Mechanical Systems and Signal Processing 56-57 (2015) 8191

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

journal homepage: www.elsevier.com/locate/ymssp

Operational modal analysis applied to the concert harp

B. Chomette a,b,n, J-L. Le Carrou a,b
a
b

Sorbonne Universits, UPMC Univ Paris 06, UMR 7190, Institut Jean le Rond d'Alembert, F-75005 Paris, France
CNRS, UMR 7190, Institut Jean le Rond d'Alembert, F-75005 Paris, France

a r t i c l e i n f o

abstract

Article history:
Received 23 April 2014
Received in revised form
13 October 2014
Accepted 23 October 2014
Available online 20 November 2014

Operational modal analysis (OMA) methods are useful to extract modal parameters of
operating systems. These methods seem to be particularly interesting to investigate the
modal basis of string instruments during operation to avoid certain disadvantages due to
conventional methods. However, the excitation in the case of string instruments is not
optimal for OMA due to the presence of damped harmonic components and low noise in
the disturbance signal. Therefore, the present study investigates the least-square complex
exponential (LSCE) and the modified least-square complex exponential methods in the
case of a string instrument to identify modal parameters of the instrument when it is
played. The efficiency of the approach is experimentally demonstrated on a concert harp
excited by some of its strings and the two methods are compared to a conventional modal
analysis. The results show that OMA allows us to identify modes particularly present in
the instrument's response with a good estimation especially if they are close to the
excitation frequency with the modified LSCE method.
& 2014 Elsevier Ltd. All rights reserved.

Keywords:
Operational modal analysis
String instrument
LSCE method

1. Introduction
Acoustic musical instruments are designed to amplify vibrations produced by an oscillator. At low frequencies, this
amplification is mostly governed, for string instruments, by the structural modes of the soundboard. These modes are thus
characteristic for each instrument and are often discussed to understand the operation of the instruments [1,2], for
comparing instruments [37] or to synthesise the sound [8,9]. Classical methods are generally used to obtain structural
modes: the instrument is either impacted by a hammer or excited by a shaker. These excitation methods can induce some
experimental problems, such as gluing a force sensor on the soundboard or causing damage to the instrument with the
hammer. An other solution is to use the instrument excitation system that is to say its strings. Therefore, the end goal is to
develop a method to be used when the instrument is being played, and this paper focuses on identifying modal parameters
by using recent methods such as operational modal analysis (OMA).
OMA has been extensively developed in the past decades particularly when the input of structures is unknown or
difficult to measure in their ambient environments. For example, Brincker et al. [10] use a decomposition technique in a set
of single degree of freedom systems to accurately identify close modes without knowing the input exciting the system. De
Vivo [11] applies OMA in order to identify the modal shapes of a large structure exposed to wind excitation. Xu [12] uses
OMA using non-contact excitations and sensors. Agneni et al. [13] devote a special attention to the case of structures with

n
Corresponding author at: Sorbonne Universits, UPMC Univ Paris 06, Institut Jean le Rond d'Alembert, Case courrier 162, 4 place Jussieu, F-75005 Paris,
France. Tel.: 33 1 44 27 51 51.
E-mail address: baptiste.chomette@upmc.fr (B. Chomette).

http://dx.doi.org/10.1016/j.ymssp.2014.10.011
0888-3270/& 2014 Elsevier Ltd. All rights reserved.

82

B. Chomette, J.-L. Le Carrou / Mechanical Systems and Signal Processing 56-57 (2015) 8191

closely spaced coupled modes and propose also the use of the power spectra densities [14] to estimate modal parameters in
the frequency domain and calculate the correlation function in the time domain. Peeters and de Roeck [15] propose an
approach based on stochastic subspace identification in order to avoid correlation function calculation. Numerous time
domain OMA can be based on the natural excitation technique (NExT) [1618] which is conventionally used to identify
modal parameters using ambient vibrations. Brincker et al. [19] propose a method to compute the cross correlation and the
auto correlation function to identify modal parameters. For example, Shen and Zheng [20] experimentally use the crosscorrelation technique to extract modal parameters on the response-only data in the case of an airplane model. In the NExT
technique, the main assumption of the input excitation is that the excitation must be assimilated to a stationary white-noise.
That is why OMA cannot be applied in theory in the case of structures characterised by the presence of harmonic excitation.
In this particular case, other methods can be applied. Brincker et al. [21] propose an indicator to separate structural and
harmonic modes. Harmonic components can be removed from the response or taken into account in the procedure.
Manzato [22] and Agneni [23] propose to remove harmonics to improve the OMA. Mohanty and Rixen propose some new
algorithms to avoid perturbations due to filters and to have a more efficient and robust procedure. These new methods are
based on the least square complex exponential (LSCE) method [24,25], on the eigensystem realization algorithm (ERA)
method [26] or on the single station time domain (SSTD) method [27] in order to apply OMA in the case of a disturbance
including harmonic components particularly for rotating machines. These methods consider that the harmonic frequencies
are close to the eigenfrequencies of the structure and include no-damped harmonic components in the identification
procedure. Therefore, these methods seem particularly appropriated to identify eigenfrequencies of string instruments in
low-frequencies. Indeed, instrument makers design musical instruments so that the soundboard's eigenfrequencies are
close to the excitation frequencies of the string and the anti-node's locations match the strings attachment position to
radiate the sound. This is particularly the case for plucked string instruments such as the guitar [28] or the harp [29,30].
This paper deals with the application of OMA methods in order to identify the modal parameters of a concert harp. This
musical instrument is particularly suitable for a first practical test because the instrumentalist does not affect the structural
vibration of the harp when it is played unlike some other instruments. With a classical guitar for instance, the comparison of
the modal basis obtained with OMA and with a classical modal analysis method would not be possible since the
instrumentalist body and arms can affect the instrument modal parameters, especially the modal damping. Therefore, in
order to study the efficiency of the method, the modal basis identification of the concert harp's soundboard should be more
suitable. For this instrument, the high damping values and the close modal frequencies lead to use a time domain OMA
method. Moreover, it is difficult to measure the input of the instrument in playing conditions and harmonic excitations due
to strings are present in the disturbance. Consequently the classical modal operational procedures can fail in the
identification of modal parameters. The novelty of this study is to apply OMA on a structure with damped harmonic
excitations containing low noise level. After a brief presentation of the method, the efficiency of the approach is
experimentally demonstrated on a concert harp excited with different strings of different frequencies.
2. Identication algorithm
In this section, the different steps of the implementation of the LSCE and the modified LSCE are detailed. The objective is
to identify modal parameters of the harp when the instrumentalist is playing.
2.1. LSCE algorithm
First, data are measured when the strings are played one after another. Consequently, a single reference can be used for
each identification corresponding to each string excitation. A more complete description can be found in [16,24,25]. These
two algorithms use the Prony method [31]. In the case of a white noise excitation, the correlation function Rij between the
signal measured at i and a reference at j is similar to the response at i of a system submitted to an impulse excitation applied
at j, as shown in the NExT method [16]. In the case of a string excitation, the structure response is the sum of the solution of
the homogeneous equation and the particular solution corresponding to the damped forced response due to the harmonic
components of the string excitation. In the transient state, the response of the structure includes almost all the structural
modes (of which the nodes are not close to the played string attachment point) and the string modes. That is why the
structure response in transient state can be assimilated to the response at i of a system submitted to an impulse excitation
applied at j. In this section, Rij denotes the correlation function or the structure response. This assumption can be written as
follows using a sampling step t
2N

Rij kt C 0rij esr kt ;

r1

where sr denotes the complex poles of the system and C 0rij is the constant associated to the rth mode. This assumption is
particularly verified for a structure excited by a string in free oscillations. Indeed, each decaying sinusoid is related to a
structural or a string mode. Due to the conjugate complex form of the poles
V r esr kt

B. Chomette, J.-L. Le Carrou / Mechanical Systems and Signal Processing 56-57 (2015) 8191

83

are the roots of a 2N polynomial equation defined with

0 1 V 1r 2N  1 V r2N  1 V 2N
r 0;

where k are the coefficient of the polynomial. These coefficients can be determined using Eqs. (1) and (3) multiplying by k
and summing up these values from k 02N
!



2N
2N
2N
2N
2N
0
0
k
sr kt
k Rij kt k C rij e
C rij k V r 0:
4
k0

k0

r1

r1

k0

Eq. (4) gives

0 Rij 0 1 Rij t 2N  1 Rij 2N  1t  Rij 2N t:

The 2N coefficients k can be finally obtained using Eq. (5) written for 2N successive time samples to build up the linear
system
8
1
>
0 R0ij 1 R1ij 2N  1 R2N
 R2N
>
ij
ij
>
>
>
1
< R1 R2 R2N
 R2N
0 ij
1 ij
2N ij
ij
6
;
>

>
>
>
>
1
4N  2
1
: 0 R2N
1 R2N
 R4N
ij
ij 4N  2 Rij
ij
where k Rkij k Rij kt. In the case of L 42N sample times to average measures, p sensors and p correlation functions
calculated using the same reference j, system (6) becomes over-determined
8
2N  1
0 R01j 1 R11j 2N  1 R1j
R2N
>
1j
>
>
>
>
1
>
>

R11j 1 R21j 2N R2N

R2N
0
1j
>
1j
>
>
>
>

>
>
>
L 2N  2
>
> 0 RL1j 1 1 RL1j L 2N  2 R1j
RL1j 2N  1
>
<

7
;
>
0
1
2N  1
2N
>
>

R
>
0
1
2N

1
pj
pj
pj
pj
>
>
>
>
1
>
0 R1pj 1 R2pj 2N R2N
R2N
>
pj
pj
>
>
>
>

>
>
>
>
L 2N  2
: RL  1 RL
R
RLpj 2N  1
0 pj
1 pj
L 2N  2 pj
and can be solved using the least square method. This method permits the improvement of the robustness of the LSCE
algorithm. The poles of the identified system are finally obtained by founding the roots of Eq. (3) using calculated k and
Eq. (2)
sr

1   
ln V r 7 i argV r :

2.2. Modified LSCE algorithm

When the excitation frequency is close to a structural mode, classical OMA methods can fail the identification of modal
parameters as explained in [24]. One way to make this identification more robust is to include the known harmonic
excitations directly in the algorithm. This step is the basis for the modified LSCE method [24]. Thus, undamped poles are
directly included in the identification process allowing an accurate identification of both structural and string modes. In fact,
this method allows the separation of physical poles with physical damping values corresponding to structural modes from
mathematical poles corresponding to the excitation with no damping. In the case of the modified LSCE method, the poles
corresponding to virtual modes having zero damping can be written
s
r 7 i;

under an harmonic excitation whose pulsation and Prony's polynomial equation has two conjugate extra roots
s
r t :
V
r e

10

In the case of m harmonic frequencies, Eq. (7) can be written in symbolic form including the harmonic components
1 to m
(
A b1 C b2 E
;
11
B b1 D b2 F

84

B. Chomette, J.-L. Le Carrou / Mechanical Systems and Signal Processing 56-57 (2015) 8191

where A is a L  2m matrix, b1 a 2m  1 vector, C a L  2N  2m matrix, E a L  1 vector, B a 2m  2m matrix, D a

2m  2N  2m matrix and F a 2m  1 vector. A and B are defined with
2 0
2
3
3
1
1
R1j

R2m

R2N
R2m
1j
1j
1j
6
6
7
7
6
6
7
7

6
6
7
7
6 RL  1 RL 2m  2 7
6 RL 2m  1 RL 2N  2 7
6 1j
6 1j
7
7
1j
1j
6
6
7
7
7 C6
7:

A6
12
6
6
7
7
2m  1 7
2N  1 7
6 R0
6 R2m

R
6 pj
6
7
7
pj
pj
pj
6
6
7
7
6
6
7
7

4
4
5
5
L 2N  2
RLpj 1 RLpj 2m  2
RLpj 2m  1 Rpj
C and D are defined with
2

0
61
6
6
B6
6
60
4

sin 1 2m  1t

cos 1 2m  1t 7
7
7
7

7
sin m 2m  1t 7
5
cos m 2m  1t

sin 1 2mt
6 cos 2mt
1
6
6

D6
6
6 sin 2mt
4
m
cos m 2mt

sin 1 2N  1t

cos 1 2N 1t 7
7
7
7:

7
sin m 2N  1t 7
5
cos m 2N  1t

13

The vectors b1 , b2 , E and F are defined as follows:

2
2
6
b1 4

2m  1

7
5;

6
b2 4

2m

2N  1

7
5;

6
E4

R2N

RL 2N  1

3
7
5;

sin 1 2Nt

6 cos 2Nt 7
1
6
7
6
7
7:

F6
6
7
6 sin 2N t 7
4
5
m
cos m 2Nt

14

b1 can be obtained with the second equation of (11)

b1 B  1 F Db2

15

and b2 with the first equation of (11) using the least square method
C  AB  1 b2 E  AB  1 F:

16

b1 and b2 finally give k and the poles of the system are computed by finding the roots of the Prony equation as in the
classical LSCE method.
We chose to use the modified LSCE method because it allows the inclusion of harmonic components in the algorithm and
therefore the using of the specifications of a string excitation. In the case of string instruments, the string modes and the
corresponding harmonics depend on the string tuning and are theoretically known. In the case of an out-of tune string, the
modified LSCE method gives the same results than the classical method. The algorithm finds an additional zero damping
pole with no interaction with the other poles. Identified poles should finally be sorted to extract only the physical modes
from all the identified poles. One way is to use stability diagrams that represent the evolution of the identified frequency
and damping according to the order estimation. Physical poles can thus be selected using frequency and damping
convergence criteria as explained in Section 3.3. The classical and modified LSCE algorithm is applied in this study on a
concert harp using the experimental set-up presented in Section 3.

3. Experimental results
3.1. Experimental set-up
The two methods previously detailed are applied to a particular vibro-acoustical system: a concert harp (Fig. 1(a)). In the
experiment, 10 accelerometers are positioned on the soundboard of the harp as shown in Fig. 1(c): six along the central line
where the strings are mounted and four on the widest part of the soundboard. Moreover, one accelerometer is located at the
bottom of the played string (the excitation system) for reference. The fundamental frequency, the note and the number of 13
strings played are detailed in Table 1. Moreover, when a string is played, the others are damped with paper (Fig. 1(b)) to
avoid any string's sympathetic vibrations [32]. Note that the dynamic behaviour of the strings and of the soundboard of the
harp is weakly non-linear and the excitation level does not change the results found in the paper. Nonetheless, a higher level
of excitation should improve the noise to signal ratio and decrease the noisy part of the signal, but is not effective close to
the resonances. All the presented results were here achieved using the mezzo forte level.

B. Chomette, J.-L. Le Carrou / Mechanical Systems and Signal Processing 56-57 (2015) 8191

85

Fig. 1. Schema (a) and photo (b) of the experimental set-up; (c) description of the accelerometers and of the played strings location.

Table 1
Number, note and fundamental frequency of each played string.
String number
Note
Fundamental frequency (Hz)

40
Ab1
52.03

36
Eb2
78.03

34
Gb2
92.34

32
Bb2
116.6

30
Db3
138.5

29
Eb3
153

28
Fb3
162.3

27
Gb3
185.5

26
Ab3
208.2

25
Bb3
232.9

24
Cb4
247.4

23
Db4
273.6

22
Eb4
310.9

3.2. Classical modal analysis

In order to compare OMA results, a classical modal analysis is performed on the harp's soundboard by impact testing. The
structure is impacted using a hammer as close as possible to accelerometer A2 (see Fig. 1(c)). All strings are damped. The
modal identification is computed using the least-squares complex frequency (LSCF)-domain algorithm implemented in
Modan software developed at the FEMTO-ST institute (http://modan.dma-softs.com/).
Only seven consecutive modes in the frequency range 30250 Hz are identified because of the high modal density above
200 Hz as shown by the typical measurement at point 2 in Fig. 3. In this figure, a synthesized response function at the same
point is also plotted to validate the modal identification. The frequency and the damping factor of the identified modes are
detailed in Table 2 and their modal shapes are shown in Fig. 2.
These results are similar to those previously obtained on another concert harp [33]. Modes 1 and 2 correspond to a global
motion of the instrument. Mode 5 is like a simply supported beam mode. Modes 6 and 7 have similar modal shapes and are
thus defined as two vibro-acoustic modes (T1 and A0). These two modes are associated to coupled motions of the bending
vibration of the soundboard and to the oscillations of the air pistons located in the sound holes. They correspond to the first
two modes of a Helmholtz resonator with yielding walls. Modes 8 and 9 are other bending modes of a quasi-clamped plate
with probably a lack of spatial informations to discern them from modes 6 and 7. Note that some peaks present in the
frequency response, at 62 Hz or 82 Hz for instance, can be explained by the fact that some strings might not be perfectly
damped.
3.3. Operational modal analysis
LSCE and modified LSCE algorithms are applied on the concert harp where each string of Table 1 is used as an excitation
source. Measured signals are acquired during 40 s with a sampling frequency equal to 10,240 Hz. These signals are then
decimated down to 5120 Hz to limit the identification order and consequently the calculation time as well as to limit the
frequency range of interest. Note that a eighth-order lowpass Chebyshev Type I filter with a cutoff frequency of 2048 Hz

86

B. Chomette, J.-L. Le Carrou / Mechanical Systems and Signal Processing 56-57 (2015) 8191

Table 2
Frequency and damping identified using LSCF estimator.
Mode
Frequency (Hz)
Damping (%)

1
53.3
3.5

2
56.4
3.2

5
106.1
1.6

6
147.2
1.1

7
185.1
4.5

8
190.6
1.6

9
213.6
3.1

Fig. 2. Modal shape obtained using the LSCF method.

25

1 2

30

Magnitude [dB]

35

40

10

11
12
13

45
50
55
60

50

100

150

200

250

300

200

250

300

Frequency [Hz]
Phase [rad]

0
1
2
3
0

50

100

150
Frequency [Hz]

Fig. 3. Collocated measured (solid line) and synthesized FRF (dotted line) at point 2 (ref. 1 dB: 10  5 m.s  2.N  1). The gray area corresponds to the frequency
range in which modes have been identified. Framed numbers are associated to identified modes by the classical modal analysis.

0.2

20

0.15

30

0.1

40

0.05
50
0
60

0.05

70

0.1
0.15

80

50

100

150

200

250

300

Fig. 4. Time response (a) and power spectrum density (b) calculated using sensor A1 during 0.4 s in the case of an excitation using string 28.

is used, and all phase distortion is removed by filtering the input sequence in both the forward and reverse directions [34].
This choice of acquisition time allows the use of more samples in the time signal and to have better averaged results.
For example, the time response and the power spectrum density of the A1 sensor in the case of an excitation using string 28 are,

B. Chomette, J.-L. Le Carrou / Mechanical Systems and Signal Processing 56-57 (2015) 8191

87

Pseudo spectra [dB]

respectively, presented in Fig. 4(a) and (b). These two figures show that a large number of structural modes are excited by
the string and can be observed on the power spectrum density. The time response in transient state includes almost all the
modes adding the fundamental frequency of the string (162.3 Hz for string 28) and other harmonic components. In Fig. 5,
the excitation spectrum measured at the bottom of each considered string in the frequency range 0400 Hz is shown.

0
100
200
40
36
34
32
30
29
28
27
26
25
24
23
22

500

ing
Str

400
300

200

y
enc

[Hz

u
req

100

Fig. 5. Power spectra density of the reference output in the case of different harmonic string excitations (strings 2240).

20

0.07

20

0.07

30

0.06

30

0.06

0.05

40

0.05

40

0.04

0.04
50

50

0.03

0.03
60

0.02

60

0.02

0.01

70

80

20

0.07

20

0.07

30

0.06

30

0.06

70
80

50

100

150

200

250

300

0.05

40

0.01
50

100

150

200

250

300

0.05

40

0.04

0.04
50

50

0.03

0.03
60

0.02

70
80

50

100

150

200

250

300

60

0.01

70

80

0.02
0.01
50

100

150

200

250

300

Fig. 6. Stability diagrams (order 215225) and power spectra density (in gray) for three excitation strings: string 25 (a and b) and string 40 (c and d).
For each excitation string the two algorithms are used: (a) and (c) for the LSCE and (b) and (d) for the modified LSCE. Poles are selected using frequency (17)
and damping (18) convergence criteria.

88

B. Chomette, J.-L. Le Carrou / Mechanical Systems and Signal Processing 56-57 (2015) 8191

300

0.07
0.06

250

300

0.07
0.06

250

0.05

0.05
200

200
0.04
150

0.03

100

0.02
0.01

50
22 23 24 25 26 27 28 29 30 32 34 36 40

0.04
150

0.03

100

0.02
0.01

50
22 23 24 25 26 27 28 29 30 32 34 36 40

Fig. 7. Identified poles of the system selected using frequency (17) and damping (18(a)) criteria using LSCF (dotted line), LSCE (a) and modified LSCE (b)
algorithm (: harmonics of each string) based on the correlation function calculation (o) and on the sensor output signals ( ).

Since only the played string oscillates, its harmonic frequencies are mostly present in the instrument response. Far from
these excitation frequencies, the signal-to-noise ratio is found to be extremely unfavourable, at about 5 dB. Note that the
presence of a peak close to 100 Hz in all spectra suggests that a mode is particularly present in the soundboard response.
The classical and modified LSCE algorithms are then applied on correlation signals measured on the 11 sensors. As
explained in Section 2, poles are selected from a stability diagram operated with a frequency and damping convergence
criterion. The frequency criterion is defined for the rth mode on the algorithm iterations with
n

f r max  f r min
o1%;
n
f r max

17

where f r nr =2 are the undamped frequencies. The damping criterion is also defined with
n

r max  r min
o 10%:
r max

18

Stability diagrams corresponding to the different excitation strings are represented in Fig. 6. In these figures, the estimated
poles verifying the frequency and damping convergence criteria in the frequency range 30300 Hz are also plotted for the
two OMA methods. It is clear that the results depend on the methods: some poles are identified using one method but not
using the other one. This is particularly true when there is a coincident frequency between the string modes and the
soundboard modes.
A synthesis of the results is shown in Fig. 7 where the string's number is indicated in abscissa and the frequency in
ordinate. Each pole is represented with a mark or with a circle whose color corresponds to its damping value. The circle
points out the poles identified using the correlation function calculation and the mark point out the poles identified using
the sensor output signal, as explained in Section 2. Dotted straight lines indicate the identified poles using the LSCF method.
These results show that the LSCE method can be directly applied on the sensor output signal without any correlation
function calculation. However, we can see differences in the identified modal parameters that are probably due to the nature
of the excitation which is not a white noise. This application shows that the two approaches can be used to compare and
validate the results. Quantitative results of the frequency and the damping of each identified mode using the output are
given in Tables 3 and 4. Some poles can be added by strings which are not perfectly muffled. Indeed, if a string is not
perfectly muffled, it can be excited by the structure vibrations. Consequently, an additional pole with low damping (close to
zero) can appear on the response of the structure.
4. Discussion
The results presented in the previous section show that operational modal analysis, such as LSCE and modified LSCE
methods, allow the identification of modes of a structure harmonically excited. In this section, the two OMA methods are
first compared. Then, the OMA methods and the classical modal analysis are confronted.
4.1. Comparison LSCE and modified LSCE algorithms
A simple comparison between results shown in Tables 3 and 4 demonstrates that there are less identified modes with
LSCE than with the modified LSCE method. Most of the missing modes in the case of the LSCE method are those whose
frequency is close to string modes, as shown in the stability diagrams presented in Fig. 6. Mode 10 at 233.6 Hz (with
damping equal to 0.5%) is identified with excitation string 25 whose fundamental frequency is at 232.9 Hz. Mode 5 at

Table 3
Identified frequency and modal damping with the LSCE method based on the sensor output signals. Poles are selected using frequency (17) and damping (18) convergence criteria.

String
22
(310.9 Hz)
23
(273.6 Hz)
24
(247.4 Hz)
25
(232.9 Hz)
26
(208.2 Hz)
27
(185.5 Hz)
28
(162.3 Hz)
29
(153.0 Hz)
30
(138.5 Hz)
32
(116.6 Hz)
34
(92.34 Hz)
36
(78.03 Hz)
40
(52.03 Hz)

f (Hz)
(%)
Mode
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)

53.3
3.5
1

56.4
3.2
2
56.8
4.2

57.3
3.3

57.3
4.4

52.0
4.8

52.4
3.7

89.1
3.5

106.1
1.6
5
106.6
1.5
106.7
1.5
106.7
1.3
106.6
1.4
106.7
1.5
106.7
1.4
106.7
1.5
106.5
1.6
106.6
1.6
106.5
1.4
106.6
1.5

147.2
1.1
6
147.7
1.7
147.3
1.9
147.8
2.2
147.4
1.8
147.6
1.9
147.7
1.8
147.5
2.1
147.5
2.1
148.6
2.2
148.5
1.9

185.1
4.5
7

190.6
1.6
8
189.3
1.6
190.1
1.9

213.6
3.1
9

148.0
1.9
147.7
1.9

188.7
2.1
189.3
2.0
188.2
3.9
187.2
3.8

10

11

12

13

10

11
250.2
2.1

12
258.8
1.4
260.9
1.9

13
273.4
2.3

223.8
3.7

249.8
2.1
250.2
2.6
250.3
2.4

215.9
4.2

213.4
2.3

275.6
2.2

275.3
1.8
277.1
2.1
271.4
1.9

188.4
2.4

187.7
3.6
229.4
2.7

248.9
2.1

B. Chomette, J.-L. Le Carrou / Mechanical Systems and Signal Processing 56-57 (2015) 8191

LSCF

Mode

89

90

B. Chomette, J.-L. Le Carrou / Mechanical Systems and Signal Processing 56-57 (2015) 8191

Table 4
Identified frequency and modal damping with modified LSCE method based on the sensor output signals and selected using frequency (17) and damping
(18) convergence criteria.
String

Mode 1

22
(310.9 Hz)
23
(273.6 Hz)
24
(247.4 Hz)
25
(232.9 Hz)
26
(208.2 Hz)
27
(185.5 Hz)
28
(162.3 Hz)
29
(153.0 Hz)
30
(138.5 Hz)
32
(116.6 Hz)
34
(92.34 Hz)
36
(78.03 Hz)
40
(52.03 Hz)

f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz)
(%)
f (Hz) 52.5
(%)
3.7
f (Hz)
(%)

56.8
4.2

100.7
3.3

106.6
1.5
106.7
1.5
106.7
1.3
106.6
1.4
106.7
1.5
106.7
1.4
106.7
1.5
106.6
1.6
106.6
1.6
106.5
1.5
106.6
1.5

147.7
1.7
147.4
1.9
147.8
2.2
147.4
1.8
147.5
1.8
147.8
1.8
147.4
2.1

57.3
3.3

89.0
5.0

8
189.3
1.6
190.1
1.9
189.4
2.8

189.0
2.2

212.6
5.0
218.2
4.0

10

233.6
0.5

11

12

13

250.2
2.1

258.8
1.4

273.4
2.3

249.6
2.3
250.2
2.6
249.6
2.3

275.5
2.3

188.2
3.8

185.3
4.4
148.4
1.9

188.4
2.4

277.2
2.0

186.8
3.9
106.7
1.0

147.5
2.0

106.7 Hz (with damping equal to 1.0%) is identified with excitation string 40 whose first harmonic is at 104.7 Hz. For mode 7
at 185.1 Hz (with damping equal to 4.5%), the identification seems to be more difficult probably due to its high damping
value. Nonetheless, this mode may be identified with the modified LSCE method applied to strings 30 and 36. Apart from
theses modes, results obtained from the two methods are comparable with possibly a slight variation in frequency and
damping satisfying the convergence criteria.
The review of the results for both methods shows that the identified modal basis is more complete when the modal
identification is performed with the modified LSCE method than with the LSCE method. The modified LSCE method is
eventually found to be more adapted than the LSCE one to identify structural modes of a plucked string instrument. Note
that the number of identified modes with OMA may vary with the excitation string independently of the used algorithm.
This is due to the capacity of the excitation string (with its own characteristics and its location) to excite structural modes.
Note that we also tested the SSTD method to compare the results. With this method which requires only one sensor for
the computation, identified modes are found to be very close to the results obtain with the others method but fewer modes
are identified (35) in comparison with the LSCE and the modified LSCE methods.
4.2. Modal basis comparison
The identified poles with the LSCF method are shown in Table 2 and are compared to those identified with the modified
LSCE method. The relative error between modal frequencies obtained with LSCE, the modified LSCE and LSCF is generally
below 3%. Whereas the difference between damping identified using the three algorithms seems to be significant as, for
instance, for modes 6, 8 and 9. However, with the LSCF damping identification, it is not so significant. Indeed, as shown in
Fig. 3, the measured and synthesised FRF are not perfectly superimposed especially around peaks 6, 8 and 9, indicating that
the LSCF damping identification is not perfect. Nevertheless, results from the modified LSCE are overall close to the LSCF
results in particular when the excitation string's frequency matches the modal frequency of the harp's structure. This can be
explained by the signal to noise ratio improvement around the excitation frequency, as shown for modes 1, 6, 7, 8 and 9 with
excitation strings 40, 29, 34 (with its first harmonic) 27 and 26, respectively. Note that the modified LSCE method allows the
identification of other modes than those identified by LSCF (such as modes 3, 4, and 1013) in the frequency range. This
method seems to be more suitable when there is a high modal overlap or when the modes magnitude have a low level in
the instrument response.
5. Conclusion
This study offers to apply new OMA methods, especially the LSCE and the modified LSCE algorithms, to identify modal
parameters (frequency and damping) of a concert harp. The novelty of the proposed approach is to apply these methods,

B. Chomette, J.-L. Le Carrou / Mechanical Systems and Signal Processing 56-57 (2015) 8191

91

initially developed for rotating machines, in the case of string excitations that also contain harmonic components. This
approach does not need any additional excitation system that could induce some experimental problems and allows the
identification of modal parameters using the excitation system included in the harp that is to say the strings. The results of
this paper demonstrates that the OMA methods help the identification of more modes particularly in the case of high
frequencies where the forced excitation brings a better signal to noise ratio. However, the identified modal parameters can
include some string modes in the case of not completely muffled strings. The OMA methods are moreover particularly
efficient in the case of low damped modes. The comparison between the classical LSCE and the modified LSCE algorithm
shows also that the modified LSCE algorithm allows the identification of more modes especially in the case of an excitation
frequency close to a structural mode. Indeed, classical LSCE methods can fail to identify frequency and damping in this case.
Future researches will focus on the concert harp's mode shapes identification using OMA methods. From musical acoustic
perspectives, these operational methods are very interesting because it identifies only modes present in the instrument
response during play. In order to compare instruments, the present study focusses only on the modes which are responsible
for differences in radiation. Moreover, with this method it is now possible to evaluate the influence of the instrumentalist on
the modal behaviour of the harp or of other musical instruments such as the guitar.
References
[1] W.R. Savage, E.L. Kottick, T.J. Hendrickson, K.D. Marshall, Air and structural modes of a harpsichord, J. Acoust. Soc. Am. 91 (4) (1992) 21802189.
[2] J.A. Torres, R.R. Boullosa, Influence of the bridge on the vibrations of the top plate of a classical guitar, Appl. Acoust. 70 (1112) (2009) 13711377.
[3] S. Le Moyne, S. Le Conte, F. Ollivier, J. Frelat, J.-C. Battault, S. Vaiedelich, Restoration of a 17th-century harpsichord to playable condition: a numerical
and experimental study, J. Acoust. Soc. Am. 131 (1) (2012) 888896.
[4] E. Skrodzka, A. Lapa, B.B.J. Linde, E. Rosenfeld, Modal parameters of two incomplete and complete guitars differing in the bracing pattern of the
soundboard, J. Acoust. Soc. Am. 130 (4) (2011) 21862194.
[5] S. Daltrop, A. Kotlicki, C. Waltham, N. Wolfe, B. Elie, F. Gautier, Vibroacoustic characteristics of a gothic harp, J. Acoust. Soc. Am. 131 (1) (2012) 837843.
[6] A. Chaigne, B. Cotte, R. Viggiano, Dynamical properties of piano soundboards, J. Acoust. Soc. Am. 133 (4) (2013) 24562466.
[7] H.O. Saldner, N.-E. Molin, E.V. Jansson, Vibration modes of the violin forced via the bridge and action of the soundpost, J. Acoust. Soc. Am. 100 (2)
(1996) 11681177.
[8] J. Woodhouse, On the synthesis of guitar plucks, Acta Acust. United Acust. 90(5) (2004) 928944.
[9] J. Woodhouse, Plucked guitar transients: comparison of measurements and synthesis, Acta Acust. United Acust. 90 (5) (2004) 945965.
[10] R. Brincker, L. Zhang, P. Andersen, Modal identification from ambient responses using frequency domain decomposition, in: XVII IMAC, S. Antonio, TX,
2000.
[11] A. De Vivo, C. Brutti, J.L. Leofanti, Modal shape identification of large structure exposed to wind excitation by operational modal analysis technique,
Mech. Syst. Signal Process. 39 (12) (2013) 195206.
[12] Y.F. Xu, W.D. Zhu, Operational modal analysis of a rectangular plate using non-contact excitation and measurement, J. Sound Vib. 332 (20) (2013)
49274939.
[13] A. Agneni, L. Balis Crema, G. Coppotelli, Output-only analysis of structures with closely spaced modes, Mech. Syst. Signal Process. 24 (5) (2010)
12401249.
[14] A. Agneni, L. Balis Crema, G. Coppotelli, Modal parameters directly estimated from power spectral densities or correlation functions in output-only
modal analysis, Exp. Tech., 2013, http://dx.doi.org/10.1111/ext.12067.
[15] B. Peeters, G. de Roeck, Reference-based stochastic subspace identification for output-only modal analysis, Mech. Syst. Signal Process. 13 (6) (1999)
855878.
[16] G. James, T. Carne, J. Lauffer, The natural excitation technique for modal parameter extraction from operating structures, Modal Anal.Int. J. Anal. Exp.
Modal Anal. 10 (4) (1995) 260277.
[17] H. Moncayo, J. Marulanda, P. Thomson, Identification and monitoring of modal parameters in aircraft structures using the natural excitation technique
(NExT) combined with the eigensystem realization algorithm (ERA), J. Aerosp. Eng. 23 (2) (2010) 99104.
[18] J.M. Caicedo, Practical guidelines for the natural excitation technique (NExT) and the eigensystem realization algorithm (ERA) for modal identification
using ambient vibration, Exp. Tech. 35 (4) (2011) 5258.
[19] R. Brincker, P. Andersen, N. Mooller, Estimation of correlation functions by the random decrement technique, in: Proceedings of the Ninth IMAC, 1991.
[20] F. Shen, M. Zheng, D. Shi, F. Xu, Using the cross-correlation technique to extract modal parameters on response-only data, J. Sound Vib. 259 (5) (2003)
11631179.
[21] R. Brincker, P. Andersen, N. Mooller, An indicator for separation of structural and harmonic modes in output-only modal testing, in: Proceeding of the
18th IMAC, 2000.
[22] S. Manzato, D. Moccia, B. Peeters, K. Janssens, J. White, A review of harmonic removal methods for improved operational modal analysis of wind
turbines, in: P. Sas, D. Moens, S. Jonckheere (Ed.), Proceedings Of International Conference on Noise and Vibration Engineering (ISMA2012)/
International Conference on Uncertainty in Structural Dynamics (USD2012), 2012, pp. 26752689.
[23] A. Agneni, G. Coppotelli, C. Grappasonni, A method for the harmonic removal in operational modal analysis of rotating blades, Mech. Syst. Signal
Process. 27 (2012) 604618.
[24] P. Mohanty, D. Rixen, Operational modal analysis in the presence of harmonic excitation, J. Sound Vib. 270 (12) (2004) 93109.
[25] P. Mohanty, D. Rixen, Identifying mode shapes and modal frequencies by operational modal analysis in the presence of harmonic excitation, Exp.
Mech. 45 (3) (2005) 213220.
[26] P. Mohanty, D. Rixen, Modified ERA method for operational modal analysis in the presence of harmonic excitations, Mech. Syst. Signal Process. 20 (1)
(2006) 114130.
[27] P. Mohanty, D. Rixen, Modified SSTD method to account for harmonic excitations during operational modal analysis, Mech. Mach. Theory 39 (12)
(2004) 12471255.
[28] N.H. Fletcher, T.D. Rossing, The Physics of Musical Instruments, 2nd ed., Springer-Verlag, New York Inc. (Softcover of orig).
[29] J.-L. Le Carrou, F. Gautier, E. Foltte, Experimental study of A0 and T1 modes of the concert harp, J. Acoust. Soc. Am. 121 (2007) 559567.
[30] J.-L. Le Carrou, Q. Leclere, F. Gautier, Some characteristics of the concert harp's acoustic radiation, J. Acoust. Soc. Am. 127 (5) (2010) 32033211.
[31] D. Garreau, B. Georgel, La mthode de Prony en analyse des vibrations, Trait. Signal 3 (45) (1986) 235240.
[32] J.-L. Le Carrou, F. Gautier, R. Badeau, Sympathetic string modes in the concert harp, Acta Acust. United Acust. 95 (2009) 744752.
[33] J.-L. Le Carrou, F. Gautier, E. Foltte, Experimental study of a0 and t1 modes of the concert harp, J. Acoust. Soc. Am. 121 (1) (2007) 559567.
[34] Programs for Digital Signal Processing, IEEE Press, New York. 1979.