You are on page 1of 12

___________________________________

Audio Engineering Society

Convention Paper 5320


Presented at the 110th Convention
2001 May 1215
Amsterdam, The Netherlands
This convention paper has been reproduced from the authors advance manuscript, without editing, corrections, or consideration
by the Review Board. The AES takes no responsibility for the contents. Additional papers may be obtained by sending request
and remittance to Audio Engineering Society, 60 East 42nd Street, New York, New York 10165-2520, USA; also see www.aes.org.
All rights reserved. Reproduction of this paper, or any portion thereof, is not permitted without direct permission from the
Journal of the Audio Engineering Society.

___________________________________
Verification of an approach for transient structural simulation of
loudspeakers incorporating damping
G P Geaves, J P Moore, D J Henwood* and P A Fryer
B&W Loudspeakers Ltd, Steyning, W.Sussex, UK and *University of Brighton, Brighton,
E.Sussex, UK
ABSTRACT

An approach for simulating transient structural wave propagation in loudspeakers is described. The Finite
Element Method is used for spatial discretisation and the Laplace transform for the time solution. The accuracy
of the spatial discretisation is verified by simulating the acoustic frequency response of a loudspeaker and
comparing the results with measured data. A damping model is introduced that approximates standard hysteretic
damping and yet can be used directly in both the time and frequency domains. The overall approach is verified
by comparing laser measured and simulated results of the transient structural response of a loudspeaker to an
impulse like excitation. Finally, structural energy is plotted and discussed.

1 INTRODUCTION
A considerable volume of work has been carried out concerned
with applying techniques such as the Finite and Boundary Element
Methods to the simulation of loudspeakers, for example [1-5].
Most previously reported work has been concerned with modelling
the steady state, frequency domain behavior of loudspeakers. The
modeled transient, time domain behavior has received little
attention.

Measurements carried out using laser Doppler interferometry of


the transient structural response of a loudspeaker to an impulsive
excitation clearly reveal how the interface between the cone and
surround has a large influence on the overall performance of a
loudspeaker [6]. In most cases at least some energy is reflected at
this interface. In the ideal case no reflection of energy would occur
and all the energy would be transmitted into the surround, where it
would be completely dissipated by internal material damping.
The overall aim of the work, which the results presented in this
paper form a part, is to develop a method that can be used to

simulate the transient structural response of a loudspeaker. By


doing this a better understanding of the underlying physics of wave
propagation in loudspeakers is sought which in turn will allow
better combinations of cone and surround to be identified.
The work described in this paper builds upon work presented in [7]
where an approach is described and verified that allows the
transient structural performance of loudspeaker-type structures to
be simulated. The work focused upon a simplified case where the
vibrating structure was a flat disc with an outer ring of compliant
material. Damping was not included in the model. In this paper, the
approach described in [7] is extended to incorporate an
approximation to hysteretic damping, which is the preferred
damping model for frequency domain simulation. The approach is
then verified by comparing simulated and laser measured
responses of a real loudspeaker.
The mathematical methods used throughout the paper are
described in section 2. The Finite Element Method (FEM) forms
the basis for modelling the structural performance of a

GEAVES ET AL.

TRANSIENT SIMULATION OF LOUDSPEAKERS

loudspeaker. The Boundary Element Method (BEM) is used to


calculate the corresponding acoustic frequency response function
(FRF).
To calculate the structural transient response, two different
approaches are employed. The favoured approach is to solve the
equation of motion directly in the time domain by first producing a
decoupled form of the equation of motion and then using the
Laplace transform. A major benefit of this approach is that the
solution through time is analytic and therefore exact and
independent of the time step used.
A second approach, used here for verification purposes, is to solve
the equation of motion in the frequency domain, and then to
employ the inverse Fast Fourier Transform (FFT) to obtain the
time domain solution.
In section 3, the FE and BE models used in the paper are described
in detail and in section 4 the accuracy of the FE spatial
discretisation is verified by comparing the simulated and measured
acoustic FRFs.
Damping has a great impact on the structural and acoustic
responses of the system in the time and frequency domains. A
commonly used means of incorporating damping in the frequency
domain is to use a complex value of Youngs modulus, thereby
creating a hysteretic damping model. However, this approach
cannot be used directly in the time domain. Therefore an
equivalent viscous damping model is introduced that approximates
the damping effect caused by use of a hysteretic damping model.
The implications and effects of using both of these models are
discussed throughout section 5.
A consequence of incorporating damping into the equation of
motion is that the eigenvectors and eigenvalues of the system
become complex valued. Care must be taken when interpreting the
meaning of these complex valued quantities and their meaning is
discussed throughout section 5. Examples of time domain basis
functions, the set of linearly independent functions that span the
solution space, are plotted as a means of clearly visualizing the
structural response. The equivalent modal damping factor is
introduced and is used to show how total damping in the system
changes as damping is increased in the surround.
In section 6 results from the two time domain solution methods are
compared and discrepancies are discussed. These simulated results
are then compared with data measured using a laser. Possible
sources of error are discussed.
Finally, in section 7 structural energy is considered. Attempts to
use structural energy to derive a measure of the quality of a cone
/surround combination is outlined and the problems of doing this
explained.
2 MATHEMATICAL METHODS
2.1 The Finite Element Method
The FEM is an established means of solving structural vibration
problems. Details of the FEM are given in, for example [8]. The
FEM is used to spatially discretise the familiar equation of motion
thereby allowing an approximate solution to be found. This is more
usually the frequency domain version of the equation of motion,
assuming simple harmonic motion

( 2 M + jC + K )U = G
where
M, C and K are the global mass, damping and stiffness matrices
and are of dimension n by n, where n is the number of degrees of
freedom in the FE model

TH

(1)

U
G

is the displacement vector


is the force vector and
is the angular frequency.

The more general time domain version of the equation is


&&(t ) + Cu
& (t ) + Ku (t ) = g (t )
Mu

(2)

where
&&(t ), u
& (t ), u (t )
u

are the time dependent acceleration, velocity

and displacement vectors and


g (t ) is the time dependent force vector.
Equations (1) and (2) are written using a standard viscous damping
model. An alternative, strictly only applicable in the frequency
domain, is hysteretic damping, which gives rise to

( 2 M + K (1 + j ))U = G
where

(3)

is the hysteretic damping factor or loss factor.

Hysteretic and alternative forms of damping are discussed in detail


in section 5. For the frequency domain verification presented in
section 4, equation (3) forms the basis of the solution.
The FE spatial discretisation gives rise to matrices that are
applicable in both the time and frequency domain. It is common,
once the FE spatial discretisation has been obtained, to extract
mode shapes and modal frequencies, and to obtain the steady-state
forced response using these modes or in a more direct manner, say
by using Gaussian elimination to solve equations (1) or (3).
2.2 The Boundary Element Method
The BEM is a standard technique used to obtain the acoustic field
resulting from the vibration of a structure [9]. The method is
currently used in a routine way to solve only the frequency domain
equivalent of the wave equation, the Helmholtz equation

2 (r ) + k 2 (r ) = 0

(4)

where

r is the spatial variable vector


(r ) is the velocity potential vector and
k is the wave number.
Also

p (r ) = j (r )
where p is acoustic pressure vector and is the fluid density.
Equation (4) is reformulated as an integral equation by
multiplication with a suitable Greens function and application of
Greens theorem. The BEM is then applied to solve the integral
equation. The BEM, like the FEM, is an element based technique
that similarly results in a spatial discretisation of the problem. The
major advantage of the BEM is that only the boundary of the
domain of interest, rather than the domain itself, requires
discretisation, an obvious benefit when modelling infinite domains.
The reader is referred to [9,10] for details of the BEM applied to
acoustical problems.

AES 110 CONVENTION, AMSTERDAM, NETHERLANDS, 2001 MAY 1215

GEAVES ET AL.

TRANSIENT SIMULATION OF LOUDSPEAKERS

2.3 Time domain solution methods


In this work, two methods have been employed to solve equation
(2). These are termed the direct and indirect approaches.
2.3.1 DIRECT APPROACH
The direct approach proceeds by first arriving at a decoupled
version of equation (2) and then using the Laplace transform to
solve the subsequent set of equations. Details of this method are
given in [11] and a brief outline follows.

&&(t )
v& (t ) = u

= v (t )
1

v& (t ) = M Cv (t ) M Ku(t ) + M g(t )

& (t ) = Aw (t ) + h (t )
w

0
q = X 1
1
M p

g (t )

i=12n

(10)

can be expressed in terms of elementary functions,

equations (10) can be solved using the Laplace transform

(s - d i )Yi (s ) = qi G( s )
and thus using the reverse transform

q
yi (t ) = L-1 i G (s )
s
d
i

(7)

where

I
0
A=
1
1
M
K
M
C

u(t )

and

v(t ) via w (t ) = X 1y (t )

Though this indirect approach is the more traditional for solving


problems of this type, the direct approach is preferred. This is
mainly because the direct approach offers an analytic solution
through time which therefore removes restrictions on size and
number of time/frequency steps.

and

0
h(t ) = g (t )
1
M p

However, the direct approach does require a force that can be


described in terms of elementary functions, use of modal
coordinates and care when constructing the damping matrix C.
Because of these restrictions, it may in some circumstances be
more appropriate to use the indirect method. In this paper, the
indirect approach is mainly used as a means of verifying the direct
approach.

g (t ) = g(t )p

p = [w1 ,......, wn ]T
th

equal to 1 if the i

degree of freedom has an applied

force or is zero otherwise.


If it is assumed that X is a matrix with columns formed with the
eigenvectors (i.e. the mode shapes) of A and that D is a diagonal
matrix of eigenvalues (i.e. natural frequencies) of A, then

AX = XD
Rewriting equation (7) as

1
& (t ) = AXX w (t ) + h(t )
XX 1w
1

thereby giving

i=12n

2.3.2 INDIRECT APPROACH


The indirect approach proceeds by solving the equation (1) or (3)
in the frequency domain employing the FEM with a unit force at
each frequency. The solution is then multiplied by a filter
corresponding to the required force. Finally the inverse FFT is
used to convert to the time domain.

u(t )
w (t ) =

v(t )

and putting

and

(6)

Equations (5) and (6) may be written more succinctly as

wi is

k (t ) = g(t )q

(9)

(5)

where

where

If

equation (2) can be re-written in terms of two first order


differential equations

and

y& (t ) Dy (t ) = k (t )

y& (t ) d i yi (t ) = g (t )qi

v(t ) = u& (t )

where

or

Equation (9) is now a decoupled system with

By making the following substitution

u& (t )

X (y& (t ) Dy (t )) = h(t )

X w (t ) = y (t ) i.e.

3 THE FINITE AND BOUNDARY ELEMENT MODELS


Analytic solutions for transient structural solutions exist only for
highly idealised structures, for example flat discs [7]. Therefore
verification of the approach in this case was achieved by modeling
a real loudspeaker and comparing simulated and measured
quantities.
3.1 The structural Finite Element Model
The details of the structural FE model are given in figure 1 and
table 1. The element used is this case was a three-noded
axisymmetric thin shell of revolution element based on an element
devised by Webster [12]. The element models tangential and
transverse displacements to 3rd and 5th order approximations
respectively.

(8)

transforming from physical to

modal coordinates, equation (8) becomes

TH

AES 110 CONVENTION, AMSTERDAM, NETHERLANDS, 2001 MAY 1215

GEAVES ET AL.

TRANSIENT SIMULATION OF LOUDSPEAKERS

An uncoupled solution method was employed utilising the Burton


and Miller method to stabilise the solution at frequencies
corresponding to the internal resonant frequencies of the enclosed
volume [10]. Frequencies from 100 Hz to 10 kHz were calculated.
4 FREQUENCY DOMAIN VERIFICATION
As a means of verifying the model described in section 3 the
calculated and simulated acoustic FRFs were compared. The
material properties given in table 1 were obtained mainly from
simple measurements and published data. The sensitivity of the
acoustic FRF to changes in the Youngs modulus and loss factor of
the surround and cone was established, and to a small extent, these
parameters were tuned to give the best correspondence with
measurement.

Fig. 1. The structural finite element model. The dots


represent FE nodes.
Component
description

E
(MPa)

Density
(Kg/m3)

Loss

72800
4800
0.008
Coil
(Cu wire)
5800
1000
0.003
Former
(Kapton)
3000
500
0.070
Cone
(Paper)
3.5
1200
0.400
Surround
(Rubber)
Table 1. Material and physical properties.

Thickness
(mm)
0.550
0.245
0.750
0.400

In figure 3 a comparison is made between the measured and


simulated acoustic FRFs. As the force used for the simulation was
constant with frequency and of arbitrary amplitude, a 13 dB shift in
overall level has been applied to the simulated results. In addition a
2.2 dB per octave slope has been applied to the simulated results to
model approximately the effect of the change in inductance with
frequency of the voice coil.
The low frequency behavior of the system is not accurately
predicted. This is partly because the spider has not been modeled
and the model is uncoupled (so there is no stiffening effect from
the enclosed volume) and therefore the frequency of the
fundamental mode is predicted at too low a frequency. It is also
partly because the induced back EMF has not been modeled and
therefore there is no electromagnetic damping. All these features
could be incorporated but have not been because the low frequency
behavior is not of interest in this work.

3.2 The Boundary Element Model

For the calculation of the acoustic field, the loudspeaker was set in
a spherical enclosure and a stationary phase plug was added. The
BEM models the radiating surface composed of the cone and
surround together with non-radiating surface composed of the
spherical enclosure and stationary phase plug. The model is shown
in figure 2.
Three-noded elements were used which model the geometry to a
2nd order approximation, though the variation in velocity potential
across the surface is assumed to be constant over each element.
The FE and BE nodal coordinates are available from the author for
interested readers wishing to reproduce the results given in this
paper.

Fig. 3. Comparison between measured (solid) and


simulated (dotted) acoustic FRFs. Dots represent
simulation frequencies.
Other than at low frequencies the prediction is quite accurate and
within approximately 1.5 dB of the measurement. However, most
importantly the general trend is predicted with the five main peaks
identified. It should be noted that it is not realistic to expect the
results of FE simulations to be perfect. Generally, FE is used in a
relative rather than absolute way, and this is where the strength of
the method lies. Strictly, one-off comparisons between
measurements and simulations are problematic as there will be a
natural variation between geometries, materials, construction etc.
even in nominally identical units.
The reasonable correspondence between simulation and
measurement gives confidence that the FE spatial discretisation is
sufficiently accurate.

Fig. 2. The acoustic boundary element model. The


crosses represent BE nodes.

TH

AES 110 CONVENTION, AMSTERDAM, NETHERLANDS, 2001 MAY 1215

GEAVES ET AL.

5 DAMPING
5.1 Damping models

Damping has a large effect on the response of a loudspeaker.


Therefore careful consideration is required of how it should be
implemented into any simulation. Overviews of damping in
vibration problems may be found in [13-15].
The most commonly used model for damping is viscous or VoightKelvin damping. Application of this damping model gives rise to
equations of the form given by (1) and (2). Damping is introduced
as a velocity dependent term in the time domain. This has the
consequence that the frequency domain damping force is linearly
dependent on frequency. However this linear dependence with
frequency often does not match measured damping values.
An alternative damping model which is applicable for frequency
domain simulation is hysteretic damping. Use of a hysteretic
damping model gives rise to equation (3). Application of hysteretic
damping results in a complex stiffness matrix and a damping force
that is independent of frequency.
Hysteretic damping gives the best match with experimental
measurement of damping in materials commonly used in
loudspeakers and is the preferred and familiar model when
carrying out frequency domain simulation. More importantly, as
was shown in section 4, use of hysteretic damping results in
reasonably accurate simulations. Therefore, hysteretic damping is
the preferred model to use in the time domain also. The problem is
that because the model relies on complex notation it is not
applicable directly in the time domain.
A more subtle problem is detailed in [14], where it is shown that
hysteretic damping when applied to a simplified one-dimensional
case gives rise to an acausal system. This is mainly as a result of
non-zero damping at 0 Hz, which gives rise to a complex DC
(static) response. This complex DC response will result in a
discontinuity in displacement response, if a real valued time
response is constructed in the normal way by forming a conjugate
response for negative frequencies and then using an inverse
Fourier transform.
This acausality problem naturally extends to more complex, multidegree of freedom systems. Strictly, this means that hysteretic
damping is not an appropriate time domain description of damping.
However, it will be demonstrated in section 6, that the effect of
acausality is small for systems of the type considered here which
are characterised by moderate levels of damping.
It is clear from inspection of equations (1) and (3) that viscously
and hysteretically damped systems can be made to be equivalent at
a single frequency. This condition will occur if the viscous
damping matrix C = K/.
The matching of damping levels at a single frequency is of little
use. However, if a frequency dependent viscous damping matrix C
is constructed instead of one constant with frequency, matching at
a set of non-zero frequencies is possible. A modal decomposition
can be used to regroup equation (3) to identify the natural
frequencies. It is then possible to match approximately the
damping in viscous and hysteretic systems at these natural
frequencies. This is considered to be acceptable as damping mainly
has an effect on the response at these natural frequencies, and the
more lightly damped the structure the more accurate this
approximation becomes.
This then presents us with the concept of an equivalent viscous
damping model which is applicable in both the time and frequency
domain and that, at least for relatively lightly damped structures,
approximates to hysteretic damping, except at frequencies below

TH

TRANSIENT SIMULATION OF LOUDSPEAKERS

the fundamental. The description given here of how a C matrix is


developed that approximates to hysteretic damping is necessarily
brief and qualitative in nature. A detailed mathematical description
and consideration of the consequences of this approach is given in
[11].
5.2 Basis functions for the time domain solution
The traditional (classical) approach for incorporating damping into
the equation of motion is to extract real valued eigenvectors and
eigenvalues by setting C to zero in equation (1), apply a modal
damping factor to each mode and then carry out forced solution
using superposition [16]. This approach will give a real solution in
the frequency domain and is applicable in the time domain also.
However, this approach suffers from a number of problems.
Firstly, measured modal damping factors are required, which will
be different for each structure and secondly the effect of damping
on modal frequencies is not included.

Use of localized material damping, whether it be viscous,


hysteretic or equivalent viscous in nature, requires a non-classical
approach to damping, and results in complex eigenvalues and
eigenvectors. The forced frequency domain solution obtained by
superposition is complex as well.
Complex modal quantities require some care when interpreting
their meaning. A detailed interpretation of the meaning of
complex eigenvalues and eigenvectors in the context of this work
is given in [11]. More general descriptions are given in [17-19].
Equation (7) yields a set of 2n eigenvectors and eigenvalues, both
of which are complex. The eigenvectors occur in pairs with
corresponding conjugate eigenvalues. Thus

i = i + j i

and

*i = i j i

i=1.n

and it can be shown that the eigenvectors [11] have the following
form

x i = a i + jb i

and

~
xi = a i jb i

i=1.n

(11)

where

x i is the ith eigenvector corresponding to i and


~
x is the ith eigenvector corresponding to * .
i

The kth (spatial) components of equations (11) may be written as

xik = rik e j ik
~
xik = rik e -j ik

and
i, k=1.n

(12)

where

rik = aik2 + bik2


and

b
ik = tan 1 ik
aik

The basis for the solution space (the set of linearly independent
functions from which any solution can be constructed) for the time

AES 110 CONVENTION, AMSTERDAM, NETHERLANDS, 2001 MAY 1215

GEAVES ET AL.

TRANSIENT SIMULATION OF LOUDSPEAKERS

domain equation (2) can be developed by considering the kth


components of

e i t xi

and

*
e it ~
xi

which may be written as

e ( i t ) rik e j ( i t + ik )

and

e ( i t ) rik e -j ( i t + ik )

i=1n

(13)

These are a complex form of the basis functions. However the


(real) time domain solution space must be spanned by real basis
functions. These real basis functions may be derived from equation
(13) by linear combination

e ( i t ) rik cos( i t + ik )

and

e ( i t ) rik sin ( i t + ik )

i=1n

(14)

Each basis function is composed of the same magnitude terms as in


equations (12), a decay term given by the real part of the
eigenvalue (which is negative) and an oscillatory component with
a natural frequency given by the imaginary part of the eigenvalue.
For a system with spatially uniform material damping it can be
shown that [11]

Therefore an approximate equivalent hysteretic damping factor for


each mode can be derived

2 i
~i =
i
This value is comparable with

(15)

the hysteretic damping factor.

The difference between the two factors is that is a material


damping value that can vary from element to element where as

~i is a

modal damping factor. Thus it can be regarded as a single

approximate measure of overall damping in a system at each


individual natural frequency i .
5.3 Examples of time domain basis functions and
equivalent modal damping factors
In this section, the time domain basis functions defined by
equation (14) are graphically represented and discussed together
with the equivalent modal damping factor defined by equation
(15).

The time domain basis functions defined in equations (14) form


pairs. Either component can be plotted as they differ by only a 90
phase angle. In figures (4a-7a) these functions are represented with
the axial component of displacement plotted as a function of node
number and time. In figures (4b-7b) the equivalent modal damping
factors plotted as a function of frequency at each natural frequency
as shown.

TH

Four cases are considered with hysteretic damping values in the


surround of 0.0, 0.2, 0.4 and 0.8 (figures 4-7 respectively) but with
material properties otherwise defined by table (1). In each of the
cases two modes are illustrated. Firstly a mode that occurs at
approximately 938 Hz in the standard case with 0.4 surround
damping (figure 6), which is an example of a mode with motion
predominantly in the surround. This mode has little effect on the
acoustic FRF (figure 3). And secondly a mode occurring at
approximately 3836 Hz in the standard case (figure 6) case, which
is a mode with significant motion in the cone which contributes to
a large peak in the acoustic FRF. The frequencies of these two
modes are indicated on the equivalent modal damping factor plots
as solid diamonds.
These plots give a lot of information about the vibrational behavior
of the system. The t=0 slice of the plots can be regarded as a
representation of the familiar steady state, frequency domain mode.
The relative displacements of each part of the system can be
clearly seen as well as the decay of vibration through time. They
also show how the equivalent overall damping in the system
changes as more damping is added to the surround.
Inspection of the equivalent damping factors in each of the four
cases reveals that as surround damping increases, the general level
of damping increases for most modes. However, the modes that
correspond to the four main peaks in the acoustic FRF (figure 3)
above 3 kHz are largely unaffected by increased surround damping
levels. This would seem to suggest that at these frequencies the
cone and surround are not well coupled and vibrate to a large
extent as separate entities. This could be interpreted as meaning
there is a large mismatch in mechanical impedance between the
two components.
Also of note is the change in frequency of the modes as the
surround loss factor is increased from 0.0 to 0.8 The surround
mode decreases in frequency (from 962 Hz to 858 Hz) and the
cone mode increases in frequency (from 3771 Hz to 3876 Hz). It
can be shown for a single degree of freedom system that viscous
damping results in a decrease in modal frequency and that
hysteretic damping results in an increase in modal frequency
compared to an undamped case [11]. The cause of the
inconsistency with the simplified theory in this case is not known.
In practice it is difficult to establish what influence damping has on
modal frequency as the level of control of material properties and
accuracy required in measurement is not currently achievable.
However, the changes in frequency are small for low to moderate
levels of damping.
6 TIME DOMAIN VERIFICATION
In section 2, two approaches were outlined that allow transient
structural simulation to be carried out. Both approaches rely upon
the structural FEM to spatially discretise the structure. The first
approach solves the equation of motion directly in the time
domain. The second approach solves the equation of motion in the
frequency domain and then utilises the inverse FFT to arrive at the
time domain solution.

In section 3 it was shown how, by considering the simulated and


measured acoustic FRFs, that the FE spatial discretisation gives
acceptable accuracy.
In this section results obtained from the two transient simulation
approaches are compared to each other and to data measured with
a laser.

AES 110 CONVENTION, AMSTERDAM, NETHERLANDS, 2001 MAY 1215

GEAVES ET AL.

TRANSIENT SIMULATION OF LOUDSPEAKERS

(a)

(a)

(b)
Fig. 4. (a) Two time basis functions and (b) equivalent
modal damping values when 0.0 hysteretic damping factor
applied to surround

TH

(b)
Fig. 5. (a) Two time basis functions and (b) equivalent
modal damping values when 0.2 hysteretic damping
factor applied to surround.

AES 110 CONVENTION, AMSTERDAM, NETHERLANDS, 2001 MAY 1215

GEAVES ET AL.

TRANSIENT SIMULATION OF LOUDSPEAKERS

(a)

(a)

(b)
Fig. 6. (a) Two time basis functions and (b)
equivalent modal damping values when 0.4 hysteretic
damping factor (standard case) applied to the surround

(b)
Fig. 7. (a) Two time basis functions and (b)
equivalent modal damping values when 0.8 hysteretic
damping factor applied to the surround

TH

AES 110 CONVENTION, AMSTERDAM, NETHERLANDS, 2001 MAY 1215

GEAVES ET AL.

TRANSIENT SIMULATION OF LOUDSPEAKERS

6.1 The force

The force chosen for this exercise was a raised cosine pulse with a
100 s period (figure 8). This force was chosen firstly because it is
impulse-like. Secondly, it is necessary to use a signal that is band
limited in some way. In this case the 6 dB down point is at 10 kHz.
This band limitation is required because a given spatial FE mesh is
only able to accurately model a forced response up to a certain
frequency. Though this frequency can be extended by using a finer
mesh, arbitrarily high frequencies cannot be modelled by any given
FE mesh.

(a)

Fig. 8.Time domain and frequency domain


representations of the force 100 s raised cosine
pulse.
6.2 Time domain verification
Time domain verification is carried out in two stages.
(b)

Firstly, the direct time domain approach, utilising equivalent


viscous damping is validated against the indirect approach using
the inverse FFT from the frequency domain, utilising standard
hysteretic damping. Acausality aside, this can be regarded as the
right answer in the sense that it is what is to be approximated in the
time domain. This comparison is intended to show the difference
between the two damping models used. It is also intended to
demonstrate that the decoupling procedure and Laplace method for
the direct solution were implemented correctly.
Secondly, the simulated responses are compared to data measured
using a laser.

Fig. 9. (a) direct approach (Laplace transform) using


equivalent viscous damping (b) indirect approach (inverse
FFT) using standard hysteretic damping.
A more detailed comparison is made in figures 10(a-c), where
slices are taken from the plots shown in figures 9(a-b). In this way
the time histories of single points are compared. The responses
compare well in general showing only very small differences. The
responses in the surround (figure 10c) show the greatest
discrepancies. The relatively high levels of material damping in the
surround are thought to be the cause of this increased difference.

6.2.1 COMPARISON BETWEEN DIRECT AND INDIRECT


SIMULATIONS

The acceleration resulting from application of the 100 s raised


cosine pulse described in section 6.1, plotted as a function of time
and spatial coordinates for both direct and indirect solution
approaches is shown in figures 9(a-b). Only the axial component is
shown and the motion from the neck of the cone to the edge of the
surround is plotted. Clearly, both sets of results are very similar.

TH

AES 110 CONVENTION, AMSTERDAM, NETHERLANDS, 2001 MAY 1215

GEAVES ET AL.

TRANSIENT SIMULATION OF LOUDSPEAKERS

(a)

Fig. 11. Comparison between time histories obtained


using the direct (dotted) and indirect (solid) methods for
a point in the surround showing the full length of the
time interval.
6.2.2 COMPARISON BETWEEN
LASER MEASURED DATA

SIMULATIONS

AND

In figure 12 the transient structural response of the loudspeaker to


an impulse measured using laser Doppler interferometry is shown.
These should be compared to the simulated results shown in
figures 9a-b. Clearly the simulated results compare well
qualitatively with the measured data with the main features and
general trend predicted.
(b)

Fig. 12. Laser measured response


(c)
Fig. 10. Comparison between time histories obtained
using the direct (dotted) and indirect (solid) methods
for points (a) at the neck of the cone, (b) near the edge
of the cone and (c) in the surround.

As discussed in section 5.1, strictly the hysteretic damping model


is not appropriate to use in the time domain as its use results in an
acausal, physically unrealisable response. However, the effect of
acausality is small and not obviously apparent in figures 10(a-c). In
figure 11, which shows the entire time interval defined by the
Fourier transform data used, the effect of acausality more clearly
manifests itself as the rising response at the end of the time
interval.
While there are differences between responses obtained by the
direct and indirect approach generally these are not significant in
this case. Therefore it is concluded that equivalent viscous
damping is a sufficiently accurate approximation to hysteretic
damping and one that is directly applicable to the time domain. It
is also concluded that the direct approach is valid.

TH

However in a quantitative way the simulations could be improved.


The following possible sources of error are suggested (and others
may exist).
The material data could be improved and the suspected frequency
dependence of Youngs modulus and loss factor included.
Non-linear effects are not included.
The effect of the back EMF and inductive rise with frequency is
not included in the transient simulations.
The effect of air is not included, though this is likely to be small in
this case.
Inaccuracy in position and angle made with the vibrating surface of
the laser and noise in the laser measuring system.
It should be noted that the transient simulations discussed in
section 6.2.1 and the acoustic FRF simulation in section 4 are two
results obtained from the same input data and using largely
overlapping methods for the structural calculations. As far as

AES 110 CONVENTION, AMSTERDAM, NETHERLANDS, 2001 MAY 1215

10

GEAVES ET AL.

TRANSIENT SIMULATION OF LOUDSPEAKERS

acoustic simulation is concerned the results shown in section 4 are


relatively accurate. However, in this section a more punishing
comparison is made between simulated and measured data, where
a 3D plot on a linear scale is considered as opposed to a 2D graph
on a decibel scale.
7 ENERGY

In figure 13 structural energy of each element is plotted as a


function of element number (for cone and surround elements) and
time. In addition graphs of total, surround and cone energy plotted
as a function of time are shown. The details of how energy can be
derived when using the FEM are given in [7 and 11]. These plots
are included as an example of the way simulated results can be
analysed and visualised in ways that are not readily achievable
with measured data. These energy plots very clearly show how
energy reverberates throughout the structure and decays over
time.

narrowband and of short duration. These are clearly mutually


exclusive properties.
An alternative, less specific, method is perhaps to consider the rate
at which the cone or total energy level decays. Cone energy is
favoured here as it is the major contributor to radiated output and
does not include surround energy due to the systems fundamental
resonance. This is a circa 20Hz component and is readily observed
on an appropriate time scale as shown in figure 13b. In terms of a
quality factor, either cone or total energy decay could be regarded
as analogous to the measured reverberation time of a room, such as
RT60, and therefore considered very relevant. However, contrary
to these expectations no link has as yet been drawn between this
and more conventional observations such as the smoothness of the
acoustic FRF.
Work is on going in this area.
8 CONCLUDING DISCUSSION

A finite element representation of a realistic loudspeaker has been


verified in the frequency domain by comparing simulated and
measured acoustic FRFs. Damping models have been discussed.
Equivalent viscous damping has been introduced as a viable
damping model that approximates hysteretic damping and that is
directly applicable in the time domain. The complex eigenvectors
and eigenvalues that result from utilising this non-classical
damping model have been discussed and the time domain basis
functions plotted for a number of cases. The equivalent modal
damping factor has been discussed and it has been demonstrated
that increasing surround damping does little to improve the overall
damping of the structure at higher frequencies. It is postulated that
this is because the cone and surround are not well matched over a
broad band in this case.
(a)

(b)
Fig. 13. Energy resulting from 100 s raised cosine
pulse over (a) 1 ms and (b) 50 ms

Work has been focused upon determining a parameter from these


energy distributions with the intention of relating this with some
general measure of system quality. Two approaches have been
investigated:
On initial inspection, and at a fundamental level, it is of interest to
consider the interaction of wave energy at the cone/surround
boundary. The aim of this is to determine a reflection/transmission
co-efficient which can be related to the energy transferred into (and
hence dissipated in) the surround. However, for a realistic structure
of this geometry and stiffness, arriving at this type of quantity is
inherently problematic. This is mainly because it is not possible to
isolate a single cone/surround interaction. For this to be achieved
in a dispersive medium requires the applied signal to be both

TH

The direct approach utilising equivalent viscous damping has been


demonstrated to give comparable results to an indirect approach
utilising hysteretic damping for structural time domain simulations.
Data measured using a laser has shown that at least qualitatively
the transient simulations are good. Finally energy plots, derived
from the FE system matrices has been plotted and discussed.
It is concluded that an approach based on the FEM, equivalent
viscous damping and the Laplace transform is a suitable, reliable
and robust scheme for carrying out time domain structural
simulation. However, in some circumstances, say when the
response to a force is required that cannot be expressed in terms of
elementary functions or in a system with heavy damping, the
indirect approach utilising hysteretic damping may be more
appropriate. However, in heavily damped cases, the lack of
causality inherent when using an hysteretic model may become an
issue.
It should be noted that this paper has focused mainly on
implementing an existing and familiar damping model commonly
used in the frequency domain, namely hysteretic damping, directly
into time domain simulations. It is readily accepted that this
damping model may not be the best to use in the time domain and
is not ideal in the frequency domain. However, it is hoped that this
model will prove to be a usefully accurate approximation.
Further work will focus upon improving the simulated results,
testing the approach on other models and interpreting and
visualising data. In particular work will focus upon trying to obtain
links between the structural transient data and the quality of a
design.

AES 110 CONVENTION, AMSTERDAM, NETHERLANDS, 2001 MAY 1215

11

GEAVES ET AL.

9 REFERENCES

[1] Shepherd, I.C. and Alfredson,R.J. (1985), An improved model


of direct-radiator loudspeakers, J. Audio. Eng. Soc., 33(5).

TRANSIENT SIMULATION OF LOUDSPEAKERS

[19] Ahmida, K.M. and Arruda, J.R.F. (2000), Interpreting


complex modes as a wave propagation phenomenon, Proc.
ISMA25 Noise and Vibration Engineering Conference, Leuven,
Belgium, Sept 13-15, 2000.

[2] Kaizer, A.J.M. and Leeuwestein, A. (1988), Calculation of the


sound radiation of a nonrigid loudspeaker diaphragm using the
finite-element method, J. Audio. Eng. Soc., 36(7/8).
[3] Wright, J.R. (1994), Radiation impedance calculation by finite
element analysis, Inst. Acoustics Bulletin, 19(6).
[4] Geaves, G.P. (1996), Design and validation of a system for
selecting optimized midrange loudspeaker diaphragm profiles, J.
Audio. Eng. Soc., 44(3).
[5] Karjalainen, M., Ikonen, V., Jdrvinen, A., Maijala, P., Savioja,
L., Suutala, A., Backman, J. and Pohjolainen, S. (1998),
Comparison of numerical simulation models and measured lowfrequency behavior of a loudspeaker, 104th Convetion of the AES,
preprint(4722).
[6] Fryer, P.A. (1991), Laser techniques in loudspeaker design
including the impulse progression plot, Acoustics Letters, 14(7).
[7] Henwood, D.J., Moore, J.P., Geaves, G.P. and Fryer, P.A.
(2000), Towards the transient modelling of loudspeaker
diaphragm/surround boundaries, Presented at the 108th Convention
of the AES, preprint(5166).
[8] Petyt, M. (1990), Introduction to Finite Element Vibration
Analysis, Cambridge University Press.
[9] Ciskowski, R.D. and Brebbia, C.A. (eds.) (1991), Boundary
Element Methods in Acoustics, Computational Mechanics
Publications.
[10] Kirkup, S.M. (1989), Solution of Exterior Acoustic Problems
by the Boundary Element Method, PhD, Brighton Polytechnic.
[11] Henwood, D.J. (2001), Approximating hysteretic damping in
the time domain, B&W White Paper and submitted to J. Sound
Vib.
[12] Webster, J.J. (1967), Free vibrations of shells of revolution
using ring elements, Int. J. Mech Sci., 9.
[13] Cremer, L., Heckl, M. and Ungar, E.E (1973), Structure
Borne Sound, Chapter 3 Damping, Spinger-Verlag.
[14] Crandall, S.H (1970), The role of damping in vibration theory.
J. Sound Vib 11(1).
[15] Bert, C.W. (1973), Material damping: an introductory review
of mathematical measures and experimental techniques, J. Sound
Vib. 29(2).
[16] Bathe, K.J. and Wilson, E.L. (1976), Numerical Methods in
Finite Element Analysis, Prentice Hall.
[17] Mitchell, L.D. (1990), Complex modes: a review, Proc. 8th
Int. Modal Analysis Conference, Kissimmee, USA.
[18] Lallement, G. and Inman, D.J. (1995), A tutorial on complex
eignevalues, Proc. 13th Int. Modal Analysis Conference, Nashville,
USA.

TH

AES 110 CONVENTION, AMSTERDAM, NETHERLANDS, 2001 MAY 1215

12

You might also like