Sei sulla pagina 1di 39

Comparison

of Numerical

Simulation

Models and Measured Low-Frequency

Behavior of a Loudspeaker

4722 (P3-3)

Matti Karjalainen Helsinki University of Technology Espoo Finland Antti Jdrvinen Helsinki University of Technology Espoo Finland Lauri Savioja Helsinki University of Technology Espoo Finland JuhaBackman Nokia Mobile Phones Salo Finland

Veijo Ikonen Tampere University of Technology Tampere Finland Panu Maijala Helsinki University of Technology Espoo Finland Antti Suutala Tampere University of Technology Tampere Finland Seppo Pohjolainen Tampere University of Technology Tampere Finland

Presented at the 104th Convention 1998 May 16-19 Amsterdam

AUDIO

Thispreprint has been reproducedfrom the author's advance manuscript,withoutediting, corrections or considerationby the Review Board. The AES takesno responsibility for the contents. Additionalpreprintsmay be obtainedby sendingrequestand remittance to the Audio EngineeringSociety, 60 East 42nd St., New York,New York 10165-2520, USA. Al/rights reserved.Reproduction of thispreprint, or any portion thereof, is not permitted withoutdirect permissionfrom the Journalof the Audio Engineering Society.

AN AUDIO ENGINEERING SOCIETY PREPRINT

Comparison of Numerical Simulation Models and Measured Low-Frequency Behavior of a Loudspeaker


Matti Karjalainen 1, Veijo Ikonen 2, Antti J_irvinen l, Panu Maijala l, Lauri Savioja l, Antti Suutala 2, Juha Backman 1,3, and Seppo Pohjolainen 2 1Helsinki University of Technology Espoo, Finland 2 Tampere University of Technology Tampere, Finland 3Nokia Mobile Phones, $alo, Finland
matt i. karj alainen_hut, f i, i77309_cc, tut. f i, antt i. j arvinen_hut, fi f i

panu.maij ala_hut, fi, lauri, savioj a_hut, fi, s 129948_alpha. cc. tut. j uha. backman_nmp, nokia, eom, seppo, pohj olainencc, tut. f i http://acoustics, hut .fi/

ABSTRACT
The vibroacoustic behavior below lkHz of a prototype closed-box loudspeaker has been studied in detail by comparing measurements and element model simulations. Sound fields were measured using a microphone array of 90 electret capsules and vibrations using a laser vibrometer and accelerometers. Simulations have been carried out using analytical, finite and boundary element, and finite difference methods. The enclosure conditions were varied from fixed wall case buried in sand to free-standing empty box and to free-standing damped box. Two positions of the driver in the front plate were examined (only the 'rim' position is documented here). The applicability of each modeling technique is discussed 1.
1Special thanks for support are due to'. Kaarina Melkas, Nokia Research Center, Tampere, Finland Jorma Salmi, Gradient Oy, J_rvenpg_&, Finland Aki M&kivirta and Ari Varla, Genelec Oy, Iisalmi, Finland Jukka Linjama, VTT Manufacturing Technology, Espoo, Finland Technology Development Centre Finland (TEKES) 1

INTRODUCTION
The design of a loudspeaker has traditionally been an iterative process based on approximate rules, experience from prior designs, and finally trial and error by constructing and modifying prototypes. Computer-based methods have helped in exploring basic features of driver to enclosure matching and crossover network design. However, not much have been published on the use of more advanced computer-based methods and tools in loudspeaker design. The detailed behavior of a loudspeaker consisting of an enclosure and driver(s) is very complex and escapes analytical mathematical solutions. Approximate (semianalytic) approaches may turn out to be useful, however, especially within a limited frequency range and when the geometry of the system is simple enough, e.g., a shoebox design. At low to mid frequencies, lumped element models may be used both for the electroacoustic and vibroacoustic subsystems. In more complex cases the system should be considered as a complex electro-vibro-acoustic system that is not easily partitioned to any simple submodels. The progress in computer-based numeric simulation of complex spatially distributed systems using various element methods has raised the question of how useful they might be for practical loudspeaker design [1]. These modeling techniques include the finite element method (FEM), the boundary element method (BEM), and the finite difference time domain (FDTD) method. In advanced forms they can be used for simulating any linear and time-invariant (and with limitations nonlinear) vibroacoustic systems at low frequencies. Low frequencies means here that the element size in the model mesh, and thus the number of spatially discrete elements, limits the highest useful frequency of the simulation. Loudspeaker design at low to mid frequencies is in principle a good application for such methods. Several commercial or experimental tools are available for element-based vibroacoustic modeling and simulation, such as SYSNOISE [2], I-DEAS Vibroacoustics [3] and Comet/Acoustics [4]. Other FEM/BEM tools that are not specifically tuned to acoustic problems are ABAQUS [5], ANSYS [6], and MSC/NASTRAN [7]. The problem of using the FEM/BEM programs, at least from the point of view of loudspeaker design, is that they are expensive, need powerful computers to work fast, the construction of the model is tedious, and the availability of material data (acoustic and mechanical parameters) is poor.

Experimental programs, e.g., from academic institutions, often lack documentation and continuing support. Thus, such simulation and design tools are not widely used in loudspeaker design, and information on their usefulness as well as comparison of their properties is practically non-existing. As the progress in this field is fast, it is important to be prepared to utilize such tools whenever they turn out to be productive. Potentially, computer-based design tools promise to make the product development time faster whenever the designer can start from an approximate model and rapidly go through variations and experimentations using software modeling up to a prototype which, when actually built, works closely enough as expected. Even more ambitiously, the computer may automatically run through some optimization steps to search for the best match to given specifications and targe t criteria. In this study our interest was focused to the applicability of the elementbased simulation and modeling tools to basic loudspeaker design. We selected a case that is simple enough, yet practical and realistic. Thus we specified and constructed a closed box enclosure with a single driver element so that it was simple to vary some interesting parameters such as the driver position, the stiffness of the enclosure walls, and the damping material inside the box. Next the actual vibroacoustic behavior of our case was measured extensively with such parametric variations. A microphone array with 90 electret capsules was constructed to measure the sound field inside and outside the loudspeaker box as impulse responses of electric excitation of the driver. A laser vibrometer and accelerometers were used to obtain vibration data of the walls and the cone of the driver so that an almost complete picture of the behavior below about 1 kHz was captured in this data. The acoustical and mechanical parameters of the materials (MDF for construction and fiber wool for damping) were also measured. The next step of the study was to model the loudspeaker using various FEM, BEM, and FDTD software tools. The computational models were built first and the measured or estimated material parameters were given to the models. The simulation results of the models are shown in this article and compared with the measured behavior. Finally, the models were hand tuned to match better to the measured data. This yields new material parameters that work better than the measured ones in modeling similar cases, but may not be generalizable to very different cases. The results of simulations and measurements are also compared to simpler analytic or semianalytic models of the same loudspeaker.

After presenting the results we will discuss the usefulness of the models and how they could be improved. Directions of further studies are shown as well.

CASE STUDY: A CLOSED-BOX

LOUDSPEAKER

To study a problem of reasonably low complexity, yet interesting from a practical point of view, we designed and constructed a closed-box prototype loudspeaker of medium size (600 x 400 x 250 mm 3) and easy enough to modify and measure. The structure of the enclosure is shown in the drawings of Fig. 1. The front panel (facing up in the figure) is removable and is built in two variations, one with a driver element symmetrically in the middle and another one with an asymmetric speaker positioning (rim case), as shown in the top drawing of Fig. 1. The loudspeaker element was a 6.5 inch driver of type SEAS P17 REX. The enclosure was made of 20 mm MDF, all panels being rigidly coupled at their edges. In our simulations and measurements the behavior of the box was studied both as buried in sand and as freestanding to allow walls to vibrate. In both cases it was studied as an empty box and with damping material 1.1 (Partek mineral wool) on the back wall inside the box (Fig. 1). Model of the Loudspeaker

Conceptual

h'om a vibroacoustic point of view the loudspeaker works as follows. The driver element converts its electrical excitation into the movement of the diaphragm. This has coupling to the air outside and inside the cabinet, transmitting a wave to both parts of the system. Another vibroacoustic coupling of interest is from the interior sound field to the walls of the enclosure, making them to vibrate and, due to this vibration, to radiate external sound field in addition to the radiation of the driver diaphragm. The driver element has also a direct mechanical coupling to the front panel and through it indirectly to all other panels of the enclosure. If the walls were rigid, only the driver diaphragm vibration would be of interest. In practice, however, the vibration of enclosure walls is not negligible and should be included in detailed simulation of the system. Furthermore, the acoustic loading of cabinet interior on the driver diaphragm movement has an effect on its radiation to the external field. All these effects influence the magnitude and phase response of the loudspeaker 4 and its directivity

pattern. We have assumed, however, that the transmission of transversal waves through walls that is important in sound insulation, is not prominent in loudspeaker enclosures.

VIBROACOUSTIC

MEASUREMENT

SYSTEM
results

In order to be able to evaluate

how realistic the numeric simulation

of the loudspeaker case are, we decided to construct a system for extensive vibroacoustic measurements. It consists of an array of miniature microphones to collect acoustic responses and a combination of a laser vibrometer and vibration sensors (accelerometers) to probe the mechanical vibrations of the loudspeaker system. A computer system was programmed to collect the sound field and mechanical vibration data in the form of impulse responses 2.1 of the loudspeaker Array element excitation.

Microphone

A microphone array of 90 small electret capsules was constructed so that it/ fits to the interior of the loudspeaker cabinet, see Figs. 2 and 3. A fi'ame of metal tube was used to support row and column wires, spaced by 40 mm x 40 mm, as shown in Fig. 2. At each wire crossing an electret microphone (Hosiden 2823) and a cascaded diode was attached as depicted in Fig. 3. Digitally controlled analog multiplexers were used to select one of the column wires and one of the row wires at a time. Only a single electret capsule, activated by current through the load resistor (R), is functional at a time to capture the sound pressure field and to transduce it to the microphone preamplifier. Thus the multiplexed microphone array can be used to measure acoustic responses in the spatially distributed mesh positions, point by point, both inside and outside the box. The lower cutoff frequency of the microphone/amplifier combination was 30 Hz and the response was found flat within 1 dB in the most interesting measurement range 100 Hz - 2 kHz of our study so that only the slightly varying gains of individual capsules needed compensation. 2.2 Vibration Measurements

For vibration measurements a laser vibrometer (Polytec OFV3001) and acceleration probes were used. A mesh of 40 mm was measured, point by

point, to obtain the vibration responses of all walls. A number of points in the cone of the driver were also registered. Point mobility measurements were made by applying impact testing to the walls of the enclosure as well as isolated pieces of MDF plates corresponding to the walls of the enclosure. Following equipment were used: impact hammer with Brfiel &: Kjeer 8200 force transducer, B&:K 4393 accelerometer with two B&K 2635 charge amplifiers. HP 3565 S analyzer and STAR-software were used for modal damping determination. 2.3 Data Acquisition System and Analysis Tools en-

The measurement

system was based on the QuickSig signal processing

vironment [8], developed in the Laboratory of Acoustics and Audio Signal Processing, Helsinki University of Technology. Impulse response measurements were carried out using random phase fiat spectrum (RPFS) excitation signal of typically 8192 samples at a sampling rate of 22050 Hz, averaged typically over 10 repetitions. This is in practice equivalent to the more commonly used MLS (maximum length sequence) measurements. The frequency range of interest, from the viewpoint of element-based modeling in this study, is only up to 1-2 kHz. The signal-to-noise ratio of acoustic measurements was in all conditions better than 40 dB so that its effect to, e.g., magnitude responses is negligible. The same data acquisition system was also used in vibration measurements (except in impact testing). Further signal analysis of acoustic and vibration data was carried out in MATLAB.

MEASURED

LOUDSPEAKER

BEHAVIOR

Typical acoustic responses, as measured inside the enclosure, are shown in Fig. 4. Subplots (a) and (b) show the impulse response from driver terminals to sound pressure in one mesh point, r14c5 (row 14/column 5), fol' a sand-supported, (a) undamped vs (b) mineral wool damped, enclosure. Subplot (c) shows the magnitude responses for the sand-supported and freestanding cases without interior damping, and subplot (d) the corresponding magnitude responses for the case with 10 cm wool at the back panel. The impulse response in Fig. 4a illustrates a long ringing of interior resonances in the undamped case. The corresponding ringing is radically shorter in a damped case, see Fig. 4b. The same information is presented in 6

the frequency domain in subplots 4c and 4d. The former one shows the resonances and antiresonances in the undamped enclosure as measured in the sand supported and free-standing case, respectively. The mode frequencies exhibit strong resonances. The difference between these two curves is surprisingly small. Only minor extra effects are introduced in the free-standing case, such as seen around 200 Hz. The same is true also for the damped case of Fig. 4d except that resonances and antiresonances are effectively smoothed out. The vibration of an enclosure wall is characterized in Fig. 5 based on three different measurements. Figure 5a illustrates the accelerance (acceleration/force) of an isolated wall panel near corner, as measured by impact testing. This information can be used to estimate the parameters of MDF for vibroacoustic modeling. Figure 5b shows the corresponding behavior when the side wall (600 mm x 400 mm) is excited at point 260 mm from front panel and 297 mm from top plate. This can be compared further with the velocity of the same point as response to an electric excitation of the driver element (Fig. 5c). Only the lowest mode (200 Hz) has prominent effect to external sound field radiation.

ANALYTICAL ELING
analytical

AND

SEMIANALYTICAL

MOD-

An accurate

solution

of coupled

vibroacoustic

equations

for a

loudspeaker is out of question. Yet it is possible to try a simplified and approximate solution, especially at relatively low frequencies. In this section we will try this approach since the loudspeaker in our study has a relatively regular shape. 4.1 Analytical Modeling Techniques

The first approximation of a closed loudspeaker enclosure is obtained when the walls of the enclosure are considered to be rigid. The element is modeled as a simple piston with given velocity. The volume inside the enclosure is denoted by fL Its boundary, the walls of the enclosure, are denoted by c912-- 0_iUc0f_2, where 0f22 is the surface of the piston and cO_ 1 refers to the other wall surfaces. The spatial variable is denoted by x and the frequency f is given as angular frequency w -- 2_r/. The pressure field p(x,w) inside the enclosure as a function of frequency is given by the solution of the 7

Hehnholtz

equation

[14, Chapter

6] (1)

V2p + k2p = 0 with boundary conditions

__ _ On -- -ipowVn,

On

- 0,
--

x & cqf_l x E Of_2

(2)

Here k = _c is a wave number, c = 343 m/s is the speed of sound in air and n is the normal of the boundary pointing away from the fluid In boundary conditions _ is the directional derivative of pressure in the direction of the normal n. Ill the boundary condition for the piston area, i is the imaginary unit, po = 1.21 kg/m 3 is the density of air in equilibrium state and Vn(X, w) is the velocity of the piston in the direction of its surface's normal n. In this case the velocity vn is considered to be constant in 0fl2. This boundary value problem can be solved using Green's function Gw, which is the solution of the equation [14]

v_a_+ k_a_ = _(x- x0)


with homogenous boundary conditions OG_ On =0' xeOm

(3)
(4)

Here 5(x - xo) is the Dirac delta function and point x0 is considered as a source point. Using the eigenfunctions _N and eigenvalues kN of the Helmholtz equation (1) with boundary condition _ = 0, the Green's function call be expressed as a series [14, Chapter 9.4]

a_(_,xo) =_

_N(_)_(x0)
l_ ly x l, the eigenfunctions

For a rectangular enclosure with dimensions are of tlle form [14]

_n_._n.(x)=cos(k_x)cos(kyy)cos(kzz)

(6)

where k_ = '*/_-, ky = _-_-_ and k_ .= _f . The coefficients %, ny and n, are ? y z non-negative integers, creating triplets of numbers, that are used to index the eigenfunctions and corresponding eigenvalues k,2 n_n. (nJ_ =\t _/ 2 (ny_r_ 2 (nz_r_ 2 +k ty] +\_ ;

(7)

The solution of the boundary value problem given by Eq. 1 and Eq. 2 can be obtained by using the integral equation [14]

= f /( ,xo)a (x,xo)dxo +

pl( ,x0j

]ds0 (8)

Since Eq. 1 is homogenous, the boundary condition for the Green's function (4) is homogenous and using boundary condition for pressure (2), this equation simplifies to the form p(w, x) =/orb In order to obtain numerical -iwpoGw(x, xo)vn(w, xo)dso results from these equations, (9) the Green's func-

tion was approximated by using only a finite number of terms in the summation. The form used for the Green's function was

G (x,xo)= E

Since only a finite number of terms was used in the summation, the integration in Eq. 9 could be taken inside the summations. The coordinate system was chosen so that the enclosure was in the positive octant of the coordinate space and that the corner of the enclosure was in origin. The element, or the piston, was placed on the wall that lies on xy-plane. The dimensions of the studied enclosure were as show in Fig. 1. The circular piston element's centre was at point x_ = 125 mm, Ye = 125 mm and z = 0 (the 'rim' position). The radius of the element was r = 75 mm. The actual numerical computations were made using MATLAB. The integrations over the elements, considered as a flat piston surface, were computed numerically using MATLABs 'quad8' function. The transfer function between the piston velocity vn(w) and pressure at some measurement point xm can be computed by dividing computed pressure p(w, Xm) by velocity vn(w). In the case studied here, the frequency dependent part of the velocity can be taken out of the integral in the Eq. 9, so the transfer function can be obtained from this equation by setting vn(w) = 1. The chosen measurement point was 120 mm from the back plate and r12c2 in the microphone mesh. The frequency response of the

computed transfer function is represented in Fig. 6 in comparison with the corresponding measured response. The overall fit is good except the damping of resonances since in the analytical model the walls were assumed totally rigid. Other minor deviations could be reduced by adjustments of enclosure interior measures, except at higher frequencies above 700 Hz where for example the piston assumption of driver diaphragm movement is not valid anymore. (The deviation at very low frequencies is due to the high-pass characteristics of the measurement microphone.) 4.2 Weakly Coupled Heuristics

A full description of the coupled vibroacoustic problem of an loudspeaker enclosure using a Green's function expansion for both the acoustical wave and the mechanical bending waves is numerically very inefficient due to the large number of terms needed. However, some heuristic deductions enable us to achieve a very efficient approximation for the effect of coupling of the lowest modes, which typically are also the only ones which are of importance when studying the sound radiation by the loudspeaker enclosure [9, 10]. Here we take as our aim to develop a first-order approximation of the vibration, which implies the following assumptions about the mechanisms:

The starting points for the calculations are the acoustical modes in a rigid enclosure with locally reacting walls and mechanical modes of an enclosure in a vacuum. The vibrational properties of the driver (diaphragm mass, suspension compliance, losses, etc.) can be taken into account as a variation of the impedance of the surface, as included in the boundary conditions in the discussion above. The driving mechanism for the acoustical modes is a volume velocity source corresponding to the driver and for the mechanical modes a point force corresponding to the recoil of the driver. The coupling between acoustical and mechanical vibrations is assumed to be weak, so only first-order coupling (acoustical ---> mechanical, mechanical -+ acoustical) is taken into account. The mechanical vibrations of adjacent panels must be assumed to be strongly coupled. The field outside the enclosure can be ignored for the following reasons: outside sound pressure on the surfaces is at least one order of 10

magnitude smaller than the pressure inside at the eigenfrequencies, and there is less frequency and place dependence in the field. The bending wave equation applicable to consist of thin plates, is of the form O2u
69t2 =

to the enclosure

walls, assumed

EK 2 c94u
P OX 4

(11) coeffi-

where E is the modulus cient.

of elasticity

and K 2 a geometry-dependent

In a finite rectangular plate the boundary condition which is of interest when analysing the acoustically excited vibrations is the simply supported edge which can rotate, but cannot have transverse displacement; the other possible boundary condition, clamped edge, where the edge cannot have either rotation or transverse displacement has a significantly higher mechanical impedance, so neither the mechanical force nor the sound pressure excite these modes as efficiently. The simply supported boundary condition, which yields the lowest resonance frequency, corresponds to a situation where adjacent enclosure surfaces move to opposite directions, thus enabling the rotation of the edge, and there is no torque on the edge. A clamped boundary condition requires that there is something to provide the torque needed to prevent the edge from rotating, which is possible only if the adjacent surfaces moves in the same direction (Fig. 7). The modes corresponding to the clamped boundary conditions also have significantly higher frequencies than the simply supported modes, the lowest modal frequency of completely clamped plate being about twice that of a simply supported plate [11]. The situation where clamped modes were the most appropriate description would be the one where the enclosure vibration is driven by a homogeneous pressure field or by a purely onedimensional standing wave, but this situation arises only at low frequencies where there are no clamped eigenfrequencies. In the following discussion we assume the enclosure to be rectangular. For simply supported edges the modal shapes are given by equation = A sin (m + X)_x sin (_ + 1)_y L_ Ly where z is the displacement and Lx and Ly are plate dimensions
11

(12) and m and

n non-negative

integers.

The eigenfrequencies

are defined by equation

0.453 cLh [\ L_ / +, Ly } L_Ly wave where h is the thickness of plate and cz is the speed of longitudinal in the plate given by equation

CL= i p(1-E Y2)

(la)

where l_ is the Poisson coefficient. Thus, even if the modal shapes of the simply supported plate are similar to those of the acoustical modes of a rectangular cavity, the relationships between modal frequencies are fundamentally different, and the modal density at low frequencies (i.e., in the first octaves above the lowest mode) is very low as compared to the acoustical mode density. The assumptions made of the modal shape have implications on the coupling between various modes. If losses are small, thc vibration can be efficiently transmitted from one enclosure surface to another at the edge only if modal indices corresponding to the wave component along the edges are equal. Similar orthogonality is valid also for the coupling of mechanical modes corresponding to the simply supported edges and the acoustical modes. Bending modes corresponding to other edge boundary conditions need also hyperbolic functions for their description, so their inner product with the acoustical modes is non-zero also for unequal modal indices, so all the modes have some, although small, coupling. However, as stated earlier, the practical significance of these modes is small. The plane wave decomposition for the sound field inside the enclosure enables the enclosure surface vibrations to be described as a superposition of vibrations excited by plane waves. To achieve this we must determine the impedance of each mode and the force distribution caused by the incident wave. The vibration velocity can be then formally written as their quotient

[12]:
m Zm The modal impedances can be determined by writing the standing wave as a superposition of travelling waves in opposite directions, and using this decomposition to describe the modal impedance as a sum of travelling-wave impedances. The exiting force is given by the sound pressure. 12

The importance of the orthogonality of the coupling is that instead of using the full Green's function expansion for both the acoustical and mechanical modes, it is necessary to only determine the modal frequencies of the bending wave modes with the modal indices corresponding to the most significant acoustical modes. Another interesting problem, discussed here only briefly, is the effect of the coupling to the modal frequencies. An analysis of losslessly coupled simple vibrating systems [13] indicates that coupling increases the frequency difference between the two resonators. Similar results hold also for systems consisting of coupled acoustical and mechanical standing waves. The coupling strength, which determines the amount of change in the resonance frequencies, can be determined from the bandwidths and initial frequency differences of the resonances.

VIBROACOUSTIC

MODELING

TECHNIQUES

Numerical modeling is, in principle, a solution to any problem that can be formulated precisely enough. There are, however, limitations of computational resources such as finite memory, accuracy, and computation time that restrict the applicability of element-based numerical methods. Although they are becoming more and more relaxed with rapid development of computer hardware, yet they will remain one of the limiting factors. Another and in practice a very important restriction is the accuracy of available material (and structure) parameters. Acoustic properties of absorbent materials are seldom known precisely, and even less information is available about the dynamic parameters of enclosure construction materials. These materials are not very homogeneous so that the variation range of parameters should be known, not only the values from a single sample measurement. In this chapter we will present the basic principles of techniques for element-based modeling that have been applied in our study. These methods include the finite element method (FEM), the boundary element method (BEM), and the finite difference time domain method (FDTD), especially its waveguide mesh formulation.

13

5.1

Finite

Element

Method

in Acoustics

The finite element method (FEM) is a popular method for solving partial differential equations (PDEs). A PDE is transformed into an integral equation, the solntion domain f is discretized with a mesh and the solution is approximated 5.1.1 at the nodes of the mesh by means of element functions. of FEM

Derivation

In this chapter a finite element method for solving the following PDE is presented. The internal acoustic field of a loudspeaker box is modeled by using inhomogeneous Helmholtz equation (16). The MDF walls of the box are acoustically very hard and they have been modeled using an impedance boundary condition (17). Because the simulation is carried out at low frequencies, the loudspeaker element can be modeled as a simple piston by means of a velocity boundary condition (18). V2p + k2p = O, Bp -piw On - Z(w) p' x _ f x _ Of! x E 092 (16) (17) (18)

Bp
On -piwv,, Here p is fluid density, Z(w) acoustical impedance and v, normal velocky. From now on, boundary conditions (17) and (18) are combined as Bp -piw con-Z 'ppi_vv' p e Of (19)

It should be noted that in this notation both Z and v are functions of place and frequency, Z = Z(x,_v) and v = v,(x,_v). Instead of looking for an exact solution of this equation and the associated boundary conditions an approximate solution is searched for. First the so called weak formulation of Eq. 16 is derived [15]: both sides of the equation are multiplied with an arbitrary test function w _ V, where V is a suitable function space, in fact H _(fi) [17], and then the equation is integrated over _:

+
After Green's theorem [17] is applied

=0

(20)

14

The boundary condition integral equation

can now be taken into account,

which leads to an

fa(-Vp.

Vw+k2pw)d_-]oa_-_-p+icov /pice

) wdF=0

(22)

The original PDE can now be replaced by its weak formulation. An Ndimensional subspace of V, VN, is chosen and the weak formulation of PDE is projected into this subspace. This means that f_ is divided into finite elements and the geometry of the domain is described with the vertices of the elements. Then a basis function _biof subspace VN is chosen for each node xi, i = 1,..., N, such that _i(xi) = 1, i(xj) = 0, j _ i. The solution for Eq. 22 is approximated as
N

p _ Z pj(ce)_j(x)
j=l

(23)

This is the Galerkin method. Because Eq. 22 is valid for all the test functions v, it is also valid for basis functions qbj. Substituting the trial Eq. 23 and basis functions into Eq. 22, a system of linear equations to solve the unknowns pi, (j = 1,..., N) is obtained:
N j=l

Z [fo(-V(pj,j)vi+k2(pj j) ,)dig]
j=l

\zpj_j +

and after simplification

j_ =l

[f_(V_j. _i--k2(/_j_i)pjdig

"]- fo_l

Piceq_JqJiPjdPz ]

= fOa2 icevq_i dr 5.1.2 Matrix representation of FEM

(25)

Let us define the following acoustic mass, damping and stiffness matrices M = (M/j), C = (C/j), K = (Kij), source vector F = (fi) and pressure vector P = (Pi) respectively as: Kij = fa V4i.
15

Vj dig

Cij =

_al 2_ P i_j

dF

(26)

fi

faa2 v_i dF form (27) N

Eq. 25 can now be expressed in a matrix

NP + iwCP - co2MP = ipwF This is a system of linear equations which can be solved for pi, i = 1,..., on all frequencies of interest using standard linear algebra. 5.2 Boundary Element Method

The boundary element method (BEM) is another approach to solving PDEs. The PDE is transformed into an integral equation which consists of boundary integrals only. As a consequence, the three-dimensional acoustical problem is reduced to a two-dimensional one. When the problem is discretized, a system of linear equations is obtained. In addition to boundary nodes, BEM can be used to calculate the solution for Eq. 16 in an arbitrary point of region a. 5.2.1 Derivation of BEM for Acoustic Problem

In this chapter a direct boundary element method, or collocation method, for solving the Hehnholtz equation (16-18) is presented [18], [19]. Assuming that the equation has a solution in a, the following integral equation is valid for all functions p* regular enough:

fn(V2p+
Applying Green's theorem

k2p)p * da = 0

(28)

leads to

f (v"p +

k2p)p * da

_-/o k2pda+ ,
Applying Green's theorem

Opp,

c r-fo w w* dx
' da

(29)

once more results in _o..O* (30)

_ rja Vp. Vp* da = - j_oap_ n dP +/apV2p 16

Based on Eq. 29 and Eq. 30 the basic integral equation written as: fa (V2p* + k2p*)pd_ = _a _,(POP*_nn Onn pop ,\) dF

of BEM can be

(31) operator

Next p* is chosen to be the Green's function of the differential V 2 + k2. This means that p* is the solution of the PDE V2p * + k2p* = -507 )

(32)

in an infinite domain, 5 is the Dirac delta function. Because of the properties of the delta function, Eq. 31 can be used to calculate a value for p, at any point r/E _h

p(v)= -

_,(POP*_nn _nnOpp / *h

(33)

In the theory of BEM the Green's function is often called the fundamental solution. For three-dimensional Helmholtz operator the fundamental solution associated with point r/is
e-lkll_-xll

p;(x) -

4_11 v_

xl[

(34)

It can be assumed that p is almost constant in a small neighbourhood of the boundary point _, Uc(_), and the left side of Eq. 31 can be approximated:

ffi(Vp*( ) +
where

k2p*(_) )p(_) df_ _ C(_)

(35)

(_) = lim_0 Ju_(e)(V2 /" p* (_) + k2p*(_))p(_) d_

(36)

If the boundary is smooth enough, it can be proved that C(_) = -p(_). At points _ of boundary I' = 09 can Eq. 31 be formulated as: Op* l_p(_) + _P_n dI'= _ _-_PnP d* I' (37) This equation consists only of values of p and its normal derivative at the boundary I'. This equation is solved in the same way as in FEM: the boundary is approximated with surface elements and the solution is searched at nodes. Let 0_2 be divided into N disjoint parts:
N

=Zr, i=1
17

(38)

For simplicity,

the following notation

is used:

ap, Q, Op*
At every boundary point xi Eq. 37 is

(39)

12 p(xi) + j_ =l /r J pQ* dF = j_ =l fr.J Qp* dF

(40)

Here p* is the fundamental solution associated with node xi and Q* = op .[ respectively. It depends on the selected boundary elements how the On integrals over boundary parts Fi are computed. If constant elements are used, Eq. 40 becomes simpler:
N N J Q;dF= jZ =l

l_p(xi) + jZ =lp(xJ)/r Finally, the boundary condition

Q(x,)/r

-/ p;dF

(41)

(19) is substituted

into Eq. 41:

l_p(xi + )j _ p(xj) fr Q? dF = j_ --_-p(xj)- piwv N N ), =l J = l (_pi w and the system of linear equations
_p(xi)

frjp dr
N):
J P_ dP

(42)

is obtained

(i = 1,...,

+ jY_ N .p(xJ) fr J ( Q_ +-_piw pij ,\ dC= j_Ar piWV/r =l =l


representation of BEM

(43)

5.2.2

Matrix

With the notations

(i,j = 1,..., Hij -Hii

N)

piw ,\ fr i (Q* +-_-pij dP, i _ j


fPi N j=l

(Q*

+ -F piwp,) ,\ ar + 1
p'dF
at j

(44)

fi =

E-vk

Pi = p(xi) Eq. 41 can be written in matrix form HP = piwF (45)

Using the fundamental solution Pl and the values of P Eq. 16 can be solved at every inner point r/6 _: P(tl) _-' - jr N [p(xJ) /r Q: + Piw (_1 p(xj) =l 18 + v_Ofp 'r:dP]f ] (46)

5.3

Coupled

FEM/BEM

When the coupling between the structure, i.e., the loudspeaker box and the acoustic field is taken into consideration, the situation becomes more complex. FEM can be used to simulate the vibration of walls under acoustical excitation and this structm'al model can be coupled with acoustical FEM/BEM model. In practice the coupling is described with a coupling matrix T. Here a structural FEM model has been coupled with an acoustical FEM model KP + iwC P -- w w M P /fsU 22 M sU = piw -TPF + pw2TTU (47)

Ks and /VI_ are the structural stiffness and mass matrices and U is the structural displacement vector. When the structural FEM-model is coupled with an acoustical BEM model, the following system of matrix equations is obtained: HU P KsU - w2Ms 5.4 Finite difference schemes --- piw -TPF + pw2TrU (48)

Finite difference time domain (FDTD) methods are found a possible solution for acoustic problems such as room acoustics simulation [20, 21]. Here we study its applicability to loudspeaker modeling. The main principle in the finite difference methods is that derivatives are replaced by corresponding differences [22]. There are various techniques available but for the wave equation it is suitable to use the so called centerscheme, such that dp(t) _ p(t + At) - p(t - At) dt 2At For this purpose the wave equation is presented in the time domain: (49)

c2V2= P a 2P
After applying the previous discretization technique twice both for space and time, in a one-dimensional into the following form:

(50)
(shown in Eq. 49) case Eq. 50 results

p(x + Ax, t) - 2p, t) + p(x - Ax, t) cZ = Ax 2 p(x, t + At) -- 2p, t) + pC, t - at) At 2 19

(51)

where the sound pressure p is a function of both time and place. This scheme call easily be expanded also to higher dimensions. Spatial dimensions mw by separated and discretized individually. Thus in a threedimensional case, to the left-hand side of Eq. 51 similar terms are added concerning spatial differences Ay and Az. The difference scheme in Eq. 51 is explicit. In practice it means that the sound pressure values for the next time step t + At can be calculated purely from the data of time t and earlier. As the finite difference schemes are often calculated in the time domain the results can be visualized easily and the propagation of wavefronts in the space under study are clearly seen. Another advantage of time domain calculation of impulse responses is the ability to use the results directly for auralization purposes, i.e., the simulation results can be easily listened to. There are also drawbacks in the finite difference schemes. Traditionally the space discretization has been done such that resulting elements are cube shaped in a rectangular mesh. That causes both dispersion and magnitude error at higher frequencies. Due to that limitation the valid frequency range of the FDTD method is somewhat lower than in the corresponding FEM. In practice for an FDTD grid at least 10 nodes per wavelength are needed.
5.4.1 Waveguide Mesh Method

The waveguide mesh method is an FDTD scheme. Its background is in digital signal processing. The method was first developed for physical mod-eling of musical instruments [23]. The method is computationally efficien_ and with one-dimensional systems, such as flutes or strings, even real-time applications are easily possible [24, 25]. A waveguide mesh is a regular army of discrete space digital 1-D waveguides arranged along each perpendicular dimension, interconnected at their crossings as illustrated in Fig. 8 which represents a two-dimensional waveguide mesh. Two conditions must be satisfied at a lossless junction connecting 2N lines of equal impedance [26]: 1. the stun of inputs equals the sum of outputs,
/=2N /=2N

(flows add to zero),

E i=1

= E p;i=1

(52)
(con(50)

2. the signals in each crossing waveguide are equal at the junction, tinuity of impedance). Pi = Pi, Vi, j 20

where p/+ represents the incoming signal in the digital waveguide i and pi is the outgoing signal in the same waveguide. The actual value of a waveguide is the sum of its input and output. Pi = P+ + Pi (54)

Since that value is the same in all waveguides connected to the node this value is also the value of the node p. The digital waveguide between two nodes implements a unit delay, such that what goes out from a waveguide gets in to its opposite end at the next time step. P_ ('_) = P_,opposin_ (n - 1) Based on these conditions of an N-dimensional a difference equation mesh: (55)

can be derived for the nodes

rectangular

1 2N

pk( )=
where p position of k. As (p+,p-) needed. This derived Eq. 51. Az = 1

- l) -

- 2)

(56)

represents the sound pressure at a junction at time step n, k is the of the junction to be calculated and I represents all the neighbors one can see in this formulation the incoming and outgoing signals have been eliminated and only the actual value p of a node is waveguide mesh equation is equivalent to a difference equation from the wave equation by discretizing time and space as shown in The discretization is done such that At = 1 time step, Ax = Ay = grid unit and the wave propagation speed: 1 Ax

cThe real update frequency

at
mesh is:
CrealV_

(57)

of a three-dimensional

A- dx

(5s)

where c_.,at represents the speed of sound in the medium and dx is the actual unit distance Ax between two nodes. That same fi'equency is also the sampling frequency of the resulting impulse response. Boundary conditions are presented as relative impedances to the air such that impedance 1 represents the impedance of air corresponding to an opening. Another choice for setting the boundary conditions is by using
21

digital filters as presented in [27]. In Fig. 8 the boundaries are filters having a transfer function H(z). When compared to FEM models this is an advantage especially in more complex cases where non-linear or timevariant boundaries are needed. A more detailed study on deriving the mesh equations and boundary conditions is presented in [28]. In the waveguide mesh method the error caused by cube-shaped elements (as described in the previous section) can be reduced, e.g., by using tetrahedral elements [29] or some interpolation technique [30]. In this study we have used cube-shaped elements.

COMPARISON ED BEHAVIOR

OF MEASURED

AND SIMULAT-

In this section we show the results of simulating the behavior of the closed box loudspeaker using element-based numerical modeling tools. 6.1 FEM and BEM simulations

The internal acoustic field of the loudspeaker box has been simulated by using both finite and boundary element methods. All calculations have been carried out using vibroacoustic software SYSNOISE, rev. 5.3, and a 300 MHz DEC Alpha workstation. FEM models have 702 nodes and BEM models have 394 nodes. Both models use the same mesh but for BEM all internal nodes have been removed. It is often required that an element mesh should have at least six nodes per wavelength. In this case the models used are valid up to 1100 Hz. Both FEM and BEM simulations of an empty enclosure, Figs. 9 and 10, show accurate results in comparison with measured behavior of the loudspeaker, except at frequencies above 600-700 Hz. Magnitude responses in a point 120 mm from back plane and mesh position r12c2 (see Fig. 2) are shown. Although it is possible to use FEM modeling of absorbent materials in SYSNOISE, it did not yield satisfactory results and instead the damping material has been modeled using impedance boundary condition (Eq. 17). It was necessary to use frequency-dependent impedances, which were tuned by hand. Figure 11 shows that the simulation is fairly successful with the 5 cm wool case, but the stronger damping effect of the 10 cm wool (not 22

shown here) might need another approach. In addition to the backplane of the cabin, the absorbing boundary condition was used on the side walls to the height of the mineral wool. The coupling effect between structure and fluid was not completely modeled. The vibration of a side plate in an undamped enclosure due to interior sound field was simulated using SYSNOISE. Three kinds of plate support principles were tried: clamped, simply supported, and free edges. None of these yielded results accurate enough. In section 4.2 simple support was suggested but our simulations matched best to measurements (see Fig. 12) when the free-edge assumption was applied, which is not easily motivated. The damping of structural vibrations was also found inadequate. Thus the boundary conditions and the material parameters for the structural FEM model as well as mechanical coupling from the driver element need further attention. However, it seems possible to use coupled element modeling techniques with these simulations. 6.2 Waveguide mesh simulations

The simulations made in this study used a three-dimensional waveguide mesh covering the interior space of the loudspeaker. Currently this method is capable of simulating only uncoupled systems, and thus only the inside sound field of the loudspeaker was simulated. The loudspeaker was modeled as a rectangular cabinet and the element was a cylinder acting like a piston sound source. The simulations were made with a 10 mm spatial discretization resulting in ca. 65.000 mesh nodes. The simulations were carried out on an SGI Octane workstation. In Fig. 13 an example of visualized time-domain simulation is shown. The figure presents a two-dimensional slice inside the enclosure 230 mm from the back plate. The excitation has been a Gaussian pulse and the driver element was located at the rim position. In the figure the primary wavefront is approaching the bottom of the cabin and behind that the first reflections from the side walls can be seen. Using the FDTD method it is easy to visualize the temporal evolution of the sound field inside and outside the loudspeaker cabinet. Figures 14 and 15 show the results of waveguide mesh simulations of the loudspeaker interior in the same point that was used in section 6.1 for FEM and BEM. Boundary conditions were varied so that in Fig. 14 the empty loudspeaker was given relative impedance of the walls set to frequency23

independent value of 100. The results match response fairly well up to about 550 Hz.

the measured

magnitude backthickof the of the thick useful

When absorption material was added the relative impedance of the plane of the enclosure was changed to be close to 1 depending on the ness of the absorbing material. At the same time also the location backplane ;vas changed such that it was at the height of the surface mineral wool. Figure 15 shows the simulation result when 50 mm mineral wool was at the back wall. The response curve shows a match with measured response up to about 350 Hz.

DISCUSSION

AND CONCLUSIONS

The aim of this study has been to apply element-based vibroacoustic simulations to the modeling of a closed-box loudspeaker in order to test the applicability of these methods in loudspeaker design. Simulation results are validated by comparing them with measured behavior of a real speaker. Good agreement of modeled and measured as well as semianalytic results in simple configurations, such as in Figs. 6, 9, and 10, confirms general validity of the approach and the measurements. First the measurement system, especially the electret microphone array, was found very important for obtaining extensive and reliable data. This data is stored on a CD-ROM for further experiments and analysis work. The first observation is that semianalytic and simplified models may yield surprisingly accurate simulation results if the enclosure is simple enough, as shown in Fig. 6 and ref. [10]. The second finding was that all elmnent-based methods yielded accurate enough internal sound field simulations at low frequencies (below 500-600 Hz) for an undamped enclosure. The anomalies at higher frequencies may be due to inaccuracies of model parameters, the non-piston behavior of the driver element, the relatively small size of FEM/BEM-meshes (5 cm discretization), and the fact that the magnet of the driver was not included in the models. FEM and BEM based models outperformed the accuracy of the waveguide mesh (difference method) in these simulations. In principle the waveguide mesh should be as accurate up to frequencies of about 5 kHz. One problem with this method was the regular mesh with 1 cm discretization whereby points of computation (including the point of observation) were discretized to the nearest available point in the mesh. The sinmlation of a more realistic case, that of a damped enclosure 24

(Figs. 11 and 15) were not as accurate. We did not succeed to use SYSNOISE in a straightforward way of modeling the coupling of the air space and the damping wool. The wool was given as an equivalent impedance condition and its placement only on one wall required some hand4uning of the model parameters to obtain a fairly good match to measured sound field. The FEM/BEM method yielded better results than the waveguide mesh method since it had a frequency-dependent complex impedance of the damping material, while the waveguide mesh technique used a simple resistive impedance. The vibration behavior of the enclosure plates due to driver excitation was simulated by the FEM/BEM method in SYSNOISE. Some of the simulations were encouraging although the results are not accurate enough and the model only partially included the vibroacoustic couplings within the system. This problem is important since the wall vibrations at low frequencies are of interest to the overall radiation of the loudspeaker. Especially the lowest modes should be simulated accurately enough. The internal sound field is of less interest in a closed box than in a vented box structure since it can radiate only through walls or due to driver cone loading. The simulation of the latter effect requires that the acoustic impedance matching of the driver and the enclosure is known. Finally, only the external sound field is of interest to the listener. It can be solved with the element methods if the driver cone behavior as well as wall vibrations are modeled accurately enough. In this study we did not try this since the piston model of the driver diaphragm is not accurate enough and the wall vibrations are not known precisely enough. The computational efficiency of the simulations was fairly good: a typical run-time was about 15 minutes for the uncoupled FEM/BEM problems and a few minutes for waveguide methods. Although even faster simulation would be wished for fast experimentation, such a speed is certainly tolerable. As a conclusion, the simulations with element-based models in our study have shown that the vibroacoustic behavior of a relatively simple loudspeaker can be simulated precisely and efficiently enough for the sound fields. Yet some crucial behavior, especially the mechanical vibration of walls and tile detailed behavior of the driver cone are not detailed enough to be useful. Our next step is to develop these crucial parts further, to extend the modeling to vented box loudspeakers, and to compute the external sound field based on this knowledge. 25

REFERENCES
[1] Bank G., and W,'ight J., "Loudspeaker Enclosures" in Loudspeaker and Headphone Handbook, (ed. J. Borwick), Focal Press, Oxford, 1997. [2] SYSNOISE [3] I-DEAS home page: http://www, lmsintl, com.

(SDRC) home page: http://www, home page: http://www,

sdrc. corn autoa, corn corn corn corn

[4] Comet/Acoustics [5] ABAQUS

home page: http://www.hks, ansys,

[6] ANSYS home page: http://www, [7] MSC/NASTRAN [8] Karjalainen gramming, 1990.

home page: http ://www.raacsch.

M., "DSP Software Integration by Object-Oriented ProA Case Study of QuickSig," IEEE ASSP Magazine, April

[9] Backman J., "Effect of Panel Damping on Loudspeaker Enclosure Vibration,'' AES 101st Convention, Los Angeles, Nov. 8 - 11, 1996, preprint n:o 4395 (C-5). [10] Backman J., "Computing the Mechanical and Acoustical Resonances in a Loudspeaker Enclosure," AES 102nd Convention, Munich, March 22 - 25, 1997, preprint n:o 4471 (J6). [11] Blevins R. D., Formulas for Natural Frequency and Mode Shape, Van Nostrand Reinhold Company, New York, 1979, pp. 258 & 261. [12] Fahy F., Sound and Structural 1985, pp. 221 - 226. Vibration, Academic Press, London

[13] Richardson E. G., Technical Aspects of Sound, vol. I, Elsevier Publishing Company, Amsterdam, 1953, pp. 3 - 4. [14] Morse P. M., and Ingard K. U., Theoretical acoustics, Princeton versity Press, USA, 1965. Uni-

[15] Zienkiewicz O. C., and Taylor R_ L., The Finite Element Method Vol. i: Basic formulation and linear problems, 4. ed. McGraw-Hill, London, 1989.

26

[16] Zienkiewicz O. C., and Taylor R. L., The Finite Element Method Vol. 2: Solid and fluid mechanics, dynamics and non-linearity, 4. ed. McGrawHill, London, 1991. [17] Schatz A. H., Thom_e V., and Wendland W. L., Mathematical Theory Bd. 15),

of Finite and Boundary Element Methods, (DMV Seminar; Birkh/iuser, Basel, 1990.

[18] Brebbia C. A., Zelles J. C. F., and Wrobel L. C., Boundary Element Techniques: Theory and Applications in Engineering, Springer Verlag, Berlin, 1984. [19] Brebbia C. A., and Ciskowsld, R. D. (ed.) Boundary Element Methods in Acoustics, Computational Mechanics Publications, Southampton, 1991. [20] Botteldooren D., "Finite-difference time-domain simulation of lowfrequency room acoustic problems," The Journal of the Acoustical Society of America, vol. 98, no. 6, pp. 3302-3308, 1995. [21] Savioja L., Backman J., J/irvinen A., and Takala T., "Waveguide mesh method for low-frequency simulation of room acoustics," in Proc. 15th Int. Congr. Acoust., Trondheim, Norway, June 1995, vol. 2, pp. 637640. [22] Strikwerda J., Finite Difference Schemes and Partial Differential Equations, Wadsworth&Brooks, Pacific Grove, CA, 1989. [23] Smith J. O., "Physical modeling using digital waveguides," Music J., vol. 16, no. 4, pp. 74-87, 1992 Winter. Computer

[24] Viilim/iki V., Huopaniemi J., Karjalainen M., and Janosy Z., "Physical modeling of plucked string instruments with application to real-time sound synthesis," Journal of the Audio Engineering Society, vol. 44, no. 5, pp. 331-353, 1996 May. [25] Jaffe D., and Smith J. O., "Extensions of the Karplus-Strong plucked string algorithm," Computer Music J., vol. 7, no. 2, pp. 56-69, 1983 Summer, Reprinted in C. Roads (ed.), The Music Machine (MIT Press, Cambridge, MA, USA, 1989), pp. 481-49.

27

[26] Van Duyne S., and Smith J. O., "Physical modeling with the 2-d digital waveguide mesh," in Proc. 1993 Int. Computer Music Conf., Tokyo, Japan, Sept. 1993, pp. 40-47. [27] Huopaniemi J., Savioja L., and Karjalainen M., "Modeling of reflec-

tions and air absorption in acoustical spaces -- a digital filter design approach," in Proceedings of IEEE 1997 Workshop on Applications of Signal Processing to Audio and Acoustics_ Mohonk, New Paltz_ New York, Oct. 19-22 1997. [28] Savioja L., Karjalainen M., and Takala T., "DSP formulation of a finite difference method for room acoustics simulation," in Proc. 1996 IEEE Nordic Signal Processing Symp., Espoo, Finland, Sept. 1996, pp. 455-458. [29] Van Duyne S., and Smith J. O., "The tetrahedral digital waveguide mesh," in Proc. 1995 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, New Paltz, NY, USA, Oct. 1995. [30] Savioja L., and V_limSki V., "Improved discrete-time modeling of multi-dimensional wave propagation using the interpolated digital waveguide mesh," in Proceedings of the International Conference on Acoustics, Speech and Signal Processing, Mfinchen, Germany, April 19-24 1997, vol. 1, pp. 459-462.

28

The element is centralized on the front plate. In the case of rim speaker, the dimensions The material of the plates is 20mm MDF. are according
I

[ [ _

_ 120,/ 1_4t

--' _.,]

to next picture.

\_ {

x 5/ _

speaker 110

speaker (rim case)

The inner dimensions

of the cabinet are 600mm, 400mm and 250 mm.

Figure 1: Mounting of the loudspeaker (asymmetric position) and the closed box loudspeaker with microphone array and damping wool inside.

29

Microphone o -_ J _ _

grid: .o -_ -" 1 :2 _ r0c0

_,_ distance between microphonesis 40mm

_,

'4
r14c5 metal frame _ single microphone

Figure

2: The structure

and dimensions

of the electret

microphone

array.

Electret microphone

capsules r ....... 2 + Bias

_-

_}-

_-;_'_

Preamplifier

....
: : o "_ \ oo

__

Digitallycontrolled analog switches

Figure 3: Principle of the electret the amplifier circuit.

microphone

array,

the multiplexer,

and

30

0.05 ,kl,

a) undamped

0.05

b) damped

0 -0,05 0 dB

.........................

0 -0.05

_'

0.1 I I

0,2 _l /_\

0,3 sec I A I A A

0 I

0.1 I

0.2 I

0.3 sec I

C) undamped

20 __a -20
I

nd_k_

/__/_'__

/__

100 dB i

200 i sand

300 J

400 i

500 i

600 i

700 i

800 i

900 f/Hz i

d)damped

0 -20 20-_
I

fr e_
I I I I I I I I

100 Figure

200

300

400

500 inside

600

700

800 (a) and

900 f/Hz (b) impulse

4: Examples

of responses

the enclosure:

responses from the electric excitation of the driver element to sound pressure in microphone mesh point r14c5 (see Fig. 2), 12 cm from back plane for (a) empty enclosure in sand; (c) magnitude ure in sand for damped in sand and (b) damped enclosure with 10 cm wool, response in the same position for undamped enclos(d) magnitude response and free standing. in the same position

and free standing; enclosure in sand

31

200
I

400

600
I

800
I

1000
I

1200

........................

z .........................

-20 0 40 200 400 ,

_ 600 I

i 800

L 1000 1200

-20 0 Figure 200 5: Results 400 of vibration 600 measurements; 800 top: 1000 accelerance 1200 (accelera-

tion/force) of an isolated side plate with impact test at a corner of the plate; middle: accelerance of the side plate of the freely supported undamped enclosure at point 260 mm from front panel and 297 mm point, from top plate (driver side); bottom: to electrical excitation side plate veloc!ty at the same of the driver element. as a response

32

'

'

I'

'

'1

'

,l

!
II

'
I

!
II

' i
I i

ii

I i! ,I i!
II

i i
0 1O0 200 300 400 500 600 700 800 900 Hz

Figure 6: Analytic modeling: magnitude response (diaphragm velocity to sound pressure) in point 120 mm from back plate and r12c2 in the microphone mesh in sand-supported undampedenclosure as measured (solid line) and computed from analytic modeling with rigid walls (dash-dot line).

Figure 7: Lowest vibrational mode corresponding to a clamped (left) and a simply supported (right) boundary condition.

33

H(z_

! t

_.r'--"_ _

, t t

Figure 8: A two-dimensional waveguide mesh, which consists of onedimensional digital waveguides interconnected at their crossings. At boundaries there are filters H(z) implementing the reflection characteristics of each surface.

34

dB

.,o
-20 -30 -40..............

'A'I..._..I . . ,... . .......... . . . ,_,^ . ....... . . _..,. . . . :.l:...r_/ ,.^....... .... [A.._
........ , _!1.:_ .,/il
............ -....

'1" "!! l.
: ' .!! i........ i :............. (!_{' ! ......
600 760 800

_1:i..\ .i,,
r : .......

' i...........

._o ....................... i....... i..................... 'ii'_/': ,


'6(_(30 200' 360 4()0 500 9()0 HZ

Figure 9: FEM modeling: magnitude response (diaphragm velocity to sound pressure) in point 120 mm from back plate and r12c2 in the microphone mesh as measured in sand-supported undamped enclosure as computed from FEM modeling (solid line) and measured (dash-dot line) response.
dB o ' ;...._:... _! _ _ . . i.........
-" :

, ; ...............
....... ',

'_:

/._ - , ! ii) ' _' _ i .'


,_ - -,I-.; .

::
: ,I........ :r

! .......

i.......

:
: i '\' I ......

: ,

-20

. _..

. ..,. ..........

.i ..

-_o
-40 :

:
:

!\
...... i i

! !
j....

.-_o......;............................. i...... i ...I.i ................


-6

_B 0

260

3()0

460

500

'

600

'

700

'

800

'

900

'

Hz

Figure sound phone puted sponse.

10: BEM modeling: magnitude response (diaphragm velocity to pressure) in point 120 mm from back plate and r12c2 in the micromesh as measured in sand-supported undamped enclosure as comfrom BEM modeling (solid line) and measured (dash-dot line) re-

35

dB

-10 ......... .......... : /

/ i ' -20 ,. '.'_'"' i ..... ! ........ ! -30 ..... : .....

'.il ..... '......... : ....... i ............... !i _ ! : i ' ..... i' ...... i........ :........ i........ ii i ! ! .... i_ '"!....... i........ ' ........ i" i i i ". ' )......... i........... i........... ! _.' ' i /.^ :....._, : : : . ........ :._.. :... _...... 400
i / : ......... : ..... /::: .........

'......

......... ! i ....... i '. ....... i :......... ..:


:........

-40 ........................... ....... 1O0 _..... 2[)0 i....... 300


i

:..... 500
i

i........ 600
i

i........ 700
i

i...... 800
i

i........ 900
i

Hz

Figure 11: FEM modeling: magnitude response (diaphragm velocity to sound pressure) in point 120 mm from back plate and r12c2 in the microphone mesh as measured in damped enclosure with 50 mm mineral wool as computed response. fi'om FEM modeling (solid line) and measured (dash-dot line)

_'

:. ..... ':"'

i ....i'i ....... .

..... ..............

.... i/ :'i:"
"'1 I' ' : I :i '

....
'

-30Ii.I

...........

:'_"

'501

250

3()0

_:'_!....t.!?, './! _
400

5()0

6()0

7()0

8()0

950

HZ

Figure 12: Vibration modeling: magnitude response, diaphragm velocity to side wall velocity, at point 260 mm from front plate and 297 mm fi'om top plate (driver side) in undamped enclosure: computed (solid line) and measured (dash-dot line) response. 36

.'!"...... . . . ,'. ..; . .: ... .,':

.... '.., : '"....:. .. ,........ .....

"i,,'
. .,..' ...,.".."

i
"' ' '..', '.'.'..'

: ..........
.''......

/ii .' ;.. : i :!....


' '" "'"

....... _".,...... ?:
'""''".i

! ? i i: ..... 1 ' ._' !.' .......

..:'.

_ ............ ......... /_o ." .: .........

-l_J,'

oq...i' I ..": ......

..... :,, :7_?_:_!4 .'...... ;::':

i.",'_?_ ,.';;,!_f_,_

,..'

..' ." ......... :.......... ..' /- 0.2


:,'.... ,., ._'/0.3
'.

'::....... ,. .."
.? '..:?'.....,.., ..'"

-0,1

'

" 0,2

'.:"'.

,..../n,5

0.3

Figure 13: Finite difference time domain simulation of a wavefront inside the enclosure. A two-dimensional slice at the height of 230 mm from back plate is visualized. The primary wavefront is approaching the bottom of the enclosure and behind that the first reflections from the sidewalls can be seen.

37

dE

.....

.,o ......_ .... l..i X, ........ f ..::_.._:, ..: ........... ..., , :........ _ ... .l_,..v,.,..,.[ ..... ^... ' / ' ! \,_l,,: :,' : I/ _: ':I II ._o . . . ._. .-. x : .......... , i . . ._ _ :::_ . . ...... .:, ., ....... 7 .i_ .. > I.. ,. .,\ ._^ .
-4o ....... :......... i W!.. ......... : : ...... ! ........... i.(...,' ......... -:..._ .......
: '6hi 0''_0 200 i i 30 i0 . 40 i0 5()0 600 700 O00 O00 HZ

-so......... i........ ; ......... : ................................. v.. ................ :

Figure 14: Waveguide mesh modeling: magnitude response (diaphragm velocity to sound pressure) in sand-supported undamped enclosure in point 120 mm from back plate and r12c2 in the microphone mesh as computed from modeling (solid line) and measured (dash-dot line) response.
dB
0 ......... ; /%

!
; ................ : ; /'1 -...............................................

-lo ..... : / ........


>.':

/ .......
/!

!....... _ .......... ;.......... ;.........


i i !

-2o..... :...... !",' '.'" .......... :: :__< '_ i ,: :_'!_ ",_- _


_30

i ......... :........ : :: ....

:1

-40

..... !..... !..

; ...................... ;.................... i ...........

'5_

O0

200 i

300 i

400

500 i

600

700

800

900

Hz

Figure 15: Waveguide mesh modeling: magnitude response (diaphragm velocity to sound pressure) in point 120 mm from back plate and r12c2 in the microphone mesh in damped enclosure with 50 mm mineral wool as computed from modeling (solid line) and measured
38

(dash-dot

line) response.

Potrebbero piacerti anche