System Level Simulation of Acoustic Devices

A. Wilde, P. Schneider
Fraunhofer Institut fr Integrierte Schaltungen, Auenstelle EAS, Dresden, Germany

ABSTRACT
The Transmission Line Matrix method (TLM) provides an easy to use and fast scheme to calculate acoustic fields. Thus
it is possible to couple this method to system simulation tools in order to investigate the behavior of acoustic systems and
the interaction between acoustic domain and control electronics at an early design stage. These investigations are very
important in MEMS design, especially for ultrasonic devices, transducers and SAW filters. As a proof of concept an active
noise control system was designed and the performance was simulated in a 2D acoustic environment. The results clearly
show the feasibility of acoustic field calculations along with system level simulations of electronic devices and transducers. Using this scheme acoustic devices may be simulated and optimized under realistic acoustic conditions.
Keywords: Acoustics, system level simulation, Transmission Line Matrix (TLM), Simulator Coupling, Saber, ultrasonic
application transducers

1. INTRODUCTION
The ongoing miniaturization of micro-machined acoustic systems on the basis of highly sophisticated production processes results in extremely high prototyping costs at the design stage of a system. Therefore a system level simulation is necessary to optimize and verify the system performance. For electronic and several mechanical components tools like
SPICE, Saber or VHDL-AMS simulators provide efficient and easy-to-use means for such system level simulations. However, in many cases important effects occur in the acoustic domain which can not be described by simple equations. For
system level simulations of acoustic devices a method for the rapid solution of the acoustic domain problems is needed
which may be coupled to appropriate simulation tools easily. For wave propagation problems the Transmission Line Matrix (TLM) method is established as a simple and very efficient algorithm for the solution of 2 and 3 dimensional problems.
For that reason it is well suited for the coupling with system level simulators.
The theoretical foundations of the TLM method date back to the 40s when Whinnery et al. and Kron developed equivalent
circuit models made up of discrete elements for Maxwells equation to study wave guide models using network analyzers13. With the advent of more powerful computers the TLM-method was made popular by Johns and Beurle4 in 70s. Since
then the method has been applied successfully to many problems which are described by wave or diffusion equations. The
main difference between the TLM approach and standard schemes such as Finite Difference or Finite Element methods
is as follows: Usually a physical problem is described precisely by a partial differential equation which subsequently is
discretized. The solution to this equation is then approximated numerically. The TLM approach is to approximate the
physical process (e.g. wave propagation) by discrete physical models, which can then be calculated exactly. This approach
has several advantages. The physical models may be programmed intuitively, all discretization approximations are apparent, and the calculation may be parallelized easily by domain decomposition. Furthermore the coupling to system simulation tools is well possible as the field is discretized in lumped elements which nicely fits into the philosophy of the
simulation of discrete elements used in system simulators. The disadvantages are that TLM is best suited to equidistant
nodes and fixed time steps, which are coupled to the spatial distance of the nodes. However, in recent years TLM was
adapted for non-equidistant and curvilinear grids.
In this study the TLM scheme is coupled with the system simulation tool SABER. As a very descriptive example an active
noise control system is simulated, which consists of microphones and a loudspeaker situated in an air filled pipe. Sound
generated by an artificial source enters the pipe from the left side. The aim is to reduce the sound pressure level at the exit.
Here little effort was made to construct an efficient active noise control system, instead the performance of a given system

was to be evaluated.
The next section of this paper reviews the basic theory of the TLM method. Section 3 deals with the coupling of SABER
with TLM while section 4 describes the setup of the active noise control system. The Results section discusses the calculations done for the example system.

2. THE TLM SCHEME

The TLM scheme may be derived in two different ways. One way is to derive it as a special and simplified case of the
more general Lattice Boltzmann schemes for the solution of the Navier-Stokes equations as was demonstrated by Chopard
et al.5. Though this approach opens the way to inclusion of many nonlinear effects of sound propagation it is not very intuitive. Therefore another derivation shall be presented here which relies on a physical model of the field effects and was
proposed by Kagawa for the acoustic field6,7.
The basic idea is to follow Huygens principle that states, that each point on a wave front can be considered as a source of
secondary wavelets, so that the propagation can be considered as a continuous process of scattering of elementary component waves. The superposition of all scattered (elementary) waves makes up the wave. This mechanism may be discretized as follows: The fluid is replaced by a network of pipes with equal diameters which are joint at fixed distances x
(see fig. 1). The junctions are called nodes. Within the pipes acoustic pressure pulses may travel undisturbed at the speed
of sound, i.e. no damping or boundary effects are considered. The diameter of the pipes is assumed to be much smaller
than the shortest wavelength of interest so higher order pipe modes are neglected. From this follows, that if a set of acoustic pressure pulses is released at the same time at arbitrary nodes, these pulses travel through the pipes and reach their
neighboring nodes after a time interval of t = x c , with c 0 being the speed of sound in the pipe. For 2 dimensions, the
junction may be seen as a meeting of the pipe carrying the initial, incoming pressure pulse with three outgoing pipes. Thus
the acoustic pressure pulse arriving at the node sees an apparent enlargement of the cross section by a factor of three, which
means a corresponding change of the impedance. At this impedance mismatch the incoming pressure pulse is reflected
with half of the initial amplitude and a reversal of sign, while it is transmitted into the other three pipes with half the initial
amplitude and no change of sign (see figure 1). From this point on the process of traveling in the pipes and scattering at
the nodes is repeated, with four pulses arriving simultaneously at each node and each scattering into the other three pipes.
As the process of scattering and traveling is linear, superposition applies and the net result is the sum of the
component effects. The algorithm described above can formally written as follows:
f i ( x + x i, t + t ) = S ij f ( x, t )

(1)

with f i ( x + x i, t + t ) representing the amplitude of the pressure pulse that travels from the node at position x to the node
at position x + x i between times t and t + t . x i is the vector pointing to the nearest neighbor in the i-th direction, and
S ij is the scattering matrix:

S ij

1
= 1
1
1

1
1
1
1

1
1
1
1

1
1
1
1

Because of the simple structure of the scattering matrix eq. (1) may also be written as
1
f i ( x + x i, t + t ) = --- f j ( x, t ) f i ( x, t )
2 j

(2)

where i is defined such that xi = x i . The acoustic pressure p ak and the velocity v ak are calculated from f i ( x, t ) applying
the following formulae:

D xi

afterafterscattering
scattering

p/2
p/2

before
scattering
before
scattering

pp

-p/2
-p/2

p/2
p/2

p/2
p/2

Figure 1:

TLM: The acoustic fluid is replaced by a network of pipes with equal diameters that are joined at constant distances. A
sound wave incident at a node is scattered by the junction with sound pressure distributed as given in the figure.

1
p ak ( x, t ) = --- f i ( x, t )
2 i
x
1
x
t
i
v x, y x --------, t ----- = --------2 ------ f i ( x x i, t ) f i ( x, t )

2
2
c t

(3)
(4)

if the x i are parallel to the axes of the co-ordinate system. This scheme is numerically equivalent to a second order Finite
Difference scheme for the wave equation, which is shown as follows: The acoustic pressure at a position x and time t + t
is given as
1
p ak ( x, t + t ) = --- f i ( x, t + t )
2 i

(5)

1
p ak ( x, t + t ) = --- p ak ( x, t + t ) p ak ( x, t t ) + f i ( x, t t )
2 i

(6)

Repeated use of eq. (3) yields

This may be rearranged to give

1
p ak ( x, t + t ) 2 p ak ( x, t ) + p ak ( x, t t ) = --- p ak ( x + x i, t ) 2 p ak ( x, t )
2 i

(7)

which equivalent to the corresponding Finite Difference equation if

x
------ =
t

2c

(8)

This last result means that the speed of wave propagation c is given by the ratio of the spatial and temporal resolution.
Note that the overall wave speed c is lower by the factor 1 ( 2 ) than the speed of component pressure pulses traveling
in the pipes, which is a feature of the method that is surprising when first encountered.

3. COUPLING SABER AND TLM

From a theoretical point of view the TLM scheme is well suited for the combination with a system simulator, because an

p(0,t)

Pl(t)
Figure 2:

Z0

Z0

p(d,t)

Pr(t)

Equivalent circuit model for the lossless transmission line, which corresponds to the pipes in the TLM mesh.

equivalent circuit model exists, which makes the formulation of coupling equations quite easy. The model for the lossless
transmission line following Branin8 is
Z 0 u ( 0, t ) = P r ( t + ) p ( 0, t )

(9)

Z 0 u ( d, t ) = P l ( t ) + p ( d, t + )

(10)

where u ( x, t ) is the velocity of the fluid, Z 0 is the acoustic impedance and p ( x, t ) is the pressure at the corresponding
ends of the transmission line at time t . P l ( t ) and P r ( t ) are pressure sources as shown in fig.2, and correspond to the amplitudes of the pressure pulses given by f in the preceding section. The pressures generated by the sources are updated at
each time step using
P r ( t + ) = 2 p ( 0, t ) P l ( t )

(11)

P l ( t + ) = 2 p ( d, t ) P r ( t )

(12)

Using this equations one is free to choose either the pressure or the velocity as input while the TLM model calculates the
complementary values as output. Here, the velocity of the membrane of the loudspeaker was used as input. The resulting
force on the membrane of the speaker was calculated as the integral of pressure on the membrane area, which in turn was
computed by
p ( x, t ) = v ( x, t )Z ak + f ( x + x, t 1 )

(13)

For the coupling of the system simulator SABER to the TLM model the C interface of the behavioral description language
MAST was used. The TLM code was embedded into a MAST model and is called as a foreign routine at equidistant time
steps. The step width is equal to the internal TLM time step. To ensure the equidistant call of the TLM model an internal
clock signal was created using the MAST statements for event driven models.

4. ACTIVE NOISE CONTROL TEST SETUP

To evaluate the coupled simulation a well known problem from acoustics - active noise control - was chosen. The test
setup consists of a pipe that might the exhaust of some industrial machinery where sound enters through the left end. The
aim is to reduce the amplitude of the sound that leaves the pipe at its right end by generating a complementary sound signal
in the middle of the pipe with a suitable loudspeaker. Clearly the performance of this active noise control system is strongly dependent on the geometrical properties of the acoustic environment (dimensions of the pipe and the loudspeaker) as
well as mechanical and electrical properties of the transducers and the design of the controlling system that generates the

Table 1: Physical parameters of the test setup and the acoustic calculation
Length of pipe

2m

diameter

0.2 m

loudspeaker membrane diameter

0.2 m

distance of microphones from pipe ends

0.2 m

speed of sound

340 m/s

spatial resolution

2 mm

time step

4.2x10-6 s

fluid density

1.21 kg/m3

number of TLM nodes

1000x100

signal applied to the loudspeaker. In this simulation only the influences of the acoustic environment are taken into account.
Fig. 3 depicts the basic setup. The physical sizes and parameter of the acoustic calculation are given in table 1.
For active noise control the pressure signal from microphone 1 is delayed and lowpass filtered. Then it is amplified and
applied to the speaker, where the velocity of the membrane is taken to be proportional to the electrical signal (ideal loudspeaker). The electrical signal of the loudspeaker is delayed and subtracted from the original microphone signal. This corrected microphone signal is used as input for the forward delay block (see. Fig. 3). For the given setup delaying and
filtering was implemented using control blocks from SABER. Then next step in design would be to use electrical circuitry
for the active noise control.

5. RESULTS
The special properties of wave propagation in TLM meshes are well described in literature5,7,9 and therefore were not subject to tests in this study.
To evaluate the method several calculations were performed, where a sound wave was fed into the left end of pipe. The
pressure signal calculated at the microphone position was passed to Saber. Inside Saber the controlling system was simulated, which resulted in a calculated velocity of the loudspeaker membrane. This velocity was given to the TLM subroutine
microphone 1
(sound detection)

microphone 2
(reference)

sound
excitation

delay

+
_

lowpass

amplifier

loudspeaker
(active noise control)

t
delay

t
Figure 3:

The active noise control test setup. It consists of an acoustic domain and some electronic control systems.
In the acoustic domain a pipe is simulated with a loudspeaker in the middle and microphones at each end.

which controlled the boundary conditions at the position of the loudspeaker membrane accordingly. The signals from the
two microphones and the loudspeaker velocity where used to monitor the system performance. Additionally, inside the
TLM subroutine snapshots of the pressure field in the pipe were taken every ten time steps, which where used to generate
an animation of the wave propagation in the pipe. No efforts were made to apply more realistic models of the transducers
to account for frequency response, membrane weights etc. in order to keep the example simple. However, for further design steps more realistic models of the transducers could be applied easily.
For this simple test case only the time delays and the amplification of the microphone signal had to be adjusted. Furthermore, to prevent oscillations the parameters of the lowpass had to be set up.
In order to get sufficient input information a gaussian shaped pressure pulse was released at the left end of the pipe and
the system evolution was calculated for the following 0.02 s. The system performance was monitored by comparison of
the power spectral densities of the input pressure pulse to that of the signal from the right microphone (Fig. 4).
Below frequencies of 4kHz the active noise control provides a decrease of sound pressure on the right end of the pipe.
Significant damping values can be observed below 1 kHz. Fig. 5 shows the performance of the system with a continuos
sound excitation. The sound input is a mixture of sine waves of different frequencies in a frequency range from 10 Hz to
500 Hz.
For this test signal a significant reduction of sound pressure at the right end of the pipe can be observed. This reduction is
also to be seen in the pressure distribution in the pipe. In Fig. 6 a snapshot of the two dimensional pressure distribution
for a sinusoidal excitation f=300 Hz is depicted.

Figure 4:

Power spectral density of gaussian shaped input impulse and system response at microphone 2.

Figure 5:

Sound pressure signal at left end of the pipe (input) and at microphone 2 (right end of the pipe)

area of sound wave superposition

(sound amplification)

Figure 6:

area of sound wave superposition

(sound cancellation)

Snapshot of the sound pressure distribution in the pipe

Due to the superposition of sound excitation and sound generated by the speaker in the left area of the pipe sound amplitudes increase. In the right area a sound cancellation occurs which is the main goal of active noise control.

6. CONCLUSIONS AND FUTURE WORK

The goal was to establish a computational framework that allows to simulate complex devices with acoustic subsystems
and electronics. This was achieved by coupling the system simulation tools SABER with the TLM method, which was
implemented as subroutine in C and linked to SABER. The main application fields of this coupling with regard to MEMS
will be the design of:
micromechanical sensor and actuators for acoustic and especially ultrasonic applications,
controllers for complex systems which include acoustic subsystems
For demonstration of the approach system simulations of an active noise control system in a pipe were carried out. These
simulations served to support the design of the structure of the active noise control unit, to define the parameters of the
blocks and to verify the performance of the system. Although most of the computation time was consumed by TLM, due
to its high efficiency system simulation were possible in a reasonable time (about 400s computing time for 100ms real
time on a SUN Blade1000, 750Mhz CPU, 1GB RAM).
Future applications of the coupled tools is the design of micromechanical transducer arrays, where complex problems of
parasitic surface waves shall be studied in 3D. These arrays are used in medical imaging and have two main modes of
single transducer
and control electronics

mode select

impulse
generator

closed loop
control

measurement

Transducer electronics

Figure 7:

Sketch of the micromechanical ultrasound transducer array with a snapshot of the pressure flied. Above the transducer array
sound waves spread into the fluid while at the same time surface waves travel along the array. The surface waves exhibit
lower propagation speeds than sound waves and dispersion.

operation: Sending an ultrasonic signal and receiving a back scattered signal from which the image is computed. A serious
problem is the disturbance during the receiving phase by surface waves that are generated while sending the ultrasonic
pulse. First 2D test calculations clearly show the ability of the system to reproduce the system behavior with respect to
surface waves10, as shown in Fig.7. The aim is to develop mechanical means or closed loop controllers (see fig.7) that
suppress surface waves.

ACKNOWLEDGMENTS
The presented work was done in projects funded by the German Government: EKOSAS - Entwurf komplexer SensorAktor-Systeme (Ref.-No.: 16SV1161/7) and the German DFG-Sonderforschungsbereich 358 Automated System
Design at the Dresden University of Technology.
.

REFERENCES
1. J. Whinnery and S. Ramo, A new approach to the solution of high-frequency problems, Proc IRE 32, pp. 284-288,
1944.
2. J. Whinnery and S. Ramo, Network analyser studies of electromagnetic cavity resonators, Proc IRE 32, pp. 360-367,
1944.
3. G. Kron, Equivalent circuit of the field equations of Maxwell," Proc IRE 32, pp. 289-299, 1944
4. P. Johns and R. Beurle, Numerical solution of two-dimensional scattering problems using transmission-line matrix,"
Proc. IEE 118, pp. 1203-1208, 1971.
5. B. Chopard and M. Droz, Cellular Automata Modeling of Physical Systems, Cambridge University Press, 1998.
6. Y. Kagawa and T. Yamabuchi, Finite-element equivalent circuits for acoustic field," Journal of the Acoustical Society
of America 64, pp. 1196-1200, 1978.
7. Y. Kagawa, T. Tsuchiya, B. Fujii, and K. Fuchioka, Discrete Huygens model approach to sound wave propagation,"
Journal of Sound and Vibration 218(3), pp. 419-444, 1998.
8. F. Branin, Transient Analysis of Lossless Transmission Lines," Proceedings of the IEEE 55(11), pp. 2012-2013, 1967
9. C. Christopoulos, The Transmission-Line Modeling Method, IEEE Press / Oxford University Press, 1995
10. A. Wilde, P.-C. Eccardt, and W. OConnor, Modeling acoustic transducer surface waves by Transmission Line Matrix method, Proceedings of the 19th CADFEM Users Meeting, Potsdam Oct. 17.-19., 2001