Physical synthesis of six-string guitar plucks using the ...
-
Upload
khangminh22 -
Category
Documents
-
view
0 -
download
0
Transcript of Physical synthesis of six-string guitar plucks using the ...
HAL Id hal-03604430httpshalarchives-ouvertesfrhal-03604430
Submitted on 10 Mar 2022
HAL is a multi-disciplinary open accessarchive for the deposit and dissemination of sci-entific research documents whether they are pub-lished or not The documents may come fromteaching and research institutions in France orabroad or from public or private research centers
Lrsquoarchive ouverte pluridisciplinaire HAL estdestineacutee au deacutepocirct et agrave la diffusion de documentsscientifiques de niveau recherche publieacutes ou noneacutemanant des eacutetablissements drsquoenseignement et derecherche franccedilais ou eacutetrangers des laboratoirespublics ou priveacutes
Physical synthesis of six-string guitar plucks using theUdwadia-Kalaba modal formulation
V Debut J Antunes
To cite this versionV Debut J Antunes Physical synthesis of six-string guitar plucks using the Udwadia-Kalaba modalformulation Journal of the Acoustical Society of America Acoustical Society of America 2020 148(2) pp575-587 101121100001635 hal-03604430
Physical synthesis of six-string guitar plucks usingthe Udwadia-Kalaba modal formulation
V Debut1 2 a) and J Antunes1 2 b1INET-md Departamento de Ciencias Musicais Faculdade de Ciencias Sociais e Humanas FCSH
Universidade Nova de Lisboa 1069-061 Lisbon Portugal2Centro de Ciencias e Tecnologias Nucleares Instituto Superior Tecnico Universidade de Lisboa
Estrada Nacional 10 Km 1397 2695-066 Bobadela LRS Portugal
(Dated May 14 2020)
Extending previous work by the authors this paper develops a time domain synthesismethod for classical guitar based on substructuring concepts and using the Udwadia-Kalaba(U-K) modeling strategy Adopting a modal description of the dynamics of the separateflexible subsystems in terms of their unconstrained modes and enforcing coupling constraintconditions for the assembly the result is an explicit dynamical modal formulation for thecoupled system that directly lends itself to time-stepping methods for simulation Theguitar model couples six strings through a body model of an actual instrument includestwo string polarizations and the string geometrical nonlinear effects as well as for thestringfret interaction as the instrument is played Details are given for putting all thevibrating components together in a satisfying manner and a specific strategy is exploredto allow for a non-rigid fret using flexible-dissipative-inertial constraints The reliability ofthe approach is demonstrated with simulation examples that confirm the features one wouldexpect regarding the dynamical behaviour of classical guitars Finally a pragmatic approachis made to calculate the radiated sound by convolution combining the computed bridgeforce with a measured vibro-acoustic impulse response of the instrument which proved togive satisfactory sounding results
ccopy2020 Acoustical Society of America [httpdxdoiorg(DOI number)]
[XYZ] Pages 1ndash13
I INTRODUCTION1
The sound produced by musical instruments is the2
result of complex vibratory events usually involving3
multiple interaction between structural components and4
the player Beyond describing the excitation process the5
geometry and the dynamical behavior of the separate6
components any physical model of musical instrument7
also requires to develop dedicated methods that enforce8
contact conditions between the various subsystems to9
make them vibrate consistently In practice this is a10
rather complex issue that involves considerable modeling11
and computational difficulties and which can be critical12
in view of computational efficiency and accuracy of13
simulation Among the methods developed in vibration14
analysis to assemble structural components we can15
distinguish predominantly two standard techniques each16
one with specific merits and drawbacks Penalty17
methods which approximate the contact condition in18
terms of an inter-penetration force are often easy to19
implement but may lead to large computational expense20
when impacting near-rigid obstacles since the penalty21
parameters then impose small computational time scales22
a)vincentdebutfcshunlptb)jantunesctntecnicoulisboapt
andor the need for iterative numerical schemes On the23
other hand Lagrange multipliers allow constraints to be24
applied in a mathematically exact manner but demand25
the introduction of extra variables thus increasing the26
order of the system of equations127
Recently the present authors23 demonstrated that28
reliable simulations of flexible systems can be carried29
out by formulating their dynamics using the Udwadia30
and Kalaba4 equations which were originally proposed31
for modeling constrained discrete dynamical systems32
The extension was achieved using a modal approach33
for the continuum in terms of the unconstrained34
modes of the components and the resulting modal U-K35
formulation was found successful for systems involving36
point-constraints either linear or intermittent thus37
enabling the dynamical computations of many systems38
of practical interest Regarding the very modeling39
challenging systems that are musical instruments it40
seems that the modal U-K approach could be used41
profitably for physics-based sound synthesis purposes42
The physics of guitar plucks has been covered by43
many authors notably Valette5 and Woodhouse6 and44
models of musical strings have now reached a fine45
level of detail including frequency-dependent losses46
flexural dispersion and nonlinear features For obtaining47
the actual solutions different formulations have been48
proposed for solving the partial differential equations49
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 1
numerically using wave propagation and reflection50
functions7 finite difference8 or the modal approach9 and51
also deriving energy preserving schemes10 to guarantee52
stability of the numerical solutions Modal methods53
- which will be employed here - have formed the54
basis of successful modeling strategies both for linear55
and nonlinear string vibration and with or without56
constraints11ndash16 with the advantages of being simple to57
implement and of dealing in an exact manner with some58
complex physical phenomena ie string damping and59
dispersion Despite the good deal of work on string60
modeling most of the models of pluck instruments in the61
literature remain highly simplified cases of the underlying62
physics and rarely address the entire problem taking63
account the full set of strings that drives the instrument64
body and that in turn produces propagating pressure65
waves In this view the modeling work by Bader1766
that employs finite difference by Mansour et al18 that67
uses a wave-based approach and by Debut et al14 that68
makes use of a modal representation certainly pertain to69
the small group of studies that deal with most complete70
models of actual multi-stringed instruments71
Following the work by Antunes and Debut2 the task72
of this paper is to show how flexible and versatile the73
modal U-K approach can be for the physically based74
sound synthesis of stringed instruments by exploring75
the issues of assembling the structural components in76
a realistic manner and illustrating the ability of the77
technique to achieve reliable simulations While the78
technique was presented for a crude model of guitar this79
paper refines and extends our previous work in several80
respects First the model to be built here involves all the81
major components of a classical guitar that technically82
raises the issue of multiple-point coupling and makes the83
implementation more complex It includes six strings84
vibrating in both polarizations and coupled together at85
the bridge by the body motions one or more pressing86
fingers on the fingerboard to control the frequency of the87
played notes as well as other relevant features of vibrating88
string namely wave dispersion frequency-dependent89
damping and nonlinear geometrical effects The issue90
of coupling the subsystems is dealt in regard to the U-91
K formalism by means of additional constraining terms92
in the dynamical equation and hence contrasts with the93
more traditional approach that uses penalty formulation94
to model contact in musical instruments111219ndash21 In95
order to model realistic finger interaction a specific96
strategy is also developed and explored to account for97
the possible stiffness damping and inertial properties98
of the constraints thus bypassing the non-deformable99
nature of common kinematic constraints Finally to100
achieve the end goal of synthesis the resulting sound101
is computed using a simple and pragmatic approach102
Based on convolution it uses the computed bridge force103
and a measured impulse response that characterizes the104
vibro-acoustic behavior of the guitar body and hence105
gets around the problem of producing satisfactory sounds106
without performing complicated numerical calculation of107
sound radiation108
The outline of this paper is as follows The U-K109
theory and its extension to deal with flexible systems110
is briefly described Then we give details of the guitar111
model and develop the dynamical equations for the112
separate subsystems and their coupling The final section113
is devoted to present numerical results obtained by time-114
domain simulations with intention of illustrating the115
dynamical behavior of the fully coupled model in view116
of the physics of classical guitars117
II THEORETICAL FORMULATION118
A Basic U-K formulation119
In general mathematical terms Udwadia and Kalaba120
proposed the following standard form for the study of the121
dynamics of constrained discrete mechanical systems4122
Mx = F(x x t) + Fc(x x t) (1)
which expresses the system response x as the result123
of the application of the constraint-independent forces124
F (external and internal including the stiffness and125
damping force terms) acting on it and some additional126
forces Fc stemming from a set of constraints which127
both are considered as known functions of x x and t128
The U-K formulation then benefits from an alternative129
expression of the usual constraint equations ψi(x x t) =130
0 (i = 1 m) obtained by differentiation with respect131
to time written as a matrix-vector constraint system in132
terms of accelerations as133
A(x x t) x(t) = b(x x t) (2)
where A is the constraint matrix and b is a known vectorTheir main result is then to provide an explicit expressionfor the constrained dynamics x(t) and the constraintforce vector Fc(t) at each instant given by
x = xu + Mminus12B+(bminusAxu) (3a)
Fc(t) = M12B+(bminusAxu) (3b)
denoting B+ the Moore-Penrose pseudo-inverse of B =134
AMminus12 The vector xu represents the dynamical135
response of the unconstrainted system solution of136
xu = Mminus1F(x x t) (4)
while the second term in the right-side hand of Eq137
(3a) accounts for the influence of the constraints on138
the system since the unconstrained system must further139
comply with the physical constraints The superlative140
elegance of the U-K formulation lies in the fact that141
it expresses the dynamics of the constrained system142
through a single dynamical equation (3a) that takes143
constraints into account and allows if needed the144
knowledge of the constraining force through (3b) No145
additional variables such as Lagrange multipliers are146
needed thus avoiding difficulties in the computation147
of the solutions as for Differential-Algebraic system of148
equations Notably Eqs (3) may be applied to linear149
2 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
or nonlinear conservative or dissipative systems and150
may also be efficiently solved using a suitable time-step151
integration scheme for a given excitation152
B Modal U-K formulation153
Extension to constrained continuous multibody154
systems defined through their unconstrained modes Φ155
was formulated in Ref 2 by transforming (3) to the156
modal space through x = Φqqq leading to the formulation157
158
qqq = qqqu +MMMminus12BBB+(bminusAAAqqqu) (5)
where qqqu are the modal accelerations of the159
unconstrained configuration MMM = ΦTMΦ is the160
modal mass matrix AAA = AΦc is the modal constraint161
matrix with Φc the matrix of all modeshapes where162
constraints are defined and BBB = AAAMMMminus12 Finally the163
modal constraining force FFF c can also be computed as164
FFF c = MMM12BBB+(bminusAAAqqqu) (6)
C Dynamically coupled subsystems165
The application of the U-K modal formulation to a166
set of subsystems coupled through a number of kinematic167
constraints is now considered In a modal description the168
dynamics of the subsystem ` (` = 1 2 L) subjected169
to an external force field can be classically written as a170
set of modal equations in terms of the vector of modal171
amplitudes qqq` and its derivatives172
MMM `qqq` +CCC`qqq` +KKK`qqq` +FFF `nl(qqq` qqq`) = FFF `ext +FFF `nl (7)
where MMM ` CCC` and KKK` are diagonal matrices of the173
modal parameters m`n c`n and k`n (n = 1 N `) defined174
according to the modeshapes φ`n and FFF `ext and FFF `nl are175
the modal forces obtained by projection of the external176
and nonlinear forces on the modal basis The modal177
accelerations of the unconstrained system to be used in178
Eq (5) are given by179
qqq`u =(MMM `)minus1
FFF ` (8)
denoting FFF ` the vector of all the constraint-independent180
modal forces181
FFF `= FFF `extminusCCC`qqq`minusKKK`qqq`minusFFF `nl(qqq` qqq`) (9)
for which it is assumed that the vectors of modal182
constrained displacements and velocities are known at183
each time-step Stacking the modal quantities of the184
unconstrained subsystems in compact vectors as185
qqqu =[qqq1uT qqqLuT
]T(10)
the unconstrained modal accelerations of the coupled186
system read finally as187
qqqu = MMMminus1
[FFFext minusCCC qqq minusKKKqqq minus FFFnl(qqq qqq)
](11)
denoting188
qqq =[qqq1T qqqLT
]T(12)
In view of Eq (10) the matrices MMM CCC and KKK areblock diagonal set up by the submatrices of the modalparameters of the subsystems assembled as
MMM = diag(MMM1 MMML) (13a)
CCC = diag(CCC1 CCCL) (13b)
KKK = diag(KKK1 KKKL) (13c)
and FFFext and FFFnl are modal vectors associated to theexternal and nonlinear interaction forces acting on thevarious subsystems written respectively as
FFFext =FFF 1
extT FFFLextTT
(14a)
FFFnl =FFF 1
nlT FFFLnlTT
(14b)
The second set of equations to be considered concerns the189
P coupling constraints at locations r`c In most practical190
situations these are amenable to linear relationships by191
appropriate differentiation with respect to time leading192
to the standard form193
AAA(qqq qqq t) qqq = b(qqq qqq t) (15)
where AAA = AΦc with Φc = diag(Φ1c Φ
Lc ) and where194
Φ`c contains the modeshape vectors of each subsystem195
at the constraint location r`c Notice that because the196
constraints are implemented at the acceleration level197
Eq (15) generally leads to error accumulation in the198
computed motion during numerical integration due to199
numerical approximations and round-off errors resulting200
in constraint drift phenomenon To compensate for these201
errors a number of constraint stabilization techniques202
have been proposed22 and can be implemented if203
necessary as recently done by the authors for systems204
with intermittent contacts3 However for the problem205
at hand there was no necessity for implementing such206
correction terms Indeed by monitoring the constraints207
as the numerical simulation proceeds we observed a208
constraint drift under 20times10minus13 m which is negligible209
III GUITAR MODEL210
The guitar model involves six strings interacting with211
a guitar body and includes further models for the pressing212
finger on the fingerboard The strings are modeled213
using a simplified nonlinear formulation considering both214
polarizations and the body behavior is expressed in terms215
of modes identified from bridge input measurements on216
a real-life guitar217
A Nonlinear string model218
The vibratory motion of the strings is modeled219
using the KirchhoffndashCarrier approach2324 which is a220
simplified manner to describe large amplitude string221
motions Essentially the model discards the dynamics222
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 3
of the longitudinal modes but retains the effect of the223
geometrical nonlinearity by means of a pseudo-external224
forcing term added to the string transverse equations225
In our context it has two advantages First the model226
is known to capture relevant nonlinear phenomena for227
musical strings1425 thus enabling the reproduction of228
typical audible nonlinear effects Second it serves as a229
convenient means of illustrating that the U-K framework230
can deal effectively with nonlinear terms231
1 Kirchhoff-Carrier simplified nonlinear string model232
We consider a stiff string of length L cross-sectionalarea S mass density ρ Youngrsquos modulus E momentof inertia I and coefficient of dissipation η subjectto an axial tension T0 vibrating in two perpendiculartransverse motions Y (x t) and Z(x t) According to theKirchhoff-Carrier assumptions a simplified model of thefree vibration of the string is given by
ρSpart2Y
partt2+ η
partY
partt+ EI
part4Y
partx4=[T0 + Tdyn(t)
]part2Ypartx2
(16a)
ρSpart2Z
partt2+ η
partZ
partt+ EI
part4Z
partx4=[T0 + Tdyn(t)
]part2Zpartx2
(16b)
which is the standard wave equations for vibrating string233
including both polarizations with an additional forcing234
term proportional to the net dynamical increase in string235
length expressed as236
Tdyn(t) =ES
2L
int L
0
[(partY (x t)
partx
)2
+
(partZ(x t)
partx
)2]dx (17)
Eq (17) shows that the Kirchhoff-Carrier model only237
requires the knowledge of the transverse motions and238
that is independent of the x coordinate As far as the239
numerical technique is concerned the form of Eq (17)240
is very convenient particularly for modal synthesis and241
contrasts with the delicate numerical resolution of the242
ldquogeometrically exactrdquo model26 by modal discretization243
due the spatio-temporal nature of the nonlinear forcing244
2 Kirchhoff-Carrier nonlinear modal terms245
Denoting X a generic direction of string motion (X246
stands for Y or Z) the modal forces associated to the247
string nonlinear behavior are obtained by integration of248
the additional force terms of Eq (16) as249
fXn (t) =
int L
0
[Tdyn(t)
part2X
partx2
]φXn (x)dx (18)
Considering a string pinned at the nut and free at the250
bridge the modeshapes φXn (x) are given by251
φXn (x) = sin
[(2nminus 1)πx
2L
](19)
Substituting Eq (19) in Eq (18) and using a modalexpansion for the string motion the nonlinear modal
terms are finally given by
fXn (t)=minusESπ4
128L3(2nminus 1)2qXn (t)(
NYsumm=1
(2mminus1)2[qYm(t)
]2+
NZsumm=1
(2mminus1)2[qZm(t)
]2)(20)
with NY and NZ the sizes of the two string modal basis252
B The instrument body253
In view of Eq (7) the dynamic behaviour of the254
body is described by its modal properties extracted255
from input admittance measurements In contrast256
to modal computations based on a full model of the257
instrument body this is a direct approach that avoids258
complications of modeling in particular the frequency-259
dependent dissipative effects which significantly affect260
the computed sounds but are very difficult to model261
properly and that readily lend the synthesis method262
to be applied to any stringed instruments Since both263
polarizations of string motions are accounted a correct264
modeling of the stringbody coupling would demand265
the knowledge of both the in-plane and out-of-plane266
body modeshapes at the bridge However a simpler267
model is taken here by considering that only string268
motions normal to the soundboard couple to the body269
motions and also by neglecting the in-plane motions of270
the soundboard This means that the string motions in271
the plane parallel to the soundboard are unable to radiate272
sound and that the sound radiation is produced only273
by the normal motions of the soundboard Of course274
none of these choices is entirely satisfactory for accurate275
synthesis because in-plane and out-of-plane motions are276
coupled and also because in-plane motions couple the277
top and back plates via the ribs However since278
radiation is dominated by the out-of-plane vibration279
there is no doubt that in-plane motions are of second-280
order significance In practice the main advantage of281
our assumptions is to make the body characterization282
considerably simpler involving only transfer function283
measurements in the direction normal to the soundboard284
and avoiding the delicate measurements of the cross285
terms of the transfer function matrix286
The test guitar was a high-quality concert guitar287
built by Friederich number 694 During the tests it288
was positioned in the vertical position clamped to a289
rigid support by the neck and softly restrained in the290
lower bout with all the tuned strings damped (see291
Fig 1) The excitation and vibratory response were292
measured using a miniature force sensor (Kistler type293
9211) and a light-weight accelerometer (BampK 4375)294
respectively attached to the tie block using a thin layer295
of bee-wax Modal data were collected at two bridge296
locations close to the attachment points of the lowest297
and highest strings and a simple linear interpolation298
between the two transfer functions was performed to299
provide the unmeasured transfer functions at the other300
points where the strings make contact with the bridge301
4 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
Figure 1 Set-up for transfer function measurements at the
guitar body bridge
Modal identification was then achieved using the set of302
impulse responses expressed in terms of velocity and303
implementing the Eigensystem Realization Algorithm27304
The modal parameters for the first-ten identified modes305
are listed in Table I Fig 2 shows an example of306
an impulse response and its transfer function measured307
at the bridge end of the lowest string together with308
the corresponding reconstructed functions showing that309
the estimation was generally reliable up to 700 Hz310
Fig 3 also shows the identified mode shapes of the311
body at the bridge which exhibit typical motions of312
guitar modes28 The modes show a small degree of313
complexity but the effect of the imaginary components314
on the transfer function is small and can be neglected315
for our purpose Note that modes have been normalized316
according to max(|real(φBn (rs))|) = 1 (s = 1 6)317
Table I Identified modal parameters of the guitar body
Mode n fn (Hz) ζn () mn (Kg)
1 894 15 0402 1826 182 0083 2299 064 0204 3102 089 0475 4035 055 1066 5114 075 0587 5442 053 1828 6139 061 0699 6698 069 03810 8484 059 081
C Constraining forces and modal constraints318
1 Stringbody constraints319
Assuming a rigid transmission at the points wherethe strings make contact with the bridge the modalequations of the constrained system require to satisfya set of constrained modal equations stemming fromthe condition that at the bridge the strings and thebridge undergo the same motion Because the bodyresponse is here described only in the direction normal
0 005 01 015
Time [s]
-20
0
20
h (
t )
[ (
ms
)N
]
100 200 300 400 500 600 700 800 900
Frequency [ Hz ]
10-2
100
| H
( f
) |
[ (
ms
)N
]
Figure 2 Measured (green) and reconstructed (red) impulse
responses and transfer functions close to the lowest string
Gray dotted lines stand for the identified modal frequencies
E A D G B E
-1
0
1
1(
xB
)
f1 = 8939 Hz
E A D G B E
-1
0
1
2(
xB
)
f2 = 18262 Hz
E A D G B E
-1
0
1
3(
xB
)
f3 = 22986 Hz
E A D G B E
-1
0
1
4(
xB
)
f4 = 31021 Hz
E A D G B E
-1
0
1
5(
xB
)
f5 = 40352 Hz
E A D G B E
-1
0
1
6(
xB
)
f6 = 51143 Hz
E A D G B E
-1
0
1
7(
xB
)
f7 = 54417 Hz
E A D G B E
-1
0
1
8(
xB
)
f8 = 61386 Hz
E A D G B E
-1
0
1
9(
xB
)
f9 = 66978 Hz
Figure 3 Identified modeshapes of the guitar body at the
stringbridge contact points Real (blue) and imaginary
(magenta) components
to the soundboard the string motion Ys(xb t) in thisplane must be the same as the body motion at thestring contact point Yb(rs t) while its motion Zs(xb t)in the direction parallel to the soundboard must be nilFormally for each string s (s = 1 S) this results inthe conditions
Ys(xb t)minus Yb(rs t) = 0 (21a)
Zs(xb t) = 0 (21b)
or in terms of modal amplitudesΦYs
b
TqqqYs(t)minus
ΦBrsTqqqB(t) = 0 (22a)
ΦZs
b
TqqqZs(t) = 0 (22b)
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 5
where the modeshape vectors are written as
ΦYs
b =φYs1 (xb) φ
Ys
NYs(xb)
T(23a)
ΦZs
b =φZs1 (xb) φ
Zs
NZs(xb)
T(23b)
ΦBrs =φB1 (rs) φ
BNB
(rs)T
(23c)
Defining a vector of modal coordinates qqq built by stacking320
the individual modal displacements of the subsystems ie321
the strings including both polarizations and the body322
qqq = qqqY qqqZ qqqBT (24)
where qqqY and qqqZ are partitioned into terms associatedwith the different strings as
qqqY =qqqY1T qqqY6T
T(25a)
qqqZ =qqqZ1T qqqZ6T
T(25b)
and323
qqqB = qB1 qBNBT (26)
the constraint equations fit the matrix form324
AAAb qqq = 000 (27)
with the stringbody constraint matrix AAAb written as325
AAAb =
[ΦΦΦYb 000 minusΦΦΦB
000 ΦΦΦZb 000
](28)
where ΦΦΦYb and ΦΦΦZb are matrices of the modeshape vectors326
of the strings taken at the bridge location xb built as327
ΦΦΦYb =
ΦY1
b
T000 000
000
ΦY2
b
T 000
000 000
ΦY6
b
T
(29)
and similarly for the perpendicular direction and ΦΦΦB328
stands for the modeshapes of the body at the stringbody329
contact points rs330
ΦΦΦB =
ΦBr1T
ΦBr6T (30)
2 Stringfinger coupling at the fingerboard331
Modeling a stopping string for playing different332
notes can be thought in several ways according to the333
type of sound to simulate as well as to the degree of334
simplicitycomplexity aimed for the model The simplest335
model was taken in Ref 2 imposing a rigid kinematical336
constraint at a single location If such model could affect337
the tuning of the string effectively it is nonphysical338
in that part of the vibrational energy flows beyond the339
contact point (see Ref 2 for discussion) This simple340
approach is here extended to allow for multiple-point341
rigid coupling or for a non-rigid interaction as shown in342
Fig 4343
a Multiple-point rigid coupling This first optiondirectly enters into the realm of the standard kinematicalconstraints common in multibody theory The stringmotions Ys and Zs at the multiple finger locations xFf
(f = 1 F ) should be nil at all times so that
Ys(xFf t)=0 rArr
ΦYs
Ff
TqqqYs(t) = 0 (31a)
Zs(xFf t)=0 rArr
ΦZs
Ff
TqqqZs(t) = 0 (31b)
where the modeshape vectors at the finger locations are
ΦYs
Ff=φYs1 (xFf
) φYs
NYs(xFf
)T
(32a)
ΦZs
Ff=φZs1 (xFf
) φZs
NYs(xFf
)T
(32b)
Eq (31) can be rearranged in a matrix form as344
AAAf qqq = 000 (33)
where the modal constrained matrix is345
AAAf =
[ΦΦΦYs
F 000
000 ΦΦΦZs
F
](34)
with346
ΦΦΦYs
F =
ΦYs
F1
T
ΦYs
FF
T ΦΦΦZs
F =
ΦZs
F1
T
ΦZs
FF
T (35)
b Non-rigid coupling To extend the model to347
allow for non-rigid contact requires a little more care348
since a flexible-dissipative-inertial model for coupling349
subsystems cannot be directly formulated as kinematic350
constraints Nevertheless they can be implemented351
by assigning one or more additional flexible-dissipative-352
inertial elementary subsystems which are constrained to353
follow the string motions at the finger location(s) exactly354
as the body modes are included This modeling strategy355
somehow creates a bridge between computationally356
efficient multibody modeling providing ldquorigidrdquo kinematic357
constraints and the versatile but computationally slower358
modeling strategy through penalty-based constraints359
Non-rigid connections on the fingerboard cantherefore be obtained by extending the guitar modelusing a set of mass-spring-damper oscillators at locationsxFf
(f = 1 F ) and enforcing coupling as
Ys(xFf t)minusY Ff
f (t)= 0 (36a)
Zs(xFf t)minusZFf
f (t)= 0 (36b)
where YFf
f and ZFf
f are the responses of the fingercontact points in the two perpendicular directionsReturning to Eq (11) the vector and matrices mustnow be partitioned as
qqqu =qqqYsu T qqqZs
u T qqqFsu T
T(37a)
MMM = diag(MMMYsMMMZsMMMFs
)(37b)
CCC = diag(CCCYsCCCZsCCCFs
)(37c)
KKK = diag(KKKYsKKKZsKKKFs
)(37d)
6 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
where the vectors and matrices for the strings are as360
beforeMMMFs CCCFs andKKKFs are block diagonal matrices of361
the inertia damping and stiffness properties associated362
with the ldquofingerrdquo subsystem acting on string s including363
both directions of string motions364
qqqFsu =
Y F1
f Y FF
f ZF1
f ZFF
f
T(38)
In terms of modal coordinates Eqs (36) becomeΦYs
Ff
TqqqYs(t)minus Y Ff
f (t) = 0 (39a)ΦZs
Ff
TqqqZs(t)minus ZFf
f (t) = 0 (39b)
which can be rearranged as365
AAAf qqq = 000 (40)
with the modal constrained matrix AAAf written as366
AAAf =
[ΦΦΦYF 000 minusI 000
000 ΦΦΦZF 000 minusI
](41)
Figure 4 Rigid kinematical constraints (left) and flexible-
dissipative-inertial constraints (right) at multiple locations
D Dynamical formulation of the fully coupled model367
The formulation (11) readily applies to the full guitar368
model including the non-rigid constraint model for the369
stringfret interaction by stacking the unconstrained370
modal displacements into a vector371
qqqu = qqqYu qqqZu qqqBu qqqFu T (42)
where qqqYu qqqZu and qqqBu are as before and372
qqqFu=qqqY1
ufT qqqY6
ufT qqqZ1
ufT qqqZ6
ufTT
(43)
In view of the definition of qqq the assembled inertia MMM373
stiffnessKKK and dampingCCC matrices which are uncoupled374
are given by375
MMM=
MMMY 000 000 000
000 MMMZ 000 000
000 000 MMMB 000
000 000 000 MMMF
CCC=
CCCY 000 000 000
000 CCCZ 000 000
000 000 CCCB 000
000 000 000 CCCF
KKK=
KKKY 000 000 000
000 KKKZ 000 000
000 000 KKKB 000
000 000 000 KKKF
(44)
with MMM ` KKK` and CCC` the inertia stiffness and dampingmatrices for each individual subsystem ` and the vectorsFFFext and FFFnl write
FFFext =FFFYextT FFFZextT000000
T(45a)
FFFnl =FFFYnlT FFFZnlT000000
T(45b)
where FFFYext and FFFZext correspond to the plucking action376
of the player expressed for both planes of polarization377
and FFFYnl and FFFZnl are the modal vectors associated to the378
string nonlinear effects with entries given by Eq (20)379
The form of the modal constraint equation is now380
AAAqqq = 000 (46)
with381
AAA =
[AAAbAAAf
]=
ΦΦΦYb 000 minusΦΦΦB 000 000 000
000 ΦΦΦZb 000 000 000 000
ΦYFΦYFΦYF 000 000 000 minusFFFY 000
000 ΦZFΦZFΦZF 000 000 000 minusFFFZ
(47)
where ΦYFΦYFΦYF and ΦZFΦZFΦZF are block matrices with entries given382
by Eq (35) and FFFY and FFFZ are block diagonal matrices383
built with either the identity or null matrix depending on384
the constraints enforced on each string The form of (47)385
shows that the interface forces and constraint conditions386
couple the Y -motions of the strings and the body as well387
as coupling between the finger and the strings in each388
direction of motions Notice that coupling between both389
polarizations of string motions is not apparent in (47)390
although it is formulated in the nonlinear forces (45b)391
IV ILLUSTRATIVE COMPUTATIONS392
A System parameters393
For given initial conditions the system of equations394
comprising (5) with the unconstrained accelerations (11)395
and the constraint matrix (15) and using the specific396
matrices given by Eqs (44) and (47) can be integrated in397
time to produce the transient response The time-domain398
simulations were performed based on the identified body399
modal data and using standard string properties of400
classical guitars Assuming non-ideal strings with fixed-401
free ends the modeshapes are given by Eq (19) and402
their modal frequencies can be approximated by403
fYn = fZn =cT2πpn
(1 +
B
2T0p2n
)(48)
where cT is the velocity of transverse waves B is the404
bending stiffness of the string and T0 its tensioning405
pn = (2n minus 1)π2L and n the mode number Modal406
damping values are introduced following the pragmatic407
formulation given in Ref 6408
ζYn = ζZn =1
2
T0(ηF + ηAωn
)+ηBBp
2n
T0 +Bp2n(49)
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 7
where ηF ηA and ηB are the loss coefficients related to409
ldquointernal frictionrdquo ldquoair viscous dampingrdquo and ldquobending410
dampingrdquo The length of the strings is L=065 m411
and their density bending stiffness tensioning and loss412
coefficients are taken from Ref 29 The size of the modal413
basis of each string is defined by the maximum frequency414
of the computations (10000 Hz) which proved a sensible415
compromise for convergence to realistic simulations416
resulting in 800 modes for the set of strings To417
obtain suitable values for the elasticity parameter ES418
we perform simple vibratory tests on string samples419
Stretching a string vertically and suspending a mass to420
its free end one can easily measure the longitudinal421
vibration of the string-mass oscillator and obtain its422
natural frequency Assuming the string acting as a423
spring with no mass and reminding that the natural424
frequency of the longitudinal vibration is given by f =425
12π
radicESML then an estimate of ES is obtained The426
simulation results were obtained by time-step integration427
of the ordinary modal equations using the velocity-428
Verlet numerical scheme30 an integrator that combines429
simplicity second order accuracy with good numerical430
stability - see implementation details in Ref 2 We431
adopted a time-step of 10minus5s and assumed all initial432
modal displacements and velocities nil initially433
B Computational results434
Pluck excitations were simulated by considering a435
point-wise external force at xe=09L with a linear force436
ramp applied to one or more strings during 10 ms and437
by assuming null excitation force at subsequent instants438
1 Dynamical behaviour of the fully coupled model439
Results in Fig 5 were obtained for the open string440
E2 excited in the direction normal to the soundboard441
first considering a rigidly mounted string then coupling442
one string and finally all six strings to the instrument443
body Assuming no dynamics for the body the response444
spectra of string motion at the bridge is virtually nil445
according to the pinned boundary condition When446
coupling the instrument body to the string the response447
becomes obviously different and shows the classical series448
of near harmonic string partials together with smaller449
peaks from the body resonances While the first modal450
frequency of the unconstrained string is 41 Hz a close451
look at Fig 5 shows that the constraint at the bridge452
yields the fundamental frequency for the tuned string453
with pinned ends (f = 82 Hz) thus confirming the454
correct implementation of the bridge coupling Coupling455
the other five strings results in new peaks in the456
spectrum extending over the frequency range of the body457
modes and with an amplitude difference of about 100 dB458
compared to the most excited modes which clearly relate459
to the non-excited strings As found in some stringed460
instruments this dynamical coupling of all the strings via461
the bridge can lead to sympathetic vibrations which can462
further results in slight interference beats in the sound463
due to the inharmonicity produced by the bridge coupling464
and of the string partials465
0 200 400 600 800 1000 1200 1400 1600 180010
-40
10-20
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 180010
-20
10-10
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 1800
Frequency [ Hz ]
10-20
10-10
100
| Y
6(f
) |
Figure 5 Response spectra of string motion at the bridge
Top rigidly mounted string middle single stringbody
coupling bottom six-stringbody coupled model Force
amplitude 5 N Linear model
0 02 04 06 08 1
Time [ s ]
0
05
1
15
2
Mo
dal en
erg
y [
J ]
10-3
STRINGSFRETBODYTOTAL
Figure 6 Instantaneous modal energies Oblique excitation
with amplitudes of 2 N and 1 N in the Y and Z directions
Linear six-stringbody coupled model Conservative model
Figs 6 and 7 then serve at checking other466
implementation details Simulations pertained to the467
string E2 constrained at xF = 02509L by a single ldquostiffrdquo468
fret modeled with mF=5 10minus3 Kg cF=1 Nsm and469
kF=106 Nm plucked in the oblique direction The470
modal energies E` of the subsystems ` shown in these471
plots were computed from the results of the simulations472
as follows in terms of the modal parameters and the473
8 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
0 02 04 06 08 10
05
1
15
[ J ]
10-3 STRINGS
Y-dir
Z-dir
0 02 04 06 08 10
05
1
15
[ J ]
10-5 FRET
Y-dir
Z-dir
0 02 04 06 08 10
05
1
[ J ]
10-4 BODY
0 02 04 06 08 1
Time [ s ]
0
05
1
15
10-3 FULL SYSTEM
Figure 7 Instantaneous modal energies Same plucking
conditions as in Fig 6 Linear six-stringbody coupled model
Non-conservative model
modal displacements and velocities474
E`(t) =
Nsumn=1
[1
2m`n(ω`n)2
[q`n(t)
]2+
1
2m`n
[q`n(t)
]2](50)
One important feature seen in Fig 6 is that the total475
energy remains roughly constant over a conservative476
simulation while energies flow back and forth between477
the subsystems thus giving confidence in the numerical478
model for simulations of reasonable duration For479
the non-conservative case shown in Fig 7 the large480
difference in energy decay for the two polarizations481
confirms the different stringbody coupling with the482
directions of string motion Note that both the body483
and the fret oscillator are dominated by periodic energy484
exchanges at the frequency of the played note as the wave485
travels back and forth between bridge and fret Finally486
looking closely at the modal energy of the fret one can487
see smaller superimposed periodic disturbances related488
to the reflections between fret and nut illustrating that489
energy still flows through the modeled fret490
Finally we compare in Fig 8 results by varying491
the excitation amplitude in order to illustrate the string492
nonlinearity The simulations were obtained for the493
string E2 using the experimentally determined value494
of 32 kN for the flexibility ES which is comparable495
to the value given by Chaigne31 Fig 8 shows the496
temporal variations of string tension with an estimate497
of the playing frequency computed from time domain498
evaluations of the zero crossing frequency The general499
features of large amplitude string motions are clearly500
visible with a quasi-static increase of tension and501
superimposed oscillations at twice the string fundamental502
frequency and a small frequency variation of the played503
note which compares well with measurements32504
1 2 3 4 5 6 7
716
718
72
722
724
T (
t )
[ N
] Fexc
= 1 N
Fexc
= 5 N
1 2 3 4 5 6 7
Time [ s ]
822
824
826
828
f (
t )
[ H
z ]
Figure 8 String tension (top) and instantaneous frequency of
string motion (bottom) Nonlinear rigid body model
2 Parameter study of the stringfinger coupling model505
We now illustrate the different strategies to account506
for a stopping finger on the fingerboard and explore the507
influence on the coupled system of the finger model508
In contrast with the approach in Ref 2 the strategy509
in terms of auxiliary oscillators rigidly coupled to the510
string seems more realistic and is undoubtedly more511
versatile allowing different types of contacts to be512
modeled by varying the oscillator parameters However513
one could expect the finger dynamics to perturb the514
coupled system with changes in frequency and damping515
for its modes similar to what observed for a multi-modal516
system coupled to a system exhibiting resonance To517
give some insights Figs 9-11 show the effects of varying518
the finger mechanical parameters Simulations pertain to519
the string E2 constrained at xF = 01857L by a finger of520
mass mf = 5 10minus3 Kg and ignore the body dynamics521
The first test consists in varying the stiffness of the522
constraint assuming no damping Fig 9 shows the523
frequency spectra of the coupling force at the rigid bridge524
where the peaks of the string modes can be seen clearly as525
well as the resonance frequency of the auxiliary oscillator526
which moves in frequency with the stiffness and slightly527
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 9
modulates the other peaks A close look shows that528
the spectra are affected much more sensitively below529
the resonance of the finger oscillator with amplitude530
variations analogous to a system driven in the stiffness-531
controlled region Decreasing the stiffness causes a rise532
in amplitude of the undesirable modes while increasing533
the stiffness lower their amplitudes As observed for the534
stringbody coupling33 inharmonicity is also created by535
the stiffness termination which in turn affects the tuning536
of the note With the increase of stiffness perturbations537
of the string modes are reduced and tuning is improved538
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
4 Nm
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
5 Nm
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
kf = 10
6 Nm
Figure 9 Spectra of the coupling force at the rigid bridge
Stiffness is varied F=1 c1 = 0 Nsm
Fig 10 then shows the effect of varying the damping539
keeping the stiffness constant Not surprisingly if540
damping increases the amplitude of the coupled541
modes close to the oscillator resonance decrease and542
the frequency region affected becomes wider This543
significantly affects the time envelope of the pluck note544
that becomes smoother and has a shorter-time decay545
Finally the effect of the number of contacts is studied546
using the rigid contact model which is convenient for547
this purpose Fig 11 gives a picture of the spatial548
distribution of the string energy by plotting the standard549
deviation of string motion as a function of its length550
As seen the ldquopassiverdquo region is affected with great551
sensitivity by the number of contacts and surprisingly552
increasing the contact points does not necessarily reduce553
the undesirable transfer of energy Computations show554
that the effectiveness of the multi-point rigid constraints555
approach is strongly problem-dependent and varies in a556
non-obvious way with several factors including the finger557
width the spacing between contact points the relative558
string length from the ends and the excitation spectrum559
Taken in combination these results and noticing560
that the perturbations induced by each finger oscillator561
are additive due to the linear coupling a robust and562
physically sound strategy to simulate the pressing finger563
would be to implement a set of flexible-dissipative564
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
cf = 1 Nsm
cf = 10 Nsm
0 1 2 3 4 5
Time [ s ]
-2
0
2
4
Y (
t )
10-3
Figure 10 Spectra of the coupling force at the rigid
bridge (up) and time history of the string motion (bottom)
Damping is varied F=1 k1 = 105 Nm
constraints imposing a first stiff constraint in order565
to ensure a precise tuning followed by several soft566
constraints that would control the waveform and spectral567
content of the decaying response depending on their568
number and damping properties For illustration569
the spectrograms of two plucks for different sets of570
parameters are presented in Fig 12571
0 01 02 03 04 05 06
x [ m ]
10-5
10-2
ST
D (
Y(t
) )
F=1
F=2
F=3
F=5
016 017 018 019 02
x [ m ]
10-5
ST
D (
Y(t
) )
Figure 11 Standard deviation of the string motion Number
of contacts is varied Vertical dotted lines represent the
constraints locations Finger width (0015m) is kept constant
C Simulated acoustic radiation572
As an alternative to computing sounds from full573
vibro-acoustic models that usually require the use of574
massive computational resources it is tempting to apply575
10 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
Figure 12 Spectrograms of the force at the rigid bridge Up
F = 3 bottom F = 5 Parameter values kF1=106 Nm
cF1=1 Nsm and kFj =104 Nm cFj =10 Nsm (j = 2 F )
Constraint at xF = 01857L (sound files available)
simpler techniques that could be efficient to generate576
convincing sounds from the vibratory responses The577
method used here is to treat the guitar body as a linear578
input-output system and to compute the pressure by579
convolution benefiting from the local excitation of the580
body by the strings and using a vibro-acoustic impulse581
response of the test guitar This approach does not by582
any means simulate the sound response everywhere in the583
pressure field but it has the merit of adding the influence584
of the radiating properties of the body at a very low585
computational cost and of providing sounds of greater586
perceptual quality than the commonly used vibratory587
signals at the bridge The pressure at location rp can be588
computed by the convolution of the time-history of the589
stringbridge interaction force FY (rb t) and the vibro-590
acoustic impulse response function hY (rb rp t) between591
the bridge and microphone by592
p(rp t) = hY (rb rp t) lowast FY (rb t) (51)
The extension to account for the effects of all the593
strings is readily obtained by linear superposition594
principle summing the contributions from each string595
Measurements of the vibro-acoustic impulse response596
were performed under the same experimental conditions597
as for transfer function measurements using an598
additional condenser microphone Impulsive excitation599
was applied in the perpendicular direction to the600
soundboard and the acoustic pressure was measured601
at a distance of 1 m from the central axis of the602
guitar in a quiet room with all strings damped (see603
Fig 1) The impulse response was computed by inverse604
Fourier transform of the complex transfer function built605
from the force and response signals described in the606
frequency domain Results in Fig 13 pertain to the607
simulation of a G major arpeggio obtained with the608
two lowest and the top strings constrained using five609
auxiliary flexible-inertial constraints on each string The610
effect of the radiation properties of the body on the611
computed vibratory signals is supported when comparing612
the spectrograms in Fig 13 which clearly shows the613
spectral filtering of the vibration signal614
Figure 13 Spectrograms of the bridge force (top) and sound
pressure (bottom) Nonlinear six-stringbody coupled model
Arpeggio of a G major chord with three strings stopped using
a set of auxiliary oscillators (sound files available)
V CONCLUSIONS615
Following our recent work showing the potential616
of the modal Udwadia-Kalaba equations to physically617
simulate musical instruments we presented in this paper618
a computational approach that allows efficient time-619
domain simulations of realistic guitar plucks In this620
work we further generalized our simple stringbody621
coupled model toward the development of a more622
realistic guitar model by including the complete set623
of strings the two polarizations of string motions the624
string geometric nonlinearity the stringfret interaction625
that controls the playing frequency and the body626
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 11
dynamics of an actual instrument The dynamics of627
the various subsystems were formulated in terms of628
their unconstrained modes and coupling were achieved629
by enforcing several constraint conditions One of the630
developments of the model was the introduction of631
flexible-dissipative-inertial supplementary equations as632
an alternative to the common kinematical constraints633
which here improve the role of the stringfingerfret634
interaction at the fingerboard This approach can635
also benefit the modeling of other complex dynamical636
systems outside the realm of physical instrument637
modeling To improve the realism of the synthesis638
we also included crudely the radiation properties of639
the instrument body into the computational model640
through an ad hoc procedure Sound pressure was641
computed adopting a hybrid approach by convoluting642
the computed bridge force signals with a vibro-acoustic643
impulse response measured on the test guitar Such644
vibro-acoustic response can be measured in an anechoic645
chamber or else include the acoustics of any specific646
room This approach appeared not only straightforward647
to compute realistic sounds of string instruments for648
which the radiating components are subjected to point-649
wise excitation but also very efficient compared to heavy650
computational sound radiation techniques651
ACKNOWLEDGMENTS652
The authors thanks Paulo Vaz de Carvalho for653
lending the guitar and interesting discussions This work654
is funded by national funds through the FCT ndash Fundacao655
para a Ciencia e a Tecnologia IP under the Norma656
Transitoria - DL 572016CP1453CT0007657
REFERENCES658
1JH Ginsberg Mechanical and Structural Vibrations Theory659
and Applications chapter 9 Wiley 2001660
2J Antunes and V Debut Dynamical computation of constrained661
flexible systems using a modal Udwadia-Kalaba formulation662
Application to musical instruments J Acoust Soc Am663
141764 2017664
3J Antunes V Debut L Borsoi X Delaune and P Piteau665
A modal Udwadia-Kalaba formulation for the vibro-impact666
modelling of continuous flexible systems with intermittent667
contacts Procedia Engineering 199322ndash329 2017668
4FE Udwadia and RE Kalaba Analytical dynamics - A new669
approach Cambridge University Press 1996670
5C Valette The mechanics of the vibrating string Mechanics of671
musical instruments Springer 1995672
6J Woodhouse On the synthesis of guitar plucks Acta Acust673
united Ac 90928ndash944 2004674
7M E McIntyre RT Schumacher and J Woodhouse On the675
oscillations of musical instruments Journal of the Acoustical676
Society of America 741325ndash1345 1983677
8A Chaigne On the use of finite differences for musical678
synthesis Application to plucked stringed instruments Journal679
drsquoAcoustique 5181ndash211 1992680
9JM Adrien The missing link Modal synthesis In G DePoli681
A Picalli and C Roads editors Representations of musical682
sounds MIT Press Cambridge 1991683
10S Bilbao Numerical Sound Synthesis Finite Difference684
Schemes and Simulation in Musical Acoustics Wiley Publishing685
2009686
11L Trautmann and R Rabenstein Multirate simulations687
of string vibrations includingnonlinear fret-string interactions688
usingthe functional transformation method EURASIP Journal689
on Applied Signal Processing 7949ndash963 2004690
12V Chatziioannou and M van Walstijn Energy conserving691
schemes for the simulation of musical instrument contact692
dynamics J Sound Vib 339262ndash279 2015693
13S Bilbao and A Torin Numerical modeling and sound synthesis694
for articulated stringfretboard interactions Acta Acust united695
Ac 63336ndash347 2015696
14V Debut M Carvalho M Marques and J Antunes Physics-697
based modeling techniques of a twelve-string portuguese guitar698
a three-dimensional non-linear time-domain computational699
approach for the multiple-stringsbridgesoundboard coupled700
dynamics Appl Acoust 1083ndash18 2016701
15M van Walstijn and J Bridges Simulation of distributed contact702
in string instruments A modal expansion approach In 24th703
European Signal Processing Conference (Eusipco 2016) pages704
1023ndash1027 2013705
16C Issanchou S Bilbao JL Le Carrou C Touze and O Doare706
A modal-based approach to the nonlinear vibration of strings707
against a unilateral obstacle simulations and experiments in the708
pointwise case J Sound Vib 393229ndash251 2017709
17R Bader Computational Mechanics of the Classical Guitar710
Springer-Verlag Berlin Heidelberg 2005711
18H Mansour J Woodhouse and G Scavone Enhanced wave-712
based modelling of musical strings Part 1 Plucked strings Acta713
Acust united Ac 1021094ndash1107 2016714
19O Inacio J Antunes and MCM Wright Computational715
modeling of string-body interaction for the violin family and716
simulation of wolf notes J Sound Vib 310260ndash286 2007717
20S Bilbao A Torin and V Chatziioannou Numerical modeling718
of collisions in musical instruments Acta Acust united Ac719
101155ndash173 2015720
21C Issanchou JL Le Carrou C Touze B Fabre and O Doare721
Stringfrets contacts in the electric bass sound Simulations and722
experiments App Acoust 129217ndash228 2018723
22D J Braun and M Goldfarb Eliminating constraint drift in the724
numerical simulation of constrained dynamical systems Comput725
Methods in Appl Mech Eng 198(37)3151 ndash 3160 2009726
23GF Carrier On the nonlinear vibration problem of the elastic727
string Q Appl Math 3157ndash165 1945728
24G Kirchhoff Vorlesungen uber Mechanik Teubner 1883729
25J Pakarinen V Valimaki and M Karjalainen Physics-based730
methods for modeling nonlinear vibrating strings Acta Acust731
united Ac 91312ndash325 2005732
26PMC Morse and KU Ingard Theoretical Acoustics733
International series in pure and applied physics Princeton734
University Press 1968735
27J Juang Applied System Identification PTR Prentice-Hall736
Inc New Jersey 1994737
28B E Richardson The acoustical development of the guitar J738
Catgut Acoust Soc 51ndash10 1994739
29J Woodhouse Plucked guitar transients Comparison of740
measurements and synthesis Acta Acust united Ac 90945ndash741
965 2004742
30M Tuckerman B J Berne and G J Martyna Reversible743
multiple time scale molecular dynamics J Chem Phys744
97(3)1990ndash2001 1992745
31A Chaigne Viscoelastic properties of nylon guitar strings J746
Catgut Acoust Soc Series II 121ndash27 1991747
32T Tolonen V Valimaki and M Karjalainen Modeling of748
tension modulation nonlinearity in plucked strings IEEE Trans749
Speech Audio Process 8(3)300ndash310 2000750
33A Chaigne and J Kergomard Acoustics of Musical Instruments751
Modern Acoustics and Signal Processing Springer New York752
2016753
12 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
754
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 13
Physical synthesis of six-string guitar plucks usingthe Udwadia-Kalaba modal formulation
V Debut1 2 a) and J Antunes1 2 b1INET-md Departamento de Ciencias Musicais Faculdade de Ciencias Sociais e Humanas FCSH
Universidade Nova de Lisboa 1069-061 Lisbon Portugal2Centro de Ciencias e Tecnologias Nucleares Instituto Superior Tecnico Universidade de Lisboa
Estrada Nacional 10 Km 1397 2695-066 Bobadela LRS Portugal
(Dated May 14 2020)
Extending previous work by the authors this paper develops a time domain synthesismethod for classical guitar based on substructuring concepts and using the Udwadia-Kalaba(U-K) modeling strategy Adopting a modal description of the dynamics of the separateflexible subsystems in terms of their unconstrained modes and enforcing coupling constraintconditions for the assembly the result is an explicit dynamical modal formulation for thecoupled system that directly lends itself to time-stepping methods for simulation Theguitar model couples six strings through a body model of an actual instrument includestwo string polarizations and the string geometrical nonlinear effects as well as for thestringfret interaction as the instrument is played Details are given for putting all thevibrating components together in a satisfying manner and a specific strategy is exploredto allow for a non-rigid fret using flexible-dissipative-inertial constraints The reliability ofthe approach is demonstrated with simulation examples that confirm the features one wouldexpect regarding the dynamical behaviour of classical guitars Finally a pragmatic approachis made to calculate the radiated sound by convolution combining the computed bridgeforce with a measured vibro-acoustic impulse response of the instrument which proved togive satisfactory sounding results
ccopy2020 Acoustical Society of America [httpdxdoiorg(DOI number)]
[XYZ] Pages 1ndash13
I INTRODUCTION1
The sound produced by musical instruments is the2
result of complex vibratory events usually involving3
multiple interaction between structural components and4
the player Beyond describing the excitation process the5
geometry and the dynamical behavior of the separate6
components any physical model of musical instrument7
also requires to develop dedicated methods that enforce8
contact conditions between the various subsystems to9
make them vibrate consistently In practice this is a10
rather complex issue that involves considerable modeling11
and computational difficulties and which can be critical12
in view of computational efficiency and accuracy of13
simulation Among the methods developed in vibration14
analysis to assemble structural components we can15
distinguish predominantly two standard techniques each16
one with specific merits and drawbacks Penalty17
methods which approximate the contact condition in18
terms of an inter-penetration force are often easy to19
implement but may lead to large computational expense20
when impacting near-rigid obstacles since the penalty21
parameters then impose small computational time scales22
a)vincentdebutfcshunlptb)jantunesctntecnicoulisboapt
andor the need for iterative numerical schemes On the23
other hand Lagrange multipliers allow constraints to be24
applied in a mathematically exact manner but demand25
the introduction of extra variables thus increasing the26
order of the system of equations127
Recently the present authors23 demonstrated that28
reliable simulations of flexible systems can be carried29
out by formulating their dynamics using the Udwadia30
and Kalaba4 equations which were originally proposed31
for modeling constrained discrete dynamical systems32
The extension was achieved using a modal approach33
for the continuum in terms of the unconstrained34
modes of the components and the resulting modal U-K35
formulation was found successful for systems involving36
point-constraints either linear or intermittent thus37
enabling the dynamical computations of many systems38
of practical interest Regarding the very modeling39
challenging systems that are musical instruments it40
seems that the modal U-K approach could be used41
profitably for physics-based sound synthesis purposes42
The physics of guitar plucks has been covered by43
many authors notably Valette5 and Woodhouse6 and44
models of musical strings have now reached a fine45
level of detail including frequency-dependent losses46
flexural dispersion and nonlinear features For obtaining47
the actual solutions different formulations have been48
proposed for solving the partial differential equations49
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 1
numerically using wave propagation and reflection50
functions7 finite difference8 or the modal approach9 and51
also deriving energy preserving schemes10 to guarantee52
stability of the numerical solutions Modal methods53
- which will be employed here - have formed the54
basis of successful modeling strategies both for linear55
and nonlinear string vibration and with or without56
constraints11ndash16 with the advantages of being simple to57
implement and of dealing in an exact manner with some58
complex physical phenomena ie string damping and59
dispersion Despite the good deal of work on string60
modeling most of the models of pluck instruments in the61
literature remain highly simplified cases of the underlying62
physics and rarely address the entire problem taking63
account the full set of strings that drives the instrument64
body and that in turn produces propagating pressure65
waves In this view the modeling work by Bader1766
that employs finite difference by Mansour et al18 that67
uses a wave-based approach and by Debut et al14 that68
makes use of a modal representation certainly pertain to69
the small group of studies that deal with most complete70
models of actual multi-stringed instruments71
Following the work by Antunes and Debut2 the task72
of this paper is to show how flexible and versatile the73
modal U-K approach can be for the physically based74
sound synthesis of stringed instruments by exploring75
the issues of assembling the structural components in76
a realistic manner and illustrating the ability of the77
technique to achieve reliable simulations While the78
technique was presented for a crude model of guitar this79
paper refines and extends our previous work in several80
respects First the model to be built here involves all the81
major components of a classical guitar that technically82
raises the issue of multiple-point coupling and makes the83
implementation more complex It includes six strings84
vibrating in both polarizations and coupled together at85
the bridge by the body motions one or more pressing86
fingers on the fingerboard to control the frequency of the87
played notes as well as other relevant features of vibrating88
string namely wave dispersion frequency-dependent89
damping and nonlinear geometrical effects The issue90
of coupling the subsystems is dealt in regard to the U-91
K formalism by means of additional constraining terms92
in the dynamical equation and hence contrasts with the93
more traditional approach that uses penalty formulation94
to model contact in musical instruments111219ndash21 In95
order to model realistic finger interaction a specific96
strategy is also developed and explored to account for97
the possible stiffness damping and inertial properties98
of the constraints thus bypassing the non-deformable99
nature of common kinematic constraints Finally to100
achieve the end goal of synthesis the resulting sound101
is computed using a simple and pragmatic approach102
Based on convolution it uses the computed bridge force103
and a measured impulse response that characterizes the104
vibro-acoustic behavior of the guitar body and hence105
gets around the problem of producing satisfactory sounds106
without performing complicated numerical calculation of107
sound radiation108
The outline of this paper is as follows The U-K109
theory and its extension to deal with flexible systems110
is briefly described Then we give details of the guitar111
model and develop the dynamical equations for the112
separate subsystems and their coupling The final section113
is devoted to present numerical results obtained by time-114
domain simulations with intention of illustrating the115
dynamical behavior of the fully coupled model in view116
of the physics of classical guitars117
II THEORETICAL FORMULATION118
A Basic U-K formulation119
In general mathematical terms Udwadia and Kalaba120
proposed the following standard form for the study of the121
dynamics of constrained discrete mechanical systems4122
Mx = F(x x t) + Fc(x x t) (1)
which expresses the system response x as the result123
of the application of the constraint-independent forces124
F (external and internal including the stiffness and125
damping force terms) acting on it and some additional126
forces Fc stemming from a set of constraints which127
both are considered as known functions of x x and t128
The U-K formulation then benefits from an alternative129
expression of the usual constraint equations ψi(x x t) =130
0 (i = 1 m) obtained by differentiation with respect131
to time written as a matrix-vector constraint system in132
terms of accelerations as133
A(x x t) x(t) = b(x x t) (2)
where A is the constraint matrix and b is a known vectorTheir main result is then to provide an explicit expressionfor the constrained dynamics x(t) and the constraintforce vector Fc(t) at each instant given by
x = xu + Mminus12B+(bminusAxu) (3a)
Fc(t) = M12B+(bminusAxu) (3b)
denoting B+ the Moore-Penrose pseudo-inverse of B =134
AMminus12 The vector xu represents the dynamical135
response of the unconstrainted system solution of136
xu = Mminus1F(x x t) (4)
while the second term in the right-side hand of Eq137
(3a) accounts for the influence of the constraints on138
the system since the unconstrained system must further139
comply with the physical constraints The superlative140
elegance of the U-K formulation lies in the fact that141
it expresses the dynamics of the constrained system142
through a single dynamical equation (3a) that takes143
constraints into account and allows if needed the144
knowledge of the constraining force through (3b) No145
additional variables such as Lagrange multipliers are146
needed thus avoiding difficulties in the computation147
of the solutions as for Differential-Algebraic system of148
equations Notably Eqs (3) may be applied to linear149
2 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
or nonlinear conservative or dissipative systems and150
may also be efficiently solved using a suitable time-step151
integration scheme for a given excitation152
B Modal U-K formulation153
Extension to constrained continuous multibody154
systems defined through their unconstrained modes Φ155
was formulated in Ref 2 by transforming (3) to the156
modal space through x = Φqqq leading to the formulation157
158
qqq = qqqu +MMMminus12BBB+(bminusAAAqqqu) (5)
where qqqu are the modal accelerations of the159
unconstrained configuration MMM = ΦTMΦ is the160
modal mass matrix AAA = AΦc is the modal constraint161
matrix with Φc the matrix of all modeshapes where162
constraints are defined and BBB = AAAMMMminus12 Finally the163
modal constraining force FFF c can also be computed as164
FFF c = MMM12BBB+(bminusAAAqqqu) (6)
C Dynamically coupled subsystems165
The application of the U-K modal formulation to a166
set of subsystems coupled through a number of kinematic167
constraints is now considered In a modal description the168
dynamics of the subsystem ` (` = 1 2 L) subjected169
to an external force field can be classically written as a170
set of modal equations in terms of the vector of modal171
amplitudes qqq` and its derivatives172
MMM `qqq` +CCC`qqq` +KKK`qqq` +FFF `nl(qqq` qqq`) = FFF `ext +FFF `nl (7)
where MMM ` CCC` and KKK` are diagonal matrices of the173
modal parameters m`n c`n and k`n (n = 1 N `) defined174
according to the modeshapes φ`n and FFF `ext and FFF `nl are175
the modal forces obtained by projection of the external176
and nonlinear forces on the modal basis The modal177
accelerations of the unconstrained system to be used in178
Eq (5) are given by179
qqq`u =(MMM `)minus1
FFF ` (8)
denoting FFF ` the vector of all the constraint-independent180
modal forces181
FFF `= FFF `extminusCCC`qqq`minusKKK`qqq`minusFFF `nl(qqq` qqq`) (9)
for which it is assumed that the vectors of modal182
constrained displacements and velocities are known at183
each time-step Stacking the modal quantities of the184
unconstrained subsystems in compact vectors as185
qqqu =[qqq1uT qqqLuT
]T(10)
the unconstrained modal accelerations of the coupled186
system read finally as187
qqqu = MMMminus1
[FFFext minusCCC qqq minusKKKqqq minus FFFnl(qqq qqq)
](11)
denoting188
qqq =[qqq1T qqqLT
]T(12)
In view of Eq (10) the matrices MMM CCC and KKK areblock diagonal set up by the submatrices of the modalparameters of the subsystems assembled as
MMM = diag(MMM1 MMML) (13a)
CCC = diag(CCC1 CCCL) (13b)
KKK = diag(KKK1 KKKL) (13c)
and FFFext and FFFnl are modal vectors associated to theexternal and nonlinear interaction forces acting on thevarious subsystems written respectively as
FFFext =FFF 1
extT FFFLextTT
(14a)
FFFnl =FFF 1
nlT FFFLnlTT
(14b)
The second set of equations to be considered concerns the189
P coupling constraints at locations r`c In most practical190
situations these are amenable to linear relationships by191
appropriate differentiation with respect to time leading192
to the standard form193
AAA(qqq qqq t) qqq = b(qqq qqq t) (15)
where AAA = AΦc with Φc = diag(Φ1c Φ
Lc ) and where194
Φ`c contains the modeshape vectors of each subsystem195
at the constraint location r`c Notice that because the196
constraints are implemented at the acceleration level197
Eq (15) generally leads to error accumulation in the198
computed motion during numerical integration due to199
numerical approximations and round-off errors resulting200
in constraint drift phenomenon To compensate for these201
errors a number of constraint stabilization techniques202
have been proposed22 and can be implemented if203
necessary as recently done by the authors for systems204
with intermittent contacts3 However for the problem205
at hand there was no necessity for implementing such206
correction terms Indeed by monitoring the constraints207
as the numerical simulation proceeds we observed a208
constraint drift under 20times10minus13 m which is negligible209
III GUITAR MODEL210
The guitar model involves six strings interacting with211
a guitar body and includes further models for the pressing212
finger on the fingerboard The strings are modeled213
using a simplified nonlinear formulation considering both214
polarizations and the body behavior is expressed in terms215
of modes identified from bridge input measurements on216
a real-life guitar217
A Nonlinear string model218
The vibratory motion of the strings is modeled219
using the KirchhoffndashCarrier approach2324 which is a220
simplified manner to describe large amplitude string221
motions Essentially the model discards the dynamics222
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 3
of the longitudinal modes but retains the effect of the223
geometrical nonlinearity by means of a pseudo-external224
forcing term added to the string transverse equations225
In our context it has two advantages First the model226
is known to capture relevant nonlinear phenomena for227
musical strings1425 thus enabling the reproduction of228
typical audible nonlinear effects Second it serves as a229
convenient means of illustrating that the U-K framework230
can deal effectively with nonlinear terms231
1 Kirchhoff-Carrier simplified nonlinear string model232
We consider a stiff string of length L cross-sectionalarea S mass density ρ Youngrsquos modulus E momentof inertia I and coefficient of dissipation η subjectto an axial tension T0 vibrating in two perpendiculartransverse motions Y (x t) and Z(x t) According to theKirchhoff-Carrier assumptions a simplified model of thefree vibration of the string is given by
ρSpart2Y
partt2+ η
partY
partt+ EI
part4Y
partx4=[T0 + Tdyn(t)
]part2Ypartx2
(16a)
ρSpart2Z
partt2+ η
partZ
partt+ EI
part4Z
partx4=[T0 + Tdyn(t)
]part2Zpartx2
(16b)
which is the standard wave equations for vibrating string233
including both polarizations with an additional forcing234
term proportional to the net dynamical increase in string235
length expressed as236
Tdyn(t) =ES
2L
int L
0
[(partY (x t)
partx
)2
+
(partZ(x t)
partx
)2]dx (17)
Eq (17) shows that the Kirchhoff-Carrier model only237
requires the knowledge of the transverse motions and238
that is independent of the x coordinate As far as the239
numerical technique is concerned the form of Eq (17)240
is very convenient particularly for modal synthesis and241
contrasts with the delicate numerical resolution of the242
ldquogeometrically exactrdquo model26 by modal discretization243
due the spatio-temporal nature of the nonlinear forcing244
2 Kirchhoff-Carrier nonlinear modal terms245
Denoting X a generic direction of string motion (X246
stands for Y or Z) the modal forces associated to the247
string nonlinear behavior are obtained by integration of248
the additional force terms of Eq (16) as249
fXn (t) =
int L
0
[Tdyn(t)
part2X
partx2
]φXn (x)dx (18)
Considering a string pinned at the nut and free at the250
bridge the modeshapes φXn (x) are given by251
φXn (x) = sin
[(2nminus 1)πx
2L
](19)
Substituting Eq (19) in Eq (18) and using a modalexpansion for the string motion the nonlinear modal
terms are finally given by
fXn (t)=minusESπ4
128L3(2nminus 1)2qXn (t)(
NYsumm=1
(2mminus1)2[qYm(t)
]2+
NZsumm=1
(2mminus1)2[qZm(t)
]2)(20)
with NY and NZ the sizes of the two string modal basis252
B The instrument body253
In view of Eq (7) the dynamic behaviour of the254
body is described by its modal properties extracted255
from input admittance measurements In contrast256
to modal computations based on a full model of the257
instrument body this is a direct approach that avoids258
complications of modeling in particular the frequency-259
dependent dissipative effects which significantly affect260
the computed sounds but are very difficult to model261
properly and that readily lend the synthesis method262
to be applied to any stringed instruments Since both263
polarizations of string motions are accounted a correct264
modeling of the stringbody coupling would demand265
the knowledge of both the in-plane and out-of-plane266
body modeshapes at the bridge However a simpler267
model is taken here by considering that only string268
motions normal to the soundboard couple to the body269
motions and also by neglecting the in-plane motions of270
the soundboard This means that the string motions in271
the plane parallel to the soundboard are unable to radiate272
sound and that the sound radiation is produced only273
by the normal motions of the soundboard Of course274
none of these choices is entirely satisfactory for accurate275
synthesis because in-plane and out-of-plane motions are276
coupled and also because in-plane motions couple the277
top and back plates via the ribs However since278
radiation is dominated by the out-of-plane vibration279
there is no doubt that in-plane motions are of second-280
order significance In practice the main advantage of281
our assumptions is to make the body characterization282
considerably simpler involving only transfer function283
measurements in the direction normal to the soundboard284
and avoiding the delicate measurements of the cross285
terms of the transfer function matrix286
The test guitar was a high-quality concert guitar287
built by Friederich number 694 During the tests it288
was positioned in the vertical position clamped to a289
rigid support by the neck and softly restrained in the290
lower bout with all the tuned strings damped (see291
Fig 1) The excitation and vibratory response were292
measured using a miniature force sensor (Kistler type293
9211) and a light-weight accelerometer (BampK 4375)294
respectively attached to the tie block using a thin layer295
of bee-wax Modal data were collected at two bridge296
locations close to the attachment points of the lowest297
and highest strings and a simple linear interpolation298
between the two transfer functions was performed to299
provide the unmeasured transfer functions at the other300
points where the strings make contact with the bridge301
4 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
Figure 1 Set-up for transfer function measurements at the
guitar body bridge
Modal identification was then achieved using the set of302
impulse responses expressed in terms of velocity and303
implementing the Eigensystem Realization Algorithm27304
The modal parameters for the first-ten identified modes305
are listed in Table I Fig 2 shows an example of306
an impulse response and its transfer function measured307
at the bridge end of the lowest string together with308
the corresponding reconstructed functions showing that309
the estimation was generally reliable up to 700 Hz310
Fig 3 also shows the identified mode shapes of the311
body at the bridge which exhibit typical motions of312
guitar modes28 The modes show a small degree of313
complexity but the effect of the imaginary components314
on the transfer function is small and can be neglected315
for our purpose Note that modes have been normalized316
according to max(|real(φBn (rs))|) = 1 (s = 1 6)317
Table I Identified modal parameters of the guitar body
Mode n fn (Hz) ζn () mn (Kg)
1 894 15 0402 1826 182 0083 2299 064 0204 3102 089 0475 4035 055 1066 5114 075 0587 5442 053 1828 6139 061 0699 6698 069 03810 8484 059 081
C Constraining forces and modal constraints318
1 Stringbody constraints319
Assuming a rigid transmission at the points wherethe strings make contact with the bridge the modalequations of the constrained system require to satisfya set of constrained modal equations stemming fromthe condition that at the bridge the strings and thebridge undergo the same motion Because the bodyresponse is here described only in the direction normal
0 005 01 015
Time [s]
-20
0
20
h (
t )
[ (
ms
)N
]
100 200 300 400 500 600 700 800 900
Frequency [ Hz ]
10-2
100
| H
( f
) |
[ (
ms
)N
]
Figure 2 Measured (green) and reconstructed (red) impulse
responses and transfer functions close to the lowest string
Gray dotted lines stand for the identified modal frequencies
E A D G B E
-1
0
1
1(
xB
)
f1 = 8939 Hz
E A D G B E
-1
0
1
2(
xB
)
f2 = 18262 Hz
E A D G B E
-1
0
1
3(
xB
)
f3 = 22986 Hz
E A D G B E
-1
0
1
4(
xB
)
f4 = 31021 Hz
E A D G B E
-1
0
1
5(
xB
)
f5 = 40352 Hz
E A D G B E
-1
0
1
6(
xB
)
f6 = 51143 Hz
E A D G B E
-1
0
1
7(
xB
)
f7 = 54417 Hz
E A D G B E
-1
0
1
8(
xB
)
f8 = 61386 Hz
E A D G B E
-1
0
1
9(
xB
)
f9 = 66978 Hz
Figure 3 Identified modeshapes of the guitar body at the
stringbridge contact points Real (blue) and imaginary
(magenta) components
to the soundboard the string motion Ys(xb t) in thisplane must be the same as the body motion at thestring contact point Yb(rs t) while its motion Zs(xb t)in the direction parallel to the soundboard must be nilFormally for each string s (s = 1 S) this results inthe conditions
Ys(xb t)minus Yb(rs t) = 0 (21a)
Zs(xb t) = 0 (21b)
or in terms of modal amplitudesΦYs
b
TqqqYs(t)minus
ΦBrsTqqqB(t) = 0 (22a)
ΦZs
b
TqqqZs(t) = 0 (22b)
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 5
where the modeshape vectors are written as
ΦYs
b =φYs1 (xb) φ
Ys
NYs(xb)
T(23a)
ΦZs
b =φZs1 (xb) φ
Zs
NZs(xb)
T(23b)
ΦBrs =φB1 (rs) φ
BNB
(rs)T
(23c)
Defining a vector of modal coordinates qqq built by stacking320
the individual modal displacements of the subsystems ie321
the strings including both polarizations and the body322
qqq = qqqY qqqZ qqqBT (24)
where qqqY and qqqZ are partitioned into terms associatedwith the different strings as
qqqY =qqqY1T qqqY6T
T(25a)
qqqZ =qqqZ1T qqqZ6T
T(25b)
and323
qqqB = qB1 qBNBT (26)
the constraint equations fit the matrix form324
AAAb qqq = 000 (27)
with the stringbody constraint matrix AAAb written as325
AAAb =
[ΦΦΦYb 000 minusΦΦΦB
000 ΦΦΦZb 000
](28)
where ΦΦΦYb and ΦΦΦZb are matrices of the modeshape vectors326
of the strings taken at the bridge location xb built as327
ΦΦΦYb =
ΦY1
b
T000 000
000
ΦY2
b
T 000
000 000
ΦY6
b
T
(29)
and similarly for the perpendicular direction and ΦΦΦB328
stands for the modeshapes of the body at the stringbody329
contact points rs330
ΦΦΦB =
ΦBr1T
ΦBr6T (30)
2 Stringfinger coupling at the fingerboard331
Modeling a stopping string for playing different332
notes can be thought in several ways according to the333
type of sound to simulate as well as to the degree of334
simplicitycomplexity aimed for the model The simplest335
model was taken in Ref 2 imposing a rigid kinematical336
constraint at a single location If such model could affect337
the tuning of the string effectively it is nonphysical338
in that part of the vibrational energy flows beyond the339
contact point (see Ref 2 for discussion) This simple340
approach is here extended to allow for multiple-point341
rigid coupling or for a non-rigid interaction as shown in342
Fig 4343
a Multiple-point rigid coupling This first optiondirectly enters into the realm of the standard kinematicalconstraints common in multibody theory The stringmotions Ys and Zs at the multiple finger locations xFf
(f = 1 F ) should be nil at all times so that
Ys(xFf t)=0 rArr
ΦYs
Ff
TqqqYs(t) = 0 (31a)
Zs(xFf t)=0 rArr
ΦZs
Ff
TqqqZs(t) = 0 (31b)
where the modeshape vectors at the finger locations are
ΦYs
Ff=φYs1 (xFf
) φYs
NYs(xFf
)T
(32a)
ΦZs
Ff=φZs1 (xFf
) φZs
NYs(xFf
)T
(32b)
Eq (31) can be rearranged in a matrix form as344
AAAf qqq = 000 (33)
where the modal constrained matrix is345
AAAf =
[ΦΦΦYs
F 000
000 ΦΦΦZs
F
](34)
with346
ΦΦΦYs
F =
ΦYs
F1
T
ΦYs
FF
T ΦΦΦZs
F =
ΦZs
F1
T
ΦZs
FF
T (35)
b Non-rigid coupling To extend the model to347
allow for non-rigid contact requires a little more care348
since a flexible-dissipative-inertial model for coupling349
subsystems cannot be directly formulated as kinematic350
constraints Nevertheless they can be implemented351
by assigning one or more additional flexible-dissipative-352
inertial elementary subsystems which are constrained to353
follow the string motions at the finger location(s) exactly354
as the body modes are included This modeling strategy355
somehow creates a bridge between computationally356
efficient multibody modeling providing ldquorigidrdquo kinematic357
constraints and the versatile but computationally slower358
modeling strategy through penalty-based constraints359
Non-rigid connections on the fingerboard cantherefore be obtained by extending the guitar modelusing a set of mass-spring-damper oscillators at locationsxFf
(f = 1 F ) and enforcing coupling as
Ys(xFf t)minusY Ff
f (t)= 0 (36a)
Zs(xFf t)minusZFf
f (t)= 0 (36b)
where YFf
f and ZFf
f are the responses of the fingercontact points in the two perpendicular directionsReturning to Eq (11) the vector and matrices mustnow be partitioned as
qqqu =qqqYsu T qqqZs
u T qqqFsu T
T(37a)
MMM = diag(MMMYsMMMZsMMMFs
)(37b)
CCC = diag(CCCYsCCCZsCCCFs
)(37c)
KKK = diag(KKKYsKKKZsKKKFs
)(37d)
6 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
where the vectors and matrices for the strings are as360
beforeMMMFs CCCFs andKKKFs are block diagonal matrices of361
the inertia damping and stiffness properties associated362
with the ldquofingerrdquo subsystem acting on string s including363
both directions of string motions364
qqqFsu =
Y F1
f Y FF
f ZF1
f ZFF
f
T(38)
In terms of modal coordinates Eqs (36) becomeΦYs
Ff
TqqqYs(t)minus Y Ff
f (t) = 0 (39a)ΦZs
Ff
TqqqZs(t)minus ZFf
f (t) = 0 (39b)
which can be rearranged as365
AAAf qqq = 000 (40)
with the modal constrained matrix AAAf written as366
AAAf =
[ΦΦΦYF 000 minusI 000
000 ΦΦΦZF 000 minusI
](41)
Figure 4 Rigid kinematical constraints (left) and flexible-
dissipative-inertial constraints (right) at multiple locations
D Dynamical formulation of the fully coupled model367
The formulation (11) readily applies to the full guitar368
model including the non-rigid constraint model for the369
stringfret interaction by stacking the unconstrained370
modal displacements into a vector371
qqqu = qqqYu qqqZu qqqBu qqqFu T (42)
where qqqYu qqqZu and qqqBu are as before and372
qqqFu=qqqY1
ufT qqqY6
ufT qqqZ1
ufT qqqZ6
ufTT
(43)
In view of the definition of qqq the assembled inertia MMM373
stiffnessKKK and dampingCCC matrices which are uncoupled374
are given by375
MMM=
MMMY 000 000 000
000 MMMZ 000 000
000 000 MMMB 000
000 000 000 MMMF
CCC=
CCCY 000 000 000
000 CCCZ 000 000
000 000 CCCB 000
000 000 000 CCCF
KKK=
KKKY 000 000 000
000 KKKZ 000 000
000 000 KKKB 000
000 000 000 KKKF
(44)
with MMM ` KKK` and CCC` the inertia stiffness and dampingmatrices for each individual subsystem ` and the vectorsFFFext and FFFnl write
FFFext =FFFYextT FFFZextT000000
T(45a)
FFFnl =FFFYnlT FFFZnlT000000
T(45b)
where FFFYext and FFFZext correspond to the plucking action376
of the player expressed for both planes of polarization377
and FFFYnl and FFFZnl are the modal vectors associated to the378
string nonlinear effects with entries given by Eq (20)379
The form of the modal constraint equation is now380
AAAqqq = 000 (46)
with381
AAA =
[AAAbAAAf
]=
ΦΦΦYb 000 minusΦΦΦB 000 000 000
000 ΦΦΦZb 000 000 000 000
ΦYFΦYFΦYF 000 000 000 minusFFFY 000
000 ΦZFΦZFΦZF 000 000 000 minusFFFZ
(47)
where ΦYFΦYFΦYF and ΦZFΦZFΦZF are block matrices with entries given382
by Eq (35) and FFFY and FFFZ are block diagonal matrices383
built with either the identity or null matrix depending on384
the constraints enforced on each string The form of (47)385
shows that the interface forces and constraint conditions386
couple the Y -motions of the strings and the body as well387
as coupling between the finger and the strings in each388
direction of motions Notice that coupling between both389
polarizations of string motions is not apparent in (47)390
although it is formulated in the nonlinear forces (45b)391
IV ILLUSTRATIVE COMPUTATIONS392
A System parameters393
For given initial conditions the system of equations394
comprising (5) with the unconstrained accelerations (11)395
and the constraint matrix (15) and using the specific396
matrices given by Eqs (44) and (47) can be integrated in397
time to produce the transient response The time-domain398
simulations were performed based on the identified body399
modal data and using standard string properties of400
classical guitars Assuming non-ideal strings with fixed-401
free ends the modeshapes are given by Eq (19) and402
their modal frequencies can be approximated by403
fYn = fZn =cT2πpn
(1 +
B
2T0p2n
)(48)
where cT is the velocity of transverse waves B is the404
bending stiffness of the string and T0 its tensioning405
pn = (2n minus 1)π2L and n the mode number Modal406
damping values are introduced following the pragmatic407
formulation given in Ref 6408
ζYn = ζZn =1
2
T0(ηF + ηAωn
)+ηBBp
2n
T0 +Bp2n(49)
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 7
where ηF ηA and ηB are the loss coefficients related to409
ldquointernal frictionrdquo ldquoair viscous dampingrdquo and ldquobending410
dampingrdquo The length of the strings is L=065 m411
and their density bending stiffness tensioning and loss412
coefficients are taken from Ref 29 The size of the modal413
basis of each string is defined by the maximum frequency414
of the computations (10000 Hz) which proved a sensible415
compromise for convergence to realistic simulations416
resulting in 800 modes for the set of strings To417
obtain suitable values for the elasticity parameter ES418
we perform simple vibratory tests on string samples419
Stretching a string vertically and suspending a mass to420
its free end one can easily measure the longitudinal421
vibration of the string-mass oscillator and obtain its422
natural frequency Assuming the string acting as a423
spring with no mass and reminding that the natural424
frequency of the longitudinal vibration is given by f =425
12π
radicESML then an estimate of ES is obtained The426
simulation results were obtained by time-step integration427
of the ordinary modal equations using the velocity-428
Verlet numerical scheme30 an integrator that combines429
simplicity second order accuracy with good numerical430
stability - see implementation details in Ref 2 We431
adopted a time-step of 10minus5s and assumed all initial432
modal displacements and velocities nil initially433
B Computational results434
Pluck excitations were simulated by considering a435
point-wise external force at xe=09L with a linear force436
ramp applied to one or more strings during 10 ms and437
by assuming null excitation force at subsequent instants438
1 Dynamical behaviour of the fully coupled model439
Results in Fig 5 were obtained for the open string440
E2 excited in the direction normal to the soundboard441
first considering a rigidly mounted string then coupling442
one string and finally all six strings to the instrument443
body Assuming no dynamics for the body the response444
spectra of string motion at the bridge is virtually nil445
according to the pinned boundary condition When446
coupling the instrument body to the string the response447
becomes obviously different and shows the classical series448
of near harmonic string partials together with smaller449
peaks from the body resonances While the first modal450
frequency of the unconstrained string is 41 Hz a close451
look at Fig 5 shows that the constraint at the bridge452
yields the fundamental frequency for the tuned string453
with pinned ends (f = 82 Hz) thus confirming the454
correct implementation of the bridge coupling Coupling455
the other five strings results in new peaks in the456
spectrum extending over the frequency range of the body457
modes and with an amplitude difference of about 100 dB458
compared to the most excited modes which clearly relate459
to the non-excited strings As found in some stringed460
instruments this dynamical coupling of all the strings via461
the bridge can lead to sympathetic vibrations which can462
further results in slight interference beats in the sound463
due to the inharmonicity produced by the bridge coupling464
and of the string partials465
0 200 400 600 800 1000 1200 1400 1600 180010
-40
10-20
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 180010
-20
10-10
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 1800
Frequency [ Hz ]
10-20
10-10
100
| Y
6(f
) |
Figure 5 Response spectra of string motion at the bridge
Top rigidly mounted string middle single stringbody
coupling bottom six-stringbody coupled model Force
amplitude 5 N Linear model
0 02 04 06 08 1
Time [ s ]
0
05
1
15
2
Mo
dal en
erg
y [
J ]
10-3
STRINGSFRETBODYTOTAL
Figure 6 Instantaneous modal energies Oblique excitation
with amplitudes of 2 N and 1 N in the Y and Z directions
Linear six-stringbody coupled model Conservative model
Figs 6 and 7 then serve at checking other466
implementation details Simulations pertained to the467
string E2 constrained at xF = 02509L by a single ldquostiffrdquo468
fret modeled with mF=5 10minus3 Kg cF=1 Nsm and469
kF=106 Nm plucked in the oblique direction The470
modal energies E` of the subsystems ` shown in these471
plots were computed from the results of the simulations472
as follows in terms of the modal parameters and the473
8 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
0 02 04 06 08 10
05
1
15
[ J ]
10-3 STRINGS
Y-dir
Z-dir
0 02 04 06 08 10
05
1
15
[ J ]
10-5 FRET
Y-dir
Z-dir
0 02 04 06 08 10
05
1
[ J ]
10-4 BODY
0 02 04 06 08 1
Time [ s ]
0
05
1
15
10-3 FULL SYSTEM
Figure 7 Instantaneous modal energies Same plucking
conditions as in Fig 6 Linear six-stringbody coupled model
Non-conservative model
modal displacements and velocities474
E`(t) =
Nsumn=1
[1
2m`n(ω`n)2
[q`n(t)
]2+
1
2m`n
[q`n(t)
]2](50)
One important feature seen in Fig 6 is that the total475
energy remains roughly constant over a conservative476
simulation while energies flow back and forth between477
the subsystems thus giving confidence in the numerical478
model for simulations of reasonable duration For479
the non-conservative case shown in Fig 7 the large480
difference in energy decay for the two polarizations481
confirms the different stringbody coupling with the482
directions of string motion Note that both the body483
and the fret oscillator are dominated by periodic energy484
exchanges at the frequency of the played note as the wave485
travels back and forth between bridge and fret Finally486
looking closely at the modal energy of the fret one can487
see smaller superimposed periodic disturbances related488
to the reflections between fret and nut illustrating that489
energy still flows through the modeled fret490
Finally we compare in Fig 8 results by varying491
the excitation amplitude in order to illustrate the string492
nonlinearity The simulations were obtained for the493
string E2 using the experimentally determined value494
of 32 kN for the flexibility ES which is comparable495
to the value given by Chaigne31 Fig 8 shows the496
temporal variations of string tension with an estimate497
of the playing frequency computed from time domain498
evaluations of the zero crossing frequency The general499
features of large amplitude string motions are clearly500
visible with a quasi-static increase of tension and501
superimposed oscillations at twice the string fundamental502
frequency and a small frequency variation of the played503
note which compares well with measurements32504
1 2 3 4 5 6 7
716
718
72
722
724
T (
t )
[ N
] Fexc
= 1 N
Fexc
= 5 N
1 2 3 4 5 6 7
Time [ s ]
822
824
826
828
f (
t )
[ H
z ]
Figure 8 String tension (top) and instantaneous frequency of
string motion (bottom) Nonlinear rigid body model
2 Parameter study of the stringfinger coupling model505
We now illustrate the different strategies to account506
for a stopping finger on the fingerboard and explore the507
influence on the coupled system of the finger model508
In contrast with the approach in Ref 2 the strategy509
in terms of auxiliary oscillators rigidly coupled to the510
string seems more realistic and is undoubtedly more511
versatile allowing different types of contacts to be512
modeled by varying the oscillator parameters However513
one could expect the finger dynamics to perturb the514
coupled system with changes in frequency and damping515
for its modes similar to what observed for a multi-modal516
system coupled to a system exhibiting resonance To517
give some insights Figs 9-11 show the effects of varying518
the finger mechanical parameters Simulations pertain to519
the string E2 constrained at xF = 01857L by a finger of520
mass mf = 5 10minus3 Kg and ignore the body dynamics521
The first test consists in varying the stiffness of the522
constraint assuming no damping Fig 9 shows the523
frequency spectra of the coupling force at the rigid bridge524
where the peaks of the string modes can be seen clearly as525
well as the resonance frequency of the auxiliary oscillator526
which moves in frequency with the stiffness and slightly527
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 9
modulates the other peaks A close look shows that528
the spectra are affected much more sensitively below529
the resonance of the finger oscillator with amplitude530
variations analogous to a system driven in the stiffness-531
controlled region Decreasing the stiffness causes a rise532
in amplitude of the undesirable modes while increasing533
the stiffness lower their amplitudes As observed for the534
stringbody coupling33 inharmonicity is also created by535
the stiffness termination which in turn affects the tuning536
of the note With the increase of stiffness perturbations537
of the string modes are reduced and tuning is improved538
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
4 Nm
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
5 Nm
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
kf = 10
6 Nm
Figure 9 Spectra of the coupling force at the rigid bridge
Stiffness is varied F=1 c1 = 0 Nsm
Fig 10 then shows the effect of varying the damping539
keeping the stiffness constant Not surprisingly if540
damping increases the amplitude of the coupled541
modes close to the oscillator resonance decrease and542
the frequency region affected becomes wider This543
significantly affects the time envelope of the pluck note544
that becomes smoother and has a shorter-time decay545
Finally the effect of the number of contacts is studied546
using the rigid contact model which is convenient for547
this purpose Fig 11 gives a picture of the spatial548
distribution of the string energy by plotting the standard549
deviation of string motion as a function of its length550
As seen the ldquopassiverdquo region is affected with great551
sensitivity by the number of contacts and surprisingly552
increasing the contact points does not necessarily reduce553
the undesirable transfer of energy Computations show554
that the effectiveness of the multi-point rigid constraints555
approach is strongly problem-dependent and varies in a556
non-obvious way with several factors including the finger557
width the spacing between contact points the relative558
string length from the ends and the excitation spectrum559
Taken in combination these results and noticing560
that the perturbations induced by each finger oscillator561
are additive due to the linear coupling a robust and562
physically sound strategy to simulate the pressing finger563
would be to implement a set of flexible-dissipative564
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
cf = 1 Nsm
cf = 10 Nsm
0 1 2 3 4 5
Time [ s ]
-2
0
2
4
Y (
t )
10-3
Figure 10 Spectra of the coupling force at the rigid
bridge (up) and time history of the string motion (bottom)
Damping is varied F=1 k1 = 105 Nm
constraints imposing a first stiff constraint in order565
to ensure a precise tuning followed by several soft566
constraints that would control the waveform and spectral567
content of the decaying response depending on their568
number and damping properties For illustration569
the spectrograms of two plucks for different sets of570
parameters are presented in Fig 12571
0 01 02 03 04 05 06
x [ m ]
10-5
10-2
ST
D (
Y(t
) )
F=1
F=2
F=3
F=5
016 017 018 019 02
x [ m ]
10-5
ST
D (
Y(t
) )
Figure 11 Standard deviation of the string motion Number
of contacts is varied Vertical dotted lines represent the
constraints locations Finger width (0015m) is kept constant
C Simulated acoustic radiation572
As an alternative to computing sounds from full573
vibro-acoustic models that usually require the use of574
massive computational resources it is tempting to apply575
10 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
Figure 12 Spectrograms of the force at the rigid bridge Up
F = 3 bottom F = 5 Parameter values kF1=106 Nm
cF1=1 Nsm and kFj =104 Nm cFj =10 Nsm (j = 2 F )
Constraint at xF = 01857L (sound files available)
simpler techniques that could be efficient to generate576
convincing sounds from the vibratory responses The577
method used here is to treat the guitar body as a linear578
input-output system and to compute the pressure by579
convolution benefiting from the local excitation of the580
body by the strings and using a vibro-acoustic impulse581
response of the test guitar This approach does not by582
any means simulate the sound response everywhere in the583
pressure field but it has the merit of adding the influence584
of the radiating properties of the body at a very low585
computational cost and of providing sounds of greater586
perceptual quality than the commonly used vibratory587
signals at the bridge The pressure at location rp can be588
computed by the convolution of the time-history of the589
stringbridge interaction force FY (rb t) and the vibro-590
acoustic impulse response function hY (rb rp t) between591
the bridge and microphone by592
p(rp t) = hY (rb rp t) lowast FY (rb t) (51)
The extension to account for the effects of all the593
strings is readily obtained by linear superposition594
principle summing the contributions from each string595
Measurements of the vibro-acoustic impulse response596
were performed under the same experimental conditions597
as for transfer function measurements using an598
additional condenser microphone Impulsive excitation599
was applied in the perpendicular direction to the600
soundboard and the acoustic pressure was measured601
at a distance of 1 m from the central axis of the602
guitar in a quiet room with all strings damped (see603
Fig 1) The impulse response was computed by inverse604
Fourier transform of the complex transfer function built605
from the force and response signals described in the606
frequency domain Results in Fig 13 pertain to the607
simulation of a G major arpeggio obtained with the608
two lowest and the top strings constrained using five609
auxiliary flexible-inertial constraints on each string The610
effect of the radiation properties of the body on the611
computed vibratory signals is supported when comparing612
the spectrograms in Fig 13 which clearly shows the613
spectral filtering of the vibration signal614
Figure 13 Spectrograms of the bridge force (top) and sound
pressure (bottom) Nonlinear six-stringbody coupled model
Arpeggio of a G major chord with three strings stopped using
a set of auxiliary oscillators (sound files available)
V CONCLUSIONS615
Following our recent work showing the potential616
of the modal Udwadia-Kalaba equations to physically617
simulate musical instruments we presented in this paper618
a computational approach that allows efficient time-619
domain simulations of realistic guitar plucks In this620
work we further generalized our simple stringbody621
coupled model toward the development of a more622
realistic guitar model by including the complete set623
of strings the two polarizations of string motions the624
string geometric nonlinearity the stringfret interaction625
that controls the playing frequency and the body626
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 11
dynamics of an actual instrument The dynamics of627
the various subsystems were formulated in terms of628
their unconstrained modes and coupling were achieved629
by enforcing several constraint conditions One of the630
developments of the model was the introduction of631
flexible-dissipative-inertial supplementary equations as632
an alternative to the common kinematical constraints633
which here improve the role of the stringfingerfret634
interaction at the fingerboard This approach can635
also benefit the modeling of other complex dynamical636
systems outside the realm of physical instrument637
modeling To improve the realism of the synthesis638
we also included crudely the radiation properties of639
the instrument body into the computational model640
through an ad hoc procedure Sound pressure was641
computed adopting a hybrid approach by convoluting642
the computed bridge force signals with a vibro-acoustic643
impulse response measured on the test guitar Such644
vibro-acoustic response can be measured in an anechoic645
chamber or else include the acoustics of any specific646
room This approach appeared not only straightforward647
to compute realistic sounds of string instruments for648
which the radiating components are subjected to point-649
wise excitation but also very efficient compared to heavy650
computational sound radiation techniques651
ACKNOWLEDGMENTS652
The authors thanks Paulo Vaz de Carvalho for653
lending the guitar and interesting discussions This work654
is funded by national funds through the FCT ndash Fundacao655
para a Ciencia e a Tecnologia IP under the Norma656
Transitoria - DL 572016CP1453CT0007657
REFERENCES658
1JH Ginsberg Mechanical and Structural Vibrations Theory659
and Applications chapter 9 Wiley 2001660
2J Antunes and V Debut Dynamical computation of constrained661
flexible systems using a modal Udwadia-Kalaba formulation662
Application to musical instruments J Acoust Soc Am663
141764 2017664
3J Antunes V Debut L Borsoi X Delaune and P Piteau665
A modal Udwadia-Kalaba formulation for the vibro-impact666
modelling of continuous flexible systems with intermittent667
contacts Procedia Engineering 199322ndash329 2017668
4FE Udwadia and RE Kalaba Analytical dynamics - A new669
approach Cambridge University Press 1996670
5C Valette The mechanics of the vibrating string Mechanics of671
musical instruments Springer 1995672
6J Woodhouse On the synthesis of guitar plucks Acta Acust673
united Ac 90928ndash944 2004674
7M E McIntyre RT Schumacher and J Woodhouse On the675
oscillations of musical instruments Journal of the Acoustical676
Society of America 741325ndash1345 1983677
8A Chaigne On the use of finite differences for musical678
synthesis Application to plucked stringed instruments Journal679
drsquoAcoustique 5181ndash211 1992680
9JM Adrien The missing link Modal synthesis In G DePoli681
A Picalli and C Roads editors Representations of musical682
sounds MIT Press Cambridge 1991683
10S Bilbao Numerical Sound Synthesis Finite Difference684
Schemes and Simulation in Musical Acoustics Wiley Publishing685
2009686
11L Trautmann and R Rabenstein Multirate simulations687
of string vibrations includingnonlinear fret-string interactions688
usingthe functional transformation method EURASIP Journal689
on Applied Signal Processing 7949ndash963 2004690
12V Chatziioannou and M van Walstijn Energy conserving691
schemes for the simulation of musical instrument contact692
dynamics J Sound Vib 339262ndash279 2015693
13S Bilbao and A Torin Numerical modeling and sound synthesis694
for articulated stringfretboard interactions Acta Acust united695
Ac 63336ndash347 2015696
14V Debut M Carvalho M Marques and J Antunes Physics-697
based modeling techniques of a twelve-string portuguese guitar698
a three-dimensional non-linear time-domain computational699
approach for the multiple-stringsbridgesoundboard coupled700
dynamics Appl Acoust 1083ndash18 2016701
15M van Walstijn and J Bridges Simulation of distributed contact702
in string instruments A modal expansion approach In 24th703
European Signal Processing Conference (Eusipco 2016) pages704
1023ndash1027 2013705
16C Issanchou S Bilbao JL Le Carrou C Touze and O Doare706
A modal-based approach to the nonlinear vibration of strings707
against a unilateral obstacle simulations and experiments in the708
pointwise case J Sound Vib 393229ndash251 2017709
17R Bader Computational Mechanics of the Classical Guitar710
Springer-Verlag Berlin Heidelberg 2005711
18H Mansour J Woodhouse and G Scavone Enhanced wave-712
based modelling of musical strings Part 1 Plucked strings Acta713
Acust united Ac 1021094ndash1107 2016714
19O Inacio J Antunes and MCM Wright Computational715
modeling of string-body interaction for the violin family and716
simulation of wolf notes J Sound Vib 310260ndash286 2007717
20S Bilbao A Torin and V Chatziioannou Numerical modeling718
of collisions in musical instruments Acta Acust united Ac719
101155ndash173 2015720
21C Issanchou JL Le Carrou C Touze B Fabre and O Doare721
Stringfrets contacts in the electric bass sound Simulations and722
experiments App Acoust 129217ndash228 2018723
22D J Braun and M Goldfarb Eliminating constraint drift in the724
numerical simulation of constrained dynamical systems Comput725
Methods in Appl Mech Eng 198(37)3151 ndash 3160 2009726
23GF Carrier On the nonlinear vibration problem of the elastic727
string Q Appl Math 3157ndash165 1945728
24G Kirchhoff Vorlesungen uber Mechanik Teubner 1883729
25J Pakarinen V Valimaki and M Karjalainen Physics-based730
methods for modeling nonlinear vibrating strings Acta Acust731
united Ac 91312ndash325 2005732
26PMC Morse and KU Ingard Theoretical Acoustics733
International series in pure and applied physics Princeton734
University Press 1968735
27J Juang Applied System Identification PTR Prentice-Hall736
Inc New Jersey 1994737
28B E Richardson The acoustical development of the guitar J738
Catgut Acoust Soc 51ndash10 1994739
29J Woodhouse Plucked guitar transients Comparison of740
measurements and synthesis Acta Acust united Ac 90945ndash741
965 2004742
30M Tuckerman B J Berne and G J Martyna Reversible743
multiple time scale molecular dynamics J Chem Phys744
97(3)1990ndash2001 1992745
31A Chaigne Viscoelastic properties of nylon guitar strings J746
Catgut Acoust Soc Series II 121ndash27 1991747
32T Tolonen V Valimaki and M Karjalainen Modeling of748
tension modulation nonlinearity in plucked strings IEEE Trans749
Speech Audio Process 8(3)300ndash310 2000750
33A Chaigne and J Kergomard Acoustics of Musical Instruments751
Modern Acoustics and Signal Processing Springer New York752
2016753
12 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
754
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 13
numerically using wave propagation and reflection50
functions7 finite difference8 or the modal approach9 and51
also deriving energy preserving schemes10 to guarantee52
stability of the numerical solutions Modal methods53
- which will be employed here - have formed the54
basis of successful modeling strategies both for linear55
and nonlinear string vibration and with or without56
constraints11ndash16 with the advantages of being simple to57
implement and of dealing in an exact manner with some58
complex physical phenomena ie string damping and59
dispersion Despite the good deal of work on string60
modeling most of the models of pluck instruments in the61
literature remain highly simplified cases of the underlying62
physics and rarely address the entire problem taking63
account the full set of strings that drives the instrument64
body and that in turn produces propagating pressure65
waves In this view the modeling work by Bader1766
that employs finite difference by Mansour et al18 that67
uses a wave-based approach and by Debut et al14 that68
makes use of a modal representation certainly pertain to69
the small group of studies that deal with most complete70
models of actual multi-stringed instruments71
Following the work by Antunes and Debut2 the task72
of this paper is to show how flexible and versatile the73
modal U-K approach can be for the physically based74
sound synthesis of stringed instruments by exploring75
the issues of assembling the structural components in76
a realistic manner and illustrating the ability of the77
technique to achieve reliable simulations While the78
technique was presented for a crude model of guitar this79
paper refines and extends our previous work in several80
respects First the model to be built here involves all the81
major components of a classical guitar that technically82
raises the issue of multiple-point coupling and makes the83
implementation more complex It includes six strings84
vibrating in both polarizations and coupled together at85
the bridge by the body motions one or more pressing86
fingers on the fingerboard to control the frequency of the87
played notes as well as other relevant features of vibrating88
string namely wave dispersion frequency-dependent89
damping and nonlinear geometrical effects The issue90
of coupling the subsystems is dealt in regard to the U-91
K formalism by means of additional constraining terms92
in the dynamical equation and hence contrasts with the93
more traditional approach that uses penalty formulation94
to model contact in musical instruments111219ndash21 In95
order to model realistic finger interaction a specific96
strategy is also developed and explored to account for97
the possible stiffness damping and inertial properties98
of the constraints thus bypassing the non-deformable99
nature of common kinematic constraints Finally to100
achieve the end goal of synthesis the resulting sound101
is computed using a simple and pragmatic approach102
Based on convolution it uses the computed bridge force103
and a measured impulse response that characterizes the104
vibro-acoustic behavior of the guitar body and hence105
gets around the problem of producing satisfactory sounds106
without performing complicated numerical calculation of107
sound radiation108
The outline of this paper is as follows The U-K109
theory and its extension to deal with flexible systems110
is briefly described Then we give details of the guitar111
model and develop the dynamical equations for the112
separate subsystems and their coupling The final section113
is devoted to present numerical results obtained by time-114
domain simulations with intention of illustrating the115
dynamical behavior of the fully coupled model in view116
of the physics of classical guitars117
II THEORETICAL FORMULATION118
A Basic U-K formulation119
In general mathematical terms Udwadia and Kalaba120
proposed the following standard form for the study of the121
dynamics of constrained discrete mechanical systems4122
Mx = F(x x t) + Fc(x x t) (1)
which expresses the system response x as the result123
of the application of the constraint-independent forces124
F (external and internal including the stiffness and125
damping force terms) acting on it and some additional126
forces Fc stemming from a set of constraints which127
both are considered as known functions of x x and t128
The U-K formulation then benefits from an alternative129
expression of the usual constraint equations ψi(x x t) =130
0 (i = 1 m) obtained by differentiation with respect131
to time written as a matrix-vector constraint system in132
terms of accelerations as133
A(x x t) x(t) = b(x x t) (2)
where A is the constraint matrix and b is a known vectorTheir main result is then to provide an explicit expressionfor the constrained dynamics x(t) and the constraintforce vector Fc(t) at each instant given by
x = xu + Mminus12B+(bminusAxu) (3a)
Fc(t) = M12B+(bminusAxu) (3b)
denoting B+ the Moore-Penrose pseudo-inverse of B =134
AMminus12 The vector xu represents the dynamical135
response of the unconstrainted system solution of136
xu = Mminus1F(x x t) (4)
while the second term in the right-side hand of Eq137
(3a) accounts for the influence of the constraints on138
the system since the unconstrained system must further139
comply with the physical constraints The superlative140
elegance of the U-K formulation lies in the fact that141
it expresses the dynamics of the constrained system142
through a single dynamical equation (3a) that takes143
constraints into account and allows if needed the144
knowledge of the constraining force through (3b) No145
additional variables such as Lagrange multipliers are146
needed thus avoiding difficulties in the computation147
of the solutions as for Differential-Algebraic system of148
equations Notably Eqs (3) may be applied to linear149
2 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
or nonlinear conservative or dissipative systems and150
may also be efficiently solved using a suitable time-step151
integration scheme for a given excitation152
B Modal U-K formulation153
Extension to constrained continuous multibody154
systems defined through their unconstrained modes Φ155
was formulated in Ref 2 by transforming (3) to the156
modal space through x = Φqqq leading to the formulation157
158
qqq = qqqu +MMMminus12BBB+(bminusAAAqqqu) (5)
where qqqu are the modal accelerations of the159
unconstrained configuration MMM = ΦTMΦ is the160
modal mass matrix AAA = AΦc is the modal constraint161
matrix with Φc the matrix of all modeshapes where162
constraints are defined and BBB = AAAMMMminus12 Finally the163
modal constraining force FFF c can also be computed as164
FFF c = MMM12BBB+(bminusAAAqqqu) (6)
C Dynamically coupled subsystems165
The application of the U-K modal formulation to a166
set of subsystems coupled through a number of kinematic167
constraints is now considered In a modal description the168
dynamics of the subsystem ` (` = 1 2 L) subjected169
to an external force field can be classically written as a170
set of modal equations in terms of the vector of modal171
amplitudes qqq` and its derivatives172
MMM `qqq` +CCC`qqq` +KKK`qqq` +FFF `nl(qqq` qqq`) = FFF `ext +FFF `nl (7)
where MMM ` CCC` and KKK` are diagonal matrices of the173
modal parameters m`n c`n and k`n (n = 1 N `) defined174
according to the modeshapes φ`n and FFF `ext and FFF `nl are175
the modal forces obtained by projection of the external176
and nonlinear forces on the modal basis The modal177
accelerations of the unconstrained system to be used in178
Eq (5) are given by179
qqq`u =(MMM `)minus1
FFF ` (8)
denoting FFF ` the vector of all the constraint-independent180
modal forces181
FFF `= FFF `extminusCCC`qqq`minusKKK`qqq`minusFFF `nl(qqq` qqq`) (9)
for which it is assumed that the vectors of modal182
constrained displacements and velocities are known at183
each time-step Stacking the modal quantities of the184
unconstrained subsystems in compact vectors as185
qqqu =[qqq1uT qqqLuT
]T(10)
the unconstrained modal accelerations of the coupled186
system read finally as187
qqqu = MMMminus1
[FFFext minusCCC qqq minusKKKqqq minus FFFnl(qqq qqq)
](11)
denoting188
qqq =[qqq1T qqqLT
]T(12)
In view of Eq (10) the matrices MMM CCC and KKK areblock diagonal set up by the submatrices of the modalparameters of the subsystems assembled as
MMM = diag(MMM1 MMML) (13a)
CCC = diag(CCC1 CCCL) (13b)
KKK = diag(KKK1 KKKL) (13c)
and FFFext and FFFnl are modal vectors associated to theexternal and nonlinear interaction forces acting on thevarious subsystems written respectively as
FFFext =FFF 1
extT FFFLextTT
(14a)
FFFnl =FFF 1
nlT FFFLnlTT
(14b)
The second set of equations to be considered concerns the189
P coupling constraints at locations r`c In most practical190
situations these are amenable to linear relationships by191
appropriate differentiation with respect to time leading192
to the standard form193
AAA(qqq qqq t) qqq = b(qqq qqq t) (15)
where AAA = AΦc with Φc = diag(Φ1c Φ
Lc ) and where194
Φ`c contains the modeshape vectors of each subsystem195
at the constraint location r`c Notice that because the196
constraints are implemented at the acceleration level197
Eq (15) generally leads to error accumulation in the198
computed motion during numerical integration due to199
numerical approximations and round-off errors resulting200
in constraint drift phenomenon To compensate for these201
errors a number of constraint stabilization techniques202
have been proposed22 and can be implemented if203
necessary as recently done by the authors for systems204
with intermittent contacts3 However for the problem205
at hand there was no necessity for implementing such206
correction terms Indeed by monitoring the constraints207
as the numerical simulation proceeds we observed a208
constraint drift under 20times10minus13 m which is negligible209
III GUITAR MODEL210
The guitar model involves six strings interacting with211
a guitar body and includes further models for the pressing212
finger on the fingerboard The strings are modeled213
using a simplified nonlinear formulation considering both214
polarizations and the body behavior is expressed in terms215
of modes identified from bridge input measurements on216
a real-life guitar217
A Nonlinear string model218
The vibratory motion of the strings is modeled219
using the KirchhoffndashCarrier approach2324 which is a220
simplified manner to describe large amplitude string221
motions Essentially the model discards the dynamics222
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 3
of the longitudinal modes but retains the effect of the223
geometrical nonlinearity by means of a pseudo-external224
forcing term added to the string transverse equations225
In our context it has two advantages First the model226
is known to capture relevant nonlinear phenomena for227
musical strings1425 thus enabling the reproduction of228
typical audible nonlinear effects Second it serves as a229
convenient means of illustrating that the U-K framework230
can deal effectively with nonlinear terms231
1 Kirchhoff-Carrier simplified nonlinear string model232
We consider a stiff string of length L cross-sectionalarea S mass density ρ Youngrsquos modulus E momentof inertia I and coefficient of dissipation η subjectto an axial tension T0 vibrating in two perpendiculartransverse motions Y (x t) and Z(x t) According to theKirchhoff-Carrier assumptions a simplified model of thefree vibration of the string is given by
ρSpart2Y
partt2+ η
partY
partt+ EI
part4Y
partx4=[T0 + Tdyn(t)
]part2Ypartx2
(16a)
ρSpart2Z
partt2+ η
partZ
partt+ EI
part4Z
partx4=[T0 + Tdyn(t)
]part2Zpartx2
(16b)
which is the standard wave equations for vibrating string233
including both polarizations with an additional forcing234
term proportional to the net dynamical increase in string235
length expressed as236
Tdyn(t) =ES
2L
int L
0
[(partY (x t)
partx
)2
+
(partZ(x t)
partx
)2]dx (17)
Eq (17) shows that the Kirchhoff-Carrier model only237
requires the knowledge of the transverse motions and238
that is independent of the x coordinate As far as the239
numerical technique is concerned the form of Eq (17)240
is very convenient particularly for modal synthesis and241
contrasts with the delicate numerical resolution of the242
ldquogeometrically exactrdquo model26 by modal discretization243
due the spatio-temporal nature of the nonlinear forcing244
2 Kirchhoff-Carrier nonlinear modal terms245
Denoting X a generic direction of string motion (X246
stands for Y or Z) the modal forces associated to the247
string nonlinear behavior are obtained by integration of248
the additional force terms of Eq (16) as249
fXn (t) =
int L
0
[Tdyn(t)
part2X
partx2
]φXn (x)dx (18)
Considering a string pinned at the nut and free at the250
bridge the modeshapes φXn (x) are given by251
φXn (x) = sin
[(2nminus 1)πx
2L
](19)
Substituting Eq (19) in Eq (18) and using a modalexpansion for the string motion the nonlinear modal
terms are finally given by
fXn (t)=minusESπ4
128L3(2nminus 1)2qXn (t)(
NYsumm=1
(2mminus1)2[qYm(t)
]2+
NZsumm=1
(2mminus1)2[qZm(t)
]2)(20)
with NY and NZ the sizes of the two string modal basis252
B The instrument body253
In view of Eq (7) the dynamic behaviour of the254
body is described by its modal properties extracted255
from input admittance measurements In contrast256
to modal computations based on a full model of the257
instrument body this is a direct approach that avoids258
complications of modeling in particular the frequency-259
dependent dissipative effects which significantly affect260
the computed sounds but are very difficult to model261
properly and that readily lend the synthesis method262
to be applied to any stringed instruments Since both263
polarizations of string motions are accounted a correct264
modeling of the stringbody coupling would demand265
the knowledge of both the in-plane and out-of-plane266
body modeshapes at the bridge However a simpler267
model is taken here by considering that only string268
motions normal to the soundboard couple to the body269
motions and also by neglecting the in-plane motions of270
the soundboard This means that the string motions in271
the plane parallel to the soundboard are unable to radiate272
sound and that the sound radiation is produced only273
by the normal motions of the soundboard Of course274
none of these choices is entirely satisfactory for accurate275
synthesis because in-plane and out-of-plane motions are276
coupled and also because in-plane motions couple the277
top and back plates via the ribs However since278
radiation is dominated by the out-of-plane vibration279
there is no doubt that in-plane motions are of second-280
order significance In practice the main advantage of281
our assumptions is to make the body characterization282
considerably simpler involving only transfer function283
measurements in the direction normal to the soundboard284
and avoiding the delicate measurements of the cross285
terms of the transfer function matrix286
The test guitar was a high-quality concert guitar287
built by Friederich number 694 During the tests it288
was positioned in the vertical position clamped to a289
rigid support by the neck and softly restrained in the290
lower bout with all the tuned strings damped (see291
Fig 1) The excitation and vibratory response were292
measured using a miniature force sensor (Kistler type293
9211) and a light-weight accelerometer (BampK 4375)294
respectively attached to the tie block using a thin layer295
of bee-wax Modal data were collected at two bridge296
locations close to the attachment points of the lowest297
and highest strings and a simple linear interpolation298
between the two transfer functions was performed to299
provide the unmeasured transfer functions at the other300
points where the strings make contact with the bridge301
4 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
Figure 1 Set-up for transfer function measurements at the
guitar body bridge
Modal identification was then achieved using the set of302
impulse responses expressed in terms of velocity and303
implementing the Eigensystem Realization Algorithm27304
The modal parameters for the first-ten identified modes305
are listed in Table I Fig 2 shows an example of306
an impulse response and its transfer function measured307
at the bridge end of the lowest string together with308
the corresponding reconstructed functions showing that309
the estimation was generally reliable up to 700 Hz310
Fig 3 also shows the identified mode shapes of the311
body at the bridge which exhibit typical motions of312
guitar modes28 The modes show a small degree of313
complexity but the effect of the imaginary components314
on the transfer function is small and can be neglected315
for our purpose Note that modes have been normalized316
according to max(|real(φBn (rs))|) = 1 (s = 1 6)317
Table I Identified modal parameters of the guitar body
Mode n fn (Hz) ζn () mn (Kg)
1 894 15 0402 1826 182 0083 2299 064 0204 3102 089 0475 4035 055 1066 5114 075 0587 5442 053 1828 6139 061 0699 6698 069 03810 8484 059 081
C Constraining forces and modal constraints318
1 Stringbody constraints319
Assuming a rigid transmission at the points wherethe strings make contact with the bridge the modalequations of the constrained system require to satisfya set of constrained modal equations stemming fromthe condition that at the bridge the strings and thebridge undergo the same motion Because the bodyresponse is here described only in the direction normal
0 005 01 015
Time [s]
-20
0
20
h (
t )
[ (
ms
)N
]
100 200 300 400 500 600 700 800 900
Frequency [ Hz ]
10-2
100
| H
( f
) |
[ (
ms
)N
]
Figure 2 Measured (green) and reconstructed (red) impulse
responses and transfer functions close to the lowest string
Gray dotted lines stand for the identified modal frequencies
E A D G B E
-1
0
1
1(
xB
)
f1 = 8939 Hz
E A D G B E
-1
0
1
2(
xB
)
f2 = 18262 Hz
E A D G B E
-1
0
1
3(
xB
)
f3 = 22986 Hz
E A D G B E
-1
0
1
4(
xB
)
f4 = 31021 Hz
E A D G B E
-1
0
1
5(
xB
)
f5 = 40352 Hz
E A D G B E
-1
0
1
6(
xB
)
f6 = 51143 Hz
E A D G B E
-1
0
1
7(
xB
)
f7 = 54417 Hz
E A D G B E
-1
0
1
8(
xB
)
f8 = 61386 Hz
E A D G B E
-1
0
1
9(
xB
)
f9 = 66978 Hz
Figure 3 Identified modeshapes of the guitar body at the
stringbridge contact points Real (blue) and imaginary
(magenta) components
to the soundboard the string motion Ys(xb t) in thisplane must be the same as the body motion at thestring contact point Yb(rs t) while its motion Zs(xb t)in the direction parallel to the soundboard must be nilFormally for each string s (s = 1 S) this results inthe conditions
Ys(xb t)minus Yb(rs t) = 0 (21a)
Zs(xb t) = 0 (21b)
or in terms of modal amplitudesΦYs
b
TqqqYs(t)minus
ΦBrsTqqqB(t) = 0 (22a)
ΦZs
b
TqqqZs(t) = 0 (22b)
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 5
where the modeshape vectors are written as
ΦYs
b =φYs1 (xb) φ
Ys
NYs(xb)
T(23a)
ΦZs
b =φZs1 (xb) φ
Zs
NZs(xb)
T(23b)
ΦBrs =φB1 (rs) φ
BNB
(rs)T
(23c)
Defining a vector of modal coordinates qqq built by stacking320
the individual modal displacements of the subsystems ie321
the strings including both polarizations and the body322
qqq = qqqY qqqZ qqqBT (24)
where qqqY and qqqZ are partitioned into terms associatedwith the different strings as
qqqY =qqqY1T qqqY6T
T(25a)
qqqZ =qqqZ1T qqqZ6T
T(25b)
and323
qqqB = qB1 qBNBT (26)
the constraint equations fit the matrix form324
AAAb qqq = 000 (27)
with the stringbody constraint matrix AAAb written as325
AAAb =
[ΦΦΦYb 000 minusΦΦΦB
000 ΦΦΦZb 000
](28)
where ΦΦΦYb and ΦΦΦZb are matrices of the modeshape vectors326
of the strings taken at the bridge location xb built as327
ΦΦΦYb =
ΦY1
b
T000 000
000
ΦY2
b
T 000
000 000
ΦY6
b
T
(29)
and similarly for the perpendicular direction and ΦΦΦB328
stands for the modeshapes of the body at the stringbody329
contact points rs330
ΦΦΦB =
ΦBr1T
ΦBr6T (30)
2 Stringfinger coupling at the fingerboard331
Modeling a stopping string for playing different332
notes can be thought in several ways according to the333
type of sound to simulate as well as to the degree of334
simplicitycomplexity aimed for the model The simplest335
model was taken in Ref 2 imposing a rigid kinematical336
constraint at a single location If such model could affect337
the tuning of the string effectively it is nonphysical338
in that part of the vibrational energy flows beyond the339
contact point (see Ref 2 for discussion) This simple340
approach is here extended to allow for multiple-point341
rigid coupling or for a non-rigid interaction as shown in342
Fig 4343
a Multiple-point rigid coupling This first optiondirectly enters into the realm of the standard kinematicalconstraints common in multibody theory The stringmotions Ys and Zs at the multiple finger locations xFf
(f = 1 F ) should be nil at all times so that
Ys(xFf t)=0 rArr
ΦYs
Ff
TqqqYs(t) = 0 (31a)
Zs(xFf t)=0 rArr
ΦZs
Ff
TqqqZs(t) = 0 (31b)
where the modeshape vectors at the finger locations are
ΦYs
Ff=φYs1 (xFf
) φYs
NYs(xFf
)T
(32a)
ΦZs
Ff=φZs1 (xFf
) φZs
NYs(xFf
)T
(32b)
Eq (31) can be rearranged in a matrix form as344
AAAf qqq = 000 (33)
where the modal constrained matrix is345
AAAf =
[ΦΦΦYs
F 000
000 ΦΦΦZs
F
](34)
with346
ΦΦΦYs
F =
ΦYs
F1
T
ΦYs
FF
T ΦΦΦZs
F =
ΦZs
F1
T
ΦZs
FF
T (35)
b Non-rigid coupling To extend the model to347
allow for non-rigid contact requires a little more care348
since a flexible-dissipative-inertial model for coupling349
subsystems cannot be directly formulated as kinematic350
constraints Nevertheless they can be implemented351
by assigning one or more additional flexible-dissipative-352
inertial elementary subsystems which are constrained to353
follow the string motions at the finger location(s) exactly354
as the body modes are included This modeling strategy355
somehow creates a bridge between computationally356
efficient multibody modeling providing ldquorigidrdquo kinematic357
constraints and the versatile but computationally slower358
modeling strategy through penalty-based constraints359
Non-rigid connections on the fingerboard cantherefore be obtained by extending the guitar modelusing a set of mass-spring-damper oscillators at locationsxFf
(f = 1 F ) and enforcing coupling as
Ys(xFf t)minusY Ff
f (t)= 0 (36a)
Zs(xFf t)minusZFf
f (t)= 0 (36b)
where YFf
f and ZFf
f are the responses of the fingercontact points in the two perpendicular directionsReturning to Eq (11) the vector and matrices mustnow be partitioned as
qqqu =qqqYsu T qqqZs
u T qqqFsu T
T(37a)
MMM = diag(MMMYsMMMZsMMMFs
)(37b)
CCC = diag(CCCYsCCCZsCCCFs
)(37c)
KKK = diag(KKKYsKKKZsKKKFs
)(37d)
6 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
where the vectors and matrices for the strings are as360
beforeMMMFs CCCFs andKKKFs are block diagonal matrices of361
the inertia damping and stiffness properties associated362
with the ldquofingerrdquo subsystem acting on string s including363
both directions of string motions364
qqqFsu =
Y F1
f Y FF
f ZF1
f ZFF
f
T(38)
In terms of modal coordinates Eqs (36) becomeΦYs
Ff
TqqqYs(t)minus Y Ff
f (t) = 0 (39a)ΦZs
Ff
TqqqZs(t)minus ZFf
f (t) = 0 (39b)
which can be rearranged as365
AAAf qqq = 000 (40)
with the modal constrained matrix AAAf written as366
AAAf =
[ΦΦΦYF 000 minusI 000
000 ΦΦΦZF 000 minusI
](41)
Figure 4 Rigid kinematical constraints (left) and flexible-
dissipative-inertial constraints (right) at multiple locations
D Dynamical formulation of the fully coupled model367
The formulation (11) readily applies to the full guitar368
model including the non-rigid constraint model for the369
stringfret interaction by stacking the unconstrained370
modal displacements into a vector371
qqqu = qqqYu qqqZu qqqBu qqqFu T (42)
where qqqYu qqqZu and qqqBu are as before and372
qqqFu=qqqY1
ufT qqqY6
ufT qqqZ1
ufT qqqZ6
ufTT
(43)
In view of the definition of qqq the assembled inertia MMM373
stiffnessKKK and dampingCCC matrices which are uncoupled374
are given by375
MMM=
MMMY 000 000 000
000 MMMZ 000 000
000 000 MMMB 000
000 000 000 MMMF
CCC=
CCCY 000 000 000
000 CCCZ 000 000
000 000 CCCB 000
000 000 000 CCCF
KKK=
KKKY 000 000 000
000 KKKZ 000 000
000 000 KKKB 000
000 000 000 KKKF
(44)
with MMM ` KKK` and CCC` the inertia stiffness and dampingmatrices for each individual subsystem ` and the vectorsFFFext and FFFnl write
FFFext =FFFYextT FFFZextT000000
T(45a)
FFFnl =FFFYnlT FFFZnlT000000
T(45b)
where FFFYext and FFFZext correspond to the plucking action376
of the player expressed for both planes of polarization377
and FFFYnl and FFFZnl are the modal vectors associated to the378
string nonlinear effects with entries given by Eq (20)379
The form of the modal constraint equation is now380
AAAqqq = 000 (46)
with381
AAA =
[AAAbAAAf
]=
ΦΦΦYb 000 minusΦΦΦB 000 000 000
000 ΦΦΦZb 000 000 000 000
ΦYFΦYFΦYF 000 000 000 minusFFFY 000
000 ΦZFΦZFΦZF 000 000 000 minusFFFZ
(47)
where ΦYFΦYFΦYF and ΦZFΦZFΦZF are block matrices with entries given382
by Eq (35) and FFFY and FFFZ are block diagonal matrices383
built with either the identity or null matrix depending on384
the constraints enforced on each string The form of (47)385
shows that the interface forces and constraint conditions386
couple the Y -motions of the strings and the body as well387
as coupling between the finger and the strings in each388
direction of motions Notice that coupling between both389
polarizations of string motions is not apparent in (47)390
although it is formulated in the nonlinear forces (45b)391
IV ILLUSTRATIVE COMPUTATIONS392
A System parameters393
For given initial conditions the system of equations394
comprising (5) with the unconstrained accelerations (11)395
and the constraint matrix (15) and using the specific396
matrices given by Eqs (44) and (47) can be integrated in397
time to produce the transient response The time-domain398
simulations were performed based on the identified body399
modal data and using standard string properties of400
classical guitars Assuming non-ideal strings with fixed-401
free ends the modeshapes are given by Eq (19) and402
their modal frequencies can be approximated by403
fYn = fZn =cT2πpn
(1 +
B
2T0p2n
)(48)
where cT is the velocity of transverse waves B is the404
bending stiffness of the string and T0 its tensioning405
pn = (2n minus 1)π2L and n the mode number Modal406
damping values are introduced following the pragmatic407
formulation given in Ref 6408
ζYn = ζZn =1
2
T0(ηF + ηAωn
)+ηBBp
2n
T0 +Bp2n(49)
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 7
where ηF ηA and ηB are the loss coefficients related to409
ldquointernal frictionrdquo ldquoair viscous dampingrdquo and ldquobending410
dampingrdquo The length of the strings is L=065 m411
and their density bending stiffness tensioning and loss412
coefficients are taken from Ref 29 The size of the modal413
basis of each string is defined by the maximum frequency414
of the computations (10000 Hz) which proved a sensible415
compromise for convergence to realistic simulations416
resulting in 800 modes for the set of strings To417
obtain suitable values for the elasticity parameter ES418
we perform simple vibratory tests on string samples419
Stretching a string vertically and suspending a mass to420
its free end one can easily measure the longitudinal421
vibration of the string-mass oscillator and obtain its422
natural frequency Assuming the string acting as a423
spring with no mass and reminding that the natural424
frequency of the longitudinal vibration is given by f =425
12π
radicESML then an estimate of ES is obtained The426
simulation results were obtained by time-step integration427
of the ordinary modal equations using the velocity-428
Verlet numerical scheme30 an integrator that combines429
simplicity second order accuracy with good numerical430
stability - see implementation details in Ref 2 We431
adopted a time-step of 10minus5s and assumed all initial432
modal displacements and velocities nil initially433
B Computational results434
Pluck excitations were simulated by considering a435
point-wise external force at xe=09L with a linear force436
ramp applied to one or more strings during 10 ms and437
by assuming null excitation force at subsequent instants438
1 Dynamical behaviour of the fully coupled model439
Results in Fig 5 were obtained for the open string440
E2 excited in the direction normal to the soundboard441
first considering a rigidly mounted string then coupling442
one string and finally all six strings to the instrument443
body Assuming no dynamics for the body the response444
spectra of string motion at the bridge is virtually nil445
according to the pinned boundary condition When446
coupling the instrument body to the string the response447
becomes obviously different and shows the classical series448
of near harmonic string partials together with smaller449
peaks from the body resonances While the first modal450
frequency of the unconstrained string is 41 Hz a close451
look at Fig 5 shows that the constraint at the bridge452
yields the fundamental frequency for the tuned string453
with pinned ends (f = 82 Hz) thus confirming the454
correct implementation of the bridge coupling Coupling455
the other five strings results in new peaks in the456
spectrum extending over the frequency range of the body457
modes and with an amplitude difference of about 100 dB458
compared to the most excited modes which clearly relate459
to the non-excited strings As found in some stringed460
instruments this dynamical coupling of all the strings via461
the bridge can lead to sympathetic vibrations which can462
further results in slight interference beats in the sound463
due to the inharmonicity produced by the bridge coupling464
and of the string partials465
0 200 400 600 800 1000 1200 1400 1600 180010
-40
10-20
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 180010
-20
10-10
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 1800
Frequency [ Hz ]
10-20
10-10
100
| Y
6(f
) |
Figure 5 Response spectra of string motion at the bridge
Top rigidly mounted string middle single stringbody
coupling bottom six-stringbody coupled model Force
amplitude 5 N Linear model
0 02 04 06 08 1
Time [ s ]
0
05
1
15
2
Mo
dal en
erg
y [
J ]
10-3
STRINGSFRETBODYTOTAL
Figure 6 Instantaneous modal energies Oblique excitation
with amplitudes of 2 N and 1 N in the Y and Z directions
Linear six-stringbody coupled model Conservative model
Figs 6 and 7 then serve at checking other466
implementation details Simulations pertained to the467
string E2 constrained at xF = 02509L by a single ldquostiffrdquo468
fret modeled with mF=5 10minus3 Kg cF=1 Nsm and469
kF=106 Nm plucked in the oblique direction The470
modal energies E` of the subsystems ` shown in these471
plots were computed from the results of the simulations472
as follows in terms of the modal parameters and the473
8 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
0 02 04 06 08 10
05
1
15
[ J ]
10-3 STRINGS
Y-dir
Z-dir
0 02 04 06 08 10
05
1
15
[ J ]
10-5 FRET
Y-dir
Z-dir
0 02 04 06 08 10
05
1
[ J ]
10-4 BODY
0 02 04 06 08 1
Time [ s ]
0
05
1
15
10-3 FULL SYSTEM
Figure 7 Instantaneous modal energies Same plucking
conditions as in Fig 6 Linear six-stringbody coupled model
Non-conservative model
modal displacements and velocities474
E`(t) =
Nsumn=1
[1
2m`n(ω`n)2
[q`n(t)
]2+
1
2m`n
[q`n(t)
]2](50)
One important feature seen in Fig 6 is that the total475
energy remains roughly constant over a conservative476
simulation while energies flow back and forth between477
the subsystems thus giving confidence in the numerical478
model for simulations of reasonable duration For479
the non-conservative case shown in Fig 7 the large480
difference in energy decay for the two polarizations481
confirms the different stringbody coupling with the482
directions of string motion Note that both the body483
and the fret oscillator are dominated by periodic energy484
exchanges at the frequency of the played note as the wave485
travels back and forth between bridge and fret Finally486
looking closely at the modal energy of the fret one can487
see smaller superimposed periodic disturbances related488
to the reflections between fret and nut illustrating that489
energy still flows through the modeled fret490
Finally we compare in Fig 8 results by varying491
the excitation amplitude in order to illustrate the string492
nonlinearity The simulations were obtained for the493
string E2 using the experimentally determined value494
of 32 kN for the flexibility ES which is comparable495
to the value given by Chaigne31 Fig 8 shows the496
temporal variations of string tension with an estimate497
of the playing frequency computed from time domain498
evaluations of the zero crossing frequency The general499
features of large amplitude string motions are clearly500
visible with a quasi-static increase of tension and501
superimposed oscillations at twice the string fundamental502
frequency and a small frequency variation of the played503
note which compares well with measurements32504
1 2 3 4 5 6 7
716
718
72
722
724
T (
t )
[ N
] Fexc
= 1 N
Fexc
= 5 N
1 2 3 4 5 6 7
Time [ s ]
822
824
826
828
f (
t )
[ H
z ]
Figure 8 String tension (top) and instantaneous frequency of
string motion (bottom) Nonlinear rigid body model
2 Parameter study of the stringfinger coupling model505
We now illustrate the different strategies to account506
for a stopping finger on the fingerboard and explore the507
influence on the coupled system of the finger model508
In contrast with the approach in Ref 2 the strategy509
in terms of auxiliary oscillators rigidly coupled to the510
string seems more realistic and is undoubtedly more511
versatile allowing different types of contacts to be512
modeled by varying the oscillator parameters However513
one could expect the finger dynamics to perturb the514
coupled system with changes in frequency and damping515
for its modes similar to what observed for a multi-modal516
system coupled to a system exhibiting resonance To517
give some insights Figs 9-11 show the effects of varying518
the finger mechanical parameters Simulations pertain to519
the string E2 constrained at xF = 01857L by a finger of520
mass mf = 5 10minus3 Kg and ignore the body dynamics521
The first test consists in varying the stiffness of the522
constraint assuming no damping Fig 9 shows the523
frequency spectra of the coupling force at the rigid bridge524
where the peaks of the string modes can be seen clearly as525
well as the resonance frequency of the auxiliary oscillator526
which moves in frequency with the stiffness and slightly527
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 9
modulates the other peaks A close look shows that528
the spectra are affected much more sensitively below529
the resonance of the finger oscillator with amplitude530
variations analogous to a system driven in the stiffness-531
controlled region Decreasing the stiffness causes a rise532
in amplitude of the undesirable modes while increasing533
the stiffness lower their amplitudes As observed for the534
stringbody coupling33 inharmonicity is also created by535
the stiffness termination which in turn affects the tuning536
of the note With the increase of stiffness perturbations537
of the string modes are reduced and tuning is improved538
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
4 Nm
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
5 Nm
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
kf = 10
6 Nm
Figure 9 Spectra of the coupling force at the rigid bridge
Stiffness is varied F=1 c1 = 0 Nsm
Fig 10 then shows the effect of varying the damping539
keeping the stiffness constant Not surprisingly if540
damping increases the amplitude of the coupled541
modes close to the oscillator resonance decrease and542
the frequency region affected becomes wider This543
significantly affects the time envelope of the pluck note544
that becomes smoother and has a shorter-time decay545
Finally the effect of the number of contacts is studied546
using the rigid contact model which is convenient for547
this purpose Fig 11 gives a picture of the spatial548
distribution of the string energy by plotting the standard549
deviation of string motion as a function of its length550
As seen the ldquopassiverdquo region is affected with great551
sensitivity by the number of contacts and surprisingly552
increasing the contact points does not necessarily reduce553
the undesirable transfer of energy Computations show554
that the effectiveness of the multi-point rigid constraints555
approach is strongly problem-dependent and varies in a556
non-obvious way with several factors including the finger557
width the spacing between contact points the relative558
string length from the ends and the excitation spectrum559
Taken in combination these results and noticing560
that the perturbations induced by each finger oscillator561
are additive due to the linear coupling a robust and562
physically sound strategy to simulate the pressing finger563
would be to implement a set of flexible-dissipative564
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
cf = 1 Nsm
cf = 10 Nsm
0 1 2 3 4 5
Time [ s ]
-2
0
2
4
Y (
t )
10-3
Figure 10 Spectra of the coupling force at the rigid
bridge (up) and time history of the string motion (bottom)
Damping is varied F=1 k1 = 105 Nm
constraints imposing a first stiff constraint in order565
to ensure a precise tuning followed by several soft566
constraints that would control the waveform and spectral567
content of the decaying response depending on their568
number and damping properties For illustration569
the spectrograms of two plucks for different sets of570
parameters are presented in Fig 12571
0 01 02 03 04 05 06
x [ m ]
10-5
10-2
ST
D (
Y(t
) )
F=1
F=2
F=3
F=5
016 017 018 019 02
x [ m ]
10-5
ST
D (
Y(t
) )
Figure 11 Standard deviation of the string motion Number
of contacts is varied Vertical dotted lines represent the
constraints locations Finger width (0015m) is kept constant
C Simulated acoustic radiation572
As an alternative to computing sounds from full573
vibro-acoustic models that usually require the use of574
massive computational resources it is tempting to apply575
10 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
Figure 12 Spectrograms of the force at the rigid bridge Up
F = 3 bottom F = 5 Parameter values kF1=106 Nm
cF1=1 Nsm and kFj =104 Nm cFj =10 Nsm (j = 2 F )
Constraint at xF = 01857L (sound files available)
simpler techniques that could be efficient to generate576
convincing sounds from the vibratory responses The577
method used here is to treat the guitar body as a linear578
input-output system and to compute the pressure by579
convolution benefiting from the local excitation of the580
body by the strings and using a vibro-acoustic impulse581
response of the test guitar This approach does not by582
any means simulate the sound response everywhere in the583
pressure field but it has the merit of adding the influence584
of the radiating properties of the body at a very low585
computational cost and of providing sounds of greater586
perceptual quality than the commonly used vibratory587
signals at the bridge The pressure at location rp can be588
computed by the convolution of the time-history of the589
stringbridge interaction force FY (rb t) and the vibro-590
acoustic impulse response function hY (rb rp t) between591
the bridge and microphone by592
p(rp t) = hY (rb rp t) lowast FY (rb t) (51)
The extension to account for the effects of all the593
strings is readily obtained by linear superposition594
principle summing the contributions from each string595
Measurements of the vibro-acoustic impulse response596
were performed under the same experimental conditions597
as for transfer function measurements using an598
additional condenser microphone Impulsive excitation599
was applied in the perpendicular direction to the600
soundboard and the acoustic pressure was measured601
at a distance of 1 m from the central axis of the602
guitar in a quiet room with all strings damped (see603
Fig 1) The impulse response was computed by inverse604
Fourier transform of the complex transfer function built605
from the force and response signals described in the606
frequency domain Results in Fig 13 pertain to the607
simulation of a G major arpeggio obtained with the608
two lowest and the top strings constrained using five609
auxiliary flexible-inertial constraints on each string The610
effect of the radiation properties of the body on the611
computed vibratory signals is supported when comparing612
the spectrograms in Fig 13 which clearly shows the613
spectral filtering of the vibration signal614
Figure 13 Spectrograms of the bridge force (top) and sound
pressure (bottom) Nonlinear six-stringbody coupled model
Arpeggio of a G major chord with three strings stopped using
a set of auxiliary oscillators (sound files available)
V CONCLUSIONS615
Following our recent work showing the potential616
of the modal Udwadia-Kalaba equations to physically617
simulate musical instruments we presented in this paper618
a computational approach that allows efficient time-619
domain simulations of realistic guitar plucks In this620
work we further generalized our simple stringbody621
coupled model toward the development of a more622
realistic guitar model by including the complete set623
of strings the two polarizations of string motions the624
string geometric nonlinearity the stringfret interaction625
that controls the playing frequency and the body626
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 11
dynamics of an actual instrument The dynamics of627
the various subsystems were formulated in terms of628
their unconstrained modes and coupling were achieved629
by enforcing several constraint conditions One of the630
developments of the model was the introduction of631
flexible-dissipative-inertial supplementary equations as632
an alternative to the common kinematical constraints633
which here improve the role of the stringfingerfret634
interaction at the fingerboard This approach can635
also benefit the modeling of other complex dynamical636
systems outside the realm of physical instrument637
modeling To improve the realism of the synthesis638
we also included crudely the radiation properties of639
the instrument body into the computational model640
through an ad hoc procedure Sound pressure was641
computed adopting a hybrid approach by convoluting642
the computed bridge force signals with a vibro-acoustic643
impulse response measured on the test guitar Such644
vibro-acoustic response can be measured in an anechoic645
chamber or else include the acoustics of any specific646
room This approach appeared not only straightforward647
to compute realistic sounds of string instruments for648
which the radiating components are subjected to point-649
wise excitation but also very efficient compared to heavy650
computational sound radiation techniques651
ACKNOWLEDGMENTS652
The authors thanks Paulo Vaz de Carvalho for653
lending the guitar and interesting discussions This work654
is funded by national funds through the FCT ndash Fundacao655
para a Ciencia e a Tecnologia IP under the Norma656
Transitoria - DL 572016CP1453CT0007657
REFERENCES658
1JH Ginsberg Mechanical and Structural Vibrations Theory659
and Applications chapter 9 Wiley 2001660
2J Antunes and V Debut Dynamical computation of constrained661
flexible systems using a modal Udwadia-Kalaba formulation662
Application to musical instruments J Acoust Soc Am663
141764 2017664
3J Antunes V Debut L Borsoi X Delaune and P Piteau665
A modal Udwadia-Kalaba formulation for the vibro-impact666
modelling of continuous flexible systems with intermittent667
contacts Procedia Engineering 199322ndash329 2017668
4FE Udwadia and RE Kalaba Analytical dynamics - A new669
approach Cambridge University Press 1996670
5C Valette The mechanics of the vibrating string Mechanics of671
musical instruments Springer 1995672
6J Woodhouse On the synthesis of guitar plucks Acta Acust673
united Ac 90928ndash944 2004674
7M E McIntyre RT Schumacher and J Woodhouse On the675
oscillations of musical instruments Journal of the Acoustical676
Society of America 741325ndash1345 1983677
8A Chaigne On the use of finite differences for musical678
synthesis Application to plucked stringed instruments Journal679
drsquoAcoustique 5181ndash211 1992680
9JM Adrien The missing link Modal synthesis In G DePoli681
A Picalli and C Roads editors Representations of musical682
sounds MIT Press Cambridge 1991683
10S Bilbao Numerical Sound Synthesis Finite Difference684
Schemes and Simulation in Musical Acoustics Wiley Publishing685
2009686
11L Trautmann and R Rabenstein Multirate simulations687
of string vibrations includingnonlinear fret-string interactions688
usingthe functional transformation method EURASIP Journal689
on Applied Signal Processing 7949ndash963 2004690
12V Chatziioannou and M van Walstijn Energy conserving691
schemes for the simulation of musical instrument contact692
dynamics J Sound Vib 339262ndash279 2015693
13S Bilbao and A Torin Numerical modeling and sound synthesis694
for articulated stringfretboard interactions Acta Acust united695
Ac 63336ndash347 2015696
14V Debut M Carvalho M Marques and J Antunes Physics-697
based modeling techniques of a twelve-string portuguese guitar698
a three-dimensional non-linear time-domain computational699
approach for the multiple-stringsbridgesoundboard coupled700
dynamics Appl Acoust 1083ndash18 2016701
15M van Walstijn and J Bridges Simulation of distributed contact702
in string instruments A modal expansion approach In 24th703
European Signal Processing Conference (Eusipco 2016) pages704
1023ndash1027 2013705
16C Issanchou S Bilbao JL Le Carrou C Touze and O Doare706
A modal-based approach to the nonlinear vibration of strings707
against a unilateral obstacle simulations and experiments in the708
pointwise case J Sound Vib 393229ndash251 2017709
17R Bader Computational Mechanics of the Classical Guitar710
Springer-Verlag Berlin Heidelberg 2005711
18H Mansour J Woodhouse and G Scavone Enhanced wave-712
based modelling of musical strings Part 1 Plucked strings Acta713
Acust united Ac 1021094ndash1107 2016714
19O Inacio J Antunes and MCM Wright Computational715
modeling of string-body interaction for the violin family and716
simulation of wolf notes J Sound Vib 310260ndash286 2007717
20S Bilbao A Torin and V Chatziioannou Numerical modeling718
of collisions in musical instruments Acta Acust united Ac719
101155ndash173 2015720
21C Issanchou JL Le Carrou C Touze B Fabre and O Doare721
Stringfrets contacts in the electric bass sound Simulations and722
experiments App Acoust 129217ndash228 2018723
22D J Braun and M Goldfarb Eliminating constraint drift in the724
numerical simulation of constrained dynamical systems Comput725
Methods in Appl Mech Eng 198(37)3151 ndash 3160 2009726
23GF Carrier On the nonlinear vibration problem of the elastic727
string Q Appl Math 3157ndash165 1945728
24G Kirchhoff Vorlesungen uber Mechanik Teubner 1883729
25J Pakarinen V Valimaki and M Karjalainen Physics-based730
methods for modeling nonlinear vibrating strings Acta Acust731
united Ac 91312ndash325 2005732
26PMC Morse and KU Ingard Theoretical Acoustics733
International series in pure and applied physics Princeton734
University Press 1968735
27J Juang Applied System Identification PTR Prentice-Hall736
Inc New Jersey 1994737
28B E Richardson The acoustical development of the guitar J738
Catgut Acoust Soc 51ndash10 1994739
29J Woodhouse Plucked guitar transients Comparison of740
measurements and synthesis Acta Acust united Ac 90945ndash741
965 2004742
30M Tuckerman B J Berne and G J Martyna Reversible743
multiple time scale molecular dynamics J Chem Phys744
97(3)1990ndash2001 1992745
31A Chaigne Viscoelastic properties of nylon guitar strings J746
Catgut Acoust Soc Series II 121ndash27 1991747
32T Tolonen V Valimaki and M Karjalainen Modeling of748
tension modulation nonlinearity in plucked strings IEEE Trans749
Speech Audio Process 8(3)300ndash310 2000750
33A Chaigne and J Kergomard Acoustics of Musical Instruments751
Modern Acoustics and Signal Processing Springer New York752
2016753
12 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
754
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 13
or nonlinear conservative or dissipative systems and150
may also be efficiently solved using a suitable time-step151
integration scheme for a given excitation152
B Modal U-K formulation153
Extension to constrained continuous multibody154
systems defined through their unconstrained modes Φ155
was formulated in Ref 2 by transforming (3) to the156
modal space through x = Φqqq leading to the formulation157
158
qqq = qqqu +MMMminus12BBB+(bminusAAAqqqu) (5)
where qqqu are the modal accelerations of the159
unconstrained configuration MMM = ΦTMΦ is the160
modal mass matrix AAA = AΦc is the modal constraint161
matrix with Φc the matrix of all modeshapes where162
constraints are defined and BBB = AAAMMMminus12 Finally the163
modal constraining force FFF c can also be computed as164
FFF c = MMM12BBB+(bminusAAAqqqu) (6)
C Dynamically coupled subsystems165
The application of the U-K modal formulation to a166
set of subsystems coupled through a number of kinematic167
constraints is now considered In a modal description the168
dynamics of the subsystem ` (` = 1 2 L) subjected169
to an external force field can be classically written as a170
set of modal equations in terms of the vector of modal171
amplitudes qqq` and its derivatives172
MMM `qqq` +CCC`qqq` +KKK`qqq` +FFF `nl(qqq` qqq`) = FFF `ext +FFF `nl (7)
where MMM ` CCC` and KKK` are diagonal matrices of the173
modal parameters m`n c`n and k`n (n = 1 N `) defined174
according to the modeshapes φ`n and FFF `ext and FFF `nl are175
the modal forces obtained by projection of the external176
and nonlinear forces on the modal basis The modal177
accelerations of the unconstrained system to be used in178
Eq (5) are given by179
qqq`u =(MMM `)minus1
FFF ` (8)
denoting FFF ` the vector of all the constraint-independent180
modal forces181
FFF `= FFF `extminusCCC`qqq`minusKKK`qqq`minusFFF `nl(qqq` qqq`) (9)
for which it is assumed that the vectors of modal182
constrained displacements and velocities are known at183
each time-step Stacking the modal quantities of the184
unconstrained subsystems in compact vectors as185
qqqu =[qqq1uT qqqLuT
]T(10)
the unconstrained modal accelerations of the coupled186
system read finally as187
qqqu = MMMminus1
[FFFext minusCCC qqq minusKKKqqq minus FFFnl(qqq qqq)
](11)
denoting188
qqq =[qqq1T qqqLT
]T(12)
In view of Eq (10) the matrices MMM CCC and KKK areblock diagonal set up by the submatrices of the modalparameters of the subsystems assembled as
MMM = diag(MMM1 MMML) (13a)
CCC = diag(CCC1 CCCL) (13b)
KKK = diag(KKK1 KKKL) (13c)
and FFFext and FFFnl are modal vectors associated to theexternal and nonlinear interaction forces acting on thevarious subsystems written respectively as
FFFext =FFF 1
extT FFFLextTT
(14a)
FFFnl =FFF 1
nlT FFFLnlTT
(14b)
The second set of equations to be considered concerns the189
P coupling constraints at locations r`c In most practical190
situations these are amenable to linear relationships by191
appropriate differentiation with respect to time leading192
to the standard form193
AAA(qqq qqq t) qqq = b(qqq qqq t) (15)
where AAA = AΦc with Φc = diag(Φ1c Φ
Lc ) and where194
Φ`c contains the modeshape vectors of each subsystem195
at the constraint location r`c Notice that because the196
constraints are implemented at the acceleration level197
Eq (15) generally leads to error accumulation in the198
computed motion during numerical integration due to199
numerical approximations and round-off errors resulting200
in constraint drift phenomenon To compensate for these201
errors a number of constraint stabilization techniques202
have been proposed22 and can be implemented if203
necessary as recently done by the authors for systems204
with intermittent contacts3 However for the problem205
at hand there was no necessity for implementing such206
correction terms Indeed by monitoring the constraints207
as the numerical simulation proceeds we observed a208
constraint drift under 20times10minus13 m which is negligible209
III GUITAR MODEL210
The guitar model involves six strings interacting with211
a guitar body and includes further models for the pressing212
finger on the fingerboard The strings are modeled213
using a simplified nonlinear formulation considering both214
polarizations and the body behavior is expressed in terms215
of modes identified from bridge input measurements on216
a real-life guitar217
A Nonlinear string model218
The vibratory motion of the strings is modeled219
using the KirchhoffndashCarrier approach2324 which is a220
simplified manner to describe large amplitude string221
motions Essentially the model discards the dynamics222
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 3
of the longitudinal modes but retains the effect of the223
geometrical nonlinearity by means of a pseudo-external224
forcing term added to the string transverse equations225
In our context it has two advantages First the model226
is known to capture relevant nonlinear phenomena for227
musical strings1425 thus enabling the reproduction of228
typical audible nonlinear effects Second it serves as a229
convenient means of illustrating that the U-K framework230
can deal effectively with nonlinear terms231
1 Kirchhoff-Carrier simplified nonlinear string model232
We consider a stiff string of length L cross-sectionalarea S mass density ρ Youngrsquos modulus E momentof inertia I and coefficient of dissipation η subjectto an axial tension T0 vibrating in two perpendiculartransverse motions Y (x t) and Z(x t) According to theKirchhoff-Carrier assumptions a simplified model of thefree vibration of the string is given by
ρSpart2Y
partt2+ η
partY
partt+ EI
part4Y
partx4=[T0 + Tdyn(t)
]part2Ypartx2
(16a)
ρSpart2Z
partt2+ η
partZ
partt+ EI
part4Z
partx4=[T0 + Tdyn(t)
]part2Zpartx2
(16b)
which is the standard wave equations for vibrating string233
including both polarizations with an additional forcing234
term proportional to the net dynamical increase in string235
length expressed as236
Tdyn(t) =ES
2L
int L
0
[(partY (x t)
partx
)2
+
(partZ(x t)
partx
)2]dx (17)
Eq (17) shows that the Kirchhoff-Carrier model only237
requires the knowledge of the transverse motions and238
that is independent of the x coordinate As far as the239
numerical technique is concerned the form of Eq (17)240
is very convenient particularly for modal synthesis and241
contrasts with the delicate numerical resolution of the242
ldquogeometrically exactrdquo model26 by modal discretization243
due the spatio-temporal nature of the nonlinear forcing244
2 Kirchhoff-Carrier nonlinear modal terms245
Denoting X a generic direction of string motion (X246
stands for Y or Z) the modal forces associated to the247
string nonlinear behavior are obtained by integration of248
the additional force terms of Eq (16) as249
fXn (t) =
int L
0
[Tdyn(t)
part2X
partx2
]φXn (x)dx (18)
Considering a string pinned at the nut and free at the250
bridge the modeshapes φXn (x) are given by251
φXn (x) = sin
[(2nminus 1)πx
2L
](19)
Substituting Eq (19) in Eq (18) and using a modalexpansion for the string motion the nonlinear modal
terms are finally given by
fXn (t)=minusESπ4
128L3(2nminus 1)2qXn (t)(
NYsumm=1
(2mminus1)2[qYm(t)
]2+
NZsumm=1
(2mminus1)2[qZm(t)
]2)(20)
with NY and NZ the sizes of the two string modal basis252
B The instrument body253
In view of Eq (7) the dynamic behaviour of the254
body is described by its modal properties extracted255
from input admittance measurements In contrast256
to modal computations based on a full model of the257
instrument body this is a direct approach that avoids258
complications of modeling in particular the frequency-259
dependent dissipative effects which significantly affect260
the computed sounds but are very difficult to model261
properly and that readily lend the synthesis method262
to be applied to any stringed instruments Since both263
polarizations of string motions are accounted a correct264
modeling of the stringbody coupling would demand265
the knowledge of both the in-plane and out-of-plane266
body modeshapes at the bridge However a simpler267
model is taken here by considering that only string268
motions normal to the soundboard couple to the body269
motions and also by neglecting the in-plane motions of270
the soundboard This means that the string motions in271
the plane parallel to the soundboard are unable to radiate272
sound and that the sound radiation is produced only273
by the normal motions of the soundboard Of course274
none of these choices is entirely satisfactory for accurate275
synthesis because in-plane and out-of-plane motions are276
coupled and also because in-plane motions couple the277
top and back plates via the ribs However since278
radiation is dominated by the out-of-plane vibration279
there is no doubt that in-plane motions are of second-280
order significance In practice the main advantage of281
our assumptions is to make the body characterization282
considerably simpler involving only transfer function283
measurements in the direction normal to the soundboard284
and avoiding the delicate measurements of the cross285
terms of the transfer function matrix286
The test guitar was a high-quality concert guitar287
built by Friederich number 694 During the tests it288
was positioned in the vertical position clamped to a289
rigid support by the neck and softly restrained in the290
lower bout with all the tuned strings damped (see291
Fig 1) The excitation and vibratory response were292
measured using a miniature force sensor (Kistler type293
9211) and a light-weight accelerometer (BampK 4375)294
respectively attached to the tie block using a thin layer295
of bee-wax Modal data were collected at two bridge296
locations close to the attachment points of the lowest297
and highest strings and a simple linear interpolation298
between the two transfer functions was performed to299
provide the unmeasured transfer functions at the other300
points where the strings make contact with the bridge301
4 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
Figure 1 Set-up for transfer function measurements at the
guitar body bridge
Modal identification was then achieved using the set of302
impulse responses expressed in terms of velocity and303
implementing the Eigensystem Realization Algorithm27304
The modal parameters for the first-ten identified modes305
are listed in Table I Fig 2 shows an example of306
an impulse response and its transfer function measured307
at the bridge end of the lowest string together with308
the corresponding reconstructed functions showing that309
the estimation was generally reliable up to 700 Hz310
Fig 3 also shows the identified mode shapes of the311
body at the bridge which exhibit typical motions of312
guitar modes28 The modes show a small degree of313
complexity but the effect of the imaginary components314
on the transfer function is small and can be neglected315
for our purpose Note that modes have been normalized316
according to max(|real(φBn (rs))|) = 1 (s = 1 6)317
Table I Identified modal parameters of the guitar body
Mode n fn (Hz) ζn () mn (Kg)
1 894 15 0402 1826 182 0083 2299 064 0204 3102 089 0475 4035 055 1066 5114 075 0587 5442 053 1828 6139 061 0699 6698 069 03810 8484 059 081
C Constraining forces and modal constraints318
1 Stringbody constraints319
Assuming a rigid transmission at the points wherethe strings make contact with the bridge the modalequations of the constrained system require to satisfya set of constrained modal equations stemming fromthe condition that at the bridge the strings and thebridge undergo the same motion Because the bodyresponse is here described only in the direction normal
0 005 01 015
Time [s]
-20
0
20
h (
t )
[ (
ms
)N
]
100 200 300 400 500 600 700 800 900
Frequency [ Hz ]
10-2
100
| H
( f
) |
[ (
ms
)N
]
Figure 2 Measured (green) and reconstructed (red) impulse
responses and transfer functions close to the lowest string
Gray dotted lines stand for the identified modal frequencies
E A D G B E
-1
0
1
1(
xB
)
f1 = 8939 Hz
E A D G B E
-1
0
1
2(
xB
)
f2 = 18262 Hz
E A D G B E
-1
0
1
3(
xB
)
f3 = 22986 Hz
E A D G B E
-1
0
1
4(
xB
)
f4 = 31021 Hz
E A D G B E
-1
0
1
5(
xB
)
f5 = 40352 Hz
E A D G B E
-1
0
1
6(
xB
)
f6 = 51143 Hz
E A D G B E
-1
0
1
7(
xB
)
f7 = 54417 Hz
E A D G B E
-1
0
1
8(
xB
)
f8 = 61386 Hz
E A D G B E
-1
0
1
9(
xB
)
f9 = 66978 Hz
Figure 3 Identified modeshapes of the guitar body at the
stringbridge contact points Real (blue) and imaginary
(magenta) components
to the soundboard the string motion Ys(xb t) in thisplane must be the same as the body motion at thestring contact point Yb(rs t) while its motion Zs(xb t)in the direction parallel to the soundboard must be nilFormally for each string s (s = 1 S) this results inthe conditions
Ys(xb t)minus Yb(rs t) = 0 (21a)
Zs(xb t) = 0 (21b)
or in terms of modal amplitudesΦYs
b
TqqqYs(t)minus
ΦBrsTqqqB(t) = 0 (22a)
ΦZs
b
TqqqZs(t) = 0 (22b)
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 5
where the modeshape vectors are written as
ΦYs
b =φYs1 (xb) φ
Ys
NYs(xb)
T(23a)
ΦZs
b =φZs1 (xb) φ
Zs
NZs(xb)
T(23b)
ΦBrs =φB1 (rs) φ
BNB
(rs)T
(23c)
Defining a vector of modal coordinates qqq built by stacking320
the individual modal displacements of the subsystems ie321
the strings including both polarizations and the body322
qqq = qqqY qqqZ qqqBT (24)
where qqqY and qqqZ are partitioned into terms associatedwith the different strings as
qqqY =qqqY1T qqqY6T
T(25a)
qqqZ =qqqZ1T qqqZ6T
T(25b)
and323
qqqB = qB1 qBNBT (26)
the constraint equations fit the matrix form324
AAAb qqq = 000 (27)
with the stringbody constraint matrix AAAb written as325
AAAb =
[ΦΦΦYb 000 minusΦΦΦB
000 ΦΦΦZb 000
](28)
where ΦΦΦYb and ΦΦΦZb are matrices of the modeshape vectors326
of the strings taken at the bridge location xb built as327
ΦΦΦYb =
ΦY1
b
T000 000
000
ΦY2
b
T 000
000 000
ΦY6
b
T
(29)
and similarly for the perpendicular direction and ΦΦΦB328
stands for the modeshapes of the body at the stringbody329
contact points rs330
ΦΦΦB =
ΦBr1T
ΦBr6T (30)
2 Stringfinger coupling at the fingerboard331
Modeling a stopping string for playing different332
notes can be thought in several ways according to the333
type of sound to simulate as well as to the degree of334
simplicitycomplexity aimed for the model The simplest335
model was taken in Ref 2 imposing a rigid kinematical336
constraint at a single location If such model could affect337
the tuning of the string effectively it is nonphysical338
in that part of the vibrational energy flows beyond the339
contact point (see Ref 2 for discussion) This simple340
approach is here extended to allow for multiple-point341
rigid coupling or for a non-rigid interaction as shown in342
Fig 4343
a Multiple-point rigid coupling This first optiondirectly enters into the realm of the standard kinematicalconstraints common in multibody theory The stringmotions Ys and Zs at the multiple finger locations xFf
(f = 1 F ) should be nil at all times so that
Ys(xFf t)=0 rArr
ΦYs
Ff
TqqqYs(t) = 0 (31a)
Zs(xFf t)=0 rArr
ΦZs
Ff
TqqqZs(t) = 0 (31b)
where the modeshape vectors at the finger locations are
ΦYs
Ff=φYs1 (xFf
) φYs
NYs(xFf
)T
(32a)
ΦZs
Ff=φZs1 (xFf
) φZs
NYs(xFf
)T
(32b)
Eq (31) can be rearranged in a matrix form as344
AAAf qqq = 000 (33)
where the modal constrained matrix is345
AAAf =
[ΦΦΦYs
F 000
000 ΦΦΦZs
F
](34)
with346
ΦΦΦYs
F =
ΦYs
F1
T
ΦYs
FF
T ΦΦΦZs
F =
ΦZs
F1
T
ΦZs
FF
T (35)
b Non-rigid coupling To extend the model to347
allow for non-rigid contact requires a little more care348
since a flexible-dissipative-inertial model for coupling349
subsystems cannot be directly formulated as kinematic350
constraints Nevertheless they can be implemented351
by assigning one or more additional flexible-dissipative-352
inertial elementary subsystems which are constrained to353
follow the string motions at the finger location(s) exactly354
as the body modes are included This modeling strategy355
somehow creates a bridge between computationally356
efficient multibody modeling providing ldquorigidrdquo kinematic357
constraints and the versatile but computationally slower358
modeling strategy through penalty-based constraints359
Non-rigid connections on the fingerboard cantherefore be obtained by extending the guitar modelusing a set of mass-spring-damper oscillators at locationsxFf
(f = 1 F ) and enforcing coupling as
Ys(xFf t)minusY Ff
f (t)= 0 (36a)
Zs(xFf t)minusZFf
f (t)= 0 (36b)
where YFf
f and ZFf
f are the responses of the fingercontact points in the two perpendicular directionsReturning to Eq (11) the vector and matrices mustnow be partitioned as
qqqu =qqqYsu T qqqZs
u T qqqFsu T
T(37a)
MMM = diag(MMMYsMMMZsMMMFs
)(37b)
CCC = diag(CCCYsCCCZsCCCFs
)(37c)
KKK = diag(KKKYsKKKZsKKKFs
)(37d)
6 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
where the vectors and matrices for the strings are as360
beforeMMMFs CCCFs andKKKFs are block diagonal matrices of361
the inertia damping and stiffness properties associated362
with the ldquofingerrdquo subsystem acting on string s including363
both directions of string motions364
qqqFsu =
Y F1
f Y FF
f ZF1
f ZFF
f
T(38)
In terms of modal coordinates Eqs (36) becomeΦYs
Ff
TqqqYs(t)minus Y Ff
f (t) = 0 (39a)ΦZs
Ff
TqqqZs(t)minus ZFf
f (t) = 0 (39b)
which can be rearranged as365
AAAf qqq = 000 (40)
with the modal constrained matrix AAAf written as366
AAAf =
[ΦΦΦYF 000 minusI 000
000 ΦΦΦZF 000 minusI
](41)
Figure 4 Rigid kinematical constraints (left) and flexible-
dissipative-inertial constraints (right) at multiple locations
D Dynamical formulation of the fully coupled model367
The formulation (11) readily applies to the full guitar368
model including the non-rigid constraint model for the369
stringfret interaction by stacking the unconstrained370
modal displacements into a vector371
qqqu = qqqYu qqqZu qqqBu qqqFu T (42)
where qqqYu qqqZu and qqqBu are as before and372
qqqFu=qqqY1
ufT qqqY6
ufT qqqZ1
ufT qqqZ6
ufTT
(43)
In view of the definition of qqq the assembled inertia MMM373
stiffnessKKK and dampingCCC matrices which are uncoupled374
are given by375
MMM=
MMMY 000 000 000
000 MMMZ 000 000
000 000 MMMB 000
000 000 000 MMMF
CCC=
CCCY 000 000 000
000 CCCZ 000 000
000 000 CCCB 000
000 000 000 CCCF
KKK=
KKKY 000 000 000
000 KKKZ 000 000
000 000 KKKB 000
000 000 000 KKKF
(44)
with MMM ` KKK` and CCC` the inertia stiffness and dampingmatrices for each individual subsystem ` and the vectorsFFFext and FFFnl write
FFFext =FFFYextT FFFZextT000000
T(45a)
FFFnl =FFFYnlT FFFZnlT000000
T(45b)
where FFFYext and FFFZext correspond to the plucking action376
of the player expressed for both planes of polarization377
and FFFYnl and FFFZnl are the modal vectors associated to the378
string nonlinear effects with entries given by Eq (20)379
The form of the modal constraint equation is now380
AAAqqq = 000 (46)
with381
AAA =
[AAAbAAAf
]=
ΦΦΦYb 000 minusΦΦΦB 000 000 000
000 ΦΦΦZb 000 000 000 000
ΦYFΦYFΦYF 000 000 000 minusFFFY 000
000 ΦZFΦZFΦZF 000 000 000 minusFFFZ
(47)
where ΦYFΦYFΦYF and ΦZFΦZFΦZF are block matrices with entries given382
by Eq (35) and FFFY and FFFZ are block diagonal matrices383
built with either the identity or null matrix depending on384
the constraints enforced on each string The form of (47)385
shows that the interface forces and constraint conditions386
couple the Y -motions of the strings and the body as well387
as coupling between the finger and the strings in each388
direction of motions Notice that coupling between both389
polarizations of string motions is not apparent in (47)390
although it is formulated in the nonlinear forces (45b)391
IV ILLUSTRATIVE COMPUTATIONS392
A System parameters393
For given initial conditions the system of equations394
comprising (5) with the unconstrained accelerations (11)395
and the constraint matrix (15) and using the specific396
matrices given by Eqs (44) and (47) can be integrated in397
time to produce the transient response The time-domain398
simulations were performed based on the identified body399
modal data and using standard string properties of400
classical guitars Assuming non-ideal strings with fixed-401
free ends the modeshapes are given by Eq (19) and402
their modal frequencies can be approximated by403
fYn = fZn =cT2πpn
(1 +
B
2T0p2n
)(48)
where cT is the velocity of transverse waves B is the404
bending stiffness of the string and T0 its tensioning405
pn = (2n minus 1)π2L and n the mode number Modal406
damping values are introduced following the pragmatic407
formulation given in Ref 6408
ζYn = ζZn =1
2
T0(ηF + ηAωn
)+ηBBp
2n
T0 +Bp2n(49)
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 7
where ηF ηA and ηB are the loss coefficients related to409
ldquointernal frictionrdquo ldquoair viscous dampingrdquo and ldquobending410
dampingrdquo The length of the strings is L=065 m411
and their density bending stiffness tensioning and loss412
coefficients are taken from Ref 29 The size of the modal413
basis of each string is defined by the maximum frequency414
of the computations (10000 Hz) which proved a sensible415
compromise for convergence to realistic simulations416
resulting in 800 modes for the set of strings To417
obtain suitable values for the elasticity parameter ES418
we perform simple vibratory tests on string samples419
Stretching a string vertically and suspending a mass to420
its free end one can easily measure the longitudinal421
vibration of the string-mass oscillator and obtain its422
natural frequency Assuming the string acting as a423
spring with no mass and reminding that the natural424
frequency of the longitudinal vibration is given by f =425
12π
radicESML then an estimate of ES is obtained The426
simulation results were obtained by time-step integration427
of the ordinary modal equations using the velocity-428
Verlet numerical scheme30 an integrator that combines429
simplicity second order accuracy with good numerical430
stability - see implementation details in Ref 2 We431
adopted a time-step of 10minus5s and assumed all initial432
modal displacements and velocities nil initially433
B Computational results434
Pluck excitations were simulated by considering a435
point-wise external force at xe=09L with a linear force436
ramp applied to one or more strings during 10 ms and437
by assuming null excitation force at subsequent instants438
1 Dynamical behaviour of the fully coupled model439
Results in Fig 5 were obtained for the open string440
E2 excited in the direction normal to the soundboard441
first considering a rigidly mounted string then coupling442
one string and finally all six strings to the instrument443
body Assuming no dynamics for the body the response444
spectra of string motion at the bridge is virtually nil445
according to the pinned boundary condition When446
coupling the instrument body to the string the response447
becomes obviously different and shows the classical series448
of near harmonic string partials together with smaller449
peaks from the body resonances While the first modal450
frequency of the unconstrained string is 41 Hz a close451
look at Fig 5 shows that the constraint at the bridge452
yields the fundamental frequency for the tuned string453
with pinned ends (f = 82 Hz) thus confirming the454
correct implementation of the bridge coupling Coupling455
the other five strings results in new peaks in the456
spectrum extending over the frequency range of the body457
modes and with an amplitude difference of about 100 dB458
compared to the most excited modes which clearly relate459
to the non-excited strings As found in some stringed460
instruments this dynamical coupling of all the strings via461
the bridge can lead to sympathetic vibrations which can462
further results in slight interference beats in the sound463
due to the inharmonicity produced by the bridge coupling464
and of the string partials465
0 200 400 600 800 1000 1200 1400 1600 180010
-40
10-20
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 180010
-20
10-10
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 1800
Frequency [ Hz ]
10-20
10-10
100
| Y
6(f
) |
Figure 5 Response spectra of string motion at the bridge
Top rigidly mounted string middle single stringbody
coupling bottom six-stringbody coupled model Force
amplitude 5 N Linear model
0 02 04 06 08 1
Time [ s ]
0
05
1
15
2
Mo
dal en
erg
y [
J ]
10-3
STRINGSFRETBODYTOTAL
Figure 6 Instantaneous modal energies Oblique excitation
with amplitudes of 2 N and 1 N in the Y and Z directions
Linear six-stringbody coupled model Conservative model
Figs 6 and 7 then serve at checking other466
implementation details Simulations pertained to the467
string E2 constrained at xF = 02509L by a single ldquostiffrdquo468
fret modeled with mF=5 10minus3 Kg cF=1 Nsm and469
kF=106 Nm plucked in the oblique direction The470
modal energies E` of the subsystems ` shown in these471
plots were computed from the results of the simulations472
as follows in terms of the modal parameters and the473
8 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
0 02 04 06 08 10
05
1
15
[ J ]
10-3 STRINGS
Y-dir
Z-dir
0 02 04 06 08 10
05
1
15
[ J ]
10-5 FRET
Y-dir
Z-dir
0 02 04 06 08 10
05
1
[ J ]
10-4 BODY
0 02 04 06 08 1
Time [ s ]
0
05
1
15
10-3 FULL SYSTEM
Figure 7 Instantaneous modal energies Same plucking
conditions as in Fig 6 Linear six-stringbody coupled model
Non-conservative model
modal displacements and velocities474
E`(t) =
Nsumn=1
[1
2m`n(ω`n)2
[q`n(t)
]2+
1
2m`n
[q`n(t)
]2](50)
One important feature seen in Fig 6 is that the total475
energy remains roughly constant over a conservative476
simulation while energies flow back and forth between477
the subsystems thus giving confidence in the numerical478
model for simulations of reasonable duration For479
the non-conservative case shown in Fig 7 the large480
difference in energy decay for the two polarizations481
confirms the different stringbody coupling with the482
directions of string motion Note that both the body483
and the fret oscillator are dominated by periodic energy484
exchanges at the frequency of the played note as the wave485
travels back and forth between bridge and fret Finally486
looking closely at the modal energy of the fret one can487
see smaller superimposed periodic disturbances related488
to the reflections between fret and nut illustrating that489
energy still flows through the modeled fret490
Finally we compare in Fig 8 results by varying491
the excitation amplitude in order to illustrate the string492
nonlinearity The simulations were obtained for the493
string E2 using the experimentally determined value494
of 32 kN for the flexibility ES which is comparable495
to the value given by Chaigne31 Fig 8 shows the496
temporal variations of string tension with an estimate497
of the playing frequency computed from time domain498
evaluations of the zero crossing frequency The general499
features of large amplitude string motions are clearly500
visible with a quasi-static increase of tension and501
superimposed oscillations at twice the string fundamental502
frequency and a small frequency variation of the played503
note which compares well with measurements32504
1 2 3 4 5 6 7
716
718
72
722
724
T (
t )
[ N
] Fexc
= 1 N
Fexc
= 5 N
1 2 3 4 5 6 7
Time [ s ]
822
824
826
828
f (
t )
[ H
z ]
Figure 8 String tension (top) and instantaneous frequency of
string motion (bottom) Nonlinear rigid body model
2 Parameter study of the stringfinger coupling model505
We now illustrate the different strategies to account506
for a stopping finger on the fingerboard and explore the507
influence on the coupled system of the finger model508
In contrast with the approach in Ref 2 the strategy509
in terms of auxiliary oscillators rigidly coupled to the510
string seems more realistic and is undoubtedly more511
versatile allowing different types of contacts to be512
modeled by varying the oscillator parameters However513
one could expect the finger dynamics to perturb the514
coupled system with changes in frequency and damping515
for its modes similar to what observed for a multi-modal516
system coupled to a system exhibiting resonance To517
give some insights Figs 9-11 show the effects of varying518
the finger mechanical parameters Simulations pertain to519
the string E2 constrained at xF = 01857L by a finger of520
mass mf = 5 10minus3 Kg and ignore the body dynamics521
The first test consists in varying the stiffness of the522
constraint assuming no damping Fig 9 shows the523
frequency spectra of the coupling force at the rigid bridge524
where the peaks of the string modes can be seen clearly as525
well as the resonance frequency of the auxiliary oscillator526
which moves in frequency with the stiffness and slightly527
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 9
modulates the other peaks A close look shows that528
the spectra are affected much more sensitively below529
the resonance of the finger oscillator with amplitude530
variations analogous to a system driven in the stiffness-531
controlled region Decreasing the stiffness causes a rise532
in amplitude of the undesirable modes while increasing533
the stiffness lower their amplitudes As observed for the534
stringbody coupling33 inharmonicity is also created by535
the stiffness termination which in turn affects the tuning536
of the note With the increase of stiffness perturbations537
of the string modes are reduced and tuning is improved538
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
4 Nm
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
5 Nm
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
kf = 10
6 Nm
Figure 9 Spectra of the coupling force at the rigid bridge
Stiffness is varied F=1 c1 = 0 Nsm
Fig 10 then shows the effect of varying the damping539
keeping the stiffness constant Not surprisingly if540
damping increases the amplitude of the coupled541
modes close to the oscillator resonance decrease and542
the frequency region affected becomes wider This543
significantly affects the time envelope of the pluck note544
that becomes smoother and has a shorter-time decay545
Finally the effect of the number of contacts is studied546
using the rigid contact model which is convenient for547
this purpose Fig 11 gives a picture of the spatial548
distribution of the string energy by plotting the standard549
deviation of string motion as a function of its length550
As seen the ldquopassiverdquo region is affected with great551
sensitivity by the number of contacts and surprisingly552
increasing the contact points does not necessarily reduce553
the undesirable transfer of energy Computations show554
that the effectiveness of the multi-point rigid constraints555
approach is strongly problem-dependent and varies in a556
non-obvious way with several factors including the finger557
width the spacing between contact points the relative558
string length from the ends and the excitation spectrum559
Taken in combination these results and noticing560
that the perturbations induced by each finger oscillator561
are additive due to the linear coupling a robust and562
physically sound strategy to simulate the pressing finger563
would be to implement a set of flexible-dissipative564
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
cf = 1 Nsm
cf = 10 Nsm
0 1 2 3 4 5
Time [ s ]
-2
0
2
4
Y (
t )
10-3
Figure 10 Spectra of the coupling force at the rigid
bridge (up) and time history of the string motion (bottom)
Damping is varied F=1 k1 = 105 Nm
constraints imposing a first stiff constraint in order565
to ensure a precise tuning followed by several soft566
constraints that would control the waveform and spectral567
content of the decaying response depending on their568
number and damping properties For illustration569
the spectrograms of two plucks for different sets of570
parameters are presented in Fig 12571
0 01 02 03 04 05 06
x [ m ]
10-5
10-2
ST
D (
Y(t
) )
F=1
F=2
F=3
F=5
016 017 018 019 02
x [ m ]
10-5
ST
D (
Y(t
) )
Figure 11 Standard deviation of the string motion Number
of contacts is varied Vertical dotted lines represent the
constraints locations Finger width (0015m) is kept constant
C Simulated acoustic radiation572
As an alternative to computing sounds from full573
vibro-acoustic models that usually require the use of574
massive computational resources it is tempting to apply575
10 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
Figure 12 Spectrograms of the force at the rigid bridge Up
F = 3 bottom F = 5 Parameter values kF1=106 Nm
cF1=1 Nsm and kFj =104 Nm cFj =10 Nsm (j = 2 F )
Constraint at xF = 01857L (sound files available)
simpler techniques that could be efficient to generate576
convincing sounds from the vibratory responses The577
method used here is to treat the guitar body as a linear578
input-output system and to compute the pressure by579
convolution benefiting from the local excitation of the580
body by the strings and using a vibro-acoustic impulse581
response of the test guitar This approach does not by582
any means simulate the sound response everywhere in the583
pressure field but it has the merit of adding the influence584
of the radiating properties of the body at a very low585
computational cost and of providing sounds of greater586
perceptual quality than the commonly used vibratory587
signals at the bridge The pressure at location rp can be588
computed by the convolution of the time-history of the589
stringbridge interaction force FY (rb t) and the vibro-590
acoustic impulse response function hY (rb rp t) between591
the bridge and microphone by592
p(rp t) = hY (rb rp t) lowast FY (rb t) (51)
The extension to account for the effects of all the593
strings is readily obtained by linear superposition594
principle summing the contributions from each string595
Measurements of the vibro-acoustic impulse response596
were performed under the same experimental conditions597
as for transfer function measurements using an598
additional condenser microphone Impulsive excitation599
was applied in the perpendicular direction to the600
soundboard and the acoustic pressure was measured601
at a distance of 1 m from the central axis of the602
guitar in a quiet room with all strings damped (see603
Fig 1) The impulse response was computed by inverse604
Fourier transform of the complex transfer function built605
from the force and response signals described in the606
frequency domain Results in Fig 13 pertain to the607
simulation of a G major arpeggio obtained with the608
two lowest and the top strings constrained using five609
auxiliary flexible-inertial constraints on each string The610
effect of the radiation properties of the body on the611
computed vibratory signals is supported when comparing612
the spectrograms in Fig 13 which clearly shows the613
spectral filtering of the vibration signal614
Figure 13 Spectrograms of the bridge force (top) and sound
pressure (bottom) Nonlinear six-stringbody coupled model
Arpeggio of a G major chord with three strings stopped using
a set of auxiliary oscillators (sound files available)
V CONCLUSIONS615
Following our recent work showing the potential616
of the modal Udwadia-Kalaba equations to physically617
simulate musical instruments we presented in this paper618
a computational approach that allows efficient time-619
domain simulations of realistic guitar plucks In this620
work we further generalized our simple stringbody621
coupled model toward the development of a more622
realistic guitar model by including the complete set623
of strings the two polarizations of string motions the624
string geometric nonlinearity the stringfret interaction625
that controls the playing frequency and the body626
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 11
dynamics of an actual instrument The dynamics of627
the various subsystems were formulated in terms of628
their unconstrained modes and coupling were achieved629
by enforcing several constraint conditions One of the630
developments of the model was the introduction of631
flexible-dissipative-inertial supplementary equations as632
an alternative to the common kinematical constraints633
which here improve the role of the stringfingerfret634
interaction at the fingerboard This approach can635
also benefit the modeling of other complex dynamical636
systems outside the realm of physical instrument637
modeling To improve the realism of the synthesis638
we also included crudely the radiation properties of639
the instrument body into the computational model640
through an ad hoc procedure Sound pressure was641
computed adopting a hybrid approach by convoluting642
the computed bridge force signals with a vibro-acoustic643
impulse response measured on the test guitar Such644
vibro-acoustic response can be measured in an anechoic645
chamber or else include the acoustics of any specific646
room This approach appeared not only straightforward647
to compute realistic sounds of string instruments for648
which the radiating components are subjected to point-649
wise excitation but also very efficient compared to heavy650
computational sound radiation techniques651
ACKNOWLEDGMENTS652
The authors thanks Paulo Vaz de Carvalho for653
lending the guitar and interesting discussions This work654
is funded by national funds through the FCT ndash Fundacao655
para a Ciencia e a Tecnologia IP under the Norma656
Transitoria - DL 572016CP1453CT0007657
REFERENCES658
1JH Ginsberg Mechanical and Structural Vibrations Theory659
and Applications chapter 9 Wiley 2001660
2J Antunes and V Debut Dynamical computation of constrained661
flexible systems using a modal Udwadia-Kalaba formulation662
Application to musical instruments J Acoust Soc Am663
141764 2017664
3J Antunes V Debut L Borsoi X Delaune and P Piteau665
A modal Udwadia-Kalaba formulation for the vibro-impact666
modelling of continuous flexible systems with intermittent667
contacts Procedia Engineering 199322ndash329 2017668
4FE Udwadia and RE Kalaba Analytical dynamics - A new669
approach Cambridge University Press 1996670
5C Valette The mechanics of the vibrating string Mechanics of671
musical instruments Springer 1995672
6J Woodhouse On the synthesis of guitar plucks Acta Acust673
united Ac 90928ndash944 2004674
7M E McIntyre RT Schumacher and J Woodhouse On the675
oscillations of musical instruments Journal of the Acoustical676
Society of America 741325ndash1345 1983677
8A Chaigne On the use of finite differences for musical678
synthesis Application to plucked stringed instruments Journal679
drsquoAcoustique 5181ndash211 1992680
9JM Adrien The missing link Modal synthesis In G DePoli681
A Picalli and C Roads editors Representations of musical682
sounds MIT Press Cambridge 1991683
10S Bilbao Numerical Sound Synthesis Finite Difference684
Schemes and Simulation in Musical Acoustics Wiley Publishing685
2009686
11L Trautmann and R Rabenstein Multirate simulations687
of string vibrations includingnonlinear fret-string interactions688
usingthe functional transformation method EURASIP Journal689
on Applied Signal Processing 7949ndash963 2004690
12V Chatziioannou and M van Walstijn Energy conserving691
schemes for the simulation of musical instrument contact692
dynamics J Sound Vib 339262ndash279 2015693
13S Bilbao and A Torin Numerical modeling and sound synthesis694
for articulated stringfretboard interactions Acta Acust united695
Ac 63336ndash347 2015696
14V Debut M Carvalho M Marques and J Antunes Physics-697
based modeling techniques of a twelve-string portuguese guitar698
a three-dimensional non-linear time-domain computational699
approach for the multiple-stringsbridgesoundboard coupled700
dynamics Appl Acoust 1083ndash18 2016701
15M van Walstijn and J Bridges Simulation of distributed contact702
in string instruments A modal expansion approach In 24th703
European Signal Processing Conference (Eusipco 2016) pages704
1023ndash1027 2013705
16C Issanchou S Bilbao JL Le Carrou C Touze and O Doare706
A modal-based approach to the nonlinear vibration of strings707
against a unilateral obstacle simulations and experiments in the708
pointwise case J Sound Vib 393229ndash251 2017709
17R Bader Computational Mechanics of the Classical Guitar710
Springer-Verlag Berlin Heidelberg 2005711
18H Mansour J Woodhouse and G Scavone Enhanced wave-712
based modelling of musical strings Part 1 Plucked strings Acta713
Acust united Ac 1021094ndash1107 2016714
19O Inacio J Antunes and MCM Wright Computational715
modeling of string-body interaction for the violin family and716
simulation of wolf notes J Sound Vib 310260ndash286 2007717
20S Bilbao A Torin and V Chatziioannou Numerical modeling718
of collisions in musical instruments Acta Acust united Ac719
101155ndash173 2015720
21C Issanchou JL Le Carrou C Touze B Fabre and O Doare721
Stringfrets contacts in the electric bass sound Simulations and722
experiments App Acoust 129217ndash228 2018723
22D J Braun and M Goldfarb Eliminating constraint drift in the724
numerical simulation of constrained dynamical systems Comput725
Methods in Appl Mech Eng 198(37)3151 ndash 3160 2009726
23GF Carrier On the nonlinear vibration problem of the elastic727
string Q Appl Math 3157ndash165 1945728
24G Kirchhoff Vorlesungen uber Mechanik Teubner 1883729
25J Pakarinen V Valimaki and M Karjalainen Physics-based730
methods for modeling nonlinear vibrating strings Acta Acust731
united Ac 91312ndash325 2005732
26PMC Morse and KU Ingard Theoretical Acoustics733
International series in pure and applied physics Princeton734
University Press 1968735
27J Juang Applied System Identification PTR Prentice-Hall736
Inc New Jersey 1994737
28B E Richardson The acoustical development of the guitar J738
Catgut Acoust Soc 51ndash10 1994739
29J Woodhouse Plucked guitar transients Comparison of740
measurements and synthesis Acta Acust united Ac 90945ndash741
965 2004742
30M Tuckerman B J Berne and G J Martyna Reversible743
multiple time scale molecular dynamics J Chem Phys744
97(3)1990ndash2001 1992745
31A Chaigne Viscoelastic properties of nylon guitar strings J746
Catgut Acoust Soc Series II 121ndash27 1991747
32T Tolonen V Valimaki and M Karjalainen Modeling of748
tension modulation nonlinearity in plucked strings IEEE Trans749
Speech Audio Process 8(3)300ndash310 2000750
33A Chaigne and J Kergomard Acoustics of Musical Instruments751
Modern Acoustics and Signal Processing Springer New York752
2016753
12 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
754
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 13
of the longitudinal modes but retains the effect of the223
geometrical nonlinearity by means of a pseudo-external224
forcing term added to the string transverse equations225
In our context it has two advantages First the model226
is known to capture relevant nonlinear phenomena for227
musical strings1425 thus enabling the reproduction of228
typical audible nonlinear effects Second it serves as a229
convenient means of illustrating that the U-K framework230
can deal effectively with nonlinear terms231
1 Kirchhoff-Carrier simplified nonlinear string model232
We consider a stiff string of length L cross-sectionalarea S mass density ρ Youngrsquos modulus E momentof inertia I and coefficient of dissipation η subjectto an axial tension T0 vibrating in two perpendiculartransverse motions Y (x t) and Z(x t) According to theKirchhoff-Carrier assumptions a simplified model of thefree vibration of the string is given by
ρSpart2Y
partt2+ η
partY
partt+ EI
part4Y
partx4=[T0 + Tdyn(t)
]part2Ypartx2
(16a)
ρSpart2Z
partt2+ η
partZ
partt+ EI
part4Z
partx4=[T0 + Tdyn(t)
]part2Zpartx2
(16b)
which is the standard wave equations for vibrating string233
including both polarizations with an additional forcing234
term proportional to the net dynamical increase in string235
length expressed as236
Tdyn(t) =ES
2L
int L
0
[(partY (x t)
partx
)2
+
(partZ(x t)
partx
)2]dx (17)
Eq (17) shows that the Kirchhoff-Carrier model only237
requires the knowledge of the transverse motions and238
that is independent of the x coordinate As far as the239
numerical technique is concerned the form of Eq (17)240
is very convenient particularly for modal synthesis and241
contrasts with the delicate numerical resolution of the242
ldquogeometrically exactrdquo model26 by modal discretization243
due the spatio-temporal nature of the nonlinear forcing244
2 Kirchhoff-Carrier nonlinear modal terms245
Denoting X a generic direction of string motion (X246
stands for Y or Z) the modal forces associated to the247
string nonlinear behavior are obtained by integration of248
the additional force terms of Eq (16) as249
fXn (t) =
int L
0
[Tdyn(t)
part2X
partx2
]φXn (x)dx (18)
Considering a string pinned at the nut and free at the250
bridge the modeshapes φXn (x) are given by251
φXn (x) = sin
[(2nminus 1)πx
2L
](19)
Substituting Eq (19) in Eq (18) and using a modalexpansion for the string motion the nonlinear modal
terms are finally given by
fXn (t)=minusESπ4
128L3(2nminus 1)2qXn (t)(
NYsumm=1
(2mminus1)2[qYm(t)
]2+
NZsumm=1
(2mminus1)2[qZm(t)
]2)(20)
with NY and NZ the sizes of the two string modal basis252
B The instrument body253
In view of Eq (7) the dynamic behaviour of the254
body is described by its modal properties extracted255
from input admittance measurements In contrast256
to modal computations based on a full model of the257
instrument body this is a direct approach that avoids258
complications of modeling in particular the frequency-259
dependent dissipative effects which significantly affect260
the computed sounds but are very difficult to model261
properly and that readily lend the synthesis method262
to be applied to any stringed instruments Since both263
polarizations of string motions are accounted a correct264
modeling of the stringbody coupling would demand265
the knowledge of both the in-plane and out-of-plane266
body modeshapes at the bridge However a simpler267
model is taken here by considering that only string268
motions normal to the soundboard couple to the body269
motions and also by neglecting the in-plane motions of270
the soundboard This means that the string motions in271
the plane parallel to the soundboard are unable to radiate272
sound and that the sound radiation is produced only273
by the normal motions of the soundboard Of course274
none of these choices is entirely satisfactory for accurate275
synthesis because in-plane and out-of-plane motions are276
coupled and also because in-plane motions couple the277
top and back plates via the ribs However since278
radiation is dominated by the out-of-plane vibration279
there is no doubt that in-plane motions are of second-280
order significance In practice the main advantage of281
our assumptions is to make the body characterization282
considerably simpler involving only transfer function283
measurements in the direction normal to the soundboard284
and avoiding the delicate measurements of the cross285
terms of the transfer function matrix286
The test guitar was a high-quality concert guitar287
built by Friederich number 694 During the tests it288
was positioned in the vertical position clamped to a289
rigid support by the neck and softly restrained in the290
lower bout with all the tuned strings damped (see291
Fig 1) The excitation and vibratory response were292
measured using a miniature force sensor (Kistler type293
9211) and a light-weight accelerometer (BampK 4375)294
respectively attached to the tie block using a thin layer295
of bee-wax Modal data were collected at two bridge296
locations close to the attachment points of the lowest297
and highest strings and a simple linear interpolation298
between the two transfer functions was performed to299
provide the unmeasured transfer functions at the other300
points where the strings make contact with the bridge301
4 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
Figure 1 Set-up for transfer function measurements at the
guitar body bridge
Modal identification was then achieved using the set of302
impulse responses expressed in terms of velocity and303
implementing the Eigensystem Realization Algorithm27304
The modal parameters for the first-ten identified modes305
are listed in Table I Fig 2 shows an example of306
an impulse response and its transfer function measured307
at the bridge end of the lowest string together with308
the corresponding reconstructed functions showing that309
the estimation was generally reliable up to 700 Hz310
Fig 3 also shows the identified mode shapes of the311
body at the bridge which exhibit typical motions of312
guitar modes28 The modes show a small degree of313
complexity but the effect of the imaginary components314
on the transfer function is small and can be neglected315
for our purpose Note that modes have been normalized316
according to max(|real(φBn (rs))|) = 1 (s = 1 6)317
Table I Identified modal parameters of the guitar body
Mode n fn (Hz) ζn () mn (Kg)
1 894 15 0402 1826 182 0083 2299 064 0204 3102 089 0475 4035 055 1066 5114 075 0587 5442 053 1828 6139 061 0699 6698 069 03810 8484 059 081
C Constraining forces and modal constraints318
1 Stringbody constraints319
Assuming a rigid transmission at the points wherethe strings make contact with the bridge the modalequations of the constrained system require to satisfya set of constrained modal equations stemming fromthe condition that at the bridge the strings and thebridge undergo the same motion Because the bodyresponse is here described only in the direction normal
0 005 01 015
Time [s]
-20
0
20
h (
t )
[ (
ms
)N
]
100 200 300 400 500 600 700 800 900
Frequency [ Hz ]
10-2
100
| H
( f
) |
[ (
ms
)N
]
Figure 2 Measured (green) and reconstructed (red) impulse
responses and transfer functions close to the lowest string
Gray dotted lines stand for the identified modal frequencies
E A D G B E
-1
0
1
1(
xB
)
f1 = 8939 Hz
E A D G B E
-1
0
1
2(
xB
)
f2 = 18262 Hz
E A D G B E
-1
0
1
3(
xB
)
f3 = 22986 Hz
E A D G B E
-1
0
1
4(
xB
)
f4 = 31021 Hz
E A D G B E
-1
0
1
5(
xB
)
f5 = 40352 Hz
E A D G B E
-1
0
1
6(
xB
)
f6 = 51143 Hz
E A D G B E
-1
0
1
7(
xB
)
f7 = 54417 Hz
E A D G B E
-1
0
1
8(
xB
)
f8 = 61386 Hz
E A D G B E
-1
0
1
9(
xB
)
f9 = 66978 Hz
Figure 3 Identified modeshapes of the guitar body at the
stringbridge contact points Real (blue) and imaginary
(magenta) components
to the soundboard the string motion Ys(xb t) in thisplane must be the same as the body motion at thestring contact point Yb(rs t) while its motion Zs(xb t)in the direction parallel to the soundboard must be nilFormally for each string s (s = 1 S) this results inthe conditions
Ys(xb t)minus Yb(rs t) = 0 (21a)
Zs(xb t) = 0 (21b)
or in terms of modal amplitudesΦYs
b
TqqqYs(t)minus
ΦBrsTqqqB(t) = 0 (22a)
ΦZs
b
TqqqZs(t) = 0 (22b)
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 5
where the modeshape vectors are written as
ΦYs
b =φYs1 (xb) φ
Ys
NYs(xb)
T(23a)
ΦZs
b =φZs1 (xb) φ
Zs
NZs(xb)
T(23b)
ΦBrs =φB1 (rs) φ
BNB
(rs)T
(23c)
Defining a vector of modal coordinates qqq built by stacking320
the individual modal displacements of the subsystems ie321
the strings including both polarizations and the body322
qqq = qqqY qqqZ qqqBT (24)
where qqqY and qqqZ are partitioned into terms associatedwith the different strings as
qqqY =qqqY1T qqqY6T
T(25a)
qqqZ =qqqZ1T qqqZ6T
T(25b)
and323
qqqB = qB1 qBNBT (26)
the constraint equations fit the matrix form324
AAAb qqq = 000 (27)
with the stringbody constraint matrix AAAb written as325
AAAb =
[ΦΦΦYb 000 minusΦΦΦB
000 ΦΦΦZb 000
](28)
where ΦΦΦYb and ΦΦΦZb are matrices of the modeshape vectors326
of the strings taken at the bridge location xb built as327
ΦΦΦYb =
ΦY1
b
T000 000
000
ΦY2
b
T 000
000 000
ΦY6
b
T
(29)
and similarly for the perpendicular direction and ΦΦΦB328
stands for the modeshapes of the body at the stringbody329
contact points rs330
ΦΦΦB =
ΦBr1T
ΦBr6T (30)
2 Stringfinger coupling at the fingerboard331
Modeling a stopping string for playing different332
notes can be thought in several ways according to the333
type of sound to simulate as well as to the degree of334
simplicitycomplexity aimed for the model The simplest335
model was taken in Ref 2 imposing a rigid kinematical336
constraint at a single location If such model could affect337
the tuning of the string effectively it is nonphysical338
in that part of the vibrational energy flows beyond the339
contact point (see Ref 2 for discussion) This simple340
approach is here extended to allow for multiple-point341
rigid coupling or for a non-rigid interaction as shown in342
Fig 4343
a Multiple-point rigid coupling This first optiondirectly enters into the realm of the standard kinematicalconstraints common in multibody theory The stringmotions Ys and Zs at the multiple finger locations xFf
(f = 1 F ) should be nil at all times so that
Ys(xFf t)=0 rArr
ΦYs
Ff
TqqqYs(t) = 0 (31a)
Zs(xFf t)=0 rArr
ΦZs
Ff
TqqqZs(t) = 0 (31b)
where the modeshape vectors at the finger locations are
ΦYs
Ff=φYs1 (xFf
) φYs
NYs(xFf
)T
(32a)
ΦZs
Ff=φZs1 (xFf
) φZs
NYs(xFf
)T
(32b)
Eq (31) can be rearranged in a matrix form as344
AAAf qqq = 000 (33)
where the modal constrained matrix is345
AAAf =
[ΦΦΦYs
F 000
000 ΦΦΦZs
F
](34)
with346
ΦΦΦYs
F =
ΦYs
F1
T
ΦYs
FF
T ΦΦΦZs
F =
ΦZs
F1
T
ΦZs
FF
T (35)
b Non-rigid coupling To extend the model to347
allow for non-rigid contact requires a little more care348
since a flexible-dissipative-inertial model for coupling349
subsystems cannot be directly formulated as kinematic350
constraints Nevertheless they can be implemented351
by assigning one or more additional flexible-dissipative-352
inertial elementary subsystems which are constrained to353
follow the string motions at the finger location(s) exactly354
as the body modes are included This modeling strategy355
somehow creates a bridge between computationally356
efficient multibody modeling providing ldquorigidrdquo kinematic357
constraints and the versatile but computationally slower358
modeling strategy through penalty-based constraints359
Non-rigid connections on the fingerboard cantherefore be obtained by extending the guitar modelusing a set of mass-spring-damper oscillators at locationsxFf
(f = 1 F ) and enforcing coupling as
Ys(xFf t)minusY Ff
f (t)= 0 (36a)
Zs(xFf t)minusZFf
f (t)= 0 (36b)
where YFf
f and ZFf
f are the responses of the fingercontact points in the two perpendicular directionsReturning to Eq (11) the vector and matrices mustnow be partitioned as
qqqu =qqqYsu T qqqZs
u T qqqFsu T
T(37a)
MMM = diag(MMMYsMMMZsMMMFs
)(37b)
CCC = diag(CCCYsCCCZsCCCFs
)(37c)
KKK = diag(KKKYsKKKZsKKKFs
)(37d)
6 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
where the vectors and matrices for the strings are as360
beforeMMMFs CCCFs andKKKFs are block diagonal matrices of361
the inertia damping and stiffness properties associated362
with the ldquofingerrdquo subsystem acting on string s including363
both directions of string motions364
qqqFsu =
Y F1
f Y FF
f ZF1
f ZFF
f
T(38)
In terms of modal coordinates Eqs (36) becomeΦYs
Ff
TqqqYs(t)minus Y Ff
f (t) = 0 (39a)ΦZs
Ff
TqqqZs(t)minus ZFf
f (t) = 0 (39b)
which can be rearranged as365
AAAf qqq = 000 (40)
with the modal constrained matrix AAAf written as366
AAAf =
[ΦΦΦYF 000 minusI 000
000 ΦΦΦZF 000 minusI
](41)
Figure 4 Rigid kinematical constraints (left) and flexible-
dissipative-inertial constraints (right) at multiple locations
D Dynamical formulation of the fully coupled model367
The formulation (11) readily applies to the full guitar368
model including the non-rigid constraint model for the369
stringfret interaction by stacking the unconstrained370
modal displacements into a vector371
qqqu = qqqYu qqqZu qqqBu qqqFu T (42)
where qqqYu qqqZu and qqqBu are as before and372
qqqFu=qqqY1
ufT qqqY6
ufT qqqZ1
ufT qqqZ6
ufTT
(43)
In view of the definition of qqq the assembled inertia MMM373
stiffnessKKK and dampingCCC matrices which are uncoupled374
are given by375
MMM=
MMMY 000 000 000
000 MMMZ 000 000
000 000 MMMB 000
000 000 000 MMMF
CCC=
CCCY 000 000 000
000 CCCZ 000 000
000 000 CCCB 000
000 000 000 CCCF
KKK=
KKKY 000 000 000
000 KKKZ 000 000
000 000 KKKB 000
000 000 000 KKKF
(44)
with MMM ` KKK` and CCC` the inertia stiffness and dampingmatrices for each individual subsystem ` and the vectorsFFFext and FFFnl write
FFFext =FFFYextT FFFZextT000000
T(45a)
FFFnl =FFFYnlT FFFZnlT000000
T(45b)
where FFFYext and FFFZext correspond to the plucking action376
of the player expressed for both planes of polarization377
and FFFYnl and FFFZnl are the modal vectors associated to the378
string nonlinear effects with entries given by Eq (20)379
The form of the modal constraint equation is now380
AAAqqq = 000 (46)
with381
AAA =
[AAAbAAAf
]=
ΦΦΦYb 000 minusΦΦΦB 000 000 000
000 ΦΦΦZb 000 000 000 000
ΦYFΦYFΦYF 000 000 000 minusFFFY 000
000 ΦZFΦZFΦZF 000 000 000 minusFFFZ
(47)
where ΦYFΦYFΦYF and ΦZFΦZFΦZF are block matrices with entries given382
by Eq (35) and FFFY and FFFZ are block diagonal matrices383
built with either the identity or null matrix depending on384
the constraints enforced on each string The form of (47)385
shows that the interface forces and constraint conditions386
couple the Y -motions of the strings and the body as well387
as coupling between the finger and the strings in each388
direction of motions Notice that coupling between both389
polarizations of string motions is not apparent in (47)390
although it is formulated in the nonlinear forces (45b)391
IV ILLUSTRATIVE COMPUTATIONS392
A System parameters393
For given initial conditions the system of equations394
comprising (5) with the unconstrained accelerations (11)395
and the constraint matrix (15) and using the specific396
matrices given by Eqs (44) and (47) can be integrated in397
time to produce the transient response The time-domain398
simulations were performed based on the identified body399
modal data and using standard string properties of400
classical guitars Assuming non-ideal strings with fixed-401
free ends the modeshapes are given by Eq (19) and402
their modal frequencies can be approximated by403
fYn = fZn =cT2πpn
(1 +
B
2T0p2n
)(48)
where cT is the velocity of transverse waves B is the404
bending stiffness of the string and T0 its tensioning405
pn = (2n minus 1)π2L and n the mode number Modal406
damping values are introduced following the pragmatic407
formulation given in Ref 6408
ζYn = ζZn =1
2
T0(ηF + ηAωn
)+ηBBp
2n
T0 +Bp2n(49)
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 7
where ηF ηA and ηB are the loss coefficients related to409
ldquointernal frictionrdquo ldquoair viscous dampingrdquo and ldquobending410
dampingrdquo The length of the strings is L=065 m411
and their density bending stiffness tensioning and loss412
coefficients are taken from Ref 29 The size of the modal413
basis of each string is defined by the maximum frequency414
of the computations (10000 Hz) which proved a sensible415
compromise for convergence to realistic simulations416
resulting in 800 modes for the set of strings To417
obtain suitable values for the elasticity parameter ES418
we perform simple vibratory tests on string samples419
Stretching a string vertically and suspending a mass to420
its free end one can easily measure the longitudinal421
vibration of the string-mass oscillator and obtain its422
natural frequency Assuming the string acting as a423
spring with no mass and reminding that the natural424
frequency of the longitudinal vibration is given by f =425
12π
radicESML then an estimate of ES is obtained The426
simulation results were obtained by time-step integration427
of the ordinary modal equations using the velocity-428
Verlet numerical scheme30 an integrator that combines429
simplicity second order accuracy with good numerical430
stability - see implementation details in Ref 2 We431
adopted a time-step of 10minus5s and assumed all initial432
modal displacements and velocities nil initially433
B Computational results434
Pluck excitations were simulated by considering a435
point-wise external force at xe=09L with a linear force436
ramp applied to one or more strings during 10 ms and437
by assuming null excitation force at subsequent instants438
1 Dynamical behaviour of the fully coupled model439
Results in Fig 5 were obtained for the open string440
E2 excited in the direction normal to the soundboard441
first considering a rigidly mounted string then coupling442
one string and finally all six strings to the instrument443
body Assuming no dynamics for the body the response444
spectra of string motion at the bridge is virtually nil445
according to the pinned boundary condition When446
coupling the instrument body to the string the response447
becomes obviously different and shows the classical series448
of near harmonic string partials together with smaller449
peaks from the body resonances While the first modal450
frequency of the unconstrained string is 41 Hz a close451
look at Fig 5 shows that the constraint at the bridge452
yields the fundamental frequency for the tuned string453
with pinned ends (f = 82 Hz) thus confirming the454
correct implementation of the bridge coupling Coupling455
the other five strings results in new peaks in the456
spectrum extending over the frequency range of the body457
modes and with an amplitude difference of about 100 dB458
compared to the most excited modes which clearly relate459
to the non-excited strings As found in some stringed460
instruments this dynamical coupling of all the strings via461
the bridge can lead to sympathetic vibrations which can462
further results in slight interference beats in the sound463
due to the inharmonicity produced by the bridge coupling464
and of the string partials465
0 200 400 600 800 1000 1200 1400 1600 180010
-40
10-20
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 180010
-20
10-10
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 1800
Frequency [ Hz ]
10-20
10-10
100
| Y
6(f
) |
Figure 5 Response spectra of string motion at the bridge
Top rigidly mounted string middle single stringbody
coupling bottom six-stringbody coupled model Force
amplitude 5 N Linear model
0 02 04 06 08 1
Time [ s ]
0
05
1
15
2
Mo
dal en
erg
y [
J ]
10-3
STRINGSFRETBODYTOTAL
Figure 6 Instantaneous modal energies Oblique excitation
with amplitudes of 2 N and 1 N in the Y and Z directions
Linear six-stringbody coupled model Conservative model
Figs 6 and 7 then serve at checking other466
implementation details Simulations pertained to the467
string E2 constrained at xF = 02509L by a single ldquostiffrdquo468
fret modeled with mF=5 10minus3 Kg cF=1 Nsm and469
kF=106 Nm plucked in the oblique direction The470
modal energies E` of the subsystems ` shown in these471
plots were computed from the results of the simulations472
as follows in terms of the modal parameters and the473
8 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
0 02 04 06 08 10
05
1
15
[ J ]
10-3 STRINGS
Y-dir
Z-dir
0 02 04 06 08 10
05
1
15
[ J ]
10-5 FRET
Y-dir
Z-dir
0 02 04 06 08 10
05
1
[ J ]
10-4 BODY
0 02 04 06 08 1
Time [ s ]
0
05
1
15
10-3 FULL SYSTEM
Figure 7 Instantaneous modal energies Same plucking
conditions as in Fig 6 Linear six-stringbody coupled model
Non-conservative model
modal displacements and velocities474
E`(t) =
Nsumn=1
[1
2m`n(ω`n)2
[q`n(t)
]2+
1
2m`n
[q`n(t)
]2](50)
One important feature seen in Fig 6 is that the total475
energy remains roughly constant over a conservative476
simulation while energies flow back and forth between477
the subsystems thus giving confidence in the numerical478
model for simulations of reasonable duration For479
the non-conservative case shown in Fig 7 the large480
difference in energy decay for the two polarizations481
confirms the different stringbody coupling with the482
directions of string motion Note that both the body483
and the fret oscillator are dominated by periodic energy484
exchanges at the frequency of the played note as the wave485
travels back and forth between bridge and fret Finally486
looking closely at the modal energy of the fret one can487
see smaller superimposed periodic disturbances related488
to the reflections between fret and nut illustrating that489
energy still flows through the modeled fret490
Finally we compare in Fig 8 results by varying491
the excitation amplitude in order to illustrate the string492
nonlinearity The simulations were obtained for the493
string E2 using the experimentally determined value494
of 32 kN for the flexibility ES which is comparable495
to the value given by Chaigne31 Fig 8 shows the496
temporal variations of string tension with an estimate497
of the playing frequency computed from time domain498
evaluations of the zero crossing frequency The general499
features of large amplitude string motions are clearly500
visible with a quasi-static increase of tension and501
superimposed oscillations at twice the string fundamental502
frequency and a small frequency variation of the played503
note which compares well with measurements32504
1 2 3 4 5 6 7
716
718
72
722
724
T (
t )
[ N
] Fexc
= 1 N
Fexc
= 5 N
1 2 3 4 5 6 7
Time [ s ]
822
824
826
828
f (
t )
[ H
z ]
Figure 8 String tension (top) and instantaneous frequency of
string motion (bottom) Nonlinear rigid body model
2 Parameter study of the stringfinger coupling model505
We now illustrate the different strategies to account506
for a stopping finger on the fingerboard and explore the507
influence on the coupled system of the finger model508
In contrast with the approach in Ref 2 the strategy509
in terms of auxiliary oscillators rigidly coupled to the510
string seems more realistic and is undoubtedly more511
versatile allowing different types of contacts to be512
modeled by varying the oscillator parameters However513
one could expect the finger dynamics to perturb the514
coupled system with changes in frequency and damping515
for its modes similar to what observed for a multi-modal516
system coupled to a system exhibiting resonance To517
give some insights Figs 9-11 show the effects of varying518
the finger mechanical parameters Simulations pertain to519
the string E2 constrained at xF = 01857L by a finger of520
mass mf = 5 10minus3 Kg and ignore the body dynamics521
The first test consists in varying the stiffness of the522
constraint assuming no damping Fig 9 shows the523
frequency spectra of the coupling force at the rigid bridge524
where the peaks of the string modes can be seen clearly as525
well as the resonance frequency of the auxiliary oscillator526
which moves in frequency with the stiffness and slightly527
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 9
modulates the other peaks A close look shows that528
the spectra are affected much more sensitively below529
the resonance of the finger oscillator with amplitude530
variations analogous to a system driven in the stiffness-531
controlled region Decreasing the stiffness causes a rise532
in amplitude of the undesirable modes while increasing533
the stiffness lower their amplitudes As observed for the534
stringbody coupling33 inharmonicity is also created by535
the stiffness termination which in turn affects the tuning536
of the note With the increase of stiffness perturbations537
of the string modes are reduced and tuning is improved538
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
4 Nm
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
5 Nm
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
kf = 10
6 Nm
Figure 9 Spectra of the coupling force at the rigid bridge
Stiffness is varied F=1 c1 = 0 Nsm
Fig 10 then shows the effect of varying the damping539
keeping the stiffness constant Not surprisingly if540
damping increases the amplitude of the coupled541
modes close to the oscillator resonance decrease and542
the frequency region affected becomes wider This543
significantly affects the time envelope of the pluck note544
that becomes smoother and has a shorter-time decay545
Finally the effect of the number of contacts is studied546
using the rigid contact model which is convenient for547
this purpose Fig 11 gives a picture of the spatial548
distribution of the string energy by plotting the standard549
deviation of string motion as a function of its length550
As seen the ldquopassiverdquo region is affected with great551
sensitivity by the number of contacts and surprisingly552
increasing the contact points does not necessarily reduce553
the undesirable transfer of energy Computations show554
that the effectiveness of the multi-point rigid constraints555
approach is strongly problem-dependent and varies in a556
non-obvious way with several factors including the finger557
width the spacing between contact points the relative558
string length from the ends and the excitation spectrum559
Taken in combination these results and noticing560
that the perturbations induced by each finger oscillator561
are additive due to the linear coupling a robust and562
physically sound strategy to simulate the pressing finger563
would be to implement a set of flexible-dissipative564
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
cf = 1 Nsm
cf = 10 Nsm
0 1 2 3 4 5
Time [ s ]
-2
0
2
4
Y (
t )
10-3
Figure 10 Spectra of the coupling force at the rigid
bridge (up) and time history of the string motion (bottom)
Damping is varied F=1 k1 = 105 Nm
constraints imposing a first stiff constraint in order565
to ensure a precise tuning followed by several soft566
constraints that would control the waveform and spectral567
content of the decaying response depending on their568
number and damping properties For illustration569
the spectrograms of two plucks for different sets of570
parameters are presented in Fig 12571
0 01 02 03 04 05 06
x [ m ]
10-5
10-2
ST
D (
Y(t
) )
F=1
F=2
F=3
F=5
016 017 018 019 02
x [ m ]
10-5
ST
D (
Y(t
) )
Figure 11 Standard deviation of the string motion Number
of contacts is varied Vertical dotted lines represent the
constraints locations Finger width (0015m) is kept constant
C Simulated acoustic radiation572
As an alternative to computing sounds from full573
vibro-acoustic models that usually require the use of574
massive computational resources it is tempting to apply575
10 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
Figure 12 Spectrograms of the force at the rigid bridge Up
F = 3 bottom F = 5 Parameter values kF1=106 Nm
cF1=1 Nsm and kFj =104 Nm cFj =10 Nsm (j = 2 F )
Constraint at xF = 01857L (sound files available)
simpler techniques that could be efficient to generate576
convincing sounds from the vibratory responses The577
method used here is to treat the guitar body as a linear578
input-output system and to compute the pressure by579
convolution benefiting from the local excitation of the580
body by the strings and using a vibro-acoustic impulse581
response of the test guitar This approach does not by582
any means simulate the sound response everywhere in the583
pressure field but it has the merit of adding the influence584
of the radiating properties of the body at a very low585
computational cost and of providing sounds of greater586
perceptual quality than the commonly used vibratory587
signals at the bridge The pressure at location rp can be588
computed by the convolution of the time-history of the589
stringbridge interaction force FY (rb t) and the vibro-590
acoustic impulse response function hY (rb rp t) between591
the bridge and microphone by592
p(rp t) = hY (rb rp t) lowast FY (rb t) (51)
The extension to account for the effects of all the593
strings is readily obtained by linear superposition594
principle summing the contributions from each string595
Measurements of the vibro-acoustic impulse response596
were performed under the same experimental conditions597
as for transfer function measurements using an598
additional condenser microphone Impulsive excitation599
was applied in the perpendicular direction to the600
soundboard and the acoustic pressure was measured601
at a distance of 1 m from the central axis of the602
guitar in a quiet room with all strings damped (see603
Fig 1) The impulse response was computed by inverse604
Fourier transform of the complex transfer function built605
from the force and response signals described in the606
frequency domain Results in Fig 13 pertain to the607
simulation of a G major arpeggio obtained with the608
two lowest and the top strings constrained using five609
auxiliary flexible-inertial constraints on each string The610
effect of the radiation properties of the body on the611
computed vibratory signals is supported when comparing612
the spectrograms in Fig 13 which clearly shows the613
spectral filtering of the vibration signal614
Figure 13 Spectrograms of the bridge force (top) and sound
pressure (bottom) Nonlinear six-stringbody coupled model
Arpeggio of a G major chord with three strings stopped using
a set of auxiliary oscillators (sound files available)
V CONCLUSIONS615
Following our recent work showing the potential616
of the modal Udwadia-Kalaba equations to physically617
simulate musical instruments we presented in this paper618
a computational approach that allows efficient time-619
domain simulations of realistic guitar plucks In this620
work we further generalized our simple stringbody621
coupled model toward the development of a more622
realistic guitar model by including the complete set623
of strings the two polarizations of string motions the624
string geometric nonlinearity the stringfret interaction625
that controls the playing frequency and the body626
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 11
dynamics of an actual instrument The dynamics of627
the various subsystems were formulated in terms of628
their unconstrained modes and coupling were achieved629
by enforcing several constraint conditions One of the630
developments of the model was the introduction of631
flexible-dissipative-inertial supplementary equations as632
an alternative to the common kinematical constraints633
which here improve the role of the stringfingerfret634
interaction at the fingerboard This approach can635
also benefit the modeling of other complex dynamical636
systems outside the realm of physical instrument637
modeling To improve the realism of the synthesis638
we also included crudely the radiation properties of639
the instrument body into the computational model640
through an ad hoc procedure Sound pressure was641
computed adopting a hybrid approach by convoluting642
the computed bridge force signals with a vibro-acoustic643
impulse response measured on the test guitar Such644
vibro-acoustic response can be measured in an anechoic645
chamber or else include the acoustics of any specific646
room This approach appeared not only straightforward647
to compute realistic sounds of string instruments for648
which the radiating components are subjected to point-649
wise excitation but also very efficient compared to heavy650
computational sound radiation techniques651
ACKNOWLEDGMENTS652
The authors thanks Paulo Vaz de Carvalho for653
lending the guitar and interesting discussions This work654
is funded by national funds through the FCT ndash Fundacao655
para a Ciencia e a Tecnologia IP under the Norma656
Transitoria - DL 572016CP1453CT0007657
REFERENCES658
1JH Ginsberg Mechanical and Structural Vibrations Theory659
and Applications chapter 9 Wiley 2001660
2J Antunes and V Debut Dynamical computation of constrained661
flexible systems using a modal Udwadia-Kalaba formulation662
Application to musical instruments J Acoust Soc Am663
141764 2017664
3J Antunes V Debut L Borsoi X Delaune and P Piteau665
A modal Udwadia-Kalaba formulation for the vibro-impact666
modelling of continuous flexible systems with intermittent667
contacts Procedia Engineering 199322ndash329 2017668
4FE Udwadia and RE Kalaba Analytical dynamics - A new669
approach Cambridge University Press 1996670
5C Valette The mechanics of the vibrating string Mechanics of671
musical instruments Springer 1995672
6J Woodhouse On the synthesis of guitar plucks Acta Acust673
united Ac 90928ndash944 2004674
7M E McIntyre RT Schumacher and J Woodhouse On the675
oscillations of musical instruments Journal of the Acoustical676
Society of America 741325ndash1345 1983677
8A Chaigne On the use of finite differences for musical678
synthesis Application to plucked stringed instruments Journal679
drsquoAcoustique 5181ndash211 1992680
9JM Adrien The missing link Modal synthesis In G DePoli681
A Picalli and C Roads editors Representations of musical682
sounds MIT Press Cambridge 1991683
10S Bilbao Numerical Sound Synthesis Finite Difference684
Schemes and Simulation in Musical Acoustics Wiley Publishing685
2009686
11L Trautmann and R Rabenstein Multirate simulations687
of string vibrations includingnonlinear fret-string interactions688
usingthe functional transformation method EURASIP Journal689
on Applied Signal Processing 7949ndash963 2004690
12V Chatziioannou and M van Walstijn Energy conserving691
schemes for the simulation of musical instrument contact692
dynamics J Sound Vib 339262ndash279 2015693
13S Bilbao and A Torin Numerical modeling and sound synthesis694
for articulated stringfretboard interactions Acta Acust united695
Ac 63336ndash347 2015696
14V Debut M Carvalho M Marques and J Antunes Physics-697
based modeling techniques of a twelve-string portuguese guitar698
a three-dimensional non-linear time-domain computational699
approach for the multiple-stringsbridgesoundboard coupled700
dynamics Appl Acoust 1083ndash18 2016701
15M van Walstijn and J Bridges Simulation of distributed contact702
in string instruments A modal expansion approach In 24th703
European Signal Processing Conference (Eusipco 2016) pages704
1023ndash1027 2013705
16C Issanchou S Bilbao JL Le Carrou C Touze and O Doare706
A modal-based approach to the nonlinear vibration of strings707
against a unilateral obstacle simulations and experiments in the708
pointwise case J Sound Vib 393229ndash251 2017709
17R Bader Computational Mechanics of the Classical Guitar710
Springer-Verlag Berlin Heidelberg 2005711
18H Mansour J Woodhouse and G Scavone Enhanced wave-712
based modelling of musical strings Part 1 Plucked strings Acta713
Acust united Ac 1021094ndash1107 2016714
19O Inacio J Antunes and MCM Wright Computational715
modeling of string-body interaction for the violin family and716
simulation of wolf notes J Sound Vib 310260ndash286 2007717
20S Bilbao A Torin and V Chatziioannou Numerical modeling718
of collisions in musical instruments Acta Acust united Ac719
101155ndash173 2015720
21C Issanchou JL Le Carrou C Touze B Fabre and O Doare721
Stringfrets contacts in the electric bass sound Simulations and722
experiments App Acoust 129217ndash228 2018723
22D J Braun and M Goldfarb Eliminating constraint drift in the724
numerical simulation of constrained dynamical systems Comput725
Methods in Appl Mech Eng 198(37)3151 ndash 3160 2009726
23GF Carrier On the nonlinear vibration problem of the elastic727
string Q Appl Math 3157ndash165 1945728
24G Kirchhoff Vorlesungen uber Mechanik Teubner 1883729
25J Pakarinen V Valimaki and M Karjalainen Physics-based730
methods for modeling nonlinear vibrating strings Acta Acust731
united Ac 91312ndash325 2005732
26PMC Morse and KU Ingard Theoretical Acoustics733
International series in pure and applied physics Princeton734
University Press 1968735
27J Juang Applied System Identification PTR Prentice-Hall736
Inc New Jersey 1994737
28B E Richardson The acoustical development of the guitar J738
Catgut Acoust Soc 51ndash10 1994739
29J Woodhouse Plucked guitar transients Comparison of740
measurements and synthesis Acta Acust united Ac 90945ndash741
965 2004742
30M Tuckerman B J Berne and G J Martyna Reversible743
multiple time scale molecular dynamics J Chem Phys744
97(3)1990ndash2001 1992745
31A Chaigne Viscoelastic properties of nylon guitar strings J746
Catgut Acoust Soc Series II 121ndash27 1991747
32T Tolonen V Valimaki and M Karjalainen Modeling of748
tension modulation nonlinearity in plucked strings IEEE Trans749
Speech Audio Process 8(3)300ndash310 2000750
33A Chaigne and J Kergomard Acoustics of Musical Instruments751
Modern Acoustics and Signal Processing Springer New York752
2016753
12 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
754
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 13
Figure 1 Set-up for transfer function measurements at the
guitar body bridge
Modal identification was then achieved using the set of302
impulse responses expressed in terms of velocity and303
implementing the Eigensystem Realization Algorithm27304
The modal parameters for the first-ten identified modes305
are listed in Table I Fig 2 shows an example of306
an impulse response and its transfer function measured307
at the bridge end of the lowest string together with308
the corresponding reconstructed functions showing that309
the estimation was generally reliable up to 700 Hz310
Fig 3 also shows the identified mode shapes of the311
body at the bridge which exhibit typical motions of312
guitar modes28 The modes show a small degree of313
complexity but the effect of the imaginary components314
on the transfer function is small and can be neglected315
for our purpose Note that modes have been normalized316
according to max(|real(φBn (rs))|) = 1 (s = 1 6)317
Table I Identified modal parameters of the guitar body
Mode n fn (Hz) ζn () mn (Kg)
1 894 15 0402 1826 182 0083 2299 064 0204 3102 089 0475 4035 055 1066 5114 075 0587 5442 053 1828 6139 061 0699 6698 069 03810 8484 059 081
C Constraining forces and modal constraints318
1 Stringbody constraints319
Assuming a rigid transmission at the points wherethe strings make contact with the bridge the modalequations of the constrained system require to satisfya set of constrained modal equations stemming fromthe condition that at the bridge the strings and thebridge undergo the same motion Because the bodyresponse is here described only in the direction normal
0 005 01 015
Time [s]
-20
0
20
h (
t )
[ (
ms
)N
]
100 200 300 400 500 600 700 800 900
Frequency [ Hz ]
10-2
100
| H
( f
) |
[ (
ms
)N
]
Figure 2 Measured (green) and reconstructed (red) impulse
responses and transfer functions close to the lowest string
Gray dotted lines stand for the identified modal frequencies
E A D G B E
-1
0
1
1(
xB
)
f1 = 8939 Hz
E A D G B E
-1
0
1
2(
xB
)
f2 = 18262 Hz
E A D G B E
-1
0
1
3(
xB
)
f3 = 22986 Hz
E A D G B E
-1
0
1
4(
xB
)
f4 = 31021 Hz
E A D G B E
-1
0
1
5(
xB
)
f5 = 40352 Hz
E A D G B E
-1
0
1
6(
xB
)
f6 = 51143 Hz
E A D G B E
-1
0
1
7(
xB
)
f7 = 54417 Hz
E A D G B E
-1
0
1
8(
xB
)
f8 = 61386 Hz
E A D G B E
-1
0
1
9(
xB
)
f9 = 66978 Hz
Figure 3 Identified modeshapes of the guitar body at the
stringbridge contact points Real (blue) and imaginary
(magenta) components
to the soundboard the string motion Ys(xb t) in thisplane must be the same as the body motion at thestring contact point Yb(rs t) while its motion Zs(xb t)in the direction parallel to the soundboard must be nilFormally for each string s (s = 1 S) this results inthe conditions
Ys(xb t)minus Yb(rs t) = 0 (21a)
Zs(xb t) = 0 (21b)
or in terms of modal amplitudesΦYs
b
TqqqYs(t)minus
ΦBrsTqqqB(t) = 0 (22a)
ΦZs
b
TqqqZs(t) = 0 (22b)
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 5
where the modeshape vectors are written as
ΦYs
b =φYs1 (xb) φ
Ys
NYs(xb)
T(23a)
ΦZs
b =φZs1 (xb) φ
Zs
NZs(xb)
T(23b)
ΦBrs =φB1 (rs) φ
BNB
(rs)T
(23c)
Defining a vector of modal coordinates qqq built by stacking320
the individual modal displacements of the subsystems ie321
the strings including both polarizations and the body322
qqq = qqqY qqqZ qqqBT (24)
where qqqY and qqqZ are partitioned into terms associatedwith the different strings as
qqqY =qqqY1T qqqY6T
T(25a)
qqqZ =qqqZ1T qqqZ6T
T(25b)
and323
qqqB = qB1 qBNBT (26)
the constraint equations fit the matrix form324
AAAb qqq = 000 (27)
with the stringbody constraint matrix AAAb written as325
AAAb =
[ΦΦΦYb 000 minusΦΦΦB
000 ΦΦΦZb 000
](28)
where ΦΦΦYb and ΦΦΦZb are matrices of the modeshape vectors326
of the strings taken at the bridge location xb built as327
ΦΦΦYb =
ΦY1
b
T000 000
000
ΦY2
b
T 000
000 000
ΦY6
b
T
(29)
and similarly for the perpendicular direction and ΦΦΦB328
stands for the modeshapes of the body at the stringbody329
contact points rs330
ΦΦΦB =
ΦBr1T
ΦBr6T (30)
2 Stringfinger coupling at the fingerboard331
Modeling a stopping string for playing different332
notes can be thought in several ways according to the333
type of sound to simulate as well as to the degree of334
simplicitycomplexity aimed for the model The simplest335
model was taken in Ref 2 imposing a rigid kinematical336
constraint at a single location If such model could affect337
the tuning of the string effectively it is nonphysical338
in that part of the vibrational energy flows beyond the339
contact point (see Ref 2 for discussion) This simple340
approach is here extended to allow for multiple-point341
rigid coupling or for a non-rigid interaction as shown in342
Fig 4343
a Multiple-point rigid coupling This first optiondirectly enters into the realm of the standard kinematicalconstraints common in multibody theory The stringmotions Ys and Zs at the multiple finger locations xFf
(f = 1 F ) should be nil at all times so that
Ys(xFf t)=0 rArr
ΦYs
Ff
TqqqYs(t) = 0 (31a)
Zs(xFf t)=0 rArr
ΦZs
Ff
TqqqZs(t) = 0 (31b)
where the modeshape vectors at the finger locations are
ΦYs
Ff=φYs1 (xFf
) φYs
NYs(xFf
)T
(32a)
ΦZs
Ff=φZs1 (xFf
) φZs
NYs(xFf
)T
(32b)
Eq (31) can be rearranged in a matrix form as344
AAAf qqq = 000 (33)
where the modal constrained matrix is345
AAAf =
[ΦΦΦYs
F 000
000 ΦΦΦZs
F
](34)
with346
ΦΦΦYs
F =
ΦYs
F1
T
ΦYs
FF
T ΦΦΦZs
F =
ΦZs
F1
T
ΦZs
FF
T (35)
b Non-rigid coupling To extend the model to347
allow for non-rigid contact requires a little more care348
since a flexible-dissipative-inertial model for coupling349
subsystems cannot be directly formulated as kinematic350
constraints Nevertheless they can be implemented351
by assigning one or more additional flexible-dissipative-352
inertial elementary subsystems which are constrained to353
follow the string motions at the finger location(s) exactly354
as the body modes are included This modeling strategy355
somehow creates a bridge between computationally356
efficient multibody modeling providing ldquorigidrdquo kinematic357
constraints and the versatile but computationally slower358
modeling strategy through penalty-based constraints359
Non-rigid connections on the fingerboard cantherefore be obtained by extending the guitar modelusing a set of mass-spring-damper oscillators at locationsxFf
(f = 1 F ) and enforcing coupling as
Ys(xFf t)minusY Ff
f (t)= 0 (36a)
Zs(xFf t)minusZFf
f (t)= 0 (36b)
where YFf
f and ZFf
f are the responses of the fingercontact points in the two perpendicular directionsReturning to Eq (11) the vector and matrices mustnow be partitioned as
qqqu =qqqYsu T qqqZs
u T qqqFsu T
T(37a)
MMM = diag(MMMYsMMMZsMMMFs
)(37b)
CCC = diag(CCCYsCCCZsCCCFs
)(37c)
KKK = diag(KKKYsKKKZsKKKFs
)(37d)
6 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
where the vectors and matrices for the strings are as360
beforeMMMFs CCCFs andKKKFs are block diagonal matrices of361
the inertia damping and stiffness properties associated362
with the ldquofingerrdquo subsystem acting on string s including363
both directions of string motions364
qqqFsu =
Y F1
f Y FF
f ZF1
f ZFF
f
T(38)
In terms of modal coordinates Eqs (36) becomeΦYs
Ff
TqqqYs(t)minus Y Ff
f (t) = 0 (39a)ΦZs
Ff
TqqqZs(t)minus ZFf
f (t) = 0 (39b)
which can be rearranged as365
AAAf qqq = 000 (40)
with the modal constrained matrix AAAf written as366
AAAf =
[ΦΦΦYF 000 minusI 000
000 ΦΦΦZF 000 minusI
](41)
Figure 4 Rigid kinematical constraints (left) and flexible-
dissipative-inertial constraints (right) at multiple locations
D Dynamical formulation of the fully coupled model367
The formulation (11) readily applies to the full guitar368
model including the non-rigid constraint model for the369
stringfret interaction by stacking the unconstrained370
modal displacements into a vector371
qqqu = qqqYu qqqZu qqqBu qqqFu T (42)
where qqqYu qqqZu and qqqBu are as before and372
qqqFu=qqqY1
ufT qqqY6
ufT qqqZ1
ufT qqqZ6
ufTT
(43)
In view of the definition of qqq the assembled inertia MMM373
stiffnessKKK and dampingCCC matrices which are uncoupled374
are given by375
MMM=
MMMY 000 000 000
000 MMMZ 000 000
000 000 MMMB 000
000 000 000 MMMF
CCC=
CCCY 000 000 000
000 CCCZ 000 000
000 000 CCCB 000
000 000 000 CCCF
KKK=
KKKY 000 000 000
000 KKKZ 000 000
000 000 KKKB 000
000 000 000 KKKF
(44)
with MMM ` KKK` and CCC` the inertia stiffness and dampingmatrices for each individual subsystem ` and the vectorsFFFext and FFFnl write
FFFext =FFFYextT FFFZextT000000
T(45a)
FFFnl =FFFYnlT FFFZnlT000000
T(45b)
where FFFYext and FFFZext correspond to the plucking action376
of the player expressed for both planes of polarization377
and FFFYnl and FFFZnl are the modal vectors associated to the378
string nonlinear effects with entries given by Eq (20)379
The form of the modal constraint equation is now380
AAAqqq = 000 (46)
with381
AAA =
[AAAbAAAf
]=
ΦΦΦYb 000 minusΦΦΦB 000 000 000
000 ΦΦΦZb 000 000 000 000
ΦYFΦYFΦYF 000 000 000 minusFFFY 000
000 ΦZFΦZFΦZF 000 000 000 minusFFFZ
(47)
where ΦYFΦYFΦYF and ΦZFΦZFΦZF are block matrices with entries given382
by Eq (35) and FFFY and FFFZ are block diagonal matrices383
built with either the identity or null matrix depending on384
the constraints enforced on each string The form of (47)385
shows that the interface forces and constraint conditions386
couple the Y -motions of the strings and the body as well387
as coupling between the finger and the strings in each388
direction of motions Notice that coupling between both389
polarizations of string motions is not apparent in (47)390
although it is formulated in the nonlinear forces (45b)391
IV ILLUSTRATIVE COMPUTATIONS392
A System parameters393
For given initial conditions the system of equations394
comprising (5) with the unconstrained accelerations (11)395
and the constraint matrix (15) and using the specific396
matrices given by Eqs (44) and (47) can be integrated in397
time to produce the transient response The time-domain398
simulations were performed based on the identified body399
modal data and using standard string properties of400
classical guitars Assuming non-ideal strings with fixed-401
free ends the modeshapes are given by Eq (19) and402
their modal frequencies can be approximated by403
fYn = fZn =cT2πpn
(1 +
B
2T0p2n
)(48)
where cT is the velocity of transverse waves B is the404
bending stiffness of the string and T0 its tensioning405
pn = (2n minus 1)π2L and n the mode number Modal406
damping values are introduced following the pragmatic407
formulation given in Ref 6408
ζYn = ζZn =1
2
T0(ηF + ηAωn
)+ηBBp
2n
T0 +Bp2n(49)
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 7
where ηF ηA and ηB are the loss coefficients related to409
ldquointernal frictionrdquo ldquoair viscous dampingrdquo and ldquobending410
dampingrdquo The length of the strings is L=065 m411
and their density bending stiffness tensioning and loss412
coefficients are taken from Ref 29 The size of the modal413
basis of each string is defined by the maximum frequency414
of the computations (10000 Hz) which proved a sensible415
compromise for convergence to realistic simulations416
resulting in 800 modes for the set of strings To417
obtain suitable values for the elasticity parameter ES418
we perform simple vibratory tests on string samples419
Stretching a string vertically and suspending a mass to420
its free end one can easily measure the longitudinal421
vibration of the string-mass oscillator and obtain its422
natural frequency Assuming the string acting as a423
spring with no mass and reminding that the natural424
frequency of the longitudinal vibration is given by f =425
12π
radicESML then an estimate of ES is obtained The426
simulation results were obtained by time-step integration427
of the ordinary modal equations using the velocity-428
Verlet numerical scheme30 an integrator that combines429
simplicity second order accuracy with good numerical430
stability - see implementation details in Ref 2 We431
adopted a time-step of 10minus5s and assumed all initial432
modal displacements and velocities nil initially433
B Computational results434
Pluck excitations were simulated by considering a435
point-wise external force at xe=09L with a linear force436
ramp applied to one or more strings during 10 ms and437
by assuming null excitation force at subsequent instants438
1 Dynamical behaviour of the fully coupled model439
Results in Fig 5 were obtained for the open string440
E2 excited in the direction normal to the soundboard441
first considering a rigidly mounted string then coupling442
one string and finally all six strings to the instrument443
body Assuming no dynamics for the body the response444
spectra of string motion at the bridge is virtually nil445
according to the pinned boundary condition When446
coupling the instrument body to the string the response447
becomes obviously different and shows the classical series448
of near harmonic string partials together with smaller449
peaks from the body resonances While the first modal450
frequency of the unconstrained string is 41 Hz a close451
look at Fig 5 shows that the constraint at the bridge452
yields the fundamental frequency for the tuned string453
with pinned ends (f = 82 Hz) thus confirming the454
correct implementation of the bridge coupling Coupling455
the other five strings results in new peaks in the456
spectrum extending over the frequency range of the body457
modes and with an amplitude difference of about 100 dB458
compared to the most excited modes which clearly relate459
to the non-excited strings As found in some stringed460
instruments this dynamical coupling of all the strings via461
the bridge can lead to sympathetic vibrations which can462
further results in slight interference beats in the sound463
due to the inharmonicity produced by the bridge coupling464
and of the string partials465
0 200 400 600 800 1000 1200 1400 1600 180010
-40
10-20
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 180010
-20
10-10
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 1800
Frequency [ Hz ]
10-20
10-10
100
| Y
6(f
) |
Figure 5 Response spectra of string motion at the bridge
Top rigidly mounted string middle single stringbody
coupling bottom six-stringbody coupled model Force
amplitude 5 N Linear model
0 02 04 06 08 1
Time [ s ]
0
05
1
15
2
Mo
dal en
erg
y [
J ]
10-3
STRINGSFRETBODYTOTAL
Figure 6 Instantaneous modal energies Oblique excitation
with amplitudes of 2 N and 1 N in the Y and Z directions
Linear six-stringbody coupled model Conservative model
Figs 6 and 7 then serve at checking other466
implementation details Simulations pertained to the467
string E2 constrained at xF = 02509L by a single ldquostiffrdquo468
fret modeled with mF=5 10minus3 Kg cF=1 Nsm and469
kF=106 Nm plucked in the oblique direction The470
modal energies E` of the subsystems ` shown in these471
plots were computed from the results of the simulations472
as follows in terms of the modal parameters and the473
8 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
0 02 04 06 08 10
05
1
15
[ J ]
10-3 STRINGS
Y-dir
Z-dir
0 02 04 06 08 10
05
1
15
[ J ]
10-5 FRET
Y-dir
Z-dir
0 02 04 06 08 10
05
1
[ J ]
10-4 BODY
0 02 04 06 08 1
Time [ s ]
0
05
1
15
10-3 FULL SYSTEM
Figure 7 Instantaneous modal energies Same plucking
conditions as in Fig 6 Linear six-stringbody coupled model
Non-conservative model
modal displacements and velocities474
E`(t) =
Nsumn=1
[1
2m`n(ω`n)2
[q`n(t)
]2+
1
2m`n
[q`n(t)
]2](50)
One important feature seen in Fig 6 is that the total475
energy remains roughly constant over a conservative476
simulation while energies flow back and forth between477
the subsystems thus giving confidence in the numerical478
model for simulations of reasonable duration For479
the non-conservative case shown in Fig 7 the large480
difference in energy decay for the two polarizations481
confirms the different stringbody coupling with the482
directions of string motion Note that both the body483
and the fret oscillator are dominated by periodic energy484
exchanges at the frequency of the played note as the wave485
travels back and forth between bridge and fret Finally486
looking closely at the modal energy of the fret one can487
see smaller superimposed periodic disturbances related488
to the reflections between fret and nut illustrating that489
energy still flows through the modeled fret490
Finally we compare in Fig 8 results by varying491
the excitation amplitude in order to illustrate the string492
nonlinearity The simulations were obtained for the493
string E2 using the experimentally determined value494
of 32 kN for the flexibility ES which is comparable495
to the value given by Chaigne31 Fig 8 shows the496
temporal variations of string tension with an estimate497
of the playing frequency computed from time domain498
evaluations of the zero crossing frequency The general499
features of large amplitude string motions are clearly500
visible with a quasi-static increase of tension and501
superimposed oscillations at twice the string fundamental502
frequency and a small frequency variation of the played503
note which compares well with measurements32504
1 2 3 4 5 6 7
716
718
72
722
724
T (
t )
[ N
] Fexc
= 1 N
Fexc
= 5 N
1 2 3 4 5 6 7
Time [ s ]
822
824
826
828
f (
t )
[ H
z ]
Figure 8 String tension (top) and instantaneous frequency of
string motion (bottom) Nonlinear rigid body model
2 Parameter study of the stringfinger coupling model505
We now illustrate the different strategies to account506
for a stopping finger on the fingerboard and explore the507
influence on the coupled system of the finger model508
In contrast with the approach in Ref 2 the strategy509
in terms of auxiliary oscillators rigidly coupled to the510
string seems more realistic and is undoubtedly more511
versatile allowing different types of contacts to be512
modeled by varying the oscillator parameters However513
one could expect the finger dynamics to perturb the514
coupled system with changes in frequency and damping515
for its modes similar to what observed for a multi-modal516
system coupled to a system exhibiting resonance To517
give some insights Figs 9-11 show the effects of varying518
the finger mechanical parameters Simulations pertain to519
the string E2 constrained at xF = 01857L by a finger of520
mass mf = 5 10minus3 Kg and ignore the body dynamics521
The first test consists in varying the stiffness of the522
constraint assuming no damping Fig 9 shows the523
frequency spectra of the coupling force at the rigid bridge524
where the peaks of the string modes can be seen clearly as525
well as the resonance frequency of the auxiliary oscillator526
which moves in frequency with the stiffness and slightly527
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 9
modulates the other peaks A close look shows that528
the spectra are affected much more sensitively below529
the resonance of the finger oscillator with amplitude530
variations analogous to a system driven in the stiffness-531
controlled region Decreasing the stiffness causes a rise532
in amplitude of the undesirable modes while increasing533
the stiffness lower their amplitudes As observed for the534
stringbody coupling33 inharmonicity is also created by535
the stiffness termination which in turn affects the tuning536
of the note With the increase of stiffness perturbations537
of the string modes are reduced and tuning is improved538
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
4 Nm
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
5 Nm
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
kf = 10
6 Nm
Figure 9 Spectra of the coupling force at the rigid bridge
Stiffness is varied F=1 c1 = 0 Nsm
Fig 10 then shows the effect of varying the damping539
keeping the stiffness constant Not surprisingly if540
damping increases the amplitude of the coupled541
modes close to the oscillator resonance decrease and542
the frequency region affected becomes wider This543
significantly affects the time envelope of the pluck note544
that becomes smoother and has a shorter-time decay545
Finally the effect of the number of contacts is studied546
using the rigid contact model which is convenient for547
this purpose Fig 11 gives a picture of the spatial548
distribution of the string energy by plotting the standard549
deviation of string motion as a function of its length550
As seen the ldquopassiverdquo region is affected with great551
sensitivity by the number of contacts and surprisingly552
increasing the contact points does not necessarily reduce553
the undesirable transfer of energy Computations show554
that the effectiveness of the multi-point rigid constraints555
approach is strongly problem-dependent and varies in a556
non-obvious way with several factors including the finger557
width the spacing between contact points the relative558
string length from the ends and the excitation spectrum559
Taken in combination these results and noticing560
that the perturbations induced by each finger oscillator561
are additive due to the linear coupling a robust and562
physically sound strategy to simulate the pressing finger563
would be to implement a set of flexible-dissipative564
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
cf = 1 Nsm
cf = 10 Nsm
0 1 2 3 4 5
Time [ s ]
-2
0
2
4
Y (
t )
10-3
Figure 10 Spectra of the coupling force at the rigid
bridge (up) and time history of the string motion (bottom)
Damping is varied F=1 k1 = 105 Nm
constraints imposing a first stiff constraint in order565
to ensure a precise tuning followed by several soft566
constraints that would control the waveform and spectral567
content of the decaying response depending on their568
number and damping properties For illustration569
the spectrograms of two plucks for different sets of570
parameters are presented in Fig 12571
0 01 02 03 04 05 06
x [ m ]
10-5
10-2
ST
D (
Y(t
) )
F=1
F=2
F=3
F=5
016 017 018 019 02
x [ m ]
10-5
ST
D (
Y(t
) )
Figure 11 Standard deviation of the string motion Number
of contacts is varied Vertical dotted lines represent the
constraints locations Finger width (0015m) is kept constant
C Simulated acoustic radiation572
As an alternative to computing sounds from full573
vibro-acoustic models that usually require the use of574
massive computational resources it is tempting to apply575
10 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
Figure 12 Spectrograms of the force at the rigid bridge Up
F = 3 bottom F = 5 Parameter values kF1=106 Nm
cF1=1 Nsm and kFj =104 Nm cFj =10 Nsm (j = 2 F )
Constraint at xF = 01857L (sound files available)
simpler techniques that could be efficient to generate576
convincing sounds from the vibratory responses The577
method used here is to treat the guitar body as a linear578
input-output system and to compute the pressure by579
convolution benefiting from the local excitation of the580
body by the strings and using a vibro-acoustic impulse581
response of the test guitar This approach does not by582
any means simulate the sound response everywhere in the583
pressure field but it has the merit of adding the influence584
of the radiating properties of the body at a very low585
computational cost and of providing sounds of greater586
perceptual quality than the commonly used vibratory587
signals at the bridge The pressure at location rp can be588
computed by the convolution of the time-history of the589
stringbridge interaction force FY (rb t) and the vibro-590
acoustic impulse response function hY (rb rp t) between591
the bridge and microphone by592
p(rp t) = hY (rb rp t) lowast FY (rb t) (51)
The extension to account for the effects of all the593
strings is readily obtained by linear superposition594
principle summing the contributions from each string595
Measurements of the vibro-acoustic impulse response596
were performed under the same experimental conditions597
as for transfer function measurements using an598
additional condenser microphone Impulsive excitation599
was applied in the perpendicular direction to the600
soundboard and the acoustic pressure was measured601
at a distance of 1 m from the central axis of the602
guitar in a quiet room with all strings damped (see603
Fig 1) The impulse response was computed by inverse604
Fourier transform of the complex transfer function built605
from the force and response signals described in the606
frequency domain Results in Fig 13 pertain to the607
simulation of a G major arpeggio obtained with the608
two lowest and the top strings constrained using five609
auxiliary flexible-inertial constraints on each string The610
effect of the radiation properties of the body on the611
computed vibratory signals is supported when comparing612
the spectrograms in Fig 13 which clearly shows the613
spectral filtering of the vibration signal614
Figure 13 Spectrograms of the bridge force (top) and sound
pressure (bottom) Nonlinear six-stringbody coupled model
Arpeggio of a G major chord with three strings stopped using
a set of auxiliary oscillators (sound files available)
V CONCLUSIONS615
Following our recent work showing the potential616
of the modal Udwadia-Kalaba equations to physically617
simulate musical instruments we presented in this paper618
a computational approach that allows efficient time-619
domain simulations of realistic guitar plucks In this620
work we further generalized our simple stringbody621
coupled model toward the development of a more622
realistic guitar model by including the complete set623
of strings the two polarizations of string motions the624
string geometric nonlinearity the stringfret interaction625
that controls the playing frequency and the body626
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 11
dynamics of an actual instrument The dynamics of627
the various subsystems were formulated in terms of628
their unconstrained modes and coupling were achieved629
by enforcing several constraint conditions One of the630
developments of the model was the introduction of631
flexible-dissipative-inertial supplementary equations as632
an alternative to the common kinematical constraints633
which here improve the role of the stringfingerfret634
interaction at the fingerboard This approach can635
also benefit the modeling of other complex dynamical636
systems outside the realm of physical instrument637
modeling To improve the realism of the synthesis638
we also included crudely the radiation properties of639
the instrument body into the computational model640
through an ad hoc procedure Sound pressure was641
computed adopting a hybrid approach by convoluting642
the computed bridge force signals with a vibro-acoustic643
impulse response measured on the test guitar Such644
vibro-acoustic response can be measured in an anechoic645
chamber or else include the acoustics of any specific646
room This approach appeared not only straightforward647
to compute realistic sounds of string instruments for648
which the radiating components are subjected to point-649
wise excitation but also very efficient compared to heavy650
computational sound radiation techniques651
ACKNOWLEDGMENTS652
The authors thanks Paulo Vaz de Carvalho for653
lending the guitar and interesting discussions This work654
is funded by national funds through the FCT ndash Fundacao655
para a Ciencia e a Tecnologia IP under the Norma656
Transitoria - DL 572016CP1453CT0007657
REFERENCES658
1JH Ginsberg Mechanical and Structural Vibrations Theory659
and Applications chapter 9 Wiley 2001660
2J Antunes and V Debut Dynamical computation of constrained661
flexible systems using a modal Udwadia-Kalaba formulation662
Application to musical instruments J Acoust Soc Am663
141764 2017664
3J Antunes V Debut L Borsoi X Delaune and P Piteau665
A modal Udwadia-Kalaba formulation for the vibro-impact666
modelling of continuous flexible systems with intermittent667
contacts Procedia Engineering 199322ndash329 2017668
4FE Udwadia and RE Kalaba Analytical dynamics - A new669
approach Cambridge University Press 1996670
5C Valette The mechanics of the vibrating string Mechanics of671
musical instruments Springer 1995672
6J Woodhouse On the synthesis of guitar plucks Acta Acust673
united Ac 90928ndash944 2004674
7M E McIntyre RT Schumacher and J Woodhouse On the675
oscillations of musical instruments Journal of the Acoustical676
Society of America 741325ndash1345 1983677
8A Chaigne On the use of finite differences for musical678
synthesis Application to plucked stringed instruments Journal679
drsquoAcoustique 5181ndash211 1992680
9JM Adrien The missing link Modal synthesis In G DePoli681
A Picalli and C Roads editors Representations of musical682
sounds MIT Press Cambridge 1991683
10S Bilbao Numerical Sound Synthesis Finite Difference684
Schemes and Simulation in Musical Acoustics Wiley Publishing685
2009686
11L Trautmann and R Rabenstein Multirate simulations687
of string vibrations includingnonlinear fret-string interactions688
usingthe functional transformation method EURASIP Journal689
on Applied Signal Processing 7949ndash963 2004690
12V Chatziioannou and M van Walstijn Energy conserving691
schemes for the simulation of musical instrument contact692
dynamics J Sound Vib 339262ndash279 2015693
13S Bilbao and A Torin Numerical modeling and sound synthesis694
for articulated stringfretboard interactions Acta Acust united695
Ac 63336ndash347 2015696
14V Debut M Carvalho M Marques and J Antunes Physics-697
based modeling techniques of a twelve-string portuguese guitar698
a three-dimensional non-linear time-domain computational699
approach for the multiple-stringsbridgesoundboard coupled700
dynamics Appl Acoust 1083ndash18 2016701
15M van Walstijn and J Bridges Simulation of distributed contact702
in string instruments A modal expansion approach In 24th703
European Signal Processing Conference (Eusipco 2016) pages704
1023ndash1027 2013705
16C Issanchou S Bilbao JL Le Carrou C Touze and O Doare706
A modal-based approach to the nonlinear vibration of strings707
against a unilateral obstacle simulations and experiments in the708
pointwise case J Sound Vib 393229ndash251 2017709
17R Bader Computational Mechanics of the Classical Guitar710
Springer-Verlag Berlin Heidelberg 2005711
18H Mansour J Woodhouse and G Scavone Enhanced wave-712
based modelling of musical strings Part 1 Plucked strings Acta713
Acust united Ac 1021094ndash1107 2016714
19O Inacio J Antunes and MCM Wright Computational715
modeling of string-body interaction for the violin family and716
simulation of wolf notes J Sound Vib 310260ndash286 2007717
20S Bilbao A Torin and V Chatziioannou Numerical modeling718
of collisions in musical instruments Acta Acust united Ac719
101155ndash173 2015720
21C Issanchou JL Le Carrou C Touze B Fabre and O Doare721
Stringfrets contacts in the electric bass sound Simulations and722
experiments App Acoust 129217ndash228 2018723
22D J Braun and M Goldfarb Eliminating constraint drift in the724
numerical simulation of constrained dynamical systems Comput725
Methods in Appl Mech Eng 198(37)3151 ndash 3160 2009726
23GF Carrier On the nonlinear vibration problem of the elastic727
string Q Appl Math 3157ndash165 1945728
24G Kirchhoff Vorlesungen uber Mechanik Teubner 1883729
25J Pakarinen V Valimaki and M Karjalainen Physics-based730
methods for modeling nonlinear vibrating strings Acta Acust731
united Ac 91312ndash325 2005732
26PMC Morse and KU Ingard Theoretical Acoustics733
International series in pure and applied physics Princeton734
University Press 1968735
27J Juang Applied System Identification PTR Prentice-Hall736
Inc New Jersey 1994737
28B E Richardson The acoustical development of the guitar J738
Catgut Acoust Soc 51ndash10 1994739
29J Woodhouse Plucked guitar transients Comparison of740
measurements and synthesis Acta Acust united Ac 90945ndash741
965 2004742
30M Tuckerman B J Berne and G J Martyna Reversible743
multiple time scale molecular dynamics J Chem Phys744
97(3)1990ndash2001 1992745
31A Chaigne Viscoelastic properties of nylon guitar strings J746
Catgut Acoust Soc Series II 121ndash27 1991747
32T Tolonen V Valimaki and M Karjalainen Modeling of748
tension modulation nonlinearity in plucked strings IEEE Trans749
Speech Audio Process 8(3)300ndash310 2000750
33A Chaigne and J Kergomard Acoustics of Musical Instruments751
Modern Acoustics and Signal Processing Springer New York752
2016753
12 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
754
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 13
where the modeshape vectors are written as
ΦYs
b =φYs1 (xb) φ
Ys
NYs(xb)
T(23a)
ΦZs
b =φZs1 (xb) φ
Zs
NZs(xb)
T(23b)
ΦBrs =φB1 (rs) φ
BNB
(rs)T
(23c)
Defining a vector of modal coordinates qqq built by stacking320
the individual modal displacements of the subsystems ie321
the strings including both polarizations and the body322
qqq = qqqY qqqZ qqqBT (24)
where qqqY and qqqZ are partitioned into terms associatedwith the different strings as
qqqY =qqqY1T qqqY6T
T(25a)
qqqZ =qqqZ1T qqqZ6T
T(25b)
and323
qqqB = qB1 qBNBT (26)
the constraint equations fit the matrix form324
AAAb qqq = 000 (27)
with the stringbody constraint matrix AAAb written as325
AAAb =
[ΦΦΦYb 000 minusΦΦΦB
000 ΦΦΦZb 000
](28)
where ΦΦΦYb and ΦΦΦZb are matrices of the modeshape vectors326
of the strings taken at the bridge location xb built as327
ΦΦΦYb =
ΦY1
b
T000 000
000
ΦY2
b
T 000
000 000
ΦY6
b
T
(29)
and similarly for the perpendicular direction and ΦΦΦB328
stands for the modeshapes of the body at the stringbody329
contact points rs330
ΦΦΦB =
ΦBr1T
ΦBr6T (30)
2 Stringfinger coupling at the fingerboard331
Modeling a stopping string for playing different332
notes can be thought in several ways according to the333
type of sound to simulate as well as to the degree of334
simplicitycomplexity aimed for the model The simplest335
model was taken in Ref 2 imposing a rigid kinematical336
constraint at a single location If such model could affect337
the tuning of the string effectively it is nonphysical338
in that part of the vibrational energy flows beyond the339
contact point (see Ref 2 for discussion) This simple340
approach is here extended to allow for multiple-point341
rigid coupling or for a non-rigid interaction as shown in342
Fig 4343
a Multiple-point rigid coupling This first optiondirectly enters into the realm of the standard kinematicalconstraints common in multibody theory The stringmotions Ys and Zs at the multiple finger locations xFf
(f = 1 F ) should be nil at all times so that
Ys(xFf t)=0 rArr
ΦYs
Ff
TqqqYs(t) = 0 (31a)
Zs(xFf t)=0 rArr
ΦZs
Ff
TqqqZs(t) = 0 (31b)
where the modeshape vectors at the finger locations are
ΦYs
Ff=φYs1 (xFf
) φYs
NYs(xFf
)T
(32a)
ΦZs
Ff=φZs1 (xFf
) φZs
NYs(xFf
)T
(32b)
Eq (31) can be rearranged in a matrix form as344
AAAf qqq = 000 (33)
where the modal constrained matrix is345
AAAf =
[ΦΦΦYs
F 000
000 ΦΦΦZs
F
](34)
with346
ΦΦΦYs
F =
ΦYs
F1
T
ΦYs
FF
T ΦΦΦZs
F =
ΦZs
F1
T
ΦZs
FF
T (35)
b Non-rigid coupling To extend the model to347
allow for non-rigid contact requires a little more care348
since a flexible-dissipative-inertial model for coupling349
subsystems cannot be directly formulated as kinematic350
constraints Nevertheless they can be implemented351
by assigning one or more additional flexible-dissipative-352
inertial elementary subsystems which are constrained to353
follow the string motions at the finger location(s) exactly354
as the body modes are included This modeling strategy355
somehow creates a bridge between computationally356
efficient multibody modeling providing ldquorigidrdquo kinematic357
constraints and the versatile but computationally slower358
modeling strategy through penalty-based constraints359
Non-rigid connections on the fingerboard cantherefore be obtained by extending the guitar modelusing a set of mass-spring-damper oscillators at locationsxFf
(f = 1 F ) and enforcing coupling as
Ys(xFf t)minusY Ff
f (t)= 0 (36a)
Zs(xFf t)minusZFf
f (t)= 0 (36b)
where YFf
f and ZFf
f are the responses of the fingercontact points in the two perpendicular directionsReturning to Eq (11) the vector and matrices mustnow be partitioned as
qqqu =qqqYsu T qqqZs
u T qqqFsu T
T(37a)
MMM = diag(MMMYsMMMZsMMMFs
)(37b)
CCC = diag(CCCYsCCCZsCCCFs
)(37c)
KKK = diag(KKKYsKKKZsKKKFs
)(37d)
6 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
where the vectors and matrices for the strings are as360
beforeMMMFs CCCFs andKKKFs are block diagonal matrices of361
the inertia damping and stiffness properties associated362
with the ldquofingerrdquo subsystem acting on string s including363
both directions of string motions364
qqqFsu =
Y F1
f Y FF
f ZF1
f ZFF
f
T(38)
In terms of modal coordinates Eqs (36) becomeΦYs
Ff
TqqqYs(t)minus Y Ff
f (t) = 0 (39a)ΦZs
Ff
TqqqZs(t)minus ZFf
f (t) = 0 (39b)
which can be rearranged as365
AAAf qqq = 000 (40)
with the modal constrained matrix AAAf written as366
AAAf =
[ΦΦΦYF 000 minusI 000
000 ΦΦΦZF 000 minusI
](41)
Figure 4 Rigid kinematical constraints (left) and flexible-
dissipative-inertial constraints (right) at multiple locations
D Dynamical formulation of the fully coupled model367
The formulation (11) readily applies to the full guitar368
model including the non-rigid constraint model for the369
stringfret interaction by stacking the unconstrained370
modal displacements into a vector371
qqqu = qqqYu qqqZu qqqBu qqqFu T (42)
where qqqYu qqqZu and qqqBu are as before and372
qqqFu=qqqY1
ufT qqqY6
ufT qqqZ1
ufT qqqZ6
ufTT
(43)
In view of the definition of qqq the assembled inertia MMM373
stiffnessKKK and dampingCCC matrices which are uncoupled374
are given by375
MMM=
MMMY 000 000 000
000 MMMZ 000 000
000 000 MMMB 000
000 000 000 MMMF
CCC=
CCCY 000 000 000
000 CCCZ 000 000
000 000 CCCB 000
000 000 000 CCCF
KKK=
KKKY 000 000 000
000 KKKZ 000 000
000 000 KKKB 000
000 000 000 KKKF
(44)
with MMM ` KKK` and CCC` the inertia stiffness and dampingmatrices for each individual subsystem ` and the vectorsFFFext and FFFnl write
FFFext =FFFYextT FFFZextT000000
T(45a)
FFFnl =FFFYnlT FFFZnlT000000
T(45b)
where FFFYext and FFFZext correspond to the plucking action376
of the player expressed for both planes of polarization377
and FFFYnl and FFFZnl are the modal vectors associated to the378
string nonlinear effects with entries given by Eq (20)379
The form of the modal constraint equation is now380
AAAqqq = 000 (46)
with381
AAA =
[AAAbAAAf
]=
ΦΦΦYb 000 minusΦΦΦB 000 000 000
000 ΦΦΦZb 000 000 000 000
ΦYFΦYFΦYF 000 000 000 minusFFFY 000
000 ΦZFΦZFΦZF 000 000 000 minusFFFZ
(47)
where ΦYFΦYFΦYF and ΦZFΦZFΦZF are block matrices with entries given382
by Eq (35) and FFFY and FFFZ are block diagonal matrices383
built with either the identity or null matrix depending on384
the constraints enforced on each string The form of (47)385
shows that the interface forces and constraint conditions386
couple the Y -motions of the strings and the body as well387
as coupling between the finger and the strings in each388
direction of motions Notice that coupling between both389
polarizations of string motions is not apparent in (47)390
although it is formulated in the nonlinear forces (45b)391
IV ILLUSTRATIVE COMPUTATIONS392
A System parameters393
For given initial conditions the system of equations394
comprising (5) with the unconstrained accelerations (11)395
and the constraint matrix (15) and using the specific396
matrices given by Eqs (44) and (47) can be integrated in397
time to produce the transient response The time-domain398
simulations were performed based on the identified body399
modal data and using standard string properties of400
classical guitars Assuming non-ideal strings with fixed-401
free ends the modeshapes are given by Eq (19) and402
their modal frequencies can be approximated by403
fYn = fZn =cT2πpn
(1 +
B
2T0p2n
)(48)
where cT is the velocity of transverse waves B is the404
bending stiffness of the string and T0 its tensioning405
pn = (2n minus 1)π2L and n the mode number Modal406
damping values are introduced following the pragmatic407
formulation given in Ref 6408
ζYn = ζZn =1
2
T0(ηF + ηAωn
)+ηBBp
2n
T0 +Bp2n(49)
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 7
where ηF ηA and ηB are the loss coefficients related to409
ldquointernal frictionrdquo ldquoair viscous dampingrdquo and ldquobending410
dampingrdquo The length of the strings is L=065 m411
and their density bending stiffness tensioning and loss412
coefficients are taken from Ref 29 The size of the modal413
basis of each string is defined by the maximum frequency414
of the computations (10000 Hz) which proved a sensible415
compromise for convergence to realistic simulations416
resulting in 800 modes for the set of strings To417
obtain suitable values for the elasticity parameter ES418
we perform simple vibratory tests on string samples419
Stretching a string vertically and suspending a mass to420
its free end one can easily measure the longitudinal421
vibration of the string-mass oscillator and obtain its422
natural frequency Assuming the string acting as a423
spring with no mass and reminding that the natural424
frequency of the longitudinal vibration is given by f =425
12π
radicESML then an estimate of ES is obtained The426
simulation results were obtained by time-step integration427
of the ordinary modal equations using the velocity-428
Verlet numerical scheme30 an integrator that combines429
simplicity second order accuracy with good numerical430
stability - see implementation details in Ref 2 We431
adopted a time-step of 10minus5s and assumed all initial432
modal displacements and velocities nil initially433
B Computational results434
Pluck excitations were simulated by considering a435
point-wise external force at xe=09L with a linear force436
ramp applied to one or more strings during 10 ms and437
by assuming null excitation force at subsequent instants438
1 Dynamical behaviour of the fully coupled model439
Results in Fig 5 were obtained for the open string440
E2 excited in the direction normal to the soundboard441
first considering a rigidly mounted string then coupling442
one string and finally all six strings to the instrument443
body Assuming no dynamics for the body the response444
spectra of string motion at the bridge is virtually nil445
according to the pinned boundary condition When446
coupling the instrument body to the string the response447
becomes obviously different and shows the classical series448
of near harmonic string partials together with smaller449
peaks from the body resonances While the first modal450
frequency of the unconstrained string is 41 Hz a close451
look at Fig 5 shows that the constraint at the bridge452
yields the fundamental frequency for the tuned string453
with pinned ends (f = 82 Hz) thus confirming the454
correct implementation of the bridge coupling Coupling455
the other five strings results in new peaks in the456
spectrum extending over the frequency range of the body457
modes and with an amplitude difference of about 100 dB458
compared to the most excited modes which clearly relate459
to the non-excited strings As found in some stringed460
instruments this dynamical coupling of all the strings via461
the bridge can lead to sympathetic vibrations which can462
further results in slight interference beats in the sound463
due to the inharmonicity produced by the bridge coupling464
and of the string partials465
0 200 400 600 800 1000 1200 1400 1600 180010
-40
10-20
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 180010
-20
10-10
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 1800
Frequency [ Hz ]
10-20
10-10
100
| Y
6(f
) |
Figure 5 Response spectra of string motion at the bridge
Top rigidly mounted string middle single stringbody
coupling bottom six-stringbody coupled model Force
amplitude 5 N Linear model
0 02 04 06 08 1
Time [ s ]
0
05
1
15
2
Mo
dal en
erg
y [
J ]
10-3
STRINGSFRETBODYTOTAL
Figure 6 Instantaneous modal energies Oblique excitation
with amplitudes of 2 N and 1 N in the Y and Z directions
Linear six-stringbody coupled model Conservative model
Figs 6 and 7 then serve at checking other466
implementation details Simulations pertained to the467
string E2 constrained at xF = 02509L by a single ldquostiffrdquo468
fret modeled with mF=5 10minus3 Kg cF=1 Nsm and469
kF=106 Nm plucked in the oblique direction The470
modal energies E` of the subsystems ` shown in these471
plots were computed from the results of the simulations472
as follows in terms of the modal parameters and the473
8 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
0 02 04 06 08 10
05
1
15
[ J ]
10-3 STRINGS
Y-dir
Z-dir
0 02 04 06 08 10
05
1
15
[ J ]
10-5 FRET
Y-dir
Z-dir
0 02 04 06 08 10
05
1
[ J ]
10-4 BODY
0 02 04 06 08 1
Time [ s ]
0
05
1
15
10-3 FULL SYSTEM
Figure 7 Instantaneous modal energies Same plucking
conditions as in Fig 6 Linear six-stringbody coupled model
Non-conservative model
modal displacements and velocities474
E`(t) =
Nsumn=1
[1
2m`n(ω`n)2
[q`n(t)
]2+
1
2m`n
[q`n(t)
]2](50)
One important feature seen in Fig 6 is that the total475
energy remains roughly constant over a conservative476
simulation while energies flow back and forth between477
the subsystems thus giving confidence in the numerical478
model for simulations of reasonable duration For479
the non-conservative case shown in Fig 7 the large480
difference in energy decay for the two polarizations481
confirms the different stringbody coupling with the482
directions of string motion Note that both the body483
and the fret oscillator are dominated by periodic energy484
exchanges at the frequency of the played note as the wave485
travels back and forth between bridge and fret Finally486
looking closely at the modal energy of the fret one can487
see smaller superimposed periodic disturbances related488
to the reflections between fret and nut illustrating that489
energy still flows through the modeled fret490
Finally we compare in Fig 8 results by varying491
the excitation amplitude in order to illustrate the string492
nonlinearity The simulations were obtained for the493
string E2 using the experimentally determined value494
of 32 kN for the flexibility ES which is comparable495
to the value given by Chaigne31 Fig 8 shows the496
temporal variations of string tension with an estimate497
of the playing frequency computed from time domain498
evaluations of the zero crossing frequency The general499
features of large amplitude string motions are clearly500
visible with a quasi-static increase of tension and501
superimposed oscillations at twice the string fundamental502
frequency and a small frequency variation of the played503
note which compares well with measurements32504
1 2 3 4 5 6 7
716
718
72
722
724
T (
t )
[ N
] Fexc
= 1 N
Fexc
= 5 N
1 2 3 4 5 6 7
Time [ s ]
822
824
826
828
f (
t )
[ H
z ]
Figure 8 String tension (top) and instantaneous frequency of
string motion (bottom) Nonlinear rigid body model
2 Parameter study of the stringfinger coupling model505
We now illustrate the different strategies to account506
for a stopping finger on the fingerboard and explore the507
influence on the coupled system of the finger model508
In contrast with the approach in Ref 2 the strategy509
in terms of auxiliary oscillators rigidly coupled to the510
string seems more realistic and is undoubtedly more511
versatile allowing different types of contacts to be512
modeled by varying the oscillator parameters However513
one could expect the finger dynamics to perturb the514
coupled system with changes in frequency and damping515
for its modes similar to what observed for a multi-modal516
system coupled to a system exhibiting resonance To517
give some insights Figs 9-11 show the effects of varying518
the finger mechanical parameters Simulations pertain to519
the string E2 constrained at xF = 01857L by a finger of520
mass mf = 5 10minus3 Kg and ignore the body dynamics521
The first test consists in varying the stiffness of the522
constraint assuming no damping Fig 9 shows the523
frequency spectra of the coupling force at the rigid bridge524
where the peaks of the string modes can be seen clearly as525
well as the resonance frequency of the auxiliary oscillator526
which moves in frequency with the stiffness and slightly527
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 9
modulates the other peaks A close look shows that528
the spectra are affected much more sensitively below529
the resonance of the finger oscillator with amplitude530
variations analogous to a system driven in the stiffness-531
controlled region Decreasing the stiffness causes a rise532
in amplitude of the undesirable modes while increasing533
the stiffness lower their amplitudes As observed for the534
stringbody coupling33 inharmonicity is also created by535
the stiffness termination which in turn affects the tuning536
of the note With the increase of stiffness perturbations537
of the string modes are reduced and tuning is improved538
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
4 Nm
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
5 Nm
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
kf = 10
6 Nm
Figure 9 Spectra of the coupling force at the rigid bridge
Stiffness is varied F=1 c1 = 0 Nsm
Fig 10 then shows the effect of varying the damping539
keeping the stiffness constant Not surprisingly if540
damping increases the amplitude of the coupled541
modes close to the oscillator resonance decrease and542
the frequency region affected becomes wider This543
significantly affects the time envelope of the pluck note544
that becomes smoother and has a shorter-time decay545
Finally the effect of the number of contacts is studied546
using the rigid contact model which is convenient for547
this purpose Fig 11 gives a picture of the spatial548
distribution of the string energy by plotting the standard549
deviation of string motion as a function of its length550
As seen the ldquopassiverdquo region is affected with great551
sensitivity by the number of contacts and surprisingly552
increasing the contact points does not necessarily reduce553
the undesirable transfer of energy Computations show554
that the effectiveness of the multi-point rigid constraints555
approach is strongly problem-dependent and varies in a556
non-obvious way with several factors including the finger557
width the spacing between contact points the relative558
string length from the ends and the excitation spectrum559
Taken in combination these results and noticing560
that the perturbations induced by each finger oscillator561
are additive due to the linear coupling a robust and562
physically sound strategy to simulate the pressing finger563
would be to implement a set of flexible-dissipative564
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
cf = 1 Nsm
cf = 10 Nsm
0 1 2 3 4 5
Time [ s ]
-2
0
2
4
Y (
t )
10-3
Figure 10 Spectra of the coupling force at the rigid
bridge (up) and time history of the string motion (bottom)
Damping is varied F=1 k1 = 105 Nm
constraints imposing a first stiff constraint in order565
to ensure a precise tuning followed by several soft566
constraints that would control the waveform and spectral567
content of the decaying response depending on their568
number and damping properties For illustration569
the spectrograms of two plucks for different sets of570
parameters are presented in Fig 12571
0 01 02 03 04 05 06
x [ m ]
10-5
10-2
ST
D (
Y(t
) )
F=1
F=2
F=3
F=5
016 017 018 019 02
x [ m ]
10-5
ST
D (
Y(t
) )
Figure 11 Standard deviation of the string motion Number
of contacts is varied Vertical dotted lines represent the
constraints locations Finger width (0015m) is kept constant
C Simulated acoustic radiation572
As an alternative to computing sounds from full573
vibro-acoustic models that usually require the use of574
massive computational resources it is tempting to apply575
10 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
Figure 12 Spectrograms of the force at the rigid bridge Up
F = 3 bottom F = 5 Parameter values kF1=106 Nm
cF1=1 Nsm and kFj =104 Nm cFj =10 Nsm (j = 2 F )
Constraint at xF = 01857L (sound files available)
simpler techniques that could be efficient to generate576
convincing sounds from the vibratory responses The577
method used here is to treat the guitar body as a linear578
input-output system and to compute the pressure by579
convolution benefiting from the local excitation of the580
body by the strings and using a vibro-acoustic impulse581
response of the test guitar This approach does not by582
any means simulate the sound response everywhere in the583
pressure field but it has the merit of adding the influence584
of the radiating properties of the body at a very low585
computational cost and of providing sounds of greater586
perceptual quality than the commonly used vibratory587
signals at the bridge The pressure at location rp can be588
computed by the convolution of the time-history of the589
stringbridge interaction force FY (rb t) and the vibro-590
acoustic impulse response function hY (rb rp t) between591
the bridge and microphone by592
p(rp t) = hY (rb rp t) lowast FY (rb t) (51)
The extension to account for the effects of all the593
strings is readily obtained by linear superposition594
principle summing the contributions from each string595
Measurements of the vibro-acoustic impulse response596
were performed under the same experimental conditions597
as for transfer function measurements using an598
additional condenser microphone Impulsive excitation599
was applied in the perpendicular direction to the600
soundboard and the acoustic pressure was measured601
at a distance of 1 m from the central axis of the602
guitar in a quiet room with all strings damped (see603
Fig 1) The impulse response was computed by inverse604
Fourier transform of the complex transfer function built605
from the force and response signals described in the606
frequency domain Results in Fig 13 pertain to the607
simulation of a G major arpeggio obtained with the608
two lowest and the top strings constrained using five609
auxiliary flexible-inertial constraints on each string The610
effect of the radiation properties of the body on the611
computed vibratory signals is supported when comparing612
the spectrograms in Fig 13 which clearly shows the613
spectral filtering of the vibration signal614
Figure 13 Spectrograms of the bridge force (top) and sound
pressure (bottom) Nonlinear six-stringbody coupled model
Arpeggio of a G major chord with three strings stopped using
a set of auxiliary oscillators (sound files available)
V CONCLUSIONS615
Following our recent work showing the potential616
of the modal Udwadia-Kalaba equations to physically617
simulate musical instruments we presented in this paper618
a computational approach that allows efficient time-619
domain simulations of realistic guitar plucks In this620
work we further generalized our simple stringbody621
coupled model toward the development of a more622
realistic guitar model by including the complete set623
of strings the two polarizations of string motions the624
string geometric nonlinearity the stringfret interaction625
that controls the playing frequency and the body626
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 11
dynamics of an actual instrument The dynamics of627
the various subsystems were formulated in terms of628
their unconstrained modes and coupling were achieved629
by enforcing several constraint conditions One of the630
developments of the model was the introduction of631
flexible-dissipative-inertial supplementary equations as632
an alternative to the common kinematical constraints633
which here improve the role of the stringfingerfret634
interaction at the fingerboard This approach can635
also benefit the modeling of other complex dynamical636
systems outside the realm of physical instrument637
modeling To improve the realism of the synthesis638
we also included crudely the radiation properties of639
the instrument body into the computational model640
through an ad hoc procedure Sound pressure was641
computed adopting a hybrid approach by convoluting642
the computed bridge force signals with a vibro-acoustic643
impulse response measured on the test guitar Such644
vibro-acoustic response can be measured in an anechoic645
chamber or else include the acoustics of any specific646
room This approach appeared not only straightforward647
to compute realistic sounds of string instruments for648
which the radiating components are subjected to point-649
wise excitation but also very efficient compared to heavy650
computational sound radiation techniques651
ACKNOWLEDGMENTS652
The authors thanks Paulo Vaz de Carvalho for653
lending the guitar and interesting discussions This work654
is funded by national funds through the FCT ndash Fundacao655
para a Ciencia e a Tecnologia IP under the Norma656
Transitoria - DL 572016CP1453CT0007657
REFERENCES658
1JH Ginsberg Mechanical and Structural Vibrations Theory659
and Applications chapter 9 Wiley 2001660
2J Antunes and V Debut Dynamical computation of constrained661
flexible systems using a modal Udwadia-Kalaba formulation662
Application to musical instruments J Acoust Soc Am663
141764 2017664
3J Antunes V Debut L Borsoi X Delaune and P Piteau665
A modal Udwadia-Kalaba formulation for the vibro-impact666
modelling of continuous flexible systems with intermittent667
contacts Procedia Engineering 199322ndash329 2017668
4FE Udwadia and RE Kalaba Analytical dynamics - A new669
approach Cambridge University Press 1996670
5C Valette The mechanics of the vibrating string Mechanics of671
musical instruments Springer 1995672
6J Woodhouse On the synthesis of guitar plucks Acta Acust673
united Ac 90928ndash944 2004674
7M E McIntyre RT Schumacher and J Woodhouse On the675
oscillations of musical instruments Journal of the Acoustical676
Society of America 741325ndash1345 1983677
8A Chaigne On the use of finite differences for musical678
synthesis Application to plucked stringed instruments Journal679
drsquoAcoustique 5181ndash211 1992680
9JM Adrien The missing link Modal synthesis In G DePoli681
A Picalli and C Roads editors Representations of musical682
sounds MIT Press Cambridge 1991683
10S Bilbao Numerical Sound Synthesis Finite Difference684
Schemes and Simulation in Musical Acoustics Wiley Publishing685
2009686
11L Trautmann and R Rabenstein Multirate simulations687
of string vibrations includingnonlinear fret-string interactions688
usingthe functional transformation method EURASIP Journal689
on Applied Signal Processing 7949ndash963 2004690
12V Chatziioannou and M van Walstijn Energy conserving691
schemes for the simulation of musical instrument contact692
dynamics J Sound Vib 339262ndash279 2015693
13S Bilbao and A Torin Numerical modeling and sound synthesis694
for articulated stringfretboard interactions Acta Acust united695
Ac 63336ndash347 2015696
14V Debut M Carvalho M Marques and J Antunes Physics-697
based modeling techniques of a twelve-string portuguese guitar698
a three-dimensional non-linear time-domain computational699
approach for the multiple-stringsbridgesoundboard coupled700
dynamics Appl Acoust 1083ndash18 2016701
15M van Walstijn and J Bridges Simulation of distributed contact702
in string instruments A modal expansion approach In 24th703
European Signal Processing Conference (Eusipco 2016) pages704
1023ndash1027 2013705
16C Issanchou S Bilbao JL Le Carrou C Touze and O Doare706
A modal-based approach to the nonlinear vibration of strings707
against a unilateral obstacle simulations and experiments in the708
pointwise case J Sound Vib 393229ndash251 2017709
17R Bader Computational Mechanics of the Classical Guitar710
Springer-Verlag Berlin Heidelberg 2005711
18H Mansour J Woodhouse and G Scavone Enhanced wave-712
based modelling of musical strings Part 1 Plucked strings Acta713
Acust united Ac 1021094ndash1107 2016714
19O Inacio J Antunes and MCM Wright Computational715
modeling of string-body interaction for the violin family and716
simulation of wolf notes J Sound Vib 310260ndash286 2007717
20S Bilbao A Torin and V Chatziioannou Numerical modeling718
of collisions in musical instruments Acta Acust united Ac719
101155ndash173 2015720
21C Issanchou JL Le Carrou C Touze B Fabre and O Doare721
Stringfrets contacts in the electric bass sound Simulations and722
experiments App Acoust 129217ndash228 2018723
22D J Braun and M Goldfarb Eliminating constraint drift in the724
numerical simulation of constrained dynamical systems Comput725
Methods in Appl Mech Eng 198(37)3151 ndash 3160 2009726
23GF Carrier On the nonlinear vibration problem of the elastic727
string Q Appl Math 3157ndash165 1945728
24G Kirchhoff Vorlesungen uber Mechanik Teubner 1883729
25J Pakarinen V Valimaki and M Karjalainen Physics-based730
methods for modeling nonlinear vibrating strings Acta Acust731
united Ac 91312ndash325 2005732
26PMC Morse and KU Ingard Theoretical Acoustics733
International series in pure and applied physics Princeton734
University Press 1968735
27J Juang Applied System Identification PTR Prentice-Hall736
Inc New Jersey 1994737
28B E Richardson The acoustical development of the guitar J738
Catgut Acoust Soc 51ndash10 1994739
29J Woodhouse Plucked guitar transients Comparison of740
measurements and synthesis Acta Acust united Ac 90945ndash741
965 2004742
30M Tuckerman B J Berne and G J Martyna Reversible743
multiple time scale molecular dynamics J Chem Phys744
97(3)1990ndash2001 1992745
31A Chaigne Viscoelastic properties of nylon guitar strings J746
Catgut Acoust Soc Series II 121ndash27 1991747
32T Tolonen V Valimaki and M Karjalainen Modeling of748
tension modulation nonlinearity in plucked strings IEEE Trans749
Speech Audio Process 8(3)300ndash310 2000750
33A Chaigne and J Kergomard Acoustics of Musical Instruments751
Modern Acoustics and Signal Processing Springer New York752
2016753
12 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
754
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 13
where the vectors and matrices for the strings are as360
beforeMMMFs CCCFs andKKKFs are block diagonal matrices of361
the inertia damping and stiffness properties associated362
with the ldquofingerrdquo subsystem acting on string s including363
both directions of string motions364
qqqFsu =
Y F1
f Y FF
f ZF1
f ZFF
f
T(38)
In terms of modal coordinates Eqs (36) becomeΦYs
Ff
TqqqYs(t)minus Y Ff
f (t) = 0 (39a)ΦZs
Ff
TqqqZs(t)minus ZFf
f (t) = 0 (39b)
which can be rearranged as365
AAAf qqq = 000 (40)
with the modal constrained matrix AAAf written as366
AAAf =
[ΦΦΦYF 000 minusI 000
000 ΦΦΦZF 000 minusI
](41)
Figure 4 Rigid kinematical constraints (left) and flexible-
dissipative-inertial constraints (right) at multiple locations
D Dynamical formulation of the fully coupled model367
The formulation (11) readily applies to the full guitar368
model including the non-rigid constraint model for the369
stringfret interaction by stacking the unconstrained370
modal displacements into a vector371
qqqu = qqqYu qqqZu qqqBu qqqFu T (42)
where qqqYu qqqZu and qqqBu are as before and372
qqqFu=qqqY1
ufT qqqY6
ufT qqqZ1
ufT qqqZ6
ufTT
(43)
In view of the definition of qqq the assembled inertia MMM373
stiffnessKKK and dampingCCC matrices which are uncoupled374
are given by375
MMM=
MMMY 000 000 000
000 MMMZ 000 000
000 000 MMMB 000
000 000 000 MMMF
CCC=
CCCY 000 000 000
000 CCCZ 000 000
000 000 CCCB 000
000 000 000 CCCF
KKK=
KKKY 000 000 000
000 KKKZ 000 000
000 000 KKKB 000
000 000 000 KKKF
(44)
with MMM ` KKK` and CCC` the inertia stiffness and dampingmatrices for each individual subsystem ` and the vectorsFFFext and FFFnl write
FFFext =FFFYextT FFFZextT000000
T(45a)
FFFnl =FFFYnlT FFFZnlT000000
T(45b)
where FFFYext and FFFZext correspond to the plucking action376
of the player expressed for both planes of polarization377
and FFFYnl and FFFZnl are the modal vectors associated to the378
string nonlinear effects with entries given by Eq (20)379
The form of the modal constraint equation is now380
AAAqqq = 000 (46)
with381
AAA =
[AAAbAAAf
]=
ΦΦΦYb 000 minusΦΦΦB 000 000 000
000 ΦΦΦZb 000 000 000 000
ΦYFΦYFΦYF 000 000 000 minusFFFY 000
000 ΦZFΦZFΦZF 000 000 000 minusFFFZ
(47)
where ΦYFΦYFΦYF and ΦZFΦZFΦZF are block matrices with entries given382
by Eq (35) and FFFY and FFFZ are block diagonal matrices383
built with either the identity or null matrix depending on384
the constraints enforced on each string The form of (47)385
shows that the interface forces and constraint conditions386
couple the Y -motions of the strings and the body as well387
as coupling between the finger and the strings in each388
direction of motions Notice that coupling between both389
polarizations of string motions is not apparent in (47)390
although it is formulated in the nonlinear forces (45b)391
IV ILLUSTRATIVE COMPUTATIONS392
A System parameters393
For given initial conditions the system of equations394
comprising (5) with the unconstrained accelerations (11)395
and the constraint matrix (15) and using the specific396
matrices given by Eqs (44) and (47) can be integrated in397
time to produce the transient response The time-domain398
simulations were performed based on the identified body399
modal data and using standard string properties of400
classical guitars Assuming non-ideal strings with fixed-401
free ends the modeshapes are given by Eq (19) and402
their modal frequencies can be approximated by403
fYn = fZn =cT2πpn
(1 +
B
2T0p2n
)(48)
where cT is the velocity of transverse waves B is the404
bending stiffness of the string and T0 its tensioning405
pn = (2n minus 1)π2L and n the mode number Modal406
damping values are introduced following the pragmatic407
formulation given in Ref 6408
ζYn = ζZn =1
2
T0(ηF + ηAωn
)+ηBBp
2n
T0 +Bp2n(49)
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 7
where ηF ηA and ηB are the loss coefficients related to409
ldquointernal frictionrdquo ldquoair viscous dampingrdquo and ldquobending410
dampingrdquo The length of the strings is L=065 m411
and their density bending stiffness tensioning and loss412
coefficients are taken from Ref 29 The size of the modal413
basis of each string is defined by the maximum frequency414
of the computations (10000 Hz) which proved a sensible415
compromise for convergence to realistic simulations416
resulting in 800 modes for the set of strings To417
obtain suitable values for the elasticity parameter ES418
we perform simple vibratory tests on string samples419
Stretching a string vertically and suspending a mass to420
its free end one can easily measure the longitudinal421
vibration of the string-mass oscillator and obtain its422
natural frequency Assuming the string acting as a423
spring with no mass and reminding that the natural424
frequency of the longitudinal vibration is given by f =425
12π
radicESML then an estimate of ES is obtained The426
simulation results were obtained by time-step integration427
of the ordinary modal equations using the velocity-428
Verlet numerical scheme30 an integrator that combines429
simplicity second order accuracy with good numerical430
stability - see implementation details in Ref 2 We431
adopted a time-step of 10minus5s and assumed all initial432
modal displacements and velocities nil initially433
B Computational results434
Pluck excitations were simulated by considering a435
point-wise external force at xe=09L with a linear force436
ramp applied to one or more strings during 10 ms and437
by assuming null excitation force at subsequent instants438
1 Dynamical behaviour of the fully coupled model439
Results in Fig 5 were obtained for the open string440
E2 excited in the direction normal to the soundboard441
first considering a rigidly mounted string then coupling442
one string and finally all six strings to the instrument443
body Assuming no dynamics for the body the response444
spectra of string motion at the bridge is virtually nil445
according to the pinned boundary condition When446
coupling the instrument body to the string the response447
becomes obviously different and shows the classical series448
of near harmonic string partials together with smaller449
peaks from the body resonances While the first modal450
frequency of the unconstrained string is 41 Hz a close451
look at Fig 5 shows that the constraint at the bridge452
yields the fundamental frequency for the tuned string453
with pinned ends (f = 82 Hz) thus confirming the454
correct implementation of the bridge coupling Coupling455
the other five strings results in new peaks in the456
spectrum extending over the frequency range of the body457
modes and with an amplitude difference of about 100 dB458
compared to the most excited modes which clearly relate459
to the non-excited strings As found in some stringed460
instruments this dynamical coupling of all the strings via461
the bridge can lead to sympathetic vibrations which can462
further results in slight interference beats in the sound463
due to the inharmonicity produced by the bridge coupling464
and of the string partials465
0 200 400 600 800 1000 1200 1400 1600 180010
-40
10-20
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 180010
-20
10-10
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 1800
Frequency [ Hz ]
10-20
10-10
100
| Y
6(f
) |
Figure 5 Response spectra of string motion at the bridge
Top rigidly mounted string middle single stringbody
coupling bottom six-stringbody coupled model Force
amplitude 5 N Linear model
0 02 04 06 08 1
Time [ s ]
0
05
1
15
2
Mo
dal en
erg
y [
J ]
10-3
STRINGSFRETBODYTOTAL
Figure 6 Instantaneous modal energies Oblique excitation
with amplitudes of 2 N and 1 N in the Y and Z directions
Linear six-stringbody coupled model Conservative model
Figs 6 and 7 then serve at checking other466
implementation details Simulations pertained to the467
string E2 constrained at xF = 02509L by a single ldquostiffrdquo468
fret modeled with mF=5 10minus3 Kg cF=1 Nsm and469
kF=106 Nm plucked in the oblique direction The470
modal energies E` of the subsystems ` shown in these471
plots were computed from the results of the simulations472
as follows in terms of the modal parameters and the473
8 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
0 02 04 06 08 10
05
1
15
[ J ]
10-3 STRINGS
Y-dir
Z-dir
0 02 04 06 08 10
05
1
15
[ J ]
10-5 FRET
Y-dir
Z-dir
0 02 04 06 08 10
05
1
[ J ]
10-4 BODY
0 02 04 06 08 1
Time [ s ]
0
05
1
15
10-3 FULL SYSTEM
Figure 7 Instantaneous modal energies Same plucking
conditions as in Fig 6 Linear six-stringbody coupled model
Non-conservative model
modal displacements and velocities474
E`(t) =
Nsumn=1
[1
2m`n(ω`n)2
[q`n(t)
]2+
1
2m`n
[q`n(t)
]2](50)
One important feature seen in Fig 6 is that the total475
energy remains roughly constant over a conservative476
simulation while energies flow back and forth between477
the subsystems thus giving confidence in the numerical478
model for simulations of reasonable duration For479
the non-conservative case shown in Fig 7 the large480
difference in energy decay for the two polarizations481
confirms the different stringbody coupling with the482
directions of string motion Note that both the body483
and the fret oscillator are dominated by periodic energy484
exchanges at the frequency of the played note as the wave485
travels back and forth between bridge and fret Finally486
looking closely at the modal energy of the fret one can487
see smaller superimposed periodic disturbances related488
to the reflections between fret and nut illustrating that489
energy still flows through the modeled fret490
Finally we compare in Fig 8 results by varying491
the excitation amplitude in order to illustrate the string492
nonlinearity The simulations were obtained for the493
string E2 using the experimentally determined value494
of 32 kN for the flexibility ES which is comparable495
to the value given by Chaigne31 Fig 8 shows the496
temporal variations of string tension with an estimate497
of the playing frequency computed from time domain498
evaluations of the zero crossing frequency The general499
features of large amplitude string motions are clearly500
visible with a quasi-static increase of tension and501
superimposed oscillations at twice the string fundamental502
frequency and a small frequency variation of the played503
note which compares well with measurements32504
1 2 3 4 5 6 7
716
718
72
722
724
T (
t )
[ N
] Fexc
= 1 N
Fexc
= 5 N
1 2 3 4 5 6 7
Time [ s ]
822
824
826
828
f (
t )
[ H
z ]
Figure 8 String tension (top) and instantaneous frequency of
string motion (bottom) Nonlinear rigid body model
2 Parameter study of the stringfinger coupling model505
We now illustrate the different strategies to account506
for a stopping finger on the fingerboard and explore the507
influence on the coupled system of the finger model508
In contrast with the approach in Ref 2 the strategy509
in terms of auxiliary oscillators rigidly coupled to the510
string seems more realistic and is undoubtedly more511
versatile allowing different types of contacts to be512
modeled by varying the oscillator parameters However513
one could expect the finger dynamics to perturb the514
coupled system with changes in frequency and damping515
for its modes similar to what observed for a multi-modal516
system coupled to a system exhibiting resonance To517
give some insights Figs 9-11 show the effects of varying518
the finger mechanical parameters Simulations pertain to519
the string E2 constrained at xF = 01857L by a finger of520
mass mf = 5 10minus3 Kg and ignore the body dynamics521
The first test consists in varying the stiffness of the522
constraint assuming no damping Fig 9 shows the523
frequency spectra of the coupling force at the rigid bridge524
where the peaks of the string modes can be seen clearly as525
well as the resonance frequency of the auxiliary oscillator526
which moves in frequency with the stiffness and slightly527
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 9
modulates the other peaks A close look shows that528
the spectra are affected much more sensitively below529
the resonance of the finger oscillator with amplitude530
variations analogous to a system driven in the stiffness-531
controlled region Decreasing the stiffness causes a rise532
in amplitude of the undesirable modes while increasing533
the stiffness lower their amplitudes As observed for the534
stringbody coupling33 inharmonicity is also created by535
the stiffness termination which in turn affects the tuning536
of the note With the increase of stiffness perturbations537
of the string modes are reduced and tuning is improved538
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
4 Nm
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
5 Nm
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
kf = 10
6 Nm
Figure 9 Spectra of the coupling force at the rigid bridge
Stiffness is varied F=1 c1 = 0 Nsm
Fig 10 then shows the effect of varying the damping539
keeping the stiffness constant Not surprisingly if540
damping increases the amplitude of the coupled541
modes close to the oscillator resonance decrease and542
the frequency region affected becomes wider This543
significantly affects the time envelope of the pluck note544
that becomes smoother and has a shorter-time decay545
Finally the effect of the number of contacts is studied546
using the rigid contact model which is convenient for547
this purpose Fig 11 gives a picture of the spatial548
distribution of the string energy by plotting the standard549
deviation of string motion as a function of its length550
As seen the ldquopassiverdquo region is affected with great551
sensitivity by the number of contacts and surprisingly552
increasing the contact points does not necessarily reduce553
the undesirable transfer of energy Computations show554
that the effectiveness of the multi-point rigid constraints555
approach is strongly problem-dependent and varies in a556
non-obvious way with several factors including the finger557
width the spacing between contact points the relative558
string length from the ends and the excitation spectrum559
Taken in combination these results and noticing560
that the perturbations induced by each finger oscillator561
are additive due to the linear coupling a robust and562
physically sound strategy to simulate the pressing finger563
would be to implement a set of flexible-dissipative564
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
cf = 1 Nsm
cf = 10 Nsm
0 1 2 3 4 5
Time [ s ]
-2
0
2
4
Y (
t )
10-3
Figure 10 Spectra of the coupling force at the rigid
bridge (up) and time history of the string motion (bottom)
Damping is varied F=1 k1 = 105 Nm
constraints imposing a first stiff constraint in order565
to ensure a precise tuning followed by several soft566
constraints that would control the waveform and spectral567
content of the decaying response depending on their568
number and damping properties For illustration569
the spectrograms of two plucks for different sets of570
parameters are presented in Fig 12571
0 01 02 03 04 05 06
x [ m ]
10-5
10-2
ST
D (
Y(t
) )
F=1
F=2
F=3
F=5
016 017 018 019 02
x [ m ]
10-5
ST
D (
Y(t
) )
Figure 11 Standard deviation of the string motion Number
of contacts is varied Vertical dotted lines represent the
constraints locations Finger width (0015m) is kept constant
C Simulated acoustic radiation572
As an alternative to computing sounds from full573
vibro-acoustic models that usually require the use of574
massive computational resources it is tempting to apply575
10 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
Figure 12 Spectrograms of the force at the rigid bridge Up
F = 3 bottom F = 5 Parameter values kF1=106 Nm
cF1=1 Nsm and kFj =104 Nm cFj =10 Nsm (j = 2 F )
Constraint at xF = 01857L (sound files available)
simpler techniques that could be efficient to generate576
convincing sounds from the vibratory responses The577
method used here is to treat the guitar body as a linear578
input-output system and to compute the pressure by579
convolution benefiting from the local excitation of the580
body by the strings and using a vibro-acoustic impulse581
response of the test guitar This approach does not by582
any means simulate the sound response everywhere in the583
pressure field but it has the merit of adding the influence584
of the radiating properties of the body at a very low585
computational cost and of providing sounds of greater586
perceptual quality than the commonly used vibratory587
signals at the bridge The pressure at location rp can be588
computed by the convolution of the time-history of the589
stringbridge interaction force FY (rb t) and the vibro-590
acoustic impulse response function hY (rb rp t) between591
the bridge and microphone by592
p(rp t) = hY (rb rp t) lowast FY (rb t) (51)
The extension to account for the effects of all the593
strings is readily obtained by linear superposition594
principle summing the contributions from each string595
Measurements of the vibro-acoustic impulse response596
were performed under the same experimental conditions597
as for transfer function measurements using an598
additional condenser microphone Impulsive excitation599
was applied in the perpendicular direction to the600
soundboard and the acoustic pressure was measured601
at a distance of 1 m from the central axis of the602
guitar in a quiet room with all strings damped (see603
Fig 1) The impulse response was computed by inverse604
Fourier transform of the complex transfer function built605
from the force and response signals described in the606
frequency domain Results in Fig 13 pertain to the607
simulation of a G major arpeggio obtained with the608
two lowest and the top strings constrained using five609
auxiliary flexible-inertial constraints on each string The610
effect of the radiation properties of the body on the611
computed vibratory signals is supported when comparing612
the spectrograms in Fig 13 which clearly shows the613
spectral filtering of the vibration signal614
Figure 13 Spectrograms of the bridge force (top) and sound
pressure (bottom) Nonlinear six-stringbody coupled model
Arpeggio of a G major chord with three strings stopped using
a set of auxiliary oscillators (sound files available)
V CONCLUSIONS615
Following our recent work showing the potential616
of the modal Udwadia-Kalaba equations to physically617
simulate musical instruments we presented in this paper618
a computational approach that allows efficient time-619
domain simulations of realistic guitar plucks In this620
work we further generalized our simple stringbody621
coupled model toward the development of a more622
realistic guitar model by including the complete set623
of strings the two polarizations of string motions the624
string geometric nonlinearity the stringfret interaction625
that controls the playing frequency and the body626
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 11
dynamics of an actual instrument The dynamics of627
the various subsystems were formulated in terms of628
their unconstrained modes and coupling were achieved629
by enforcing several constraint conditions One of the630
developments of the model was the introduction of631
flexible-dissipative-inertial supplementary equations as632
an alternative to the common kinematical constraints633
which here improve the role of the stringfingerfret634
interaction at the fingerboard This approach can635
also benefit the modeling of other complex dynamical636
systems outside the realm of physical instrument637
modeling To improve the realism of the synthesis638
we also included crudely the radiation properties of639
the instrument body into the computational model640
through an ad hoc procedure Sound pressure was641
computed adopting a hybrid approach by convoluting642
the computed bridge force signals with a vibro-acoustic643
impulse response measured on the test guitar Such644
vibro-acoustic response can be measured in an anechoic645
chamber or else include the acoustics of any specific646
room This approach appeared not only straightforward647
to compute realistic sounds of string instruments for648
which the radiating components are subjected to point-649
wise excitation but also very efficient compared to heavy650
computational sound radiation techniques651
ACKNOWLEDGMENTS652
The authors thanks Paulo Vaz de Carvalho for653
lending the guitar and interesting discussions This work654
is funded by national funds through the FCT ndash Fundacao655
para a Ciencia e a Tecnologia IP under the Norma656
Transitoria - DL 572016CP1453CT0007657
REFERENCES658
1JH Ginsberg Mechanical and Structural Vibrations Theory659
and Applications chapter 9 Wiley 2001660
2J Antunes and V Debut Dynamical computation of constrained661
flexible systems using a modal Udwadia-Kalaba formulation662
Application to musical instruments J Acoust Soc Am663
141764 2017664
3J Antunes V Debut L Borsoi X Delaune and P Piteau665
A modal Udwadia-Kalaba formulation for the vibro-impact666
modelling of continuous flexible systems with intermittent667
contacts Procedia Engineering 199322ndash329 2017668
4FE Udwadia and RE Kalaba Analytical dynamics - A new669
approach Cambridge University Press 1996670
5C Valette The mechanics of the vibrating string Mechanics of671
musical instruments Springer 1995672
6J Woodhouse On the synthesis of guitar plucks Acta Acust673
united Ac 90928ndash944 2004674
7M E McIntyre RT Schumacher and J Woodhouse On the675
oscillations of musical instruments Journal of the Acoustical676
Society of America 741325ndash1345 1983677
8A Chaigne On the use of finite differences for musical678
synthesis Application to plucked stringed instruments Journal679
drsquoAcoustique 5181ndash211 1992680
9JM Adrien The missing link Modal synthesis In G DePoli681
A Picalli and C Roads editors Representations of musical682
sounds MIT Press Cambridge 1991683
10S Bilbao Numerical Sound Synthesis Finite Difference684
Schemes and Simulation in Musical Acoustics Wiley Publishing685
2009686
11L Trautmann and R Rabenstein Multirate simulations687
of string vibrations includingnonlinear fret-string interactions688
usingthe functional transformation method EURASIP Journal689
on Applied Signal Processing 7949ndash963 2004690
12V Chatziioannou and M van Walstijn Energy conserving691
schemes for the simulation of musical instrument contact692
dynamics J Sound Vib 339262ndash279 2015693
13S Bilbao and A Torin Numerical modeling and sound synthesis694
for articulated stringfretboard interactions Acta Acust united695
Ac 63336ndash347 2015696
14V Debut M Carvalho M Marques and J Antunes Physics-697
based modeling techniques of a twelve-string portuguese guitar698
a three-dimensional non-linear time-domain computational699
approach for the multiple-stringsbridgesoundboard coupled700
dynamics Appl Acoust 1083ndash18 2016701
15M van Walstijn and J Bridges Simulation of distributed contact702
in string instruments A modal expansion approach In 24th703
European Signal Processing Conference (Eusipco 2016) pages704
1023ndash1027 2013705
16C Issanchou S Bilbao JL Le Carrou C Touze and O Doare706
A modal-based approach to the nonlinear vibration of strings707
against a unilateral obstacle simulations and experiments in the708
pointwise case J Sound Vib 393229ndash251 2017709
17R Bader Computational Mechanics of the Classical Guitar710
Springer-Verlag Berlin Heidelberg 2005711
18H Mansour J Woodhouse and G Scavone Enhanced wave-712
based modelling of musical strings Part 1 Plucked strings Acta713
Acust united Ac 1021094ndash1107 2016714
19O Inacio J Antunes and MCM Wright Computational715
modeling of string-body interaction for the violin family and716
simulation of wolf notes J Sound Vib 310260ndash286 2007717
20S Bilbao A Torin and V Chatziioannou Numerical modeling718
of collisions in musical instruments Acta Acust united Ac719
101155ndash173 2015720
21C Issanchou JL Le Carrou C Touze B Fabre and O Doare721
Stringfrets contacts in the electric bass sound Simulations and722
experiments App Acoust 129217ndash228 2018723
22D J Braun and M Goldfarb Eliminating constraint drift in the724
numerical simulation of constrained dynamical systems Comput725
Methods in Appl Mech Eng 198(37)3151 ndash 3160 2009726
23GF Carrier On the nonlinear vibration problem of the elastic727
string Q Appl Math 3157ndash165 1945728
24G Kirchhoff Vorlesungen uber Mechanik Teubner 1883729
25J Pakarinen V Valimaki and M Karjalainen Physics-based730
methods for modeling nonlinear vibrating strings Acta Acust731
united Ac 91312ndash325 2005732
26PMC Morse and KU Ingard Theoretical Acoustics733
International series in pure and applied physics Princeton734
University Press 1968735
27J Juang Applied System Identification PTR Prentice-Hall736
Inc New Jersey 1994737
28B E Richardson The acoustical development of the guitar J738
Catgut Acoust Soc 51ndash10 1994739
29J Woodhouse Plucked guitar transients Comparison of740
measurements and synthesis Acta Acust united Ac 90945ndash741
965 2004742
30M Tuckerman B J Berne and G J Martyna Reversible743
multiple time scale molecular dynamics J Chem Phys744
97(3)1990ndash2001 1992745
31A Chaigne Viscoelastic properties of nylon guitar strings J746
Catgut Acoust Soc Series II 121ndash27 1991747
32T Tolonen V Valimaki and M Karjalainen Modeling of748
tension modulation nonlinearity in plucked strings IEEE Trans749
Speech Audio Process 8(3)300ndash310 2000750
33A Chaigne and J Kergomard Acoustics of Musical Instruments751
Modern Acoustics and Signal Processing Springer New York752
2016753
12 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
754
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 13
where ηF ηA and ηB are the loss coefficients related to409
ldquointernal frictionrdquo ldquoair viscous dampingrdquo and ldquobending410
dampingrdquo The length of the strings is L=065 m411
and their density bending stiffness tensioning and loss412
coefficients are taken from Ref 29 The size of the modal413
basis of each string is defined by the maximum frequency414
of the computations (10000 Hz) which proved a sensible415
compromise for convergence to realistic simulations416
resulting in 800 modes for the set of strings To417
obtain suitable values for the elasticity parameter ES418
we perform simple vibratory tests on string samples419
Stretching a string vertically and suspending a mass to420
its free end one can easily measure the longitudinal421
vibration of the string-mass oscillator and obtain its422
natural frequency Assuming the string acting as a423
spring with no mass and reminding that the natural424
frequency of the longitudinal vibration is given by f =425
12π
radicESML then an estimate of ES is obtained The426
simulation results were obtained by time-step integration427
of the ordinary modal equations using the velocity-428
Verlet numerical scheme30 an integrator that combines429
simplicity second order accuracy with good numerical430
stability - see implementation details in Ref 2 We431
adopted a time-step of 10minus5s and assumed all initial432
modal displacements and velocities nil initially433
B Computational results434
Pluck excitations were simulated by considering a435
point-wise external force at xe=09L with a linear force436
ramp applied to one or more strings during 10 ms and437
by assuming null excitation force at subsequent instants438
1 Dynamical behaviour of the fully coupled model439
Results in Fig 5 were obtained for the open string440
E2 excited in the direction normal to the soundboard441
first considering a rigidly mounted string then coupling442
one string and finally all six strings to the instrument443
body Assuming no dynamics for the body the response444
spectra of string motion at the bridge is virtually nil445
according to the pinned boundary condition When446
coupling the instrument body to the string the response447
becomes obviously different and shows the classical series448
of near harmonic string partials together with smaller449
peaks from the body resonances While the first modal450
frequency of the unconstrained string is 41 Hz a close451
look at Fig 5 shows that the constraint at the bridge452
yields the fundamental frequency for the tuned string453
with pinned ends (f = 82 Hz) thus confirming the454
correct implementation of the bridge coupling Coupling455
the other five strings results in new peaks in the456
spectrum extending over the frequency range of the body457
modes and with an amplitude difference of about 100 dB458
compared to the most excited modes which clearly relate459
to the non-excited strings As found in some stringed460
instruments this dynamical coupling of all the strings via461
the bridge can lead to sympathetic vibrations which can462
further results in slight interference beats in the sound463
due to the inharmonicity produced by the bridge coupling464
and of the string partials465
0 200 400 600 800 1000 1200 1400 1600 180010
-40
10-20
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 180010
-20
10-10
100
| Y
6(f
) |
0 200 400 600 800 1000 1200 1400 1600 1800
Frequency [ Hz ]
10-20
10-10
100
| Y
6(f
) |
Figure 5 Response spectra of string motion at the bridge
Top rigidly mounted string middle single stringbody
coupling bottom six-stringbody coupled model Force
amplitude 5 N Linear model
0 02 04 06 08 1
Time [ s ]
0
05
1
15
2
Mo
dal en
erg
y [
J ]
10-3
STRINGSFRETBODYTOTAL
Figure 6 Instantaneous modal energies Oblique excitation
with amplitudes of 2 N and 1 N in the Y and Z directions
Linear six-stringbody coupled model Conservative model
Figs 6 and 7 then serve at checking other466
implementation details Simulations pertained to the467
string E2 constrained at xF = 02509L by a single ldquostiffrdquo468
fret modeled with mF=5 10minus3 Kg cF=1 Nsm and469
kF=106 Nm plucked in the oblique direction The470
modal energies E` of the subsystems ` shown in these471
plots were computed from the results of the simulations472
as follows in terms of the modal parameters and the473
8 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
0 02 04 06 08 10
05
1
15
[ J ]
10-3 STRINGS
Y-dir
Z-dir
0 02 04 06 08 10
05
1
15
[ J ]
10-5 FRET
Y-dir
Z-dir
0 02 04 06 08 10
05
1
[ J ]
10-4 BODY
0 02 04 06 08 1
Time [ s ]
0
05
1
15
10-3 FULL SYSTEM
Figure 7 Instantaneous modal energies Same plucking
conditions as in Fig 6 Linear six-stringbody coupled model
Non-conservative model
modal displacements and velocities474
E`(t) =
Nsumn=1
[1
2m`n(ω`n)2
[q`n(t)
]2+
1
2m`n
[q`n(t)
]2](50)
One important feature seen in Fig 6 is that the total475
energy remains roughly constant over a conservative476
simulation while energies flow back and forth between477
the subsystems thus giving confidence in the numerical478
model for simulations of reasonable duration For479
the non-conservative case shown in Fig 7 the large480
difference in energy decay for the two polarizations481
confirms the different stringbody coupling with the482
directions of string motion Note that both the body483
and the fret oscillator are dominated by periodic energy484
exchanges at the frequency of the played note as the wave485
travels back and forth between bridge and fret Finally486
looking closely at the modal energy of the fret one can487
see smaller superimposed periodic disturbances related488
to the reflections between fret and nut illustrating that489
energy still flows through the modeled fret490
Finally we compare in Fig 8 results by varying491
the excitation amplitude in order to illustrate the string492
nonlinearity The simulations were obtained for the493
string E2 using the experimentally determined value494
of 32 kN for the flexibility ES which is comparable495
to the value given by Chaigne31 Fig 8 shows the496
temporal variations of string tension with an estimate497
of the playing frequency computed from time domain498
evaluations of the zero crossing frequency The general499
features of large amplitude string motions are clearly500
visible with a quasi-static increase of tension and501
superimposed oscillations at twice the string fundamental502
frequency and a small frequency variation of the played503
note which compares well with measurements32504
1 2 3 4 5 6 7
716
718
72
722
724
T (
t )
[ N
] Fexc
= 1 N
Fexc
= 5 N
1 2 3 4 5 6 7
Time [ s ]
822
824
826
828
f (
t )
[ H
z ]
Figure 8 String tension (top) and instantaneous frequency of
string motion (bottom) Nonlinear rigid body model
2 Parameter study of the stringfinger coupling model505
We now illustrate the different strategies to account506
for a stopping finger on the fingerboard and explore the507
influence on the coupled system of the finger model508
In contrast with the approach in Ref 2 the strategy509
in terms of auxiliary oscillators rigidly coupled to the510
string seems more realistic and is undoubtedly more511
versatile allowing different types of contacts to be512
modeled by varying the oscillator parameters However513
one could expect the finger dynamics to perturb the514
coupled system with changes in frequency and damping515
for its modes similar to what observed for a multi-modal516
system coupled to a system exhibiting resonance To517
give some insights Figs 9-11 show the effects of varying518
the finger mechanical parameters Simulations pertain to519
the string E2 constrained at xF = 01857L by a finger of520
mass mf = 5 10minus3 Kg and ignore the body dynamics521
The first test consists in varying the stiffness of the522
constraint assuming no damping Fig 9 shows the523
frequency spectra of the coupling force at the rigid bridge524
where the peaks of the string modes can be seen clearly as525
well as the resonance frequency of the auxiliary oscillator526
which moves in frequency with the stiffness and slightly527
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 9
modulates the other peaks A close look shows that528
the spectra are affected much more sensitively below529
the resonance of the finger oscillator with amplitude530
variations analogous to a system driven in the stiffness-531
controlled region Decreasing the stiffness causes a rise532
in amplitude of the undesirable modes while increasing533
the stiffness lower their amplitudes As observed for the534
stringbody coupling33 inharmonicity is also created by535
the stiffness termination which in turn affects the tuning536
of the note With the increase of stiffness perturbations537
of the string modes are reduced and tuning is improved538
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
4 Nm
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
5 Nm
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
kf = 10
6 Nm
Figure 9 Spectra of the coupling force at the rigid bridge
Stiffness is varied F=1 c1 = 0 Nsm
Fig 10 then shows the effect of varying the damping539
keeping the stiffness constant Not surprisingly if540
damping increases the amplitude of the coupled541
modes close to the oscillator resonance decrease and542
the frequency region affected becomes wider This543
significantly affects the time envelope of the pluck note544
that becomes smoother and has a shorter-time decay545
Finally the effect of the number of contacts is studied546
using the rigid contact model which is convenient for547
this purpose Fig 11 gives a picture of the spatial548
distribution of the string energy by plotting the standard549
deviation of string motion as a function of its length550
As seen the ldquopassiverdquo region is affected with great551
sensitivity by the number of contacts and surprisingly552
increasing the contact points does not necessarily reduce553
the undesirable transfer of energy Computations show554
that the effectiveness of the multi-point rigid constraints555
approach is strongly problem-dependent and varies in a556
non-obvious way with several factors including the finger557
width the spacing between contact points the relative558
string length from the ends and the excitation spectrum559
Taken in combination these results and noticing560
that the perturbations induced by each finger oscillator561
are additive due to the linear coupling a robust and562
physically sound strategy to simulate the pressing finger563
would be to implement a set of flexible-dissipative564
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
cf = 1 Nsm
cf = 10 Nsm
0 1 2 3 4 5
Time [ s ]
-2
0
2
4
Y (
t )
10-3
Figure 10 Spectra of the coupling force at the rigid
bridge (up) and time history of the string motion (bottom)
Damping is varied F=1 k1 = 105 Nm
constraints imposing a first stiff constraint in order565
to ensure a precise tuning followed by several soft566
constraints that would control the waveform and spectral567
content of the decaying response depending on their568
number and damping properties For illustration569
the spectrograms of two plucks for different sets of570
parameters are presented in Fig 12571
0 01 02 03 04 05 06
x [ m ]
10-5
10-2
ST
D (
Y(t
) )
F=1
F=2
F=3
F=5
016 017 018 019 02
x [ m ]
10-5
ST
D (
Y(t
) )
Figure 11 Standard deviation of the string motion Number
of contacts is varied Vertical dotted lines represent the
constraints locations Finger width (0015m) is kept constant
C Simulated acoustic radiation572
As an alternative to computing sounds from full573
vibro-acoustic models that usually require the use of574
massive computational resources it is tempting to apply575
10 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
Figure 12 Spectrograms of the force at the rigid bridge Up
F = 3 bottom F = 5 Parameter values kF1=106 Nm
cF1=1 Nsm and kFj =104 Nm cFj =10 Nsm (j = 2 F )
Constraint at xF = 01857L (sound files available)
simpler techniques that could be efficient to generate576
convincing sounds from the vibratory responses The577
method used here is to treat the guitar body as a linear578
input-output system and to compute the pressure by579
convolution benefiting from the local excitation of the580
body by the strings and using a vibro-acoustic impulse581
response of the test guitar This approach does not by582
any means simulate the sound response everywhere in the583
pressure field but it has the merit of adding the influence584
of the radiating properties of the body at a very low585
computational cost and of providing sounds of greater586
perceptual quality than the commonly used vibratory587
signals at the bridge The pressure at location rp can be588
computed by the convolution of the time-history of the589
stringbridge interaction force FY (rb t) and the vibro-590
acoustic impulse response function hY (rb rp t) between591
the bridge and microphone by592
p(rp t) = hY (rb rp t) lowast FY (rb t) (51)
The extension to account for the effects of all the593
strings is readily obtained by linear superposition594
principle summing the contributions from each string595
Measurements of the vibro-acoustic impulse response596
were performed under the same experimental conditions597
as for transfer function measurements using an598
additional condenser microphone Impulsive excitation599
was applied in the perpendicular direction to the600
soundboard and the acoustic pressure was measured601
at a distance of 1 m from the central axis of the602
guitar in a quiet room with all strings damped (see603
Fig 1) The impulse response was computed by inverse604
Fourier transform of the complex transfer function built605
from the force and response signals described in the606
frequency domain Results in Fig 13 pertain to the607
simulation of a G major arpeggio obtained with the608
two lowest and the top strings constrained using five609
auxiliary flexible-inertial constraints on each string The610
effect of the radiation properties of the body on the611
computed vibratory signals is supported when comparing612
the spectrograms in Fig 13 which clearly shows the613
spectral filtering of the vibration signal614
Figure 13 Spectrograms of the bridge force (top) and sound
pressure (bottom) Nonlinear six-stringbody coupled model
Arpeggio of a G major chord with three strings stopped using
a set of auxiliary oscillators (sound files available)
V CONCLUSIONS615
Following our recent work showing the potential616
of the modal Udwadia-Kalaba equations to physically617
simulate musical instruments we presented in this paper618
a computational approach that allows efficient time-619
domain simulations of realistic guitar plucks In this620
work we further generalized our simple stringbody621
coupled model toward the development of a more622
realistic guitar model by including the complete set623
of strings the two polarizations of string motions the624
string geometric nonlinearity the stringfret interaction625
that controls the playing frequency and the body626
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 11
dynamics of an actual instrument The dynamics of627
the various subsystems were formulated in terms of628
their unconstrained modes and coupling were achieved629
by enforcing several constraint conditions One of the630
developments of the model was the introduction of631
flexible-dissipative-inertial supplementary equations as632
an alternative to the common kinematical constraints633
which here improve the role of the stringfingerfret634
interaction at the fingerboard This approach can635
also benefit the modeling of other complex dynamical636
systems outside the realm of physical instrument637
modeling To improve the realism of the synthesis638
we also included crudely the radiation properties of639
the instrument body into the computational model640
through an ad hoc procedure Sound pressure was641
computed adopting a hybrid approach by convoluting642
the computed bridge force signals with a vibro-acoustic643
impulse response measured on the test guitar Such644
vibro-acoustic response can be measured in an anechoic645
chamber or else include the acoustics of any specific646
room This approach appeared not only straightforward647
to compute realistic sounds of string instruments for648
which the radiating components are subjected to point-649
wise excitation but also very efficient compared to heavy650
computational sound radiation techniques651
ACKNOWLEDGMENTS652
The authors thanks Paulo Vaz de Carvalho for653
lending the guitar and interesting discussions This work654
is funded by national funds through the FCT ndash Fundacao655
para a Ciencia e a Tecnologia IP under the Norma656
Transitoria - DL 572016CP1453CT0007657
REFERENCES658
1JH Ginsberg Mechanical and Structural Vibrations Theory659
and Applications chapter 9 Wiley 2001660
2J Antunes and V Debut Dynamical computation of constrained661
flexible systems using a modal Udwadia-Kalaba formulation662
Application to musical instruments J Acoust Soc Am663
141764 2017664
3J Antunes V Debut L Borsoi X Delaune and P Piteau665
A modal Udwadia-Kalaba formulation for the vibro-impact666
modelling of continuous flexible systems with intermittent667
contacts Procedia Engineering 199322ndash329 2017668
4FE Udwadia and RE Kalaba Analytical dynamics - A new669
approach Cambridge University Press 1996670
5C Valette The mechanics of the vibrating string Mechanics of671
musical instruments Springer 1995672
6J Woodhouse On the synthesis of guitar plucks Acta Acust673
united Ac 90928ndash944 2004674
7M E McIntyre RT Schumacher and J Woodhouse On the675
oscillations of musical instruments Journal of the Acoustical676
Society of America 741325ndash1345 1983677
8A Chaigne On the use of finite differences for musical678
synthesis Application to plucked stringed instruments Journal679
drsquoAcoustique 5181ndash211 1992680
9JM Adrien The missing link Modal synthesis In G DePoli681
A Picalli and C Roads editors Representations of musical682
sounds MIT Press Cambridge 1991683
10S Bilbao Numerical Sound Synthesis Finite Difference684
Schemes and Simulation in Musical Acoustics Wiley Publishing685
2009686
11L Trautmann and R Rabenstein Multirate simulations687
of string vibrations includingnonlinear fret-string interactions688
usingthe functional transformation method EURASIP Journal689
on Applied Signal Processing 7949ndash963 2004690
12V Chatziioannou and M van Walstijn Energy conserving691
schemes for the simulation of musical instrument contact692
dynamics J Sound Vib 339262ndash279 2015693
13S Bilbao and A Torin Numerical modeling and sound synthesis694
for articulated stringfretboard interactions Acta Acust united695
Ac 63336ndash347 2015696
14V Debut M Carvalho M Marques and J Antunes Physics-697
based modeling techniques of a twelve-string portuguese guitar698
a three-dimensional non-linear time-domain computational699
approach for the multiple-stringsbridgesoundboard coupled700
dynamics Appl Acoust 1083ndash18 2016701
15M van Walstijn and J Bridges Simulation of distributed contact702
in string instruments A modal expansion approach In 24th703
European Signal Processing Conference (Eusipco 2016) pages704
1023ndash1027 2013705
16C Issanchou S Bilbao JL Le Carrou C Touze and O Doare706
A modal-based approach to the nonlinear vibration of strings707
against a unilateral obstacle simulations and experiments in the708
pointwise case J Sound Vib 393229ndash251 2017709
17R Bader Computational Mechanics of the Classical Guitar710
Springer-Verlag Berlin Heidelberg 2005711
18H Mansour J Woodhouse and G Scavone Enhanced wave-712
based modelling of musical strings Part 1 Plucked strings Acta713
Acust united Ac 1021094ndash1107 2016714
19O Inacio J Antunes and MCM Wright Computational715
modeling of string-body interaction for the violin family and716
simulation of wolf notes J Sound Vib 310260ndash286 2007717
20S Bilbao A Torin and V Chatziioannou Numerical modeling718
of collisions in musical instruments Acta Acust united Ac719
101155ndash173 2015720
21C Issanchou JL Le Carrou C Touze B Fabre and O Doare721
Stringfrets contacts in the electric bass sound Simulations and722
experiments App Acoust 129217ndash228 2018723
22D J Braun and M Goldfarb Eliminating constraint drift in the724
numerical simulation of constrained dynamical systems Comput725
Methods in Appl Mech Eng 198(37)3151 ndash 3160 2009726
23GF Carrier On the nonlinear vibration problem of the elastic727
string Q Appl Math 3157ndash165 1945728
24G Kirchhoff Vorlesungen uber Mechanik Teubner 1883729
25J Pakarinen V Valimaki and M Karjalainen Physics-based730
methods for modeling nonlinear vibrating strings Acta Acust731
united Ac 91312ndash325 2005732
26PMC Morse and KU Ingard Theoretical Acoustics733
International series in pure and applied physics Princeton734
University Press 1968735
27J Juang Applied System Identification PTR Prentice-Hall736
Inc New Jersey 1994737
28B E Richardson The acoustical development of the guitar J738
Catgut Acoust Soc 51ndash10 1994739
29J Woodhouse Plucked guitar transients Comparison of740
measurements and synthesis Acta Acust united Ac 90945ndash741
965 2004742
30M Tuckerman B J Berne and G J Martyna Reversible743
multiple time scale molecular dynamics J Chem Phys744
97(3)1990ndash2001 1992745
31A Chaigne Viscoelastic properties of nylon guitar strings J746
Catgut Acoust Soc Series II 121ndash27 1991747
32T Tolonen V Valimaki and M Karjalainen Modeling of748
tension modulation nonlinearity in plucked strings IEEE Trans749
Speech Audio Process 8(3)300ndash310 2000750
33A Chaigne and J Kergomard Acoustics of Musical Instruments751
Modern Acoustics and Signal Processing Springer New York752
2016753
12 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
754
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 13
0 02 04 06 08 10
05
1
15
[ J ]
10-3 STRINGS
Y-dir
Z-dir
0 02 04 06 08 10
05
1
15
[ J ]
10-5 FRET
Y-dir
Z-dir
0 02 04 06 08 10
05
1
[ J ]
10-4 BODY
0 02 04 06 08 1
Time [ s ]
0
05
1
15
10-3 FULL SYSTEM
Figure 7 Instantaneous modal energies Same plucking
conditions as in Fig 6 Linear six-stringbody coupled model
Non-conservative model
modal displacements and velocities474
E`(t) =
Nsumn=1
[1
2m`n(ω`n)2
[q`n(t)
]2+
1
2m`n
[q`n(t)
]2](50)
One important feature seen in Fig 6 is that the total475
energy remains roughly constant over a conservative476
simulation while energies flow back and forth between477
the subsystems thus giving confidence in the numerical478
model for simulations of reasonable duration For479
the non-conservative case shown in Fig 7 the large480
difference in energy decay for the two polarizations481
confirms the different stringbody coupling with the482
directions of string motion Note that both the body483
and the fret oscillator are dominated by periodic energy484
exchanges at the frequency of the played note as the wave485
travels back and forth between bridge and fret Finally486
looking closely at the modal energy of the fret one can487
see smaller superimposed periodic disturbances related488
to the reflections between fret and nut illustrating that489
energy still flows through the modeled fret490
Finally we compare in Fig 8 results by varying491
the excitation amplitude in order to illustrate the string492
nonlinearity The simulations were obtained for the493
string E2 using the experimentally determined value494
of 32 kN for the flexibility ES which is comparable495
to the value given by Chaigne31 Fig 8 shows the496
temporal variations of string tension with an estimate497
of the playing frequency computed from time domain498
evaluations of the zero crossing frequency The general499
features of large amplitude string motions are clearly500
visible with a quasi-static increase of tension and501
superimposed oscillations at twice the string fundamental502
frequency and a small frequency variation of the played503
note which compares well with measurements32504
1 2 3 4 5 6 7
716
718
72
722
724
T (
t )
[ N
] Fexc
= 1 N
Fexc
= 5 N
1 2 3 4 5 6 7
Time [ s ]
822
824
826
828
f (
t )
[ H
z ]
Figure 8 String tension (top) and instantaneous frequency of
string motion (bottom) Nonlinear rigid body model
2 Parameter study of the stringfinger coupling model505
We now illustrate the different strategies to account506
for a stopping finger on the fingerboard and explore the507
influence on the coupled system of the finger model508
In contrast with the approach in Ref 2 the strategy509
in terms of auxiliary oscillators rigidly coupled to the510
string seems more realistic and is undoubtedly more511
versatile allowing different types of contacts to be512
modeled by varying the oscillator parameters However513
one could expect the finger dynamics to perturb the514
coupled system with changes in frequency and damping515
for its modes similar to what observed for a multi-modal516
system coupled to a system exhibiting resonance To517
give some insights Figs 9-11 show the effects of varying518
the finger mechanical parameters Simulations pertain to519
the string E2 constrained at xF = 01857L by a finger of520
mass mf = 5 10minus3 Kg and ignore the body dynamics521
The first test consists in varying the stiffness of the522
constraint assuming no damping Fig 9 shows the523
frequency spectra of the coupling force at the rigid bridge524
where the peaks of the string modes can be seen clearly as525
well as the resonance frequency of the auxiliary oscillator526
which moves in frequency with the stiffness and slightly527
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 9
modulates the other peaks A close look shows that528
the spectra are affected much more sensitively below529
the resonance of the finger oscillator with amplitude530
variations analogous to a system driven in the stiffness-531
controlled region Decreasing the stiffness causes a rise532
in amplitude of the undesirable modes while increasing533
the stiffness lower their amplitudes As observed for the534
stringbody coupling33 inharmonicity is also created by535
the stiffness termination which in turn affects the tuning536
of the note With the increase of stiffness perturbations537
of the string modes are reduced and tuning is improved538
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
4 Nm
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
5 Nm
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
kf = 10
6 Nm
Figure 9 Spectra of the coupling force at the rigid bridge
Stiffness is varied F=1 c1 = 0 Nsm
Fig 10 then shows the effect of varying the damping539
keeping the stiffness constant Not surprisingly if540
damping increases the amplitude of the coupled541
modes close to the oscillator resonance decrease and542
the frequency region affected becomes wider This543
significantly affects the time envelope of the pluck note544
that becomes smoother and has a shorter-time decay545
Finally the effect of the number of contacts is studied546
using the rigid contact model which is convenient for547
this purpose Fig 11 gives a picture of the spatial548
distribution of the string energy by plotting the standard549
deviation of string motion as a function of its length550
As seen the ldquopassiverdquo region is affected with great551
sensitivity by the number of contacts and surprisingly552
increasing the contact points does not necessarily reduce553
the undesirable transfer of energy Computations show554
that the effectiveness of the multi-point rigid constraints555
approach is strongly problem-dependent and varies in a556
non-obvious way with several factors including the finger557
width the spacing between contact points the relative558
string length from the ends and the excitation spectrum559
Taken in combination these results and noticing560
that the perturbations induced by each finger oscillator561
are additive due to the linear coupling a robust and562
physically sound strategy to simulate the pressing finger563
would be to implement a set of flexible-dissipative564
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
cf = 1 Nsm
cf = 10 Nsm
0 1 2 3 4 5
Time [ s ]
-2
0
2
4
Y (
t )
10-3
Figure 10 Spectra of the coupling force at the rigid
bridge (up) and time history of the string motion (bottom)
Damping is varied F=1 k1 = 105 Nm
constraints imposing a first stiff constraint in order565
to ensure a precise tuning followed by several soft566
constraints that would control the waveform and spectral567
content of the decaying response depending on their568
number and damping properties For illustration569
the spectrograms of two plucks for different sets of570
parameters are presented in Fig 12571
0 01 02 03 04 05 06
x [ m ]
10-5
10-2
ST
D (
Y(t
) )
F=1
F=2
F=3
F=5
016 017 018 019 02
x [ m ]
10-5
ST
D (
Y(t
) )
Figure 11 Standard deviation of the string motion Number
of contacts is varied Vertical dotted lines represent the
constraints locations Finger width (0015m) is kept constant
C Simulated acoustic radiation572
As an alternative to computing sounds from full573
vibro-acoustic models that usually require the use of574
massive computational resources it is tempting to apply575
10 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
Figure 12 Spectrograms of the force at the rigid bridge Up
F = 3 bottom F = 5 Parameter values kF1=106 Nm
cF1=1 Nsm and kFj =104 Nm cFj =10 Nsm (j = 2 F )
Constraint at xF = 01857L (sound files available)
simpler techniques that could be efficient to generate576
convincing sounds from the vibratory responses The577
method used here is to treat the guitar body as a linear578
input-output system and to compute the pressure by579
convolution benefiting from the local excitation of the580
body by the strings and using a vibro-acoustic impulse581
response of the test guitar This approach does not by582
any means simulate the sound response everywhere in the583
pressure field but it has the merit of adding the influence584
of the radiating properties of the body at a very low585
computational cost and of providing sounds of greater586
perceptual quality than the commonly used vibratory587
signals at the bridge The pressure at location rp can be588
computed by the convolution of the time-history of the589
stringbridge interaction force FY (rb t) and the vibro-590
acoustic impulse response function hY (rb rp t) between591
the bridge and microphone by592
p(rp t) = hY (rb rp t) lowast FY (rb t) (51)
The extension to account for the effects of all the593
strings is readily obtained by linear superposition594
principle summing the contributions from each string595
Measurements of the vibro-acoustic impulse response596
were performed under the same experimental conditions597
as for transfer function measurements using an598
additional condenser microphone Impulsive excitation599
was applied in the perpendicular direction to the600
soundboard and the acoustic pressure was measured601
at a distance of 1 m from the central axis of the602
guitar in a quiet room with all strings damped (see603
Fig 1) The impulse response was computed by inverse604
Fourier transform of the complex transfer function built605
from the force and response signals described in the606
frequency domain Results in Fig 13 pertain to the607
simulation of a G major arpeggio obtained with the608
two lowest and the top strings constrained using five609
auxiliary flexible-inertial constraints on each string The610
effect of the radiation properties of the body on the611
computed vibratory signals is supported when comparing612
the spectrograms in Fig 13 which clearly shows the613
spectral filtering of the vibration signal614
Figure 13 Spectrograms of the bridge force (top) and sound
pressure (bottom) Nonlinear six-stringbody coupled model
Arpeggio of a G major chord with three strings stopped using
a set of auxiliary oscillators (sound files available)
V CONCLUSIONS615
Following our recent work showing the potential616
of the modal Udwadia-Kalaba equations to physically617
simulate musical instruments we presented in this paper618
a computational approach that allows efficient time-619
domain simulations of realistic guitar plucks In this620
work we further generalized our simple stringbody621
coupled model toward the development of a more622
realistic guitar model by including the complete set623
of strings the two polarizations of string motions the624
string geometric nonlinearity the stringfret interaction625
that controls the playing frequency and the body626
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 11
dynamics of an actual instrument The dynamics of627
the various subsystems were formulated in terms of628
their unconstrained modes and coupling were achieved629
by enforcing several constraint conditions One of the630
developments of the model was the introduction of631
flexible-dissipative-inertial supplementary equations as632
an alternative to the common kinematical constraints633
which here improve the role of the stringfingerfret634
interaction at the fingerboard This approach can635
also benefit the modeling of other complex dynamical636
systems outside the realm of physical instrument637
modeling To improve the realism of the synthesis638
we also included crudely the radiation properties of639
the instrument body into the computational model640
through an ad hoc procedure Sound pressure was641
computed adopting a hybrid approach by convoluting642
the computed bridge force signals with a vibro-acoustic643
impulse response measured on the test guitar Such644
vibro-acoustic response can be measured in an anechoic645
chamber or else include the acoustics of any specific646
room This approach appeared not only straightforward647
to compute realistic sounds of string instruments for648
which the radiating components are subjected to point-649
wise excitation but also very efficient compared to heavy650
computational sound radiation techniques651
ACKNOWLEDGMENTS652
The authors thanks Paulo Vaz de Carvalho for653
lending the guitar and interesting discussions This work654
is funded by national funds through the FCT ndash Fundacao655
para a Ciencia e a Tecnologia IP under the Norma656
Transitoria - DL 572016CP1453CT0007657
REFERENCES658
1JH Ginsberg Mechanical and Structural Vibrations Theory659
and Applications chapter 9 Wiley 2001660
2J Antunes and V Debut Dynamical computation of constrained661
flexible systems using a modal Udwadia-Kalaba formulation662
Application to musical instruments J Acoust Soc Am663
141764 2017664
3J Antunes V Debut L Borsoi X Delaune and P Piteau665
A modal Udwadia-Kalaba formulation for the vibro-impact666
modelling of continuous flexible systems with intermittent667
contacts Procedia Engineering 199322ndash329 2017668
4FE Udwadia and RE Kalaba Analytical dynamics - A new669
approach Cambridge University Press 1996670
5C Valette The mechanics of the vibrating string Mechanics of671
musical instruments Springer 1995672
6J Woodhouse On the synthesis of guitar plucks Acta Acust673
united Ac 90928ndash944 2004674
7M E McIntyre RT Schumacher and J Woodhouse On the675
oscillations of musical instruments Journal of the Acoustical676
Society of America 741325ndash1345 1983677
8A Chaigne On the use of finite differences for musical678
synthesis Application to plucked stringed instruments Journal679
drsquoAcoustique 5181ndash211 1992680
9JM Adrien The missing link Modal synthesis In G DePoli681
A Picalli and C Roads editors Representations of musical682
sounds MIT Press Cambridge 1991683
10S Bilbao Numerical Sound Synthesis Finite Difference684
Schemes and Simulation in Musical Acoustics Wiley Publishing685
2009686
11L Trautmann and R Rabenstein Multirate simulations687
of string vibrations includingnonlinear fret-string interactions688
usingthe functional transformation method EURASIP Journal689
on Applied Signal Processing 7949ndash963 2004690
12V Chatziioannou and M van Walstijn Energy conserving691
schemes for the simulation of musical instrument contact692
dynamics J Sound Vib 339262ndash279 2015693
13S Bilbao and A Torin Numerical modeling and sound synthesis694
for articulated stringfretboard interactions Acta Acust united695
Ac 63336ndash347 2015696
14V Debut M Carvalho M Marques and J Antunes Physics-697
based modeling techniques of a twelve-string portuguese guitar698
a three-dimensional non-linear time-domain computational699
approach for the multiple-stringsbridgesoundboard coupled700
dynamics Appl Acoust 1083ndash18 2016701
15M van Walstijn and J Bridges Simulation of distributed contact702
in string instruments A modal expansion approach In 24th703
European Signal Processing Conference (Eusipco 2016) pages704
1023ndash1027 2013705
16C Issanchou S Bilbao JL Le Carrou C Touze and O Doare706
A modal-based approach to the nonlinear vibration of strings707
against a unilateral obstacle simulations and experiments in the708
pointwise case J Sound Vib 393229ndash251 2017709
17R Bader Computational Mechanics of the Classical Guitar710
Springer-Verlag Berlin Heidelberg 2005711
18H Mansour J Woodhouse and G Scavone Enhanced wave-712
based modelling of musical strings Part 1 Plucked strings Acta713
Acust united Ac 1021094ndash1107 2016714
19O Inacio J Antunes and MCM Wright Computational715
modeling of string-body interaction for the violin family and716
simulation of wolf notes J Sound Vib 310260ndash286 2007717
20S Bilbao A Torin and V Chatziioannou Numerical modeling718
of collisions in musical instruments Acta Acust united Ac719
101155ndash173 2015720
21C Issanchou JL Le Carrou C Touze B Fabre and O Doare721
Stringfrets contacts in the electric bass sound Simulations and722
experiments App Acoust 129217ndash228 2018723
22D J Braun and M Goldfarb Eliminating constraint drift in the724
numerical simulation of constrained dynamical systems Comput725
Methods in Appl Mech Eng 198(37)3151 ndash 3160 2009726
23GF Carrier On the nonlinear vibration problem of the elastic727
string Q Appl Math 3157ndash165 1945728
24G Kirchhoff Vorlesungen uber Mechanik Teubner 1883729
25J Pakarinen V Valimaki and M Karjalainen Physics-based730
methods for modeling nonlinear vibrating strings Acta Acust731
united Ac 91312ndash325 2005732
26PMC Morse and KU Ingard Theoretical Acoustics733
International series in pure and applied physics Princeton734
University Press 1968735
27J Juang Applied System Identification PTR Prentice-Hall736
Inc New Jersey 1994737
28B E Richardson The acoustical development of the guitar J738
Catgut Acoust Soc 51ndash10 1994739
29J Woodhouse Plucked guitar transients Comparison of740
measurements and synthesis Acta Acust united Ac 90945ndash741
965 2004742
30M Tuckerman B J Berne and G J Martyna Reversible743
multiple time scale molecular dynamics J Chem Phys744
97(3)1990ndash2001 1992745
31A Chaigne Viscoelastic properties of nylon guitar strings J746
Catgut Acoust Soc Series II 121ndash27 1991747
32T Tolonen V Valimaki and M Karjalainen Modeling of748
tension modulation nonlinearity in plucked strings IEEE Trans749
Speech Audio Process 8(3)300ndash310 2000750
33A Chaigne and J Kergomard Acoustics of Musical Instruments751
Modern Acoustics and Signal Processing Springer New York752
2016753
12 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
754
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 13
modulates the other peaks A close look shows that528
the spectra are affected much more sensitively below529
the resonance of the finger oscillator with amplitude530
variations analogous to a system driven in the stiffness-531
controlled region Decreasing the stiffness causes a rise532
in amplitude of the undesirable modes while increasing533
the stiffness lower their amplitudes As observed for the534
stringbody coupling33 inharmonicity is also created by535
the stiffness termination which in turn affects the tuning536
of the note With the increase of stiffness perturbations537
of the string modes are reduced and tuning is improved538
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
4 Nm
0 1000 2000 3000 4000 5000
100
1010
| F
SB
(f)
|
kf = 10
5 Nm
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
kf = 10
6 Nm
Figure 9 Spectra of the coupling force at the rigid bridge
Stiffness is varied F=1 c1 = 0 Nsm
Fig 10 then shows the effect of varying the damping539
keeping the stiffness constant Not surprisingly if540
damping increases the amplitude of the coupled541
modes close to the oscillator resonance decrease and542
the frequency region affected becomes wider This543
significantly affects the time envelope of the pluck note544
that becomes smoother and has a shorter-time decay545
Finally the effect of the number of contacts is studied546
using the rigid contact model which is convenient for547
this purpose Fig 11 gives a picture of the spatial548
distribution of the string energy by plotting the standard549
deviation of string motion as a function of its length550
As seen the ldquopassiverdquo region is affected with great551
sensitivity by the number of contacts and surprisingly552
increasing the contact points does not necessarily reduce553
the undesirable transfer of energy Computations show554
that the effectiveness of the multi-point rigid constraints555
approach is strongly problem-dependent and varies in a556
non-obvious way with several factors including the finger557
width the spacing between contact points the relative558
string length from the ends and the excitation spectrum559
Taken in combination these results and noticing560
that the perturbations induced by each finger oscillator561
are additive due to the linear coupling a robust and562
physically sound strategy to simulate the pressing finger563
would be to implement a set of flexible-dissipative564
0 1000 2000 3000 4000 5000
Frequency [ Hz ]
100
1010
| F
SB
(f)
|
cf = 1 Nsm
cf = 10 Nsm
0 1 2 3 4 5
Time [ s ]
-2
0
2
4
Y (
t )
10-3
Figure 10 Spectra of the coupling force at the rigid
bridge (up) and time history of the string motion (bottom)
Damping is varied F=1 k1 = 105 Nm
constraints imposing a first stiff constraint in order565
to ensure a precise tuning followed by several soft566
constraints that would control the waveform and spectral567
content of the decaying response depending on their568
number and damping properties For illustration569
the spectrograms of two plucks for different sets of570
parameters are presented in Fig 12571
0 01 02 03 04 05 06
x [ m ]
10-5
10-2
ST
D (
Y(t
) )
F=1
F=2
F=3
F=5
016 017 018 019 02
x [ m ]
10-5
ST
D (
Y(t
) )
Figure 11 Standard deviation of the string motion Number
of contacts is varied Vertical dotted lines represent the
constraints locations Finger width (0015m) is kept constant
C Simulated acoustic radiation572
As an alternative to computing sounds from full573
vibro-acoustic models that usually require the use of574
massive computational resources it is tempting to apply575
10 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
Figure 12 Spectrograms of the force at the rigid bridge Up
F = 3 bottom F = 5 Parameter values kF1=106 Nm
cF1=1 Nsm and kFj =104 Nm cFj =10 Nsm (j = 2 F )
Constraint at xF = 01857L (sound files available)
simpler techniques that could be efficient to generate576
convincing sounds from the vibratory responses The577
method used here is to treat the guitar body as a linear578
input-output system and to compute the pressure by579
convolution benefiting from the local excitation of the580
body by the strings and using a vibro-acoustic impulse581
response of the test guitar This approach does not by582
any means simulate the sound response everywhere in the583
pressure field but it has the merit of adding the influence584
of the radiating properties of the body at a very low585
computational cost and of providing sounds of greater586
perceptual quality than the commonly used vibratory587
signals at the bridge The pressure at location rp can be588
computed by the convolution of the time-history of the589
stringbridge interaction force FY (rb t) and the vibro-590
acoustic impulse response function hY (rb rp t) between591
the bridge and microphone by592
p(rp t) = hY (rb rp t) lowast FY (rb t) (51)
The extension to account for the effects of all the593
strings is readily obtained by linear superposition594
principle summing the contributions from each string595
Measurements of the vibro-acoustic impulse response596
were performed under the same experimental conditions597
as for transfer function measurements using an598
additional condenser microphone Impulsive excitation599
was applied in the perpendicular direction to the600
soundboard and the acoustic pressure was measured601
at a distance of 1 m from the central axis of the602
guitar in a quiet room with all strings damped (see603
Fig 1) The impulse response was computed by inverse604
Fourier transform of the complex transfer function built605
from the force and response signals described in the606
frequency domain Results in Fig 13 pertain to the607
simulation of a G major arpeggio obtained with the608
two lowest and the top strings constrained using five609
auxiliary flexible-inertial constraints on each string The610
effect of the radiation properties of the body on the611
computed vibratory signals is supported when comparing612
the spectrograms in Fig 13 which clearly shows the613
spectral filtering of the vibration signal614
Figure 13 Spectrograms of the bridge force (top) and sound
pressure (bottom) Nonlinear six-stringbody coupled model
Arpeggio of a G major chord with three strings stopped using
a set of auxiliary oscillators (sound files available)
V CONCLUSIONS615
Following our recent work showing the potential616
of the modal Udwadia-Kalaba equations to physically617
simulate musical instruments we presented in this paper618
a computational approach that allows efficient time-619
domain simulations of realistic guitar plucks In this620
work we further generalized our simple stringbody621
coupled model toward the development of a more622
realistic guitar model by including the complete set623
of strings the two polarizations of string motions the624
string geometric nonlinearity the stringfret interaction625
that controls the playing frequency and the body626
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 11
dynamics of an actual instrument The dynamics of627
the various subsystems were formulated in terms of628
their unconstrained modes and coupling were achieved629
by enforcing several constraint conditions One of the630
developments of the model was the introduction of631
flexible-dissipative-inertial supplementary equations as632
an alternative to the common kinematical constraints633
which here improve the role of the stringfingerfret634
interaction at the fingerboard This approach can635
also benefit the modeling of other complex dynamical636
systems outside the realm of physical instrument637
modeling To improve the realism of the synthesis638
we also included crudely the radiation properties of639
the instrument body into the computational model640
through an ad hoc procedure Sound pressure was641
computed adopting a hybrid approach by convoluting642
the computed bridge force signals with a vibro-acoustic643
impulse response measured on the test guitar Such644
vibro-acoustic response can be measured in an anechoic645
chamber or else include the acoustics of any specific646
room This approach appeared not only straightforward647
to compute realistic sounds of string instruments for648
which the radiating components are subjected to point-649
wise excitation but also very efficient compared to heavy650
computational sound radiation techniques651
ACKNOWLEDGMENTS652
The authors thanks Paulo Vaz de Carvalho for653
lending the guitar and interesting discussions This work654
is funded by national funds through the FCT ndash Fundacao655
para a Ciencia e a Tecnologia IP under the Norma656
Transitoria - DL 572016CP1453CT0007657
REFERENCES658
1JH Ginsberg Mechanical and Structural Vibrations Theory659
and Applications chapter 9 Wiley 2001660
2J Antunes and V Debut Dynamical computation of constrained661
flexible systems using a modal Udwadia-Kalaba formulation662
Application to musical instruments J Acoust Soc Am663
141764 2017664
3J Antunes V Debut L Borsoi X Delaune and P Piteau665
A modal Udwadia-Kalaba formulation for the vibro-impact666
modelling of continuous flexible systems with intermittent667
contacts Procedia Engineering 199322ndash329 2017668
4FE Udwadia and RE Kalaba Analytical dynamics - A new669
approach Cambridge University Press 1996670
5C Valette The mechanics of the vibrating string Mechanics of671
musical instruments Springer 1995672
6J Woodhouse On the synthesis of guitar plucks Acta Acust673
united Ac 90928ndash944 2004674
7M E McIntyre RT Schumacher and J Woodhouse On the675
oscillations of musical instruments Journal of the Acoustical676
Society of America 741325ndash1345 1983677
8A Chaigne On the use of finite differences for musical678
synthesis Application to plucked stringed instruments Journal679
drsquoAcoustique 5181ndash211 1992680
9JM Adrien The missing link Modal synthesis In G DePoli681
A Picalli and C Roads editors Representations of musical682
sounds MIT Press Cambridge 1991683
10S Bilbao Numerical Sound Synthesis Finite Difference684
Schemes and Simulation in Musical Acoustics Wiley Publishing685
2009686
11L Trautmann and R Rabenstein Multirate simulations687
of string vibrations includingnonlinear fret-string interactions688
usingthe functional transformation method EURASIP Journal689
on Applied Signal Processing 7949ndash963 2004690
12V Chatziioannou and M van Walstijn Energy conserving691
schemes for the simulation of musical instrument contact692
dynamics J Sound Vib 339262ndash279 2015693
13S Bilbao and A Torin Numerical modeling and sound synthesis694
for articulated stringfretboard interactions Acta Acust united695
Ac 63336ndash347 2015696
14V Debut M Carvalho M Marques and J Antunes Physics-697
based modeling techniques of a twelve-string portuguese guitar698
a three-dimensional non-linear time-domain computational699
approach for the multiple-stringsbridgesoundboard coupled700
dynamics Appl Acoust 1083ndash18 2016701
15M van Walstijn and J Bridges Simulation of distributed contact702
in string instruments A modal expansion approach In 24th703
European Signal Processing Conference (Eusipco 2016) pages704
1023ndash1027 2013705
16C Issanchou S Bilbao JL Le Carrou C Touze and O Doare706
A modal-based approach to the nonlinear vibration of strings707
against a unilateral obstacle simulations and experiments in the708
pointwise case J Sound Vib 393229ndash251 2017709
17R Bader Computational Mechanics of the Classical Guitar710
Springer-Verlag Berlin Heidelberg 2005711
18H Mansour J Woodhouse and G Scavone Enhanced wave-712
based modelling of musical strings Part 1 Plucked strings Acta713
Acust united Ac 1021094ndash1107 2016714
19O Inacio J Antunes and MCM Wright Computational715
modeling of string-body interaction for the violin family and716
simulation of wolf notes J Sound Vib 310260ndash286 2007717
20S Bilbao A Torin and V Chatziioannou Numerical modeling718
of collisions in musical instruments Acta Acust united Ac719
101155ndash173 2015720
21C Issanchou JL Le Carrou C Touze B Fabre and O Doare721
Stringfrets contacts in the electric bass sound Simulations and722
experiments App Acoust 129217ndash228 2018723
22D J Braun and M Goldfarb Eliminating constraint drift in the724
numerical simulation of constrained dynamical systems Comput725
Methods in Appl Mech Eng 198(37)3151 ndash 3160 2009726
23GF Carrier On the nonlinear vibration problem of the elastic727
string Q Appl Math 3157ndash165 1945728
24G Kirchhoff Vorlesungen uber Mechanik Teubner 1883729
25J Pakarinen V Valimaki and M Karjalainen Physics-based730
methods for modeling nonlinear vibrating strings Acta Acust731
united Ac 91312ndash325 2005732
26PMC Morse and KU Ingard Theoretical Acoustics733
International series in pure and applied physics Princeton734
University Press 1968735
27J Juang Applied System Identification PTR Prentice-Hall736
Inc New Jersey 1994737
28B E Richardson The acoustical development of the guitar J738
Catgut Acoust Soc 51ndash10 1994739
29J Woodhouse Plucked guitar transients Comparison of740
measurements and synthesis Acta Acust united Ac 90945ndash741
965 2004742
30M Tuckerman B J Berne and G J Martyna Reversible743
multiple time scale molecular dynamics J Chem Phys744
97(3)1990ndash2001 1992745
31A Chaigne Viscoelastic properties of nylon guitar strings J746
Catgut Acoust Soc Series II 121ndash27 1991747
32T Tolonen V Valimaki and M Karjalainen Modeling of748
tension modulation nonlinearity in plucked strings IEEE Trans749
Speech Audio Process 8(3)300ndash310 2000750
33A Chaigne and J Kergomard Acoustics of Musical Instruments751
Modern Acoustics and Signal Processing Springer New York752
2016753
12 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
754
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 13
Figure 12 Spectrograms of the force at the rigid bridge Up
F = 3 bottom F = 5 Parameter values kF1=106 Nm
cF1=1 Nsm and kFj =104 Nm cFj =10 Nsm (j = 2 F )
Constraint at xF = 01857L (sound files available)
simpler techniques that could be efficient to generate576
convincing sounds from the vibratory responses The577
method used here is to treat the guitar body as a linear578
input-output system and to compute the pressure by579
convolution benefiting from the local excitation of the580
body by the strings and using a vibro-acoustic impulse581
response of the test guitar This approach does not by582
any means simulate the sound response everywhere in the583
pressure field but it has the merit of adding the influence584
of the radiating properties of the body at a very low585
computational cost and of providing sounds of greater586
perceptual quality than the commonly used vibratory587
signals at the bridge The pressure at location rp can be588
computed by the convolution of the time-history of the589
stringbridge interaction force FY (rb t) and the vibro-590
acoustic impulse response function hY (rb rp t) between591
the bridge and microphone by592
p(rp t) = hY (rb rp t) lowast FY (rb t) (51)
The extension to account for the effects of all the593
strings is readily obtained by linear superposition594
principle summing the contributions from each string595
Measurements of the vibro-acoustic impulse response596
were performed under the same experimental conditions597
as for transfer function measurements using an598
additional condenser microphone Impulsive excitation599
was applied in the perpendicular direction to the600
soundboard and the acoustic pressure was measured601
at a distance of 1 m from the central axis of the602
guitar in a quiet room with all strings damped (see603
Fig 1) The impulse response was computed by inverse604
Fourier transform of the complex transfer function built605
from the force and response signals described in the606
frequency domain Results in Fig 13 pertain to the607
simulation of a G major arpeggio obtained with the608
two lowest and the top strings constrained using five609
auxiliary flexible-inertial constraints on each string The610
effect of the radiation properties of the body on the611
computed vibratory signals is supported when comparing612
the spectrograms in Fig 13 which clearly shows the613
spectral filtering of the vibration signal614
Figure 13 Spectrograms of the bridge force (top) and sound
pressure (bottom) Nonlinear six-stringbody coupled model
Arpeggio of a G major chord with three strings stopped using
a set of auxiliary oscillators (sound files available)
V CONCLUSIONS615
Following our recent work showing the potential616
of the modal Udwadia-Kalaba equations to physically617
simulate musical instruments we presented in this paper618
a computational approach that allows efficient time-619
domain simulations of realistic guitar plucks In this620
work we further generalized our simple stringbody621
coupled model toward the development of a more622
realistic guitar model by including the complete set623
of strings the two polarizations of string motions the624
string geometric nonlinearity the stringfret interaction625
that controls the playing frequency and the body626
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 11
dynamics of an actual instrument The dynamics of627
the various subsystems were formulated in terms of628
their unconstrained modes and coupling were achieved629
by enforcing several constraint conditions One of the630
developments of the model was the introduction of631
flexible-dissipative-inertial supplementary equations as632
an alternative to the common kinematical constraints633
which here improve the role of the stringfingerfret634
interaction at the fingerboard This approach can635
also benefit the modeling of other complex dynamical636
systems outside the realm of physical instrument637
modeling To improve the realism of the synthesis638
we also included crudely the radiation properties of639
the instrument body into the computational model640
through an ad hoc procedure Sound pressure was641
computed adopting a hybrid approach by convoluting642
the computed bridge force signals with a vibro-acoustic643
impulse response measured on the test guitar Such644
vibro-acoustic response can be measured in an anechoic645
chamber or else include the acoustics of any specific646
room This approach appeared not only straightforward647
to compute realistic sounds of string instruments for648
which the radiating components are subjected to point-649
wise excitation but also very efficient compared to heavy650
computational sound radiation techniques651
ACKNOWLEDGMENTS652
The authors thanks Paulo Vaz de Carvalho for653
lending the guitar and interesting discussions This work654
is funded by national funds through the FCT ndash Fundacao655
para a Ciencia e a Tecnologia IP under the Norma656
Transitoria - DL 572016CP1453CT0007657
REFERENCES658
1JH Ginsberg Mechanical and Structural Vibrations Theory659
and Applications chapter 9 Wiley 2001660
2J Antunes and V Debut Dynamical computation of constrained661
flexible systems using a modal Udwadia-Kalaba formulation662
Application to musical instruments J Acoust Soc Am663
141764 2017664
3J Antunes V Debut L Borsoi X Delaune and P Piteau665
A modal Udwadia-Kalaba formulation for the vibro-impact666
modelling of continuous flexible systems with intermittent667
contacts Procedia Engineering 199322ndash329 2017668
4FE Udwadia and RE Kalaba Analytical dynamics - A new669
approach Cambridge University Press 1996670
5C Valette The mechanics of the vibrating string Mechanics of671
musical instruments Springer 1995672
6J Woodhouse On the synthesis of guitar plucks Acta Acust673
united Ac 90928ndash944 2004674
7M E McIntyre RT Schumacher and J Woodhouse On the675
oscillations of musical instruments Journal of the Acoustical676
Society of America 741325ndash1345 1983677
8A Chaigne On the use of finite differences for musical678
synthesis Application to plucked stringed instruments Journal679
drsquoAcoustique 5181ndash211 1992680
9JM Adrien The missing link Modal synthesis In G DePoli681
A Picalli and C Roads editors Representations of musical682
sounds MIT Press Cambridge 1991683
10S Bilbao Numerical Sound Synthesis Finite Difference684
Schemes and Simulation in Musical Acoustics Wiley Publishing685
2009686
11L Trautmann and R Rabenstein Multirate simulations687
of string vibrations includingnonlinear fret-string interactions688
usingthe functional transformation method EURASIP Journal689
on Applied Signal Processing 7949ndash963 2004690
12V Chatziioannou and M van Walstijn Energy conserving691
schemes for the simulation of musical instrument contact692
dynamics J Sound Vib 339262ndash279 2015693
13S Bilbao and A Torin Numerical modeling and sound synthesis694
for articulated stringfretboard interactions Acta Acust united695
Ac 63336ndash347 2015696
14V Debut M Carvalho M Marques and J Antunes Physics-697
based modeling techniques of a twelve-string portuguese guitar698
a three-dimensional non-linear time-domain computational699
approach for the multiple-stringsbridgesoundboard coupled700
dynamics Appl Acoust 1083ndash18 2016701
15M van Walstijn and J Bridges Simulation of distributed contact702
in string instruments A modal expansion approach In 24th703
European Signal Processing Conference (Eusipco 2016) pages704
1023ndash1027 2013705
16C Issanchou S Bilbao JL Le Carrou C Touze and O Doare706
A modal-based approach to the nonlinear vibration of strings707
against a unilateral obstacle simulations and experiments in the708
pointwise case J Sound Vib 393229ndash251 2017709
17R Bader Computational Mechanics of the Classical Guitar710
Springer-Verlag Berlin Heidelberg 2005711
18H Mansour J Woodhouse and G Scavone Enhanced wave-712
based modelling of musical strings Part 1 Plucked strings Acta713
Acust united Ac 1021094ndash1107 2016714
19O Inacio J Antunes and MCM Wright Computational715
modeling of string-body interaction for the violin family and716
simulation of wolf notes J Sound Vib 310260ndash286 2007717
20S Bilbao A Torin and V Chatziioannou Numerical modeling718
of collisions in musical instruments Acta Acust united Ac719
101155ndash173 2015720
21C Issanchou JL Le Carrou C Touze B Fabre and O Doare721
Stringfrets contacts in the electric bass sound Simulations and722
experiments App Acoust 129217ndash228 2018723
22D J Braun and M Goldfarb Eliminating constraint drift in the724
numerical simulation of constrained dynamical systems Comput725
Methods in Appl Mech Eng 198(37)3151 ndash 3160 2009726
23GF Carrier On the nonlinear vibration problem of the elastic727
string Q Appl Math 3157ndash165 1945728
24G Kirchhoff Vorlesungen uber Mechanik Teubner 1883729
25J Pakarinen V Valimaki and M Karjalainen Physics-based730
methods for modeling nonlinear vibrating strings Acta Acust731
united Ac 91312ndash325 2005732
26PMC Morse and KU Ingard Theoretical Acoustics733
International series in pure and applied physics Princeton734
University Press 1968735
27J Juang Applied System Identification PTR Prentice-Hall736
Inc New Jersey 1994737
28B E Richardson The acoustical development of the guitar J738
Catgut Acoust Soc 51ndash10 1994739
29J Woodhouse Plucked guitar transients Comparison of740
measurements and synthesis Acta Acust united Ac 90945ndash741
965 2004742
30M Tuckerman B J Berne and G J Martyna Reversible743
multiple time scale molecular dynamics J Chem Phys744
97(3)1990ndash2001 1992745
31A Chaigne Viscoelastic properties of nylon guitar strings J746
Catgut Acoust Soc Series II 121ndash27 1991747
32T Tolonen V Valimaki and M Karjalainen Modeling of748
tension modulation nonlinearity in plucked strings IEEE Trans749
Speech Audio Process 8(3)300ndash310 2000750
33A Chaigne and J Kergomard Acoustics of Musical Instruments751
Modern Acoustics and Signal Processing Springer New York752
2016753
12 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
754
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 13
dynamics of an actual instrument The dynamics of627
the various subsystems were formulated in terms of628
their unconstrained modes and coupling were achieved629
by enforcing several constraint conditions One of the630
developments of the model was the introduction of631
flexible-dissipative-inertial supplementary equations as632
an alternative to the common kinematical constraints633
which here improve the role of the stringfingerfret634
interaction at the fingerboard This approach can635
also benefit the modeling of other complex dynamical636
systems outside the realm of physical instrument637
modeling To improve the realism of the synthesis638
we also included crudely the radiation properties of639
the instrument body into the computational model640
through an ad hoc procedure Sound pressure was641
computed adopting a hybrid approach by convoluting642
the computed bridge force signals with a vibro-acoustic643
impulse response measured on the test guitar Such644
vibro-acoustic response can be measured in an anechoic645
chamber or else include the acoustics of any specific646
room This approach appeared not only straightforward647
to compute realistic sounds of string instruments for648
which the radiating components are subjected to point-649
wise excitation but also very efficient compared to heavy650
computational sound radiation techniques651
ACKNOWLEDGMENTS652
The authors thanks Paulo Vaz de Carvalho for653
lending the guitar and interesting discussions This work654
is funded by national funds through the FCT ndash Fundacao655
para a Ciencia e a Tecnologia IP under the Norma656
Transitoria - DL 572016CP1453CT0007657
REFERENCES658
1JH Ginsberg Mechanical and Structural Vibrations Theory659
and Applications chapter 9 Wiley 2001660
2J Antunes and V Debut Dynamical computation of constrained661
flexible systems using a modal Udwadia-Kalaba formulation662
Application to musical instruments J Acoust Soc Am663
141764 2017664
3J Antunes V Debut L Borsoi X Delaune and P Piteau665
A modal Udwadia-Kalaba formulation for the vibro-impact666
modelling of continuous flexible systems with intermittent667
contacts Procedia Engineering 199322ndash329 2017668
4FE Udwadia and RE Kalaba Analytical dynamics - A new669
approach Cambridge University Press 1996670
5C Valette The mechanics of the vibrating string Mechanics of671
musical instruments Springer 1995672
6J Woodhouse On the synthesis of guitar plucks Acta Acust673
united Ac 90928ndash944 2004674
7M E McIntyre RT Schumacher and J Woodhouse On the675
oscillations of musical instruments Journal of the Acoustical676
Society of America 741325ndash1345 1983677
8A Chaigne On the use of finite differences for musical678
synthesis Application to plucked stringed instruments Journal679
drsquoAcoustique 5181ndash211 1992680
9JM Adrien The missing link Modal synthesis In G DePoli681
A Picalli and C Roads editors Representations of musical682
sounds MIT Press Cambridge 1991683
10S Bilbao Numerical Sound Synthesis Finite Difference684
Schemes and Simulation in Musical Acoustics Wiley Publishing685
2009686
11L Trautmann and R Rabenstein Multirate simulations687
of string vibrations includingnonlinear fret-string interactions688
usingthe functional transformation method EURASIP Journal689
on Applied Signal Processing 7949ndash963 2004690
12V Chatziioannou and M van Walstijn Energy conserving691
schemes for the simulation of musical instrument contact692
dynamics J Sound Vib 339262ndash279 2015693
13S Bilbao and A Torin Numerical modeling and sound synthesis694
for articulated stringfretboard interactions Acta Acust united695
Ac 63336ndash347 2015696
14V Debut M Carvalho M Marques and J Antunes Physics-697
based modeling techniques of a twelve-string portuguese guitar698
a three-dimensional non-linear time-domain computational699
approach for the multiple-stringsbridgesoundboard coupled700
dynamics Appl Acoust 1083ndash18 2016701
15M van Walstijn and J Bridges Simulation of distributed contact702
in string instruments A modal expansion approach In 24th703
European Signal Processing Conference (Eusipco 2016) pages704
1023ndash1027 2013705
16C Issanchou S Bilbao JL Le Carrou C Touze and O Doare706
A modal-based approach to the nonlinear vibration of strings707
against a unilateral obstacle simulations and experiments in the708
pointwise case J Sound Vib 393229ndash251 2017709
17R Bader Computational Mechanics of the Classical Guitar710
Springer-Verlag Berlin Heidelberg 2005711
18H Mansour J Woodhouse and G Scavone Enhanced wave-712
based modelling of musical strings Part 1 Plucked strings Acta713
Acust united Ac 1021094ndash1107 2016714
19O Inacio J Antunes and MCM Wright Computational715
modeling of string-body interaction for the violin family and716
simulation of wolf notes J Sound Vib 310260ndash286 2007717
20S Bilbao A Torin and V Chatziioannou Numerical modeling718
of collisions in musical instruments Acta Acust united Ac719
101155ndash173 2015720
21C Issanchou JL Le Carrou C Touze B Fabre and O Doare721
Stringfrets contacts in the electric bass sound Simulations and722
experiments App Acoust 129217ndash228 2018723
22D J Braun and M Goldfarb Eliminating constraint drift in the724
numerical simulation of constrained dynamical systems Comput725
Methods in Appl Mech Eng 198(37)3151 ndash 3160 2009726
23GF Carrier On the nonlinear vibration problem of the elastic727
string Q Appl Math 3157ndash165 1945728
24G Kirchhoff Vorlesungen uber Mechanik Teubner 1883729
25J Pakarinen V Valimaki and M Karjalainen Physics-based730
methods for modeling nonlinear vibrating strings Acta Acust731
united Ac 91312ndash325 2005732
26PMC Morse and KU Ingard Theoretical Acoustics733
International series in pure and applied physics Princeton734
University Press 1968735
27J Juang Applied System Identification PTR Prentice-Hall736
Inc New Jersey 1994737
28B E Richardson The acoustical development of the guitar J738
Catgut Acoust Soc 51ndash10 1994739
29J Woodhouse Plucked guitar transients Comparison of740
measurements and synthesis Acta Acust united Ac 90945ndash741
965 2004742
30M Tuckerman B J Berne and G J Martyna Reversible743
multiple time scale molecular dynamics J Chem Phys744
97(3)1990ndash2001 1992745
31A Chaigne Viscoelastic properties of nylon guitar strings J746
Catgut Acoust Soc Series II 121ndash27 1991747
32T Tolonen V Valimaki and M Karjalainen Modeling of748
tension modulation nonlinearity in plucked strings IEEE Trans749
Speech Audio Process 8(3)300ndash310 2000750
33A Chaigne and J Kergomard Acoustics of Musical Instruments751
Modern Acoustics and Signal Processing Springer New York752
2016753
12 J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation
754
J Acoust Soc Am May 14 2020 Guitar synthesis using Udwadia-Kalaba modal formulation 13