
⳵t
⫽ f s 共 x,t 兲 , in ]0,l s 关 .
共1兲
chardson et al.2 It is also a step forward compared to the
spatial Fourier method developed by Le Pichon et al., where The internal damping is modeled by two terms: a viscoelastic
the velocity distribution on the surface of the body was mea- term with coefficient ␩ s whose essential purpose is to intro-
sured, and not calculated.3 Second, the results obtained in the duce a frequency-dependent damping in the string, and a
time domain with the present model should be compared to fluid term with coefficient R s whose purpose is to introduce
the results obtained in the frequency domain by Elejabarrieta a constant internal damping for low frequencies. The string is
et al., who used commercial software.4 In particular, the fixed at one end to the neck of the guitar, which is supposed
bracing used in our model is less accurate than in this latter to be perfectly rigid
paper. However, more emphasis is put here on the descrip- u s 共 0,t 兲 ⫽0, ᭙t⬎0. 共2兲
tion of the air sound field and on the effects of the pressure
jump on the top plate. At the bridge, the displacement of the string is assumed
Finally, as far as the numerical technique is concerned, to be given by the displacement of the plate at the contact
the present modeling is a direct extension to plates of the point between string and bridge, denoted (x 0 ,y 0 )
timpani model developed by Rhaouti et al.1 The modeling of u s 共 l s ,t 兲 ⫽u p 共 x 0 ,y 0 ,t 兲 , ᭙t⬎0. 共3兲
damping in the guitar top plate makes use also of results
obtained by Chaigne and Lambourg.5 The plucking of the string is assumed to be an imposed
force 共see Fig. 2 and Table I兲
II. PRESENTATION OF THE MODEL
f s 共 x,t 兲 ⫽g 共 x 兲 h 共 t 兲 , 共4兲
A. Description
where h(t) represents a simple idealized version of the
The body of the guitar is delimited by a surface denoted ‘‘stick–slip’’ mechanism that governs the interaction be-
⌫ which is divided into two parts: ⌫⫽ ␻ 艛⌺, where ␻ is the tween string and finger. One pulls on the string during nearly

FIG. 2. Representation of the ideal-

ized plucking force. 共a兲 Time depen-
dence. 共b兲 Space dependence.

J. Acoust. Soc. Am., Vol. 114, No. 6, Pt. 1, Dec. 2003 Derveaux et al.: Time-domain simulation of a guitar 3369
 TABLE I. Typical values of physical and numerical values used for the simulations.

String: ␳ s ⫽0.005 25 kg•m⫺1 , T⫽60 N, l s ⫽0.65 m,

R s ⫽0.75 s⫺1 , ␩ s ⫽9•10⫺8 s, (x 0 ,y 0 )⫽(65 cm,4 cm)

Soundboard:
D1 D2 D3 D4 ␳p a Rp ␩p
共MPa兲 共MPa兲 共MPa兲 共MPa兲 共kg•m⫺1兲 共mm兲 共s⫺1兲 共s兲

Plate 850 50 75 200 350 2.9 7 0.005

Bridge 80 50 900 270 400 6 7 0.005
Struts 100 60 1250 300 400 14 7 0.005

Air: c a ⫽344 m•s⫺1 , ␳ a ⫽1.21 kg•m⫺3

Numerical information:
String: 308 nodes h c ⫽2.1 mm
Plate: 3230 nodes h p ⫽1.2 cm
Pressure jump: 1260 nodes h ␭ ⫽1.2 cm
Acoustic field: ⬇400 000 cubes h a ⫽1.1 cm
Time step: ⌬t⫽2.10⫺5 s
Sampling frequency: f e ⫽50 000 Hz

再
Plucking force: f s (x,t)⫽g(x)h(t).
共1⫺cos共␲t/t1兲兲, 0⭐t⭐t1 ,
exp共⫺共x⫺x0 /␦s兲2兲 h共t兲⫽ 共1⫹cos共␲共t⫺t1兲/t2兲兲, t1⭐t⭐t2 ,
g共x兲⫽
兰0s exp共⫺共x⫺x0 /␦s兲2兲
l
0, t⬎t 2 .
x 0 ⫽55 cm, ␦ c ⫽0.006 m, t 1 ⫽0.015 s, t 2 ⫽0.000 4 s.

15 ms and releases it during 0.4 ms. This force is distributed

over a small segment of the string by means of the smooth
l
positive function g, normalized so that 兰 0s g(x)dx⫽1. De-
a␳p
⳵ 2u p
⳵t2 冉 ⳵
⳵t 冊
⫹ 1⫹ ␩ p div Div a 3 C␧= 共 ⵜគ u p 兲 ⫹a ␳ p R p
⳵up
⳵t

spite its simplicity, this excitation is in fairly good agreement ⫽F⫺ 关 p 兴 ␻ , in ␻ , 共5兲
with experiments.6 where Div is the divergence operator for tensors: (Div.␶ )
⫽ ⳵ j ␶ i j , ␧= is the plane linearized strain tensor 关 ␧ ␣␤ ( ␪គ )
⫽ 12 ( ⳵ ␤ ␪ ␣ ⫹ ⳵ ␣ ␪ ␤ ) 兴 , div and ⵜគ denote the usual divergence
C. The soundboard
operator and gradient operator for vectors, respectively 共see
It is assumed that the only vibrating part of the guitar is Appendix A for details and more explicit notations兲.
the soundboard. The other parts of the guitar body 共back, As for the string, the internal damping in the plate is
sides, neck兲 are assumed to be perfectly rigid. If necessary, modeled by two terms: a viscoelastic term with coefficient
the motion of the back could be introduced straightaway in ␩ p and a fluid term with coefficient R p . Notice, however,
this model since it is of exactly the same nature as the top that since a spectral method is used for the numerical reso-
plate. The soundboard is a thin, orthotropic wooden layer, lution of this equation, it becomes possible to introduce other
with a soundhole. The thickness is relatively small 共about 3 laws of damping, which do not need to have so simple a
mm兲 compared to the dimensions of the soundboard 共about counterpart in the time domain. This point will be developed
50 cm long兲. In terms of spectral content, our study is limited in Sec. III C.
to the bandwidth 0–3000 Hz. For both these reasons, it is not The top plate is clamped on ␥ 0
necessary to use a thick plate model, such as the Reissner–
Mindlin one. The flexural Kirchhoff–Love’s plate model is u p ⫽0 and ⳵ n u p ⫽0, on ␥0 , 共6兲
thus appropriate to describe the vibrations of the soundboard. whereas the boundary is free along the hole ␥ f 共see Appen-
The strutting system and the bridge are considered as dix A for explicit notations in particular cases兲
heterogeneities of the soundboard. As a consequence, the
following physical parameters which characterize the top = nគ 兲 •nគ ⫽0,
共M on ␥ f ,
plate are variable functions of space: density ␳ p (x,y), thick- 共7兲
= 兲 •nគ ⫹ ⳵ ␶ 关共 M
共 DivM = nគ 兲 • ␶គ 兴 ⫽0, on ␥ f ,
ness a(x,y), and rigidity tensor for orthotropic material
C(x,y). In what follows, the space dependence of these pa- where M = ⫽a 3 C␧= (ⵜ
គ u p ) is the bending moment. nគ denotes
rameters will be omitted, in order to simplify the notations. the outer normal and ␶គ the tangential vector along the bound-
Finally, the in-plane motion of the soundboard is ne- ary ␥ f . The two conditions in Eq. 共7兲 express approximately
glected compared to its flexural motion. The motion of the that force and moment vanish at the free boundary.
strutted soundboard is thus completely described by the The force density term exerted on the plate by the sur-
transverse displacement of the top plate, denoted u p (x,y,t), rounding air is 关 p 兴 ␻ , the pressure jump across the sound-
(x,y)苸 ␻ . This motion is governed by the following board. F denotes the force density exerted by the string on
equation:7–10 the bridge. It is assumed to be the normal component of the

3370 J. Acoust. Soc. Am., Vol. 114, No. 6, Pt. 1, Dec. 2003 Derveaux et al.: Time-domain simulation of a guitar
tension of the string at this point. This force is applied at the In addition to its physical meaning, this property guar-
contact point (x 0 ,y 0 ) between string and bridge antees the wellposedness of the continuous model and will
be exploited for ensuring the stability of the numerical
F共 x,y,t 兲 ⬇⫺T ⳵ x u s 共 l s ,t 兲 ␦ x 0 ,y 0 共 x,y 兲 . 共8兲
scheme, as it will be shown in the next section.
D. The acoustic field
The acoustic field is governed by the linearized Euler’s
equations III. NUMERICAL RESOLUTION

⳵p Let us begin with a brief description of the main diffi-

⫽⫺c 2a ␳ a div共 vᠪ a 兲 in ⍀, 共9a兲
⳵t culties we have to face in order to solve these equations
numerically. The 3D nature of the computational domain for
⳵ vᠪ a
␳a ⫽⫺ⵜ
គp in ⍀, 共9b兲 the sound radiation yields a large size problem. This domain
⳵t is unbounded, so that it is necessary to reduce it to a finite
where c a is the speed of the sound in air, ␳ a is the density of size problem, in order to make it computable. The geometry
air, p is the sound pressure in ⍀, and vᠪ a the acoustic velocity of the instrument is complex and includes third-order bound-
in ⍀. These equations are complemented by a condition of ary conditions 关Eq. 共7兲兴. Furthermore, the space- and time
continuity for the normal component of the velocity at the discretization of the Kirchhoff–Love equation requires an
surface ␻ of the plate adapted resolution method, because this equation includes a
fourth-order space operator and is intrinsically dispersive.
vᠪ a 共 x,y,0,t 兲 •ez ⫽ ⳵ t u p 共 x,y,t 兲 , ᭙ 共 x,y 兲 苸 ␻ , ᭙t⬎0. Last but not least, one has to derive a stable resolution
共10兲
scheme.
In addition, as the body of the guitar is assumed to be per- The problem is first rewritten as an equivalent varia-
fectly rigid, one has tional formulation, which is the first step of the finite element
vᠪ a 共 x,y,z,t 兲 •N⌫⫽0, ᭙ 共 x,y,z 兲 苸⌫, ᭙t⬎0, 共11兲 method. A variational formulation corresponds to the virtual
work principle which is naturally related to the continuous
where N⌫ denotes the outer normal to the boundary ⌺. energy identity 关Eq. 共13兲兴. The conforming space approxima-
tion of the variational formulation of the problem leads to a
E. Energy decay discrete energy identity, which is a convenient approach to
derive the stability of the resolution scheme. The main as-
The system of coupled equations 共1兲 to 共11兲 fulfills a pects of the numerical resolution are the following:
property of energy decay in the case of free oscillations. The
respective energies of the string E s , plate E p , and sound field 共i兲 The plate equation is solved by a modal decomposi-
E a , are given by tion method 共Secs. III A 1 and III C兲.

冕 冏 冏 冕 冏 冏
共ii兲 The fluid–structure interaction problem is solved with
1 ls du s 2 1 ls du s 2
E 共 t 兲⫽
s
␳s dx⫹ T dx, the use of the fictitious domain method 共Sec. III A 2兲.
2 0 dt 2 0 dx 共iii兲 One of the main originalities of the method is a stable

冕冕␳ 冏 冏
1 du p 2 coupling between a continuous time resolution and a
E p共 t 兲 ⫽ pa dx dy discrete one 共Sec. III C兲.
2 ␻ dt

冕冕
1 In what follows, for the sake of simplicity, the damping
⫹ a 3 C␧= 共 ⵜ
គ u p 兲 :␧= 共 ⵜ
គ u p 兲 dx dy, 共12兲 in the plate is described by a single fluid damping term, and
2 ␻
there is no damping in the string. It does not add formal

E a共 t 兲 ⫽
1
2
冕冕冕␳ ⍀
ᠪ a兩
a兩 v
2
dx dy dz
difficulty to take into account the other damping terms pre-
sented in Eqs. 共1兲 and 共5兲.

⫹
1
2
冕冕冕 ␳ 2
⍀ca
1
a
兩 p 兩 2 dx dy dz.
A. Variational formulation

It is easy to show that the total energy of the system E

⫽E s ⫹E p ⫹E a fulfills the following identity property: 1. The plate
dE
dt
共 t 兲⫽ 冕 ls

0
fs
du s
dt
dx⫺ 冕␩ 0
ls
s T 兩 ⳵ x,t u s 兩
2
dx
A well-known difficulty for the resolution of the
Kirchhoff–Love’s problem is due to the presence of a fourth-

冕␳ ⳵ ls order space operator. This operator would require a priori the

⫺ sR s兩 tu s兩
2
dx use of sophisticated finite elements of class C 1 , with conti-
0 nuity of the normal derivative along each edge of the plate

⫺ 冕冕␩ ␻
pa
3
C␧= 共 ⵜ
គ ⳵ t u p 兲 :␧= 共 ⵜ
គ ⳵ t u p 兲 dx dy
triangulation. To circumvent this difficulty, the space opera-
tor is split into two second-order operators. We introduce the
velocity v p ⫽ ⳵ t u p and the bending moment M = ⫽␧= (ⵜគ u p ),
⫺ 冕冕 ␳ ␻
a pR p兩 ⳵ tu p兩
2
dx dy. 共13兲 which leads to the following equivalent problem of second
order in space and first order in time:

J. Acoust. Soc. Am., Vol. 114, No. 6, Pt. 1, Dec. 2003 Derveaux et al.: Time-domain simulation of a guitar 3371
 = ⫹a ␳ p R p v p ⫽F⫺ 关 p 兴 ␻ ,
a ␳ p ⳵ t v p ⫹div DivM in ␻ , known variable ␭⫽ 关 p 兴 ⌫ ⫽ p e ⫺ p i , defined on ⌫ which de-
共14a兲 notes the pressure jump across the boundary of the instru-
ment.
= ⫺a 3 C␧= 共 ⵜ
⳵ tM គ v p 兲 ⫽0, in ␻ , 共14b兲
Multiplying Eq. 共9a兲 by a test function p * :R3 →R, and
v p ⫽ ⳵ n v p ⫽0, on ␥ 0 , 共14c兲 Eq. 共9b兲 by a test function vᠪ a* :R3 →R3 , and using the Green
formula leads to Eqs. 共16a兲 and 共16b兲. It remains to take into
to which the conditions on the free boundary given by Eq. account the boundary conditions given by Eqs. 共10兲 and 共11兲
共7兲 are added. The variational formulation of Eq. 共14兲 is ob- on the surface of the instrument, which is done in a weak
tained by multiplying Eq. 共14a兲 by a test function v * p :␻ sense 关Eq. 16共c兲兴. This leads to the following problem:
→R and by multiplying equation Eq. 共14b兲 by a test function
M= * : ␻ →R4 . Using then an integration by parts on ␻ leads Find p 共 t 兲 :R3 → R,vᠪ a 共 t 兲 ⬋R3 → R3 , and ␭(t):⌫→R such
to the following mixed formulation: that
= (t): ␻ →R4 such that
Find v p (t): ␻ →R and M d
冕␳ 1
2 p p *⫹ 冕 p * div vᠪ a ⫽0, ᭙p * , 共16a兲
d
dt
冕 a ␳ pv pv *
p⫺ 冕 = •ⵜ
DivM គ v*
p⫺ 冕⳵ = nគ 兲 • ␶គ 兴v *
␶ 关共 M p
dt R3 ac a R3

冕␳ 冕 冕
␻ ␻ ␥f
d

冕
ᠪ a •vᠪ a* ⫺
av p div vᠪ *
a⫺ 共 vᠪ a* •N
គ 兲 ␭⫽0, ᭙vᠪ *
a ,
dt R3 R3 ⌫
⫹ a ␳ pR pv pv *
p
␻ 共16b兲

⫽⫺T ⳵ x u s 共 l s ,t 兲v *
p 共 x 0 ,y 0 兲 ⫺ 冕
␻
关 p 兴 ␻v *
p , ᭙v*
p , 共15a兲 冕␻
v p␭ *⫺ 冕 ⌫
共 vᠪ a •N
គ 兲 ␭ * ⫽0, ᭙␭ * . 共16c兲

d
dt
冕 ␻
a ⫺3 C⫺1 M
= :M
= *⫹ 冕 ␻
= * •ⵜ
DivM គ vp In Eq. 共16兲, vᠪ a and p are defined in the complete domain R3 ,
and not only in the domain ⍀⫽R3 ⶿⌫. It will then be possible

冕⳵
to use a regular mesh for the approximation of the acoustic
⫹ = * nគ 兲 • ␶គ 兴v p ⫽0,
␶ 关共 M = *.
᭙M 共15b兲 field. It turns out that this choice leads in fact to a finite
␥f difference scheme, which is the major interest of the ficti-
The main interest of this formulation is that it can be ap- tious domain method. In addition, the geometry of the instru-
proximated using standard Lagrange finite elements 共see Sec. ment is well fitted, since it relies on a triangular mesh of its
III B兲. surface for the approximation of ␭, which is easy to con-
struct.

2. The fluid – structure interaction

With regards to the large size of the problem, the finite 3. The string
difference method would certainly be an efficient approach Because the plate velocity v p has been introduced in
for solving the acoustic equation in terms of computational order to solve the Kirchhoff–Love’s problem 关Eq. 共14兲兴, one
time, since it relies on the use of a regular mesh made of has to differentiate in time the boundary condition at the
small cubes. Unfortunately, this choice leads to a staircase bridge given by Eq. 共3兲. This leads naturally to introduce the
approximation of the complex geometry of the guitar which string velocity v s ⫽ ⳵ t u s . It can be shown that in order to
generates spurious diffractions.11 Furthermore, the construc- preserve the energy identity 关Eq. 共13兲兴, one also has to intro-
tion of a coupling scheme with the finite element method duce the constraint q⫽T ⳵ x u s , which appears, in particular,
used for the plate problem is not straightforward because of in the expression of the force exerted by the string on the
the incompatibility of the meshes. As an alternative, the finite plate 关Eq. 共8兲兴. The following problem of first order in space
element method allows one to take into account the shape of and first order in time is obtained:
the instrument accurately. However, this method requires the
construction of a tetrahedric mesh and leads to a quite ex- ␳ s ⳵ t v s ⫺ ⳵ x q⫽ f s , in]0,l s 关
pensive resolution scheme, with regard to computing time
and memory. ⳵ t q⫺T ⳵ x v s ⫽0, in0,l s [
In practice, we used a fictitious domain method, first 共17兲
introduced by Glowinsky,12 and which has been successfully v s 共 0,t 兲 ⫽0, ᭙t
applied to kettledrums by Rhaouti et al.1 The fictitious do-
main method preserves the advantages of both previously v s 共 l s ,t 兲 ⫽ v p 共 x 0 ,y 0 兲 , ᭙t.
mentioned methods: the efficiency of the finite difference
method in terms of computing time, and the good approxi- The variational formulation of Eq. 共17兲 is obtained by mul-
mation of the shape of the guitar. Its leading idea is to refor- tiplying the first equation by a test function v s* :]0,l s 关 →R,
mulate the problem in a domain of simple geometry which and the second by a test function q * :]0,l s 关 →R, and using
ignores the instrument. The fluid–structure interaction is integration by part. This leads to the following problem:
then taken into account via the introduction of the new un- Find v s : 关 0,l s 关 →R and q:]0,l s 关 →R such that

3372 J. Acoust. Soc. Am., Vol. 114, No. 6, Pt. 1, Dec. 2003 Derveaux et al.: Time-domain simulation of a guitar
 FIG. 3. Mesh for ( v s h ,q h ). 共a兲 Mesh of ]0,l s 关 . 共b兲 Degrees of freedom for v s . 共c兲 Degrees of freedom for q.

d
dt
冕␳
0
ls
冕⳵ * 冕
s v s v s* ⫺
0
ls
xq v s ⫽
0
ls
f s v s* , ᭙ v s* , 共18a兲
below. In the following P k denotes the polynomial functions
of one or more variables of degree less than or equal to k.
The continuous variables v s and q are approximated by
d
dt
冕
0
ls 1
T
qq * ⫹ 冕⳵ * *
ls

0
xq v s ⫺q 共 l s ,t 兲v p 共 x 0 ,y 0 兲 ⫽0, ᭙q *
discrete variables in space denoted v s h and q h , respectively.
A regular mesh of the string, made of small segments of
共18b兲 length h c is constructed 关Fig. 3共a兲兴. v s h is a piecewise con-
4. Variational formulation of the complete problem stant and is thus entirely determined by its values at the
center of each segment, which are called the degrees of free-
Considering the additional unknowns ␭ and q, which dom 共dof兲. In other words v s h is discretized using P 0 discon-
were introduced in order to solve the acoustic equation and
tinuous finite elements. q h is a continuous piecewise linear
the string equation, respectively, one has to replace
function and it is thus fully determined by its values at each
T ⳵ x u s (l s ,t) by q(l s ,t) and 关 p 兴 ␻ by ␭ 兩 ␻ in Eq. 共15a兲. Equa-
node of the mesh ( P 1 continuous finite elements兲. The de-
tion 共15a兲 is thus replaced by
grees of freedom for v s h and q h are represented in Figs. 3共b兲
d
dt
冕␳ * 冕
␻
pv pv p ⫺
␻
= •ⵜ
DivM គ v*
p⫺ 冕⳵␥f
= nគ 兲 • ␶គ 兴v *
␶ 关共 M p
and 共c兲, respectively.
Given a triangular mesh of the soundboard ␻, with the
smallest edge length denoted h p 关Fig. 4共a兲兴, the variables v p
⫹ 冕␳ * ␻
pR pv pv p
and M = are discretized by means of P 2 alike-continuous fi-
nite elements.13 This particular choice reduces the numerical

冕
dispersion and leads to the required estimation of the eigen-
⫽⫺q 共 l s ,t 兲v *
p 共 x 0 ,y 0 兲 ⫺ ␭ 兩␻v *
p , p : ␻ →R. 共19兲
᭙v* frequencies of the soundboard.13 With regard to sound syn-
␻ thesis it is an important aim to simulate the eigenfrequencies
Finally, the variational formulation of the complete of the soundboard with sufficient accuracy, in view of the
problem is given by the set of equations 共15b兲, 共16兲, 共18兲, ability of the human ear to detect small frequency variations.
共19兲. In this context, an order of magnitude of 1% seems a reason-
able target. An analysis of the performance of finite element
approximations of different orders for the plate equation is
B. Space discretization presented in Appendix B. Another advantage of using P 2
In this variational formulation, the unknowns elements lies in the elimination of the unknown M = h in the
( v s ,q, v p ,M = ,p,vᠪ a ,␭) and the test functions final scheme, as it will be shown in Sec. III C.
( v s* ,q * , v * ,M= * ,p * ᠪ
,v * ,␭ * ) are chosen in appropriate The discrete variables v p h and M = h are determined by
p a
spaces so that all integrals are defined. q has to be chosen in their values at the vertex, at the gravity center of each tri-
a space of continuous functions because of the term ⳵ x q in angle, and at the center of each edge. The degrees of freedom
Eq. 共18兲. For similar reasons, v p , M = , and ␭ will also be for v p h and M= h are represented in Figs. 4共b兲 and 共c兲, respec-
chosen in a space of continuous functions. On the other tively.
hand, the spaces for v s and for p do not require any condition To simulate the free space, the computations are re-
of continuity. Finally, vᠪ a is in a space of vector field with stricted to a box of finite size denoted C, with the help of
continuity of the normal component across the surface, be- higher-order absorbing boundary conditions defined on the
cause of the term div vᠪ a in Eq. 共16兲. faces of this cube.1,14 For the sake of conciseness, these ad-
The finite dimensional spaces chosen for the approxima- ditional equations C are not given here. Given a regular mesh
tion of the variational formulation 共15b兲, 共16兲, 共18兲, 共19兲 and of C, made of small cubes with edge length denoted h a 关Fig.
satisfying the adequate continuity conditions are described 5共a兲兴, p is discretized by means of P 0 -discontinuous finite

= h ). 共a兲 Mesh
FIG. 4. Mesh for ( v p h ,M
of ␻. 共b兲 Degrees of freedom for v p h .
共c兲 Degrees of freedom for M = h.

J. Acoust. Soc. Am., Vol. 114, No. 6, Pt. 1, Dec. 2003 Derveaux et al.: Time-domain simulation of a guitar 3373
 FIG. 5. Mesh for (p h ,vᠪ a h ). 共a兲 Mesh
of C. 共b兲 Degrees of freedom for p h .
共c兲 Degrees of freedom for vᠪ a h .

elements. The field p h is thus determined by its values at the p dph

center of each cube. The approximation of vᠪ a is made with M ha ⫹G Th vᠪ a h ⫽0, 共20f兲
dt
the first-order Raviart–Thomas finite element space.15 Each
component of vᠪ a h is linear in one direction and constant in B ␻ h v p h ⫺B ⌫ h vᠪ a h ⫽0, 共20g兲
the other two directions. vᠪ a h is entirely determined by the
where A Th denotes the transpose of a matrix A h . M sh , M qh ,
value of its normal component across each face of the cubes. MM a pa
h , M h , and M h are mass matrices. f s h denotes a vector
The degrees of freedom for p h and vᠪ a h are represented in
which approximates the plucking force exerted on the string.
Figs. 5共b兲 and 共c兲, respectively. D h , J h , H h , C h , B ␻ h , G h , and B ⌫ h are matrices which
Finally, a triangular mesh is given on the surface of the
represent the discrete operators approximating the continu-
guitar ⌫, with the smallest edge length denoted h ␭ 关see Fig.
ous operators of the variational formulation 共15b兲, 共16兲, 共18兲,
6共a兲兴. The pressure jump ␭ is discretized by means of P 1
共19兲. G h , for example, is the approximation of the three-
continuous finite elements. Its approximation ␭ h is deter-
dimensional gradient operator ⵜ គ.
mined by the values at the vertex of each triangle. The de-
grees of freedom of ␭ h are represented in Fig. 6共b兲.
The space discretization of the problem is then obtained C. Time discretization
by substituting = h , p h ,vᠪ a h ,␭ h )
( v s h ,q h , v p h ,M for
For the time discretization, a classical explicit centered
= ,p,vᠪ a ,␭) and ( v s* ,q *
( v s ,q, v p ,M h , v *
ph ,M= *
h , p *
h ,vᠪ *
ah ,␭ h* ) finite difference scheme of second order is used for the string
h
* *
for ( v s ,q * , v p ,M= * , p * ,vᠪ a ,␭ * ) in the variational formula-
* and for the acoustic field. For the plate equation, it appears
tion 共15b兲, 共16兲, 共18兲, 共19兲. The resulting differential system that the choice of an explicit finite difference scheme in time
can be written in the following matrix form: imposes the selection of a small time step ⌬t because of a
d v ph restricting stability condition of the form ⌬t⭐Ch 2p , where C
M hp = h ⫹R p M hp v p h ⫽⫺J h q h ⫺ 共 B ␻ h 兲 T␭ h ,
⫺H Th M is a constant depending on the physical parameters of the
dt plate. A natural idea would be to choose an implicit, uncon-
共20a兲 ditionally stable scheme, allowing a larger ⌬t. However, in
dM= h this case the analysis of the error made on the eigenfrequen-
MM
h ⫹H h v p h ⫽0, 共20b兲 cies of the soundboard shows that, for accuracy consider-
dt
ations, one has to choose ⌬t proportional to h 2p as for the
d v sh explicit scheme.16 For this reason, it has been finally decided
M sh ⫺D h q h ⫽ f s h , 共20c兲 to solve the plate equation analytically in time, which allows
dt
one to choose ⌬t without degrading the accuracy of the se-
dq h midiscretized spatial scheme.
M qh ⫹D Th v s h ⫺J Th v p h ⫽0, 共20d兲 In a first step, the size of the plate system is reduced by
dt
eliminating the bending moment M = h in Eqs. 共20a兲 and
dvᠪ a h 共20b兲, which leads to
M ah ⫺G h p h ⫺ 共 B ⌫ h 兲 T␭ h ⫽0, 共20e兲
dt d 2v ph d v ph
M hp 2 ⫹R p M hp ⫹K h v p h
dt dt
dq h d␭ h
⫽⫺J h ⫺B ␻T , ᭙t⭓0, 共21兲
dt h dt

where K h is the matrix defined by

K h ⫽H h 共 M M
h 兲
⫺1 T
Hh . 共22兲
The matrix MM
is computed using quadrature formula so
h
that it reduces to a 4⫻4 block diagonal matrix, without loss
FIG. 6. Mesh for ␭ h . 共a兲 Mesh of ⌫. 共b兲 Degrees of freedom for ␭ h . of accuracy. This technique, called mass lumping, permits

3374 J. Acoust. Soc. Am., Vol. 114, No. 6, Pt. 1, Dec. 2003 Derveaux et al.: Time-domain simulation of a guitar
one to invert M M h , and, consequently, permits one to com-
The scheme is chosen in order to obtain a discrete energy
pute K h easily.13 identity similar to Eq. 共13兲. In addition, we look for a scheme
For the resolution of Eq. 共21兲, one has to sample the in which almost all computations are explicit 共i.e., without
solution. Given a time step ⌬t, v p h and v̇ p h ⫽d v ph /dt are matrix inversion兲, in particular for the 3D part. These two
calculated at the successive instants of time t n⫹(1/2) ⫽(n reasons lead to computing the variables v s h , p h , and ␭ h at
⫹ 12 )⌬t. Equation 共21兲 is solved on each interval time t n⫹(1/2) , while q h and vᠪ a h are computed at time t n , and
关 t n⫺(1/2) ,t n⫹(1/2) 兴 . Since the string and acoustic equations of to discretize Eq. 共20g兲 which is now differentiated in time.
the problem are solved by means of finite differences with The following system is obtained:
time step ⌬t, the right-hand side is constant on this interval. 共 1/2兲
v sn⫹ ⫺ v sn⫺ 共 1/2兲
The following equation is thus solved at each time step: h h
M sh ⫺D h q nh ⫽ f sn ,
⌬t h
d 2v ph d v ph
M hp ⫹R p M hp ⫹K h v p h q n⫹1 ⫺q nh
dt 2 dt M qh
h
⫹D Th v sn⫹ 共 1/2兲 ⫺J Th v n⫹ 共 1/2兲
⫽0,
⌬t h p h
q n⫹1
h ⫺q n⫺1 ␭ n⫹(1/2)
h ⫺␭ n⫺(1/2)
h
⫽⫺J h ⫺B ␻T , ᠪvan⫹1 ⫺vᠪ an
2⌬t h ⌬t h h 共 1/2兲 共 1/2兲
M ah ⫺G h p n⫹ ⫺B ⌫T ␭ n⫹ ⫽0, 共26兲
n⫺ 共 1/2兲 n⫹ 共 1/2兲 ⌬t h h h
᭙tⱬ]t ,t 关,
共 1/2兲 共 1/2兲
共 1/2兲 共 1/2兲 p n⫹ ⫺ p n⫺
v p h 共 t n⫺ 共 1/2兲 兲 ⫽ v n⫺ and v̇ p h 共 t n⫺ 共 1/2兲 兲 ⫽ v̇ n⫺ h h
ph ph . 共23兲 p
M ha ⫹G h vᠪ an ⫽0,
⌬t h

The resolution of Eq. 共23兲 relies on the computation of 共 1/2兲 共 1/2兲

v n⫹
ph ⫺ v n⫺
ph ᠪvn⫹1
a ⫺vᠪ n⫺1
a
the eigenmodes of the positive definite matrix K h , and on B ␻h ⫺B ⌫ h
h h
⫽0.
the ordinary differential equation theory. In the absence of ⌬t 2⌬t
damping, the solution is given by The resolution of the linear system consisting of Eqs.

v n⫹
ph
共 1/2兲
冋
⫽ cos共 冑K h ⌬t 兲v n⫺
ph
共 1/2兲
⫹
sin共 冑K h ⌬t 兲
冑K h
v̇ n⫺
ph
共 1/2兲
册 共24兲 and 共26兲 imposes, in particular, to invert the mass ma-
p
trices M sh , M qh , M ah , and M h a . To avoid this inversion at
each time step, these matrices are reduced to diagonal matri-

冉
ces using the mass-lumping technique 关see Eq. 共22兲兴.
I⫺cos共 冑K h ⌬t 兲 q n⫹1 ⫺q n⫺1
⫹ ⫺J h
h
We seek for ( v sn⫹(1/2) ,q n⫹1 , v n⫹(1/2) , v̇ n⫹(1/2) ,vᠪ an⫹1 ,
冑K h h h ph ph h
2⌬t
p n⫹(1/2) ,␭ n⫹(1/2) ), with all other terms being known. It is

冊
h h
␭ n⫹ 共 1/2兲
⫺␭ n⫺ 共 1/2兲 easy to show that if ␭ n⫹(1/2) is known, the rest of the com-
h h
⫺B ␻T , 共24兲 putation is entirely explicit. For the computation of ␭ n⫹(1/2) ,
h ⌬t
one has to invert the matrix C ␭ , where
with a similar equation for v̇ n⫹(1/2)
ph . For our computations,
only the first 50 modes of the plate were retained, which
corresponds to a cutoff frequency nearly equal to 3 kHz.
C ␭⫽
2B ␻ h
⌬t 冋 冉
I⫹J h
1⫺cos共 冑K h ⌬t 兲
冑K h 冊册
J Th
⫺1

B ␻T
h

The introduction of the damping terms leads in fact to ⫹⌬tB ⌫ h 共 M ah 兲 ⫺1 B ⌫T . 共27兲

h
exponentially damped sinusoidal solutions for each modal
component. It is a well-known fact that the fluid and vis- C ␭ is symmetric and positive without any condition on the
coelastic terms introduced in the model 关 R p and ␩ p in Eq. time step. Furthermore, this matrix is definite, provided that
共5兲兴 lead to a damping factor ␣ n ⫽(R p ⫹ ␩ p ␻ 2n )/2 for the nth the following compatibility between the mesh of C and the
eigenmode of K h with pulsation ␻ n . At this stage, notice that mesh of ⌫ is satisfied:
it is possible to choose any other frequency dependence for
the damping factors, without requiring that it has a local h␭
⭓␣, 共28兲
expression in the time domain. This is one interest of the ha
spectral method which allows one to model the internal where ␣ is a constant. This compatibility condition, which is
damping in the plate with such a flexibility. Considering the not demonstrated here, is a consequence of the consistency
damping ratio values for wooden plate given in Ref. 5 in the of the fictious domain method.17 In practice, ␣ is approxi-
range 关0–3 kHz兴 corresponding to our case, a linear law has mately equal to 1.1.
been chosen, of the form The stability of the scheme given by Eqs. 共24兲–共26兲 is
␣ n ⫽ 共 R p ⫹ ␩ p ␻ n 兲 /2. 共25兲 guaranteed with the help of an energy identity similar to Eq.
共13兲. The details of the proof are beyond the scope of this
One must now write a finite difference scheme for the paper. This leads to the following conditions which have to
time discretization of Eqs. 共20c兲–共20g兲. The crucial point is be satisfied for ensuring the stability:

冑
to propose a stable coupling between two radically different
time-resolution techniques: a continuous one for the plate T ⌬t
共 i兲 ⭐1, 共29兲
equation and a discrete one for the string and air equations. ␳s hc

J. Acoust. Soc. Am., Vol. 114, No. 6, Pt. 1, Dec. 2003 Derveaux et al.: Time-domain simulation of a guitar 3375
 FIG. 7. Admittance curves at the
bridge. A pulse force is applied to the
lowest guitar string. The strings are
damped. 共a兲 Soundboard in vacuo,
plate thickness 2.9 mm. 共b兲 Sound-
board in vacuo, plate thickness 1.45
mm. 共c兲 Guitar coupled to the air, plate
thickness 2.9 mm. 共d兲 Guitar coupled
to the air, plate thickness 1.45 mm.

⌬t 1 guitar body in vacuo compared to a guitar body in air. This

共 ii兲 ca ⭐ . clearly shows the various effects of both radiation and cavity
ha )
field on the soundboard vibration.
Another interest of our model is the use of a variational
One should notice that these two conditions are optimal, in
formulation from which energetic quantities can be very eas-
the sense that they are identical to the usual stability condi-
tions obtained for the standard finite difference discretization ily derived. We use that property for gaining important infor-
of the uncoupled 1D and 3D wave equation. This result mation on the energy balance between strings, soundboard,
shows the robustness of the coupling scheme. and air.
The numerical parameters chosen for the numerical The simulated sound pressure is observed in the time
simulations presented in the following section and which ful- domain. Comparisons are made for two different values of
fill the conditions Eqs. 共28兲 and 共29兲 are given in Table I. box height and plate thickness, all other material and geo-
metrical parameters remaining unchanged. The results show
a clear difference between the waveforms. A spectral analy-
IV. PRELIMINARY RESULTS AND DISCUSSION sis is performed on the initial part of these simulated sounds.
This analysis shows the role of both the string–body and
The prime and most visible interest of our model is the air–body coupling in the spectral content and decay pattern
computation of the complete 3D sound field radiated by the of a guitar sound.
guitar 共see Fig. 15兲. However, it is also necessary to validate Finally, the spatial properties of the sound field are dis-
our method through comparisons with existing results of the played in terms of instantaneous directivity patterns.
literature on guitars. In what follows, we start by presenting
simulation results in terms of admittance, a quantity which is A. Admittance at the bridge
widely used for characterizing the essential coupling proper- The admittance is defined as the ratio between velocity
ties between strings and plate loaded by air in stringed in- and force. In order to compute such a quantity, we use our
struments. In addition to the validation purpose, the specific
model to simulate the velocity waveform consecutive to a
interest of the presented results on admittances is to show the
force impulse f (x,y,t) at one particular point at the bridge.
possible use of our model for comparing two different
In practice, the strings are damped and we use f (x,y,t)
soundboards mounted on the same instrument. In our ex-
ample, only the thickness is changed, all other parameters ⫽h(t) ␦ x 0 ,y 0 (x,y), where h(t) is given in Table I with t 1
remaining identical. Such a modification would certainly be ⫽t 2 ⫽5•10⫺3 s. The selected point (x 0 ,y 0 ) corresponds to
hard to realize on a real instrument. Finally, we show the the attachment point of the lowest guitar string 共6th string,
differences, in terms of admittance and decay times, for a note E2 ). Using standard FFT tools, the admittance is simply

3376 J. Acoust. Soc. Am., Vol. 114, No. 6, Pt. 1, Dec. 2003 Derveaux et al.: Time-domain simulation of a guitar
TABLE II. Analysis of simulated guitar sound pressure note E2 共open 6th quency of the rigid cavity and f 2 is the lowest structural
string兲 Damping factors and frequencies. The string partials are in bold. 共a兲
Reference 共plate thickness 2.9 mm, cavity height 10.4 cm兲. 共b兲 Large cavity
eigenfrequency of the soundboard. Here, we have f ⫹
共plate thickness 2.9 mm, cavity height 21 cm兲. 共c兲 Thin plate 共plate thickness ⫽200.6 Hz, f ⫺ ⫽96.6 Hz, f 1 ⫽117 Hz, and f 2 ⫽181.9 Hz,
1.45 mm, cavity height 10.4 cm兲. which means that Eq. 共30兲 is verified to within ⫾5%. This
discrepancy is mainly due to the fact that we do not have
共a兲 Reference 共b兲 Large cavity 共c兲 Thin plate
here an isolated system with two degrees of freedom, but a
F 共Hz兲 ␣ 共s⫺1兲 F 共Hz兲 ␣ 共s⫺1兲 F 共Hz兲 ␣ 共s⫺1兲 truncated system extracted from a continuous system. Fre-
82.4 0.4 76.7 1.6 55.5 1.6 quency f 2 is obtained from the simulation of the complete
96.6 2.1 82.6 0.4 82.6 0.4 guitar ‘‘in vacuo’’ 共see Table II兲. Frequency f 1 , like the other
164.7 0.6 164.2 1.0 135.0 4.8 acoustic modes of the uncoupled system, is obtained from
200.6 20.8 183.3 14.9 144.4 6.5 the simulation of the guitar coupled with external air and
247.8 0.8 247.8 0.7 163.0 11.2
272.3 7.9 274.7 8.3 166.4 1.7
cavity while keeping the top plate rigidly fixed. Frequencies
289.0 7.0 293.0 7.7 216.4 11.9 f ⫹ and f ⫺ are derived from a spectral analysis performed
330.6 0.7 330.6 0.7 248.4 1.0 either on the pressure or on the plate vibration, for the com-
372.9 12.9 355.6 7.2 266.7 7.4 plete guitar.
413.0 0.9 412.9 0.9 296.1 30.0 The other lowest frequencies of the uncoupled acoustic
437.6 19.0 436.2 15.0 317.0 8.7
495.6 1.3 495.5 1.4 319.0 8.3
modes are: 355, 699, 905, and 1046 Hz. It is interesting to
508.7 1.6 505.6 1.1 331.3 1.0 notice that Eq. 共30兲 also applies to other subsets of modes,
517.1 21.9 515.0 19.7 340.4 8.7 provided that these subsets are reasonably well separated
558.3 15.6 578.6 1.1 388.5 12.0 from the other modes. For the reference guitar, the coupled
578.5 1.1 656.5 11.4 413.4 0.9 共called structural–acoustic or SA兲 modes at 373 and 437.6
652.7 9.3 662.0 4.0 436.6 9.7
661.1 3.1 672.1 24.4 461.6 12.4 Hz, for example, are due to the coupling between the acous-
672.7 26.7 707.4 7.6 495.8 1.0 tic mode at 355 Hz and the structural mode at 450 Hz.
716.0 13.0 743.7 1.4 531.2 8.2 Similar conclusions can be drawn from the comparison
743.7 1.5 532.3 20.4 between Figs. 7共b兲 and 共d兲, corresponding to the thin plate,
557.4 15.7
with stronger variations of frequencies due to coupling in
578.6 1.1
648.8 16.1 case 共d兲 compared to case 共c兲. In this latter case, notice that
660.6 1.6 the modal density of the fluid–structure coupled system is
670.4 13.7 high so that the maxima of the admittance curve cannot be
693.4 39.4 well separated. This property also appears in case 共b兲, though
708.6 23.1
for higher frequencies than for case 共d兲.
720.5 29.1
735.2 23.1 Finally, one should highlight the fact that the model
744.4 2.4 yields admittance curves up to 3 kHz, as shown in Fig. 8.
This is an improvement compared to previously published
results on simulated guitars which are generally limited to
obtained from the computation of the frequency response less than 1 kHz.19 These figures show, among other things,
between velocity and force. that the average slope for the mobility of the thin plate is
In order to illustrate the capabilities of our model, Fig. 7 relatively higher than the one of the reference soundboard.
shows typical results obtained in four different situations. This suggests that the guitar with the thin soundboard is
The case 共a兲 corresponds to the reference guitar ‘‘in vacuo.’’ relatively more efficient in low frequencies compared to the
The reference guitar has a soundboard of thickness equal to high-frequency domain.
2.9 mm and a cavity height of 10.4 cm. In Fig. 7共b兲, the Detailed analysis of the simulated plate velocity at the
thickness is equal to 1.45 mm, all other parameters of the junction between string and bridge, using the matrix-pencil
guitar remaining unchanged. Figures 7共c兲 and 共d兲 correspond method,20 yields accurate values of the damping factors. Fig-
to the same configurations as 共a兲 and 共b兲, respectively, when ures 9共a兲 and 共b兲 show the effect of air loading and cavity on
the guitar vibrates in air. these factors for the soundboard vibration. For clarity, only
As expected, the modal density is higher for case 共b兲 the lowest modes, between 0 and 800 Hz, are presented. In
than for case 共a兲, and all frequencies are lowered 共see also this bandwidth, the admittance 关Fig. 7共a兲兴 shows 8 structural
Table II兲. Figure 7共c兲 clearly shows additional peaks in the modes 共in vacuo modes兲 from 182 Hz 共1,1兲-mode to 704 Hz
admittance curve, variation of the bandwidth for some peaks, 共1,4兲-mode. For these modes, the damping factors ␣ 共in s⫺1兲
and slight variations of the peak frequencies, compared to are only due to the material losses, which are assumed here
case 共a兲. These phenomena are due to the coupling of the to increase linearly with frequency 关see Eq. 共25兲兴.
soundboard with cavity and external air. The frequency For the plate coupled with air 关case 7共c兲兴, 12 structural–
change of the two lowest peaks, in particular, follow the acoustic modes 共SA modes兲 were found. At this stage, notice
well-known rule18 that the matrix-pencil method yields accurate values of the
eigenfrequencies and damping factors for the in vacuo
2
f⫹ ⫹f⫺
2
⫽ f 21 ⫹ f 22 , 共30兲
soundboard, which validates the analysis. The damping fac-
where f ⫹ and f ⫺ are the two lowest frequencies of the tors for the air–body system are compared to those obtained
coupled system, whereas f 1 is the lowest acoustic eigenfre- from the analysis of the plate admittance in vacuo. These

J. Acoust. Soc. Am., Vol. 114, No. 6, Pt. 1, Dec. 2003 Derveaux et al.: Time-domain simulation of a guitar 3377
 FIG. 8. Admittance curve at the
bridge, in the interval 关0–3000兴 Hz.
共a兲 Reference guitar in air. 共b兲 Guitar
with thin plate in air.

factors are proportional to the sound power radiated by the mechanical energy due to the plucking is transferred to the
instrument: high values of the damping coefficients corre- string. During this period of time, only a small amount of
spond to strongly radiating modes.21 For the reference case energy is transformed into quasistatic deformation of the
presented here, it can be seen, for example, that the SA plate and motion of the air. As the finger leaves the string,
modes at 201, 438, 517, and 673 Hz are efficient radiators. both plate and air energy increase almost instantaneously.
For the reference guitar, it turns out that the Q-values, The air-energy history is then dominated by almost periodic
related to the damping factors ␣ by the relation Q⫽ ␲ f / ␣ , exchange with fundamental frequency equal to the funda-
are slightly higher in our simulation than those observed ex-
mental frequency of the string. The plate-energy history is
perimentally in real guitars 共30–130 compared to 20– 80兲,
also periodic with the same fundamental frequency. For the
for the SA modes.22 This should incite us to slightly increase
the plate damping in future simulations in order to obtain case presented here, plate and air vibrations are in antiphase.
sounds closer to real guitar sounds. Also, the fluid damping The plate-energy history is also characterized by a modula-
and the visco-thermal losses at the boundaries are not taken tion at a frequency nearly equal to the 5th harmonic of the
into account in our simulations. As a consequence, the simu- fundamental. This property is due to the fact that the 5th
lated Q-values of some SA modes are high compared to real harmonic of the string 共around 413 Hz兲 is strongly coupled
guitars, as it can be observed, for example, on the lowest to an efficiently radiating structural–acoustic mode at 438
structural–acoustic mode at 96.6 Hz. Hz.
In conclusion, the admittance analysis shows that, de- After the pluck, the average magnitude of both plate and
spite some crude aspects of the modeling 共bracing, rigid air energies corresponds approximately to 1 percent of the
back, air damping兲, our system is able to reproduce the main string energy. This proportion varies of course with the cou-
properties of a real guitar. In addition, the separate modeling pling properties between string and plate, and with the radia-
of structural, acoustics, and structural–acoustic modes, re- tion efficiency of the instrument. However, the figure shown
spectively, not only validates the fictitious domain method, here corresponds to a typical order of magnitude for the low-
but also yields a powerful tool for the estimation and predic-
est open string of a guitar, and this explains why guitar notes
tion of changes in radiated sound power consecutive to struc-
usual last for several seconds. As shown in Fig. 11, simula-
tural modifications of the instrument.
tions made with the thickness of the top plate reduced by a
B. Energy exchange factor of 2, all other parameters remaining unchanged, show
Figure 10 shows the evolution of various energy quan- a significant increase in the radiated sound power. In this
tities vs time for the plucked open string E2 共fundamental case the average value of the air energy is equal to 10% of
82.6 Hz兲. the string energy. As a consequence, the string vibrations are
As the string is in contact with the finger, most of the damped more rapidly.

FIG. 9. Damping coefficients ␣ 共s⫺1兲

for the lowest modes of the plate–air
system 共0– 800 Hz兲. Comparison be-
tween structural and structural–
acoustic modes. 共a兲 Plate thickness 2.9
mm. Structural modes 共*兲; structural–
acoustic modes 共䉮兲. 共b兲 Plate thick-
ness 1.45 mm. Structural modes 共*兲;
structural–acoustic modes 共䉮兲.

3378 J. Acoust. Soc. Am., Vol. 114, No. 6, Pt. 1, Dec. 2003 Derveaux et al.: Time-domain simulation of a guitar
 FIG. 10. 共a兲 Evolution of the energy of
string, plate, air, and total energy vs
time for the plucked open string E2
共fundamental 82.6 Hz兲 during the first
100 ms of the tone. Reference guitar.
共b兲 Plate and air energy represented
with appropriate scale.

C. The sound of the guitar in time modes are strongly modified 共such as the lowest one at 96.6
Examples of guitar pressure waveforms are simulated in Hz, which becomes equal to 76.7 Hz兲. Other SA modes,
three different situations, for a pluck on the open E2 共6th兲 which are mainly imposed by the lateral dimensions of the
string. The first waveform shown in Fig. 12共a兲 corresponds cavity, remain almost unchanged 共272.3 and 289 Hz兲. For the
to the reference guitar 共see Table I兲. The waveform shown in simulated guitar sound with a thin plate, the pressure spec-
Fig. 12共b兲 is obtained for a doubling of the cavity height, all trum shows 22 SA modes between 0 and 800 Hz. This is,
other parameters remaining unchanged. Finally, the wave- again, a result in accordance with the admittance analysis. As
form shown in Fig. 12共c兲 corresponds to a top plate with a consequence of the air–plate coupling, the lowest SA
thickness divided by a factor of 2, compared to the reference mode, in particular, is significantly reduced 共55.5 Hz兲, com-
case, all other parameters remaining unchanged. For these pared to the reference case 共96.6 Hz兲.
three simulations, the tone duration is equal to 3 s. The pres- Figures 13 and 14 are aimed at summarizing the effects
sure is simulated at a specific point in free space whose co- of the various coupling in the guitar through careful exami-
ordinates are x⫽40 cm, y⫽5 cm, and z⫽60 cm 共see Fig. 1 nation of the damping factors of the string’s partials. Com-
for the definition of the axes兲. pared to Fig. 9, which illustrated the air–body coupling only,
The waveforms look clearly different. An accurate fre- these data show that the sound of the guitar is governed by
quency analysis of these waveforms has been carried out the combination of both air–body and string–body coupling.
with the help of the matrix-pencil method.20 This analysis The solid line in Fig. 13 shows the frequency-dependent
yields the frequencies and damping factors shown in Table damping on the isolated string: the quadratic function is a
II. consequence of the damping model used 共see Sec. II兲. The
As expected, the pressure spectra are composed of string coupling of the string with the plate in vacuo 共dashed line兲
modes and of the SA modes of the guitar body coupled with contributes to increase its damping factors, with a more pro-
air. All three spectra exhibit the nine lowest string modes and nounced effect for the 2nd, 4th, 7th, and 8th partials, for
a variable number of SA modes, between 0 and 800 Hz. It which the degree of coupling with the plate modes is higher
can be seen that the coupling does not affect the frequencies than for the other partials. The coupling of the plate with air
and damping factors of the string modes significantly. As modifies these factors substantially. In general, the damping
expected, the reference spectrum shows, in addition, the 12 factors increase due to radiation into the free space. How-
SA modes observed in the admittance analysis. The damping ever, since the air coupling also modifies the plate’s modes,
factors of these modes are all significantly higher than those this can in turn modify the string–plate coupling and thus
of the string modes, so that the well-known ‘‘box sound’’ in lead to a decrease of the damping factors for some string’s
the guitar transient is damped more rapidly than the string components. Figure 13 shows that it is the case here for the
components. The spectrum corresponding to the simulation 2nd, 4th, and 7th partials of the note E2 共open 6th string兲
with a large cavity shows 11 SA modes, in the same fre- coupled to the reference guitar. Similarly, Fig. 14 shows the
quency range. Due to the doubling of the height some SA effect of reducing the plate thickness on the radiation of the

FIG. 11. 共a兲 Evolution of the energy of

string, plate, air, and total energy vs
time for a plucked open string E2 共fun-
damental 82.6 Hz兲 during the first 100
ms of the tone. Soundboard thickness
1.45 mm. 共b兲 Plate and air energy rep-
resented with appropriate scale.

J. Acoust. Soc. Am., Vol. 114, No. 6, Pt. 1, Dec. 2003 Derveaux et al.: Time-domain simulation of a guitar 3379
FIG. 12. Waveforms of the pressure recorded at the point (x⫽40 cm,y⫽5 cm,z⫽60 cm) 共see Fig. 1 for the definition of the axes兲. 共a兲 Reference 共plate
thickness 2.9 mm, cavity height 10.4 cm兲. 共b兲 Large cavity 共plate thickness 2.9 mm, cavity height 21 cm兲. 共c兲 Thin plate 共plate thickness 1.45 mm, cavity
height 10.4 cm兲.

string’s components. In the frequency range under examina- be inserted in this paper. The temporal evolution described
tion, it can be seen that the thin plate is more efficient for below is better seen on a videotape and/or in animated
allowing the guitar to radiate below 400 Hz, whereas the pictures.23
reference plate is more efficient between 400 and 800 Hz. In The selected pictures are aimed at illustrating some of
the low-frequency range, this result is due to the fact that the the basic physical phenomena relative to the acoustic field
fluid–structure coupling is more pronounced for the thin generated by a guitar, during the first 45 ms of the sound. For
plate than for the reference plate. However, as the frequency this reason, the time interval between consecutive pictures is
increases, it becomes hazardous to find a general interpreta- not imposed to be kept constant. The black color corresponds
tion for the changes in damping, since the values of the fac- to a negative acoustic pressure, whereas the white color cor-
tors depend on the particular degrees of coupling between responds to a positive pressure 共see the scale兲. Figure 15共a兲
the string’s modes and SA modes. A small change of thick- 共at time 16.4 ms兲 shows the acoustic field right after the
ness may alter this coupling substantially, as is experienced release of the string: two waves of opposite signs propagate
daily by guitar makers. More work is needed here for sum- inside and outside the cavity, respectively. At time 18.4 ms
marizing the efficiency of the instrument in the medium 关Fig. 15共b兲兴, the cavity is filled with a strong, positive pres-
range, and in attempting to express the result in terms of sure field. Plate and rose contributions are in phase, which
global radiated power. yields a rather omnidirectional and homogeneous field in
front of the instrument. At time 19.2 ms, Fig. 15共c兲 shows an
example where the cavity field is not homogeneous. The ex-
D. The sound of the guitar in space
ternal field is not symmetrical, which is a consequence of the
Figure 15 shows an example of the spatial distribution of fact that the various radiating components of the instruments
sound pressure in a plane perpendicular to the guitar, in the are no longer in phase. Figure 15共d兲 共at time 21.1 ms兲 shows
direction of the strings. For obvious reasons, only a small the opposite situation compared to case 共b兲: the internal field
number of pictures which ‘‘sample’’ the acoustic field with a is negative, like the near field of the instrument which shows
rather large period of time 共typically a few milliseconds兲, can

FIG. 14. Damping factors of the string partials coupled to soundboard and
FIG. 13. Damping factors of the string partials. 共*———兲 uncoupled string, air 共䉮-•-兲 string coupled to the reference soundboard 共plate thickness 2.9
共䊐---兲 string coupled to the soundboard, 共䉮-•-兲 string coupled to sound- mm兲, 共〫———兲 string coupled to the thin soundboard 共plate thickness 1.45
board and air. mm兲.

3380 J. Acoust. Soc. Am., Vol. 114, No. 6, Pt. 1, Dec. 2003 Derveaux et al.: Time-domain simulation of a guitar
FIG. 15. Sound field radiated by the guitar in the plane y⫽0 for a plucked open string E2 共fundamental 82.6 Hz兲 during the 45 first ms. Linear scale 共Pa兲.
共a兲 16.4 ms. 共b兲 18.4 ms. 共c兲 19.2 ms. 共d兲 21.1 ms. 共e兲 23.1 ms. 共f兲 23.9 ms.

again a rather omnidirectional pattern. Similarly, Fig. 15共e兲 ments to the radiated field while reducing the local differ-
共at time 23.1 ms兲 shows the situation comparable to case 共c兲, ences in pressure, compared to the linear plot. The black
though with opposite signs. This leads to strong heterogene- color corresponds to 0 dB, while the white color corresponds
ities both in the internal and external field. The phenomena to 120 dB 共see the scale兲. Figure 16共a兲 共at time 9.6 ms兲
repeat themselves first pseudoperiodically during the tran- shows an example of almost-perfect omnidirectional pattern
sient where the ‘‘box sound’’ is present, and then become in front of the instrument. Figure 16共b兲 共at time 17.3 ms兲, on
progressively periodic. It can be seen, for example, that case the contrary, denotes a multipolar radiation. In addition, the
共f兲 共at time 23.9 ms兲 is comparable to case 共b兲. Animated diffraction of the sound field on the back side of the guitar is
pictures23 show, in addition, the propagation of the acoustic
clearly seen. Other examples of irregular directivity pattern
waves along the boundaries of the instruments. As a conse-
can be also observed in Fig. 16共c兲 共at time 28.3 ms兲 and Fig.
quence, the acoustic load on the structure varies continuously
16共d兲 共at time 32.4 ms兲, while Fig. 16共e兲 共at time 36.1 ms兲
with time.
The previous results suggest that the directivity of the shows again an almost constant directivity in the plane under
instrument is not a stationary pattern. This feature is better study. In most of the cases, the sound field inside the cavity
shown on a logarithmic scale, as seen in Fig. 16, which rep- is significantly louder than the external field.
resents instantaneous directivity patterns. The acoustic pres- It could be of interest to validate the pressure patterns
sure is represented in decibels 共dB兲. At time t the following obtained with the present model with noncontacting TV-
function is plotted: 20 log10兩 p(x,t) 兩 /p e , where p e ⫽2 holography techniques, such as those developed by Runne-
•10⫺5 Pa is the reference. This figure has the property of malm and Molin.24 This should be done not only for station-
enhancing the contributions of the various parts of the instru- ary waves, but also for transients.

FIG. 16. Instantaneous directivity patterns: sound field radiated by the guitar in the plane y⫽0 for a plucked open string E2 共fundamental 82.6 Hz兲 during the
45 first ms. dB scale. At time t, function 20 log10兩 p(x,t) 兩 /p e is plotted where p e ⫽2.10⫺5 Pa. 共a兲 9.6 ms. 共b兲 17.3 ms. 共c兲 28.3 ms. 共d兲 32.4 ms. 共e兲 36.1 ms.

J. Acoust. Soc. Am., Vol. 114, No. 6, Pt. 1, Dec. 2003 Derveaux et al.: Time-domain simulation of a guitar 3381
V. CONCLUSION coupled. This feature has been highlighted by Jansson.25
The novel aspects of the present work arise from the The present paper was mainly focused on the guitar
demanding specificities of the model: complex geometry for model and on the numerical method to solve it, with only a
the guitar, complex material for the top plate, coupling of the few selected examples showing its capabilities in terms of
top plate with the partially opened cavity and the external guitar acoustics. At present, since the air equations are ex-
free space. This relatively exhaustive model requires the as- pressed in terms of pressure and acoustic velocity, it becomes
possible to investigate other quantities of interest such as
sociation of highly elaborated numerical techniques in order
to solve it with enough accuracy. The resolution has been sound intensity and sound power. As an example, recent ex-
made possible here through the use of a spectral method for perimental work by Wang and Burroughs showed the interest
the top plate, a fictitious domain method for the fluid– of representing the complete acoustic intensity field around a
structure interaction problem and a conservative scheme for stringed instrument in order to characterize its radiating
the time discretization. This method is applicable to other properties.26 With regard to such problems, the model devel-
radiating structures composed of plates, holes, and cavities, oped here can be viewed as an interesting tool for validating
and, in particular, to other stringed musical instruments. the experiments theoretically.
Solving the vibroacoustical system in the time domain
yields, among other things, the 3D temporal evolution of the
acoustic field. From these results, important new features of APPENDIX A: THE KIRCHHOFF-LOVE’S PLATE
EQUATION
the sound radiated by the guitar can be derived. The com-
plexity and the variation in time of the directivity of the 1. Orthotropic plate equation
instrument, in particular, has been clearly shown.
We present here the Kirchhoff–Love’s plate equation in
The present method allows the separate computation of
some particular Cartesian cases. M = ⫽a 3 C␧= (ⵜ
គ u p ) denotes
the acoustic modes, assuming a completely rigid body, of the
the bending moment and C is the rigidity tensor for plates.
structural modes, assuming that the instrument vibrates in
One can write Eq. 共5兲 in the following form 共without damp-
vacuo, and of the coupled structural–acoustic modes for the
ing terms兲:7–10

冉 冊 冉 冊冉 冊
complete air–structure model. This tool is thus applicable to,
among other things, the evaluation of the relative contribu- Mxx D 1 D 2 /2 0 ⳵ xx u p
tions of structural losses and radiation losses in the sounds = ⫽ My y ⫽a 3 D 2 /2 D 3
M 0 ⳵ yyu p ,
generated by the guitar. It is well-known that such a separa- Mxy , ⳵ xy u p
0 0 D 4 /2
tion is hard to conduct experimentally.
In its present state, the guitar model is such that the ⳵ 2 u p ⳵ 2 Mxx ⳵ 2 My y ⳵ 2 Mxy
effects of structural modifications on the produced sounds, a␳p ⫹ ⫹ ⫹2
⳵t2 ⳵x2 ⳵y2 ⳵x⳵y
such as plate thickness and cavity height, can be tested in a
straightforward manner. Such simulations were systemati- ⫽F⫺ 关 p 兴 ␻ in ␻ ,
cally validated in this paper through analysis of admittance at where D 1 , D 2 , D 3 , and D 4 , are the four rigidity constants
the bridge, energy histories, and pressure signals. A good of the plate.
coherence was found between the results of this analysis, in
terms of modal frequencies and damping factors. These re-
sults are compatible with experimental results published in
2. Homogeneous isotropic plate equation
the literature devoted to the acoustics of guitars. Because we
have control here on the parameters fed into the model, it has In the homogeneous isotropic case, one has: D 1 ⫽D 3
been made possible to evaluate in detail the effects of the ⫽(D 2 /2⫹D 4 /2)⫽E/ 关 12(1⫺ ␯ 2 ) 兴 where E and ␯ are the
various sources of coupling 共string–plate, plate–air兲 in terms Young’s modulus and Poisson’s ratio, respectively. The
of frequencies and decay rates. Kirchhoff–Love’s plate operator then becomes a usual bi-
Despite these interesting features, the present version of Laplacian. Equation 共5兲 reduces to
the model suffers from some numerical and physical limita-
tions. As seen in Fig. 4, the main limitation due to numerical
reasons is the bracing, which is rather crude compared to real
a␳p
⳵ 2u p
⳵t2 冉 ⳵
冊
⫹D 1 a 3 1⫹ ␩ p ⌬ 2 u p ⫹a ␳ p R p
⳵t
⳵up
⳵t
guitars. This is due to the necessity of dealing with a more ⫽F⫺ 关 p 兴 ␻ in ␻ .
refined mesh if the purpose is to reproduce a real guitar ad-
equately. From a physical point of view, the stronger limita- 3. Free-boundary condition
tion is probably due to the underestimation of the damping in
the fluid, which might be mainly located at the rose and in We give now the expression of the free-boundary con-
the boundary layer in the box. More work is needed here for ditions in the case of a plate with a straight edge. For ex-
modeling and quantifying these effects convincingly. Other ample, if the y axis is a free boundary of the plate, the two
improvements should be investigated, such as the double po- conditions given by Eq. 共7兲 can be written
larization of the string and the vibrations of the neck. Al- ⳵ Mxx ⳵ Mxy
though this part of the instrument does not participate di- Mxx ⫽0 and ⫹2 ⫽0,
⳵x ⳵y
rectly in the radiation, its vibrations have effects on the
motion of the other parts 共strings, soundboard兲 to which it is along the edge x⫽0.

3382 J. Acoust. Soc. Am., Vol. 114, No. 6, Pt. 1, Dec. 2003 Derveaux et al.: Time-domain simulation of a guitar
 5
A. Chaigne and C. Lambourg, ‘‘Time-domain simulation of damped im-
pacted plates. I. Theory and experiments,’’ J. Acoust. Soc. Am. 109共4兲,
1422–1432 共2001兲.
6
A. Chaigne, ‘‘On the use of finite differences for musical synthesis. Ap-
plication to plucked string instruments,’’ J. Acoust. 5, 181–211 共1992兲.
7
A. W. Leissa, Vibrations of Plates 共NASA SP, Washington, D.C., 1969兲,
Vol. 160.
8
R. F. S. Hearmon, An Introduction to Applied Anisotropic Elasticity 共Ox-
ford University Press, London, 1961兲.
9
S. Ambartsumyan, Theory of Anisotropic Plates 共Technomic, Stanford,
1970兲.
10
G. Caldersmith, ‘‘Vibration of orthotropic rectangular plates,’’ Acustica
56, 144 –151 共1984兲.
11
F. Millot, F. Collino, and P. Joly, ‘‘Fictitious domain method for unsteady
problems: Application to electromagnetic scattering,’’ Mathematical and
Numerical Aspects of Wave Propagation 共Mandelieu-La Napoule, 1995兲
共SIAM, Philadelphia, 1995兲, pp. 260–269.
FIG. 17. Comparison between reference and computed eigenfrequencies of 12
R. Glowinski, T. Pan, and J. Periaux, ‘‘A fictitious domain method for
the soundboard. 共⫻兲 relative frequency f com1
/ f ref computed using P1 finite
Dirichlet problem and applications,’’ Comput. Methods Appl. Mech. Eng.
elements. 共䊐兲 relative frequency f com
1
/ f ref computed using P2-alike finite 111共3– 4兲, 283–304 共1994兲.
elements. 13
G. Cohen, P. Joly, J. E. Roberts, and N. Tordjman, ‘‘Higher order triangu-
lar finite elements with mass lumping for the wave equation,’’ SIAM 共Soc.
APPENDIX B: DISPERSION ANALYSIS OF FINITE Ind. Appl. Math.兲 J. Numer. Anal. 38共6兲, 2047–2078 共2001兲.
14
ELEMENT OF ORDER 1 AND 2 FOR THE PLATE F. Collino, ‘‘Conditions absorbantes d’ordre élevé pour des modèles de
propagation d’onde dans des domaines rectangulaires,’’ Technical Report
In order to show the interest of using of second-order No. 1790, INRIA, Rocquencourt.
finite element for the approximation of the plate equation, we 15
J. Nédélec, ‘‘Mixed finite elements in R3 ,’’ Numer. Math. 35, 315–341
compare here the values of the first 50 eigenfrequencies of 共1980兲.
16
E. Bécache, A. Chaigne, G. Derveaux, and P. Joly, ‘‘An explicit finite
the soundboard obtained using P1 or P2-alike continuous
element scheme for time-dependent Kirchhoff–Love equations,’’ in Fifth
finite elements for the discretization of the variables v p and International Conference on Mathematical and Numerical Aspects of
M = , with the same computational effort. Wave Propagation 共SIAM, Santiago, Spain, 2000兲, pp. 735–740.
17
Since the actual eigenvalues are not known analytically, P. Joly and L. Rhaouti, ‘‘Domaines fictifs, éléments finis H共div兲 et condi-
references values denoted f ref are first computed using a re- tion de Neumann: Le problème de la condition inf-sup,’’ C. R. Acad. Sci.,
Ser. I: Math. 328共12兲, 1225–1230 共1999兲.
fined mesh of the soundboard with space step h p /2 and using 18
O. Christensen and R. Vistisen, ‘‘A simple model for low-frequency guitar
P2-alike continuous finite elements. The results are repre- function,’’ J. Acoust. Soc. Am. 68, 758 –766 共1980兲.
sented in Fig. 17. It appears immediately that the error made 19
B. Richardson, T. J. W. Hill, and S. Richardson, ‘‘Input admittance and
on the estimation of the eigenfrequencies is larger with the sound field measurements of ten classical guitars,’’ in Proceedings of the
use of P 1 finite elements, since the error is greater than 5% Institute of Acoustics 共IOA, London, 2002兲, Vol. 24共2兲, pp. 1–10.
20
J. Laroche, ‘‘The use of the matrix pencil method for the spectrum analy-
for all frequencies. It is thus natural to choose second-order sis of musical signals,’’ J. Acoust. Soc. Am. 94共4兲, 1958 –1965 共1993兲.
finite elements, which leads to a better approximation of the 21
B. David and X. Boutillon, ‘‘Using vacuum to measure the acoustical
eigenfrequencies of the soundboard. efficiency,’’ in ISMA 95 共Société Franca̧ise d’acoustique, Dourdan, France,
1995兲, pp. 380–385.
22
1 B. E. Richardson, ‘‘Simple models as a basis for guitar design,’’ Catgut
L. Rhaouti, A. Chaigne, and P. Joly, ‘‘Time-domain simulation and nu-
Acoust. Soc. J. 4共5兲, 30–36 共2002兲.
merical modeling of the kettledrum,’’ J. Acoust. Soc. Am. 105共6兲, 3545– 23
3562 共1999兲. E. Bécache, A. Chaigne, G. Derveaux, and P. Joly, ‘‘Numerical simulation
2
B. E. Richardson, ‘‘Numerical modelling of stringed musical instru- of the acoustic guitar,’’ DVD, VHS, and RealPlayer document, English,
ments,’’ in SMAC 93 共Publication of the Royal Swedish Academy of Mu- 2003, http://www.inria.fr/multimedia/Videotheque-fra.html
24
sic, Stockholm, 1993兲, Vol. 79, pp. 457– 462. A. Runnemalm and N.-E. Molin, ‘‘Operating deflexion shapes of the
3
A. Le Pichon, S. Berge, and A. Chaigne, ‘‘Comparison between experi- plates and standing aerial waves in a violin and a guitar model,’’ Acustica
mental and predicted radiation of a guitar,’’ Acustica 84共1兲, 136 –145 86共5兲, 883– 890 共2000兲.
共1998兲.
25
E. Jansson, Acoustics for Violin and Guitar Makers 共Royal Institute of
4 Technology, Stockholm, Stockholm, 2002兲.
M. J. Elejabarrieta, A. Ezcurra, and C. Santamaria, ‘‘Coupled modes of the
26
resonance box of the guitar,’’ J. Acoust. Soc. Am. 111共5兲, 2283–2292 L. Wang and C. Burroughs, ‘‘Acoustic radiation from bowed violins,’’ J.
共2002兲. Acoust. Soc. Am. 110共1兲, 543–555 共2001兲.

J. Acoust. Soc. Am., Vol. 114, No. 6, Pt. 1, Dec. 2003 Derveaux et al.: Time-domain simulation of a guitar 3383