is an open access repository that collects the work of Arts et Métiers Institute of
Technology researchers and makes it freely available over the web where possible.
This is an author-deposited version published in:
https://sam.ensam.eu
Handle ID: .
http://hdl.handle.net/10985/7362
To cite this version :
Eric DUCASSE - On waveguide modeling of stiff piano string - J. Acoust. Soc. Am. - Vol. 118,
n°3, p.1776-1781 - 2005
On waveguide modeling of sti piano strings
Éri Du asse
É ole NationaleSupérieured'Arts etMétiers, C.E.R. de Bordeaux-Talen e,33405 Talen e edex, Fran e
eri .du asseensam.eu
ABSTRACT
Bensa et al. [J. A oust. So . Am. 114(2) 10951107 (2003), Se tion IV℄ re ently proposed a waveguide
model for the transverse displa ement of a sti piano string. The study des ribed here is an attempt to ast
a omplementary light on this topi , based on a ommon wave approa h instead of a modal approa h. A pair
of weakly attenuated traveling waves and a pair of fast-de aying waves both satisfy the one-dimensional wave
equation developed by Bensa et al.. These solutions have to be arefully onsidered, however, for portions of
string intera ting with the hammer felt,the bridge,or the apo d'astro bar.
PACS number: 43.75.Mn
I. INTRODUCTION
The rst attempt for synthesizing musi al sounds using physi al models was made more than 30 years ago
by Hiller and Ruiz
1
. Starting from the one-dimensional wave equation
2
of the transverse displa ement of a
string, three main approa hes are generally used for buildinga omputationalmodel: either the use of a nite
dieren e s heme (e.g. Refs. 3, 4), or a modal synthesis (e.g. Ref. 5), or the building of a Digital Waveguide
model(e.g. Refs.6,7,3). Awaveguide is onsideredhereasalterlikestru ture modelingone-dimensionalwave
propagationas purely losslessthroughoutthe length of thestring, withlossand dispersionlumped in terminating
lters
3
. The hammera tionand theree tion onditions atbothendsof thestringare alsomodeledaslumped
lters. Portions ofstring are distributed elements
6
represented by two-portnetworks. Inthis paperwe fo us on
this latter approa h.
Bensa et al. propose an improved one-dimensional wave equation [Ref. 3, Eq. (6)℄ orresponding to a well
posed modelofsti pianostrings. In Se . IV.Bthey use amodalapproa h toextra tthe waveguide parameters
from the partial dierential equation (PDE). This approa h requires the knowledge of both the length of the
string and the boundary onditions [Ref. 3, Eq. (10)℄ for the al ulation of the vibration modes asso iated
with standing waves. The waveguide parameters are then dedu ed from the hara teristi s of the rst mode
in luding the fundamental frequen y of the ideal string. The present paperis an attemptto show that a wave
approa h using omplexwave-numbers appearsasmore onvenientbe auseea hdispersiveattenuatedtraveling
wave isdire tly obtained fromthe one-dimensional wave equation,apartfrom the ree tionsat the endsof the
string. Assumingthattheone-dimensionalwaveequationislinear,time-andspa e-invariant,atwo-dimensional
Fourier-Lapla e transformation gives an algebrai equation relating the transverse displa ement of the string
(output)tothespa edistributionofexternalfor es(input)inthewave-number/frequen y domain. Thesolution
of this equation an beseen as the superposition of traveling waves whi ho ur ina waveguide model.
Somegeneralpointsaboutwaveguides, travelingwaves, and omplexwave-numbers areoutlinedinSe tionII
tohelp the readertounderstandthe followingse tions. In Se tionIII, afterthe one-dimensionalwave equation
advan edbyBensaetal.
3
isre alled,thisequationissolvedtoobtaintheresponseofaninnitestringtoapoint
impulse inthe wave-number/frequen y domain,showing that two fast-de aying traveling waves are omittedby
the modal approa h in Bensa et al.
3
. These fast-de aying waves an be negle ted only for portions of string
without sour es whi h are long enough. They should, however, be taken into a ount at the neighborhood of
II. WAVEGUIDE MODELING AND COMPLEX WAVE-NUMBERS
The question is this: howto pass from the wave equationof a one-dimensional system to awaveguide model?
A. From the one-dimensional wave equation to the transfer fun tion
1. A general one-dimensional wave equation
The vibration state of a one-dimensional system (e.g. the sti piano string) is assumed to be dened by a
fun tion
y
(e.g. the transverse displa ement of the string) of timet
and positionx
along the system.Ageneralone-dimensionalwaveequation,assumingthatitislinear,time-andspa e-invariant, anbewritten
as:
dx
X
m=0
dt
X
n=0
α
m,n
∂
m+n
y
∂ x
m
∂ t
n
(x, t) = f (x, t) ,
(1)where
f
is the spa e distribution of external for es (e.g. exerted by the hammer felt, the bridge and the apo d'astro bar).2. Time/frequen y and spa e/wave-number onversions
In the spa e/frequen y domain, ea h signal
s(x, t)
is onverted intobs(x, ω)
by Fourier transformation:bs(x, ω) =
Z
+∞
−∞
s(x, t) e
−i ω t
dt ,
(2)where
ω
is the angular frequen y andi
2
= −1
.
In the wave-number/frequen y domain,ea h signal
s(x, t)
be omesS(k, ω)
by FourierLapla e transforma-tion:S(k, ω) =
Z
+∞
−∞
bs(x, ω) e
−i k x
dx =
Z
+∞
−∞
Z
+∞
−∞
s(x, t) e
−i (ω t+k x)
dt dx ,
(3)where
k
is the omplex wave-number.3. An ordinary dierential equation in the spa e/frequen y domain
The Fourier transformationof the waveequation (1)leads toanordinary linear onstant- oe ient dierential
equation with respe t to
x
:dx
X
m=0
"
dt
X
n=0
α
m,n
(i ω)
n
#
∂
m
by
∂ x
m
(x, ω) = b
f (x, ω) .
(4)4. The transfer fun tion
In the wave-number/frequen y domain,we obtain:
"
dx
X
m=0
dt
X
n=0
α
m,n
i
m+n
ω
n
k
m
#
Y (k, ω) = F (k, ω) ⇐⇒ Y (k, ω) = H(k, ω) F (k, ω) ,
(5)where
H(k, ω)
is the transfer fun tion of the system. The denominator ofH(k, ω)
is a polynomial of orderd
x
ink
and orderd
t
inω
. The poles ofH(k, ω)
(whereH(k, ω)
is onsidered as afun tion ofk
)are the roots of a polynomialequation ommonly alled the dispersion equation.B. From the transfer fun tion to the blo k-diagram model
1. The impulse response as the sum of pairs of symmetri al traveling waves
Assuming both that the propagation model orresponds to a well-posed physi al problem ( f. e.g. Ref. 3 for
more details) and that the problem is symmetri al (i.e., un hanged by substituting
−x
forx
), the oe ientsα
2k+1,n
are zero, whi h impliesthatd
x
= 2 J
is an even number, and the transfer fun tionH(k, ω)
isrewritten as follows:H(k, ω) =
a(ω)
J
Y
j=1
[ k
2
− k
j
(ω)
2
]
=
J
X
j=1
a
j
(ω)
k
2
− k
j
(ω)
2
=
J
X
j=1
g
j
(ω)
1
k − k
j
(ω)
−
1
k + k
j
(ω)
,
(6) wherea
j
(ω) = a(ω)/
Q
J
m=1, m6=j
[ k
m
(ω)
2
− k
j
(ω)
2
]
andg
j
(ω) = a
j
(ω)/[2 k
j
(ω)]
, assumingk
j
(ω)6=0
andk
j
(ω)6=k
m
(ω)
ifj6=m
.For an innite system, the assumed boundary (no sour es at innity) and initial onditions (rest initial
ondition) are:
y|
x→+∞
= y|
x→−∞
= 0 ; y|
t<0
= 0 ; ∀n , 1 6 n 6 d
t
− 1 ,
∂
n
y
∂ t
n
t<0
= 0 .
(7)Comingba k tothe spa e/frequen y domain, Eqs.(6) and (7) lead tothe onvolution produ twith respe t
to the spatial variable:
by(x, ω) =
Z
+∞
−∞
b
f (x
0
, ω) b
h(x − x
0
, ω) d x
0
,
(8)bh(x, ω)
satisfyingbh(x, ω) =
J
X
j=1
g
j
(ω)
e
−i kj(ω) x
u(x) + e
+i kj(ω) x
u(−x)
,
(9)where
u
isthe Heavisideunit step fun tion:∀ x < 0 , u(x) = 0 and ∀ x > 0 , u(x) = 1 .
Thefun tion
bh(x, ω)
orrespondstotheFouriertransformoftheimpulseresponseh(x, t)
(alsonamedGreen's fun tion)tof (x, t) = δ(x) δ(t)
whereδ
isthe Dira impulse. Thisresponseh(x, t)
isthe ee t ofatimeimpulse (att = 0
)exerted onasingle point(atx = 0
)of the system. Theunit ofh
isthe unit ofy
perNewton and per se ond. An inverse Fouriertransformation gives:h(x, t) =
1
2 π
J
X
j=1
Z
+∞
−∞
g
j
(ω)
e
i [ ω t−kj(ω) x ]
u(x) + e
i [ ω t+kj(ω) x ]
u(−x)
d ω .
(10)Note that
h(x, t) = h(−x, t)
: the symmetry of the problem issatised. Ea hk
j
(ω)
is a omplex wave-number su hthat:k
j
(ω) = ω τ
j
(ω) − i α
j
(ω) =
ω
c
j
(ω)
− i α
j
(ω) ,
(11)where
τ
j
(ω) > 0
isthe propagationdelay perunit length (orslowness),c
j
(ω)
the phase velo ity,andα
j
(ω) > 0
the attenuation perunit length.The impulseresponse an berewritten as:
h(x, t) =
1
2 π
J
X
j=1
Z
+∞
−∞
g
j
(ω)
e
−αj(ω) x
e
i ω [ t−x/cj(ω)]
u(x) + e
+αj(ω) x
e
i ω [ t+x/cj(ω)]
u(−x)
d ω .
(12)This impulseresponse is onsequently the superpositionof
J
pairs of symmetri alde aying traveling waves starting fromthe ex itation point and propagatingin opposite dire tions.Thus, the general solution
y
is given by the two-dimensional onvolution:y(x, t) =
Z
+∞
−∞
Z
+∞
−∞
f (x
0
, t
0
) h(x − x
0
, t − t
0
) d x
0
d t
0
.
(13)2. Waveguide modeling of a region without sour es
Under the assumption that no sour es exist in a region between
x = x
min
andx = x
max
, one an demonstrate that the vibration state is the sum ofJ
de aying traveling wavesy
+
j
in the in reasingx
dire tion and ofJ
de ayingtraveling wavesy
−
j
inthe de reasingx
dire tion:∀ x , x
min
6
x 6 x
max
, y(x, t) =
J
X
j=1
y
+
j
(x, t) +
J
X
j=1
y
−
j
(x, t) ,
where c
y
j
+
(x, ω) = e
−i kj(ω) (x−xmin)
y
c
+
j
(x
min
, ω)
and c
y
j
−
(x, ω) = e
+i kj(ω) (x−x
max
)
c
y
−
j
(x
max
, ω) .
(14)
This region an bemodeled as
J
waveguidesin parallelwhi hbe omeJ
digital waveguidese.g. 6, 7
indis rete
time. Ea htransfer fun tion
e
−i kj(ω) ∆x
isgenerally designed asadelay lineinseries with adigital ausal lter.
After suitabledigital lters havebeen found, e ient time-domainsimulations an be made.
Inthe aseofpianostrings,the sour esarelo atedinthesegmentofhammerfelt/string onta tand atboth
ends. A blo k-diagrammodelis drawn inFig. 1,in ludinganobservationpoint. This gure isageneralization
of the model involved in Bensa et al. (Ref. 3, Fig. 2). Note that the boundary onditions hara terizing the
stringterminations an be hangedby onlymodifyingonelumped lteratea h end,apartfromthe waveguides
modeling the portions of string withoutsour es. PSfragrepla ements
x
0
x
1
x
0
− δx
x
0
+ δx
by(x
1
, ω)
x
0
L
c
y
+
(x, ω)
c
y
−
(x, ω)
Left termination Righ t terminationH
am
m
er
/s
tr
in
g
in
te
ra
ct
io
n
H
1
(ω)
H
1
(ω)
H
2
(ω)
H
2
(ω)
H
3
(ω)
H
3
(ω)
Figure1: Blo k-diagramofthewaveguidemodelofapianostring. Thestringsegmentintera tingwiththehammerfelt isassumedtobe
[ x
0
−δx ; x
0
+ δx ]
.x
1
isthepositionofobservation.x
= 0
isthe apod'astrobarpositionandx
= L
isthebridgeposition. Ea hH
n
(ω)
isequivalenttoJ
delayline/lterblo ksinparallel:H
1
(ω) =
J
X
j=1
e
−
i k
j
(ω) (x
0
−
δx)
,H
2
(ω) =
J
X
j=1
e
−
i k
j
(ω) (x
1
−
x
0
−
δx)
andH
3
(ω) =
J
X
j=1
e
−
i k
j
(ω) (L−x
1
)
.III. THE FOUR WAVES IN A STIFF PIANO STRING
In this se tion and the next one, we apply the prin iples given in the previous se tion to the one-dimensional
wave equation introdu ed by Bensa et al. [Ref. 3, Eq. (6)℄. The dispersion equation provides two pairs of
de ayingtraveling waves whi hneed a detailedanalysis.
A. The transfer fun tion of a sti piano string
The transverse displa ement
y
is assumed to satisfy [Ref.3, Eq. (6)℄:∂
2
y
∂ t
2
+ 2 b
1
∂ y
∂ t
− 2 b
2
∂
3
y
∂ x
2
∂ t
− c
2
∂
2
y
∂ x
2
+ κ
2
∂
4
y
∂ x
4
=
1
µ
f ,
(15)where
b
1
isthe rst oe ientof damping(duetovis osity of theair),b
2
the se onddamping oe ient (inner losses),c =
p
T /µ
the transverse wave velo ity of string4
,
T
the string tension,µ
the linear mass density of string,κ = c
p
E I/T
a stiness oe ient,E
the Young modulus,I = S r
2
the moment of inertia of the
ross-se tion,
S
the ross-se tion,andr
the radius of gyration.Inthe wave-number/frequen y domain,Eqs. (5)and (15) lead to the transfer fun tion:
H(k, ω) =
Y (k, ω)
F (k, ω)
=
1
µ [(−ω
2
+ 2 i ω b
1
) + (c
2
+ 2 i ω b
2
) k
2
+ κ
2
k
4
]
.
(16) B. Complex wave-numbersThe two (
J = 2
) wave-numbersk
s
andk
d
satisfy:k
2
s
(ω) =
c
2
2 κ
2
"
−1 − i
2 ω
c
2
b
2
+
s
1 +
4 (κ
2
− b
2
2
) ω
2
c
4
+ 4 i ω
1
c
2
b
2
− 2
κ
2
c
4
b
1
#
,
(17)k
2
d
(ω) =
c
2
2 κ
2
"
−1 − i
2 ω
c
2
b
2
−
s
1 +
4 (κ
2
− b
2
2
) ω
2
c
4
+ 4 i ω
1
c
2
b
2
− 2
κ
2
c
4
b
1
#
.
(18)TheseequationsaresimilartoEq.(11)inRef.3. Theapproa hes,however,aredierent. Inthemodalapproa h
3
β
+
andβ
−
are real-valued fun tions of the omplex frequen ys
. In the present wave approa hk
s
andk
d
are the rootsof the dispersion equation and are omplex-valuedfun tionsof the ommonangularfrequen yω
(real number).The assumptions[Ref. 3 Eqs. (24) and (25)℄
b
1
b
2
≪ c
2
; b
2
2
≪ κ
2
and b
2
1
≪ ω
2
(19) lead tok
s
(ω) ≃
c
κ
r
ξ
2
− i
κ ω
c
3
(1 + ξ)
r
2
ξ
b
1
+
c
2
ξ
2 κ
2
b
2
(20) andk
d
(ω) ≃
1
c
3
√
2 ξ (1 + ξ)
(2 ω
2
b
2
− c
2
ξ b
1
) − i
r
2
ξ
ω
c
,
(21) withξ = −1 +
r
1 +
4 κ
2
c
4
ω
2
.
(22)C. Waveguide modeling of portions of string without sour es
Therstwave-number
k
s
hara terizesaweaklyattenuateddispersivetravelingwavesimilartotheonedes ribed by Bensa et al. [Ref. 3, Eq. (34)℄. At low frequen ies [b
2
1
≪ ω
2
≪ c
4
/(4 κ
2
)
℄ the phase velo ity is lose toc
. Its numeri alvalues are between160
and420 m · s
−1
for piano tones C2, C4, and C7 (Ref. 3, Table I). The
attenuationperunit length
α
s
≃ b
1
/c
is small: itsnumeri alvalues [10 α
s
/ log(10)
℄are less than0.1 dB · m
−1
.
These ondwave-number
k
d
representsafast-de ayingwavewithaveryhighphasevelo ity. Atlow frequen- ies the phase velo ity is approximatelyc κ/(b
2
− κ
2
b
1
/c
2
)
. Its numeri al values are in the range2.6 × 10
5
to
1.62 × 10
6
m · s
−1
. The attenuationperunit lengthis
α
d
≃ c/κ
. The numeri alvalues [10 α
d
/ log(10)
℄ofα
d
are greaterthan1.1 × 10
3
dB · m
−1
. Forany length
∆x
greaterthan1.8 cm
,e
−αd
∆x
(the modulusofe
−i kd
∆x
)isless than10
−2
. This se ond wave is quasi-evanes ent i.e., its phase velo ity is almost innite and its attenuation
perunitlength ishigh. It ould represent the fa tthat apart ofthe energy of ahammerstrikewouldinstantly
propagate along the sti string around the onta t region su h as to avoidthe formation of a sharp orner, as
shown belowand in a ordan e with Cremer
8
.
Consequently, the waveguide model of a portion of string without sour es will ontain only a single delay
line/lter(wave-number
k
s
)forea hdire tionof propagation,provided that theportionofstringislongerthan a few entimeters. But the other wave hara terized byk
d
annot be negle ted at the neighborhood of the ex itation region,as shown in the following se tion.IV. The impulse response of the string
A. Spa e/frequen y domain
In the spa e/frequen y domain Eqs. (1618), (6), and (9) imply that the Fourier transform of the impulse
response is:
bh(x, ω) =
c
h
+
s
(0, ω) e
−i ks(ω) x
+ c
h
+
d
(0, ω) e
−i kd(ω) x
x > 0
c
h
+
s
(0, ω) + c
h
−
s
(0, ω) + c
h
+
d
(0, ω) + c
h
−
d
(0, ω) x = 0
c
h
−
s
(0, ω) e
+i ks(ω) x
+ c
h
−
d
(0, ω) e
+i kd(ω) x
x < 0
,
(23) wherec
h
+
s
(0, ω) = c
h
−
s
(0, ω) = g
s
(ω) =
−i
2 µ κ
2
[k
s
(ω)
2
− k
d
(ω)
2
] k
s
(ω)
c
h
+
d
(0, ω) = c
h
−
d
(0, ω) = g
d
(ω) =
i
2 µ κ
2
[k
s
(ω)
2
− k
d
(ω)
2
] k
d
(ω)
.
(24)The omplex omplian es
g
s
(ω)
andg
d
(ω)
(unit:m · N
−1
) hara terize the onversion of for e tomotion.
B. Example of a C2 piano string
The response
y
of a C2 piano string (Ref. 3, Table I) to a downward Gaussian time impulse at a single pointx = 0
isgiven by Fig.2. This response isvery losetothe negativeof theimpulse responseh
inthe bandwidth 020 kHz
. Thetransversedispla ementy
isthesumoftwoweaklyattenuatedtravelingwavesy
±
s
(wave-numbers±k
s
)andoftwofast-de ayingwavesy
±
d
(wave-numbers±k
d
)goingawaysymmetri allyfromtheex itationpoint. In agreement with Refs. 8 and 2 (Se tion 2.18), the bend is rounded appre iably by the stiness of the string.The fast-de ayingwaves inhibit the sharp orner whi h ould begenerated by the weakly attenuated traveling
waves, ifthe formerwere not onsidered. Asshown inthepreviousse tion,onlythe weaklyattenuatedtraveling
waves exist far from the ex itation point (more than
2 cm
), as well as in the neighborhood of the ex itation pointafter0.1 ms
.-
2
0
2
4
6
8
10
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
0.11
0.12
0.13
0.14
0.15
0
-
1
-
2
-
3
-
2
-
1
0
1
-
3
-
2
-
1
0
1
-
3
-
2
-
1
0
1
PSfragrepla ementst
[m
s]
x
[cm]
ϕ(t)
[×10
5
N ]
y
(x
,t
)
[m
m
]
Figure 2: Response of a C2 piano string to a downward Gaussian impulse
f
.f
(x, t) = f
0
δ(x) ϕ(t)
whereϕ(t) =
−λ e
−
π λ
2
(t−τ )
2
.f
0
= 1 N
,λ
= 2 × 10
5
s
−1
, and
τ
= 0.02 ms
. The parametersof the C2 piano stringare (Ref. 3, TableI):c
= 160.9 m · s
−1
,κ
= 0.58 m
2
· s
−1
,b
1
= 0.25 s
−1
,andb
2
= 7.5 × 10
−5
m
2
· s
−1
. Theresponsey
(plainline) isthesumofaweaklyattenuatedwavey
s
(dashedline)andofaquasi-evanes entwavey
d
(dottedline).C. Non- ausal omponents of the ausal impulse response
An important additional point is that
y
±
s
andy
±
d
are not ausal responses whereas their sum is ausal. A mathemati al explanation an be found for this by fo using on the impulse response at the ex itation point[Eq. (24)℄. The signals
h
±
s
(0, t)
andh
±
d
(0, t)
and the modulus of their Fourier transformsg
s
(ω) = c
h
±
s
(0, ω)
andg
d
(ω) = c
h
±
d
(0, ω)
are plotted inFig. 3 (inthe ase of a C2 piano string). The latter signalh
±
d
(0, t)
is nearly an even fun tion (with respe t tot
). The former signalh
±
s
(0, t)
is nearly the sum of a onstant [1/(4 µ c)
℄and of an odd fun tion.(a)
-
0.02
-
0.01
0
0.01
0.02
0.03
-
0.25
0.
0.25
0.5
0.75
1.
-
0.02
-
0.01
0
0.01
0.02
0.03
-
0.25
0.
0.25
0.5
0.75
1.
PSfragrepla ementst
[ms]
t
[ms]
c
µ
h
(0
,t
)
;
c
µ
h
±
s/
d
(0
,t
)
c
µ
h
(0
,t
)
;
c
µ
h
±
s/
d
(0
,t
)
(b)0
5
10
15
20
0
0.2
0.4
0.6
0.8
1
cg
PSfragrepla ementsf = ω/(2 π)
[kHz]
Figure 3: (a) Impulse responses of a C2 piano string and (b) the modulus of their Fourier transforms times the
angularfrequen y. The omplian e
g
s
(ω) = c
h
±
s
(0, ω)
istheFouriertransformofthesignalh
±
s
(0, t)
(weaklyattenuated waves, dashed line).g
d
(ω) = c
h
±
d
(0, ω)
is the Fourier transform ofh
±
d
(0, t)
(quasi-evanes ent waves, dotted line). The impulse responseh(0, t)
(plain line) is two times the sum ofh
±
s
(0, t)
andh
±
d
(0, t)
. Its Fourier transform isbh(0, ω) = 2 [g
s
(ω) + g
d
(ω)]
.Thisis naturalbe ause, onassumingthat
b
1
andb
2
are zero, andω
isgreaterthan zero[see Eqs. (17), (18), (22), and (24)℄,the omplian es be ome:g
s
(ω) ≃
i κ
µ c
3
√
2 ξ (1 + ξ)
and g
d
(ω) ≃
−
√
ξ
2 µ
√
2 c (1 + ξ) ω
.
(25)Consequently, the omplian e
g
s
is an imaginary-valued fun tion andg
d
is a real-valued fun tion. Be ause the signalsh
±
s
(0, t)
andh
±
d
(0, t)
are real-valued, the former should be an odd fun tion and the latter an even fun tion. ButtheFouriertransformofh
±
s
(0, t)
has asingularityatω = 0
andthe limitofh
±
s
(0, t)
,whent
tends to−∞
, has to be zero. This implies that a onstant has to be added toh
±
s
(0, t)
, giving an additional termπ δ(ω)/(2 µ c)
inits Fouriertransform.Applying the nal value theorem to the signal
h
±
s
(0, t)
, the limit ofh
±
s
(0, t)
witht
tending to innity is1/(2 µ c)
(see Fig. 3). The nal value of the impulse responseh(0, t)
issimilarly1/(µ c)
. This lastproperty is false ifb
1
> 0
: all the signals tend slowly to zero whent
tends to innity. Indeed,h
±
d
(0, t)
is lose to zero for great values oft
andh(0, t) ≃ h
+
s
(0, t) + h
−
s
(0, t)
. In summary, the two wavesh
±
s
andh
±
non-Y
PSfragrepla ementsx = x
0
x
by(x
0
, ω)
e
−i k
s
(ω) L
e
−i k
s
(ω) L
e
−i k
s
(ω) L
′
e
−i k
s
(ω) L
′
bh(0, ω)
g
s
(ω)
b
ϕ(ω)
Figure4: Blo k-diagramofaportionofstringex itedbyapointfor e. Thisfor e
ϕ(t)
isexertedbythehammerfelt atthepositionx
= x
0
. Thefor e distributionisassumedto bef
(x, t) = ϕ(t) δ(x − x
0
)
forx
in therangex
0
− L
tox
0
+ L
′
(L
andL
′
arebothgreaterthanafew entimeters).
This implies that if the for e distribution
f (x, t) = δ(x − x
0
) ϕ(t)
is assumed to represent a point-for e exerted by the hammer felt at the positionx = x
0
, then the responsey(x
0
, t)
at the ex itation point satisesby(x
0
, ω) = 2 [ g
s
(ω) + g
d
(ω) ]
ϕ(ω)
b
. This ase isrepresented inFig. 4asablo k-diagram. Similarblo k-diagrams are used forbowed strings( f. e.g. the review arti le by Smith9
, Fig. 14).
V. CONCLUSION AND PROSPECTS
The on lusion of this paperis that a omplete mathemati al treatment of the one-dimensional wave equation
(15) arefully takes into a ount all the waves provided by the dispersion equation. Fast-de aying waves an
benegle ted inthe waveguidemodelingof portions ofstring withoutsour esif they are long enough. However,
they annota urately be negle ted atthe neighborhoodof the ex itationregion (hammer) and of the xation
points(bridge, apo d'astro bar).
In parti ular if a point-for e is exerted by the hammer felt on the string, the waveguide model proposed
by Bensa et al. (Ref. 3, Fig. 2) needs to be omplemented at the ex itation point, as shown in Fig. 4 whi h
summarizes the resultsof the lastse tion. Notethat arealisti ex itationby the hammeris distributed over a
ertainwidth
4
. This extension ould bemade by aspatial onvolution. A future study may determinewhether
or not this orre tion would noti eably improvethe quality of synthesized sti-stringsounds.
This paper is an attempt toextra t as mu h information as possible from the physi almodeldened by a
one-dimensional wave equation, in a waveguide modeling ontextand froma mathemati alpoint of view. It is
hoped that this study will help to improvefuture waveguide models of strings and toenhan e the a ura y of
e ient omputationalmodelsusing Digital Waveguides
6, 7, 3
,afterasubstantial eortto designsuitabledigital
REFERENCES
1
L. Hiller and P. Ruiz: Synthesizing musi al sounds by solving the wave equation for vibrating obje ts.
J. Audio Eng. So . 19 462472(Part I) and 542551(Part II) (1971).
2
N. Flet her and T. Rossing, The Physi s of Musi al Instruments (2nd ed., ISBN 0387983740,
Springer-Verlag, New York,1999).
3
J. Bensa, S. Bilbao, R. Kronland-Martinet, and J. O. Smith III: The simulation of piano string vibration:
from physi al model to nite dieren e s hemes and digital waveguides. J. A oust. So . Am. 114(2)
10951107 (2003).
4
A. Chaigne and A. Askenfelt: Numeri al simulations of stru k strings. I. A physi al model for string using
nite dieren e methods. J.A oust. So .Am. 95(2)11121118 (1994).
5
J.-M.Adrien: TheMissing Link: ModalSynthesis. inRepresentations of Musi alSignals. MITPress, 1991,
269297.
6
J. O.Smith III: Physi almodeling using digitalwaveguides. Computer Musi J. 16(4) 7491(1992).
7
J. O.Smith III: Physi almodeling synthesis update. Computer Musi J. 20(2) 4456(1996).
8
L. Cremer, The Physi s of Violin (translation by J.S. Allen,MIT Press, Cambridge,Massa husetts, 1984).
9
J. O. Smith III: Virtual a ousti musi al instruments: Review and update, J. New Musi Resear h 33(3)