A new MIMO sine testing technique for

accelerated, high quality FRF measurements

On todays vehicles it becomes ever more challenging to measure (vibro-)acoustic frequency
response functions (FRF) for noise path contribution or experimental modal analysis. Vehicles
have more and more damping material, leading to reduced noise and vibration levels, also during
modal testing or (reciprocal) FRF measurement. The widely adopted (burst) random excitation
techniques are often unable to excite the structure at suffciently high levels resulting in signal-
to-noise ratio problems, long measurement times (many averages) and noisy FRFs. High quality
FRFs can be measured using stepped sine excitation techniques, that are able to concentrate
the excitation energy at a single frequency and excite the structure at much higher energy levels.
However, the current implementation of these techniques is rather slow with measurement times
that are a multiple of (burst) random techniques and as such are not often used. Therefore, a new
MIMO (multiple input multiple output) sine testing technique has been developed that allows to
have the high excitation levels (and resulting high quality FRF data) of a stepped sine technique,
but at drastically reduced measurement times. This new technique is compared with stepped sine
and burst random techniques with relation to data quality and measurement time using industrial
application cases.
1 Introduction
Experimental Modal Analysis (EMA) is today one of the key
technologies in structural dynamics analysis. Based on the
academic fundaments of system identifcation, it has evolved
to become a standard approach in mechanical product
development. While in the past isolated structures with low
damping were tested and analyzed, modal analysis is nowadays
also explored on complex structures with high damping such
as trimmed car bodies or complete vehicles. Applications
have expanded from structural-only modal analysis into
vibro-acoustic modal analysis, combining vibration and noise
measurements. In order to get correct modal models, accurate
frequency response function (FRF) measurements are required.
Noise path contribution or transfer path analysis [1][2] is
another key technology in todays product development where
high quality FRF functions are required. Transfer path analysis
typically requires FRFs from many different input locations,
but only a few response locations, typically microphones at
driver and passengers ears. Therefore these FRF functions are
often measured in a reciprocal way, by exciting the vehicle
at the driver and passengers ears with loudspeakers and
measuring the response of the vehicle using accelerometers
and microphones.
Today burst random excitation is probably the most popular
excitation technique to measure these FRF functions, as the
technique is fast and able to minimize leakage quite well [3].
Due to the continuous improvements in noise and vibration
levels of todays vehicles, it becomes ever more challenging to
perform accurate FRF measurements. Increased use of damping
and acoustic isolation materials greatly reduces the noise and
vibration levels, not only while driving the vehicle, but also
during testing. In combination with the low energy density of
the burst random excitation signals the total excitation energy
is distributed over the complete frequency band this means
that signal-to-noise ratios of the response channels become
very low and FRF accuracy is compromised. In many cases
one try to compensate this by measuring several hundreds of
averages, drastically increasing measurement time, and often
with little improvement in FRF quality.
Much higher excitation levels can be obtained by using
sinusoidal excitation, the most common being the discrete
stepped sine technique [3][4], where the structure is excited by
a fxed frequency sine, and response is measured after transient
effects have died out. Then the frequency is changed to the
one of the next frequency line and the process repeated. These
techniques concentrate all the energy at a single frequency,
with higher signal-to-noise ratios and high-quality FRF functions
as a consequence, but are extremely slow. Further they are able
to accurately control the excitation amplitude, and are as such
excellently suited to analyze non-linearities.
In the next section a new MIMO sine testing technique is
introduced, that keeps all the advantages of a sine excitation,
but drastically reduces measurement time. Recently, this
excitation signal received considerable attention to speed up
Ground Vibration Tests (GVT) of large aircraft [5]. In sections 3
and 4 the technique is illustrated on a midsize passenger car
for 2 different applications: reciprocal FRF measurements and
modal testing.
2 A new MIMO swept sine technique
The MIMO swept sine technique is a digital implementation of
the classical swept sine technique that excites the structure
with a slowly sweeping sine over a user-defned frequency
band [6]. A typical excitation and response signal is shown in
Figure 1. A harmonic estimator performs data reduction, and
the spectra are online built up frequency line by frequency line
(Figure 2). At the end of the test, FRF and coherence functions
are calculated from the measured spectra.
In case of multiple-input multiple-output (MIMO) test setups,
all exciters are driven by the same instantaneous frequency.
Multiple averages with uncorrelated inputs are required in order
to calculate correct FRFs. This is achieved by defning multiple
sweeps with different phase relations between the different
exciters.
2.1 Leakage-free online sine extraction
The data is measured over a user-defned number of periods.
During the measurement, the frequency of the excitation is
constant. A harmonic estimator continuously extracts amplitude
and phase from the measured time data in a least squares
sense. Simultaneously the excitation frequency is already
updated to the next value that is calculated from the current
frequency, the elapsed measurement time and the required
sweep rate, and the whole process is repeated. Leakage is
avoided, as there is no need for a Fourier transformation (FFT)
to the frequency domain.
The spectra have a user-defned frequency resolution f and
only need to contain data at discrete frequencies f
k
= k f,
where k is an integer. However, the complex spectrum values
(consisting of amplitude and phase) are extracted at
excitation frequencies , which typically do not coincide with
a discrete frequency line of the requested spectrum. Therefore,
the extracted values are averaged to yield the value at
spectral line :
(1)
where is the number of spectrum estimates available in an
interval around :
f f
.


Figure 1: Typical excitation (top) and response signal (bottom) from a swept sine test.
Figure 2: Spectra are built up frequency line by frequency line
during a MIMO swept sine test.

Figure 3: System Identifcation step prior to a swept sine test
with controlled excitation.

2.2 Control of excitation amplitudes
In the most basic swept sine test the structure is excited by the
output voltages following user-defned reference profles: output
voltage (amplitude and phase) as a function of frequency. These
output voltages are sent to the exciters exactly as the reference
profle defnes, and are not modifed based on the response of
the structure (i.e. no control).
However, one of the advantages of sine excitation is the
possibility to easily control the excitation, which is useful in
many cases. The structure can be excited at different excitation
levels to assess and analyze the non-linearity of the structure
under test. The infuence of the excitation level on the modal
parameters can be studied. Also, during modal testing (or FRF
measurements in general) usually a fat excitation force is
desired. Drops in the force autopower spectra indicate that no
energy is entering the structure and lead to bad quality FRF
functions. There are several reasons for a non-fat autopower,
which are not further discussed here. Being able to control the
excitation allows compensating for these drops and peaks in
the force autopower spectra, exciting the structure with a fat
excitation level. Excitation control can also be used to simulate
operational conditions, by making sure that the required
response levels are present at the defned response locations.
For each of the control channels a target reference profle
is defned, the required measured quantity as a function of
frequency. In case of modal testing the control channels are
typically the force cells and the target reference profle is
a fat force (in Newton) over the complete frequency range.
To avoid having a time consuming MIMO online closed loop
control, another strategy has been implemented: control of
the excitation amplitude is achieved by frst executing a so-
called system identifcation step, during which the system
FRF matrix H
CV
between the output voltages V and control
channels C(C=H
CV
V) is estimated using pseudo random
excitation and according to the MIMO H
1
estimate:

C C
(2)
where S
CV
contains the cross spectra between control channels
and voltages and S
VV
contains the auto and cross spectra
between the voltages.
The system identifcation (Figure 3) also performs an extensive
test setup validation by measuring background noise levels,
checking signal-to-noise ratios and detecting open channels
(broken cables or broken transducers). In a second step this
FRF model is used to calculate the required output voltages Vref
to reach the target values Cref at the control channels:
V
ref
= H
-1
C
ref
(3)
In order to evaluate the quality of the system matrix inversion,
the singular values of the system matrix H
CV
and the inverse
system FRFs are available for checking possible singularities
that might result in a bad system identifcation and bad control.
The results from the system identifcation are used to calculate
the maximum levels that are expected during the test on each
channel and to optimize the input ranges of the AD-converters.
During the actual sine sweep test the calculated output voltages
Vref are sent to the exciters without any closed loop control.
The complete test then runs automatically without any user
interaction.
CV
3 Case study: reciprocal FRF measurements on
midsize passenger car
Some typical acoustical transfer functions required for transfer
path analysis or noise path contribution were measured on a
midsize passengers car. As mentioned in the introduction, these
FRFs are often measured in a reciprocal way, using loudspeaker
excitation. Two LMS low-frequency volume velocity sources
were placed on the drivers and passengers seats. These
volume velocity sources have the shape of an average human
torso and the sound output at the position of a humans ear. A
total of 37 responses were measured using the LMS Scadas III
frontend at various locations in the vehicle (Figure 4):
4 microphones at 4 seats, located at driver and passengers
ears
3 microphones in the engine compartment at the air intake,
alternator and between the engine and frewall
Triaxial acceleration at the wheel knuckle of the front right
wheel
Triaxial accelerations on both active and passive side of the
engine mounts, gearbox anti-roll mount and shock tower
mount.
The measurements were performed between 20 Hz and 800
Hz with 1 Hz frequency resolution. Three averages (sweeps)
were measured using the swept sine technique, in order to be
able to evaluate FRFs as well as coherence functions. Phase
differences between the 2 volume velocity sources during the
different sweeps were 0, 90 and 180 degrees. A linear sweep
rate of 5 Hz/s was used. Results were compared with the
stepped sine and burst random techniques. The burst random
measurements were performed with 150 averages up to
1024 Hz with a 1 Hz frequency resolution, as the anti-aliasing
flter starts working from 800 Hz on. For the burst random
measurement, the amplifer gain was set as high as possible
without damaging the volume velocity sources.
For the swept sine measurement, both a fxed volume
acceleration excitation of 5 m
3
/s
2
and a fxed voltage
excitation (no control) were used. The results from the system
identifcation procedure used for control of the volume
acceleration are shown in Figure 5. From the system FRF matrix
it can be seen that the response of the volume velocity sources
to a fxed voltage input is not completely fat. The very low off-
diagonal system FRFs indicate that there is little cross coupling
between both volume velocity sources.


Figure 4: Test setup for the reciprocal FRF measurements. (Top left) 2 low frequency volume velocity sources
on drivers and passengers seat. (Top right) Triaxial response measurements on active and passive side of the
gearbox anti-roll mount. (Bottom left) Microphone between engine and frewall. (Bottom right) LMS Scadas III
frontend equipped with 40 channels.
Figure 5: Results from the system identifcation procedure. (Top) Target reference profles. (Middle) System FRF matrix between
volume acceleration and voltage output. (Bottom) Calculated voltage output for in-phase sweep (left) and out-of-phase sweep
(right).
Figure 6: Volume acceleration spectra for both volume velocity sources during the 1st sweep: (red/black) target reference volume
acceleration. (green/gray) fxed voltage output sweep (open loop). (blue/black) controlled excitation using system identifcation.
(purple/gray) burst random.
In Figure 6 the volume acceleration spectra of both volume
velocity sources from the measurement with and without
control are compared. It is clear that the volume acceleration
due to a fxed voltage output is not fat, while the volume
acceleration nicely follows the target reference profle during
the controlled measurement. Also the volume acceleration
spectra of the burst random measurement are included as a
reference, clearly indicating the much higher excitation levels of
the swept sine technique.
The measurements were executed in one of the Engineering
Services labs at the LMS facilities. Because it was not an
acoustical lab, there was quite some interference from
people passing by, doors opening and closing, etc. In order to
check the repeatability of the measurements, burst random
measurements were performed before testing started, between
the stepped sine and swept sine tests and after all tests
had been completed. Figure 7 shows these 3 burst random
measurements of the wheel knuckle lateral acceleration
towards the volume velocity source at the drivers ear.
Small differences are present due to these external noise
Figure 7: Repeatability of a wheel knuckle FRF for the burst
random measurements before, halfway and after the tests.
Figure 8: Comparison between burst random (red/black), stepped sine (blue/black) and swept sine (green/gray) FRF and
coherence. (Left) front-right microphone. (Right) Wheel knuckle FRF.
disturbances. Table 1 indicates the measurement times for
each measurement technique. The stepped sine measurement
is most time consuming. The swept sine measurement time is
a lot shorter and has the same order of magnitude as the burst
random measurement time.
Measurement technique Measurement time
Burst random (150 avgs) 2min35sec
Stepped sine (3 avgs) 50 min
Swept sine (3 avgs) 8 min
Table 1: Reciprocal FRF measurement times for the different
measurement techniques
In Figure 8 the FRF and coherence functions of the car interior
microphone at the front-right passengers ear and wheel knuckle
lateral acceleration towards the volume velocity source at
the drivers ear measured with the 3 different techniques are
compared. The microphone FRFs overlay perfectly, while there
are small differences in the wheel knuckle FRF. However, these
are of the same order as the difference between consecutive
burst random measurements (Figure 7). As expected, the
coherence functions are better when sine excitation techniques
are used.
In Figure 9 the burst random and swept sine FRF and coherence
of the active side of the gearbox anti-roll mount is shown. The
coherence of the burst random measurement is extremely
low in large areas of the frequency band. Using the swept sine
technique, coherence values are a lot higher, only at some
frequency lines dropping below 0.8. The right display of Figure
9 zooms in on the 120-250 Hz frequency band, where the
coherence of the burst random measurement is approaching 0.
In these cases one has to question the accuracy and reliability
of the measurement. In most cases it will be concluded that this
measurement is impossible with burst random excitation and
the data cannot be used. The swept sine measurement has in
the same region quite high coherence values above 0.8, and the
FRF is smoother and less noisy. This is a measurement that can
be trusted and used for further analysis.
4 Case study: modal test on a midsize passenger car
A modal test was performed on the same vehicle as was used
in the case study of section 3. The car was placed on its wheels
and excited by 2 shakers, one at the front of the chassis under
the gearbox mount, one at the back at the rear subframe mount.
Responses were measured in 3 directions at 10 locations on the
chassis and 3 on the powertrain (2 on the engine and 1 on the
gearbox). In addition 4 microphones located at the drivers and
passengers ear in the car interior were measured.
The measurements were performed between 1 Hz and 200 Hz
with 0.125 Hz frequency resolution. Three averages (sweeps)
were measured using the swept sine technique, in order to
be able to evaluate FRFs as well as coherence functions. The
car was excited with a controlled force of 25 N and 0, 90 and
180 degrees phase difference between the forces during the
different sweeps. A logarithmic sweep rate of 1 oct/min was
used. A frequency dependent sweep rate that varies linearly
over the frequency band was tried as well, with similar results
to the logarithmic sweep rate. FRFs and coherences were
compared with the stepped sine and burst random techniques.
Figure 9: Comparison between burst random (red/black) and swept sine (green/gray) FRF and coherence of the active side of the
gearbox mount. (Left) complete frequency band. (Right) Zoomed 120-250Hz.
Figure 10: Engine DOF FRF and coherence with 50 averages
(red/black) and 150 averages (green/gray).
Figure 11: Repeatability of a driving point FRF for the burst
random measurements before, halfway and after.
The burst random measurements were frst performed with 50
averages up to 256 Hz with a 0.125 Hz frequency resolution.
After data inspection coherence levels at the engine DOFs
(degrees of freedom) were very low with extremely noisy
corresponding FRFs. It was decided to continue the burst
random measurements with 150 averages. Figure 10 compares
the burst random measurement with 50 and 150 averages of a
typical engine DOF.
To check repeatability of the measurements, burst random
measurements were performed before testing started, between
the stepped sine and swept sine tests and after all tests had
been completed. Figure 11 shows a driving point FRF from
the 3 burst random measurements. It can be concluded that
there were repeatability problems at the lower frequencies. The
swept sine measurement was then repeated as well and showed
similar repeatability problems. The measurement times for the 3
techniques are listed in Table 2. The stepped sine technique was
once more the slowest technique, with burst random and swept
sine measurement time almost equal.
Figure 12: Comparison between burst random (red/black), stepped sine (blue/black) and swept sine (green/gray) FRF and
coherence. (Left) driving point FRF. (Right) DOF on chassis.
Figure 13: Comparison between burst random (red/black) and swept sine (green/gray) FRF and coherence of a DOF on the engine.
(Left) complete frequency band. (Right) Zoomed 50-100 Hz.
Measurement technique Measurement time
Burst random (150 avgs) 22 min
Stepped sine (3 avgs) 66 min
Swept sine (3 avgs) 24 min

Table 2: Modal test measurement times for the different
measurement techniques
The left display of Figure 12 compares the driving point FRF
and coherence of the shaker at the front of the car between the
different techniques. Coherences are for all 3 techniques close
to 1, with the swept sine coherence almost everywhere equal
to 1. There are some differences between the burst random
and sine techniques at the lower frequencies. This is probably
caused by the non-linearity of the suspension of the car. The
stepped sine and swept sine FRFs are very similar. At higher
frequencies (above 50 Hz), all 3 FRFs are similar. The same
conclusions can be made when looking at other DOFs, the right
display of Figure 12 shows a typical FRF and coherence from a
point on the chassis.
In Figure 13 the burst random and swept sine FRF and
coherence of one of the engine DOFs are shown. The
coherence of the burst random measurement is very low at
higher frequencies. Using the swept sine technique, coherence
values are a lot higher, only at some deep anti-resonances and
at the highest frequencies it is dropping. The right display
of Figure 13 zooms in on the 50-100 Hz frequency band,
where the coherence of the burst random measurement is
dropping severely. In these cases one has to question the
accuracy and reliability of the measurement, while the swept
sine measurement has in the same region coherence values
approaching 1 resulting in smooth and almost noise-free FRFs.
5 Conclusions
A new MIMO sine testing technique has been introduced
that uses a digital implementation of the classical swept sine
excitation [6]. The technique acquires leakage-free spectra,
which are processed into multiple-input-multiple-output FRFs.
A system identifcation approach is implemented to control
the excitation level during the test without using a time-
consuming online closed loop control scheme. During several
industrial case studies, the new technique was compared with
the traditional burst random and stepped sine techniques. It
was proven that this technique is able to measure FRF and
coherence functions of similar high quality as the stepped sine
technique, but at drastically reduced measurement times, which
are comparable with the burst random technique.
Acknowledgements
This work was carried out in the frame of the EUREKA project
E!3341 FLITE2 (Flight Test Easy - Extension) which is conducted
in cooperation with V.U.Brussels (B), K.U.Leuven (B), Lambert
Aircraft Engineering (B), INRIA (F), Sopemea (F), Onera (F),
Airbus (F), Dassault (F), AGH (PL), ILOT (PL), PZL Mielec
(PL). The fnancial support of IWT-Vlaanderen is gratefully
acknowledged.
References
[1] LMS INTERNATIONAL, Transfer Path Analysis the
Qualifcation and Quantifcation of Vibro-acoustic Transfer
Paths, Application Note, Leuven, Belgium, 1995.
[2] K. WYCKAERT AND H. VAN DER AUWERAER, Operational
analysis, Transfer Path Analysis, Modal Analysis: tools to
understand road noise problems in cars. In Proceedings of the
SAE Noise and Vibration Conference, 139-143, Traverse City,
1995.
[3] W. HEYLEN, S. LAMMENS, AND P. SAS, Modal Analysis
Theory and Testing. K.U.Leuven, Belgium, 1997.
[4] LMS INTERNATIONAL, LMS Cada-X, MIMO Stepped Sine
Acquisition Monitor, Leuven, Belgium, www.lmsintl.com, 2005.
[5] G. GLOTH AND M. SINAPIUS, Analysis of swept sine runs
during ground vibration tests of large aircraft. In Proceedings
of IFASD 2003, the International Forum on Aeroelasticity and
Structural Dynamics, Amsterdam, The Netherlands, June 2003.
[6] LMS INTERNATIONAL, LMS Test.Lab, MIMO Sine Testing,
Leuven, Belgium, www.lmsintl.com, 2005.
LMS INTERNATIONAL
Researchpark Z1, Interleuvenlaan 68
B-3001 Leuven [Belgium]
T +32 16 384 200 | F +32 16 384 350
info@lms.be | www.lmsintl.com
Worldwide For the address of your local representative, please
visit www.lmsintl.com/lmsworldwide
LMS is an engineering innovation partner for
companies in the automotive, aerospace and
other advanced manufacturing industries. LMS
enables its customers to get better products
faster to market, and to turn superior process
effciency to their strategic competitive
advantage. LMS offers a unique combination of
virtual simulation software, testing systems and
engineering services.
LMS is focused on the mission critical
performance attributes in key manufacturing
industries, including structural integrity, system
dynamics, handling, safety, reliability, comfort
and sound quality. Through our technology,
people and over 25 years of experience, LMS
has become the partner of choice for most of
the leading discrete manufacturing companies
worldwide.
LMS is certifed to ISO9001:2000 quality
standards and operates through a network
of subsidiaries and representatives in key
locations around the world.
LMS FRANCE Paris | T +33 1 69 35 19 20 | info@lmsfrance.fr
Lyon | T +33 4 37 49 13 25
LMS DEUTSCHLAND Leonberg | T +49 7152 97 97 90 | info@lms-gmbh.de
LMS ITALIANA Novara | T +39 0321 445 011 | info@lms.it
LMS UK Coventry | T +44 2476 408120 | info@lms-uk.co.uk
LMS N. AMERICA Detroit | T +1 248 952-5664 | info@lmsna.com
Los Angeles | T +1 714 891-4205
Washington D.C. / Baltimore | T +1 410 203-1200
LMS JAPAN Yokohama | T +81 45-478-4800 | info@lms-ja.co.jp
LMS KOREA Seoul | T +82 2-571-7246 | info@lmskorea.com
LMS CHINA Beijing | T +86 10 8497 6463 | info@lmschina.com
LMS INDIA CAE - Bangalore | T +91 80 511 51 512 | info.lmsindia@lms.be
Test - Chennai | T +91 44 281 73 711| info.lmsindia@lms.be
LMS RUSSIA Moscow | T +7 095 937 84 93 | info.lmsrussia@lms.be
LMS SINGAPORE Singapore | T +65 6777 3609 | info.lmssingapore@lms.be
LMS Test Division Leuven [Belgium] | T +32 16 384 200 | info@lms.be
Breda [The Netherlands] | T +31 76 573 6363
Gttingen [Germany] | T +49 551 50 48 50
Brasov [Romania] | T +40 268 4142
LMS CAE Division Leuven [Belgium] | T +32 16 384 200 | info@lms.be
Kaiserslautern [Germany] | T +49 631 30322 100
Coralville, IA [United States] | T +1 319 626 6700
Brasov [Romania] | T +40 268 4142
LMS Engineering Leuven [Belgium] | T +32 16 384 200 | info@lms.be
Services Division Kaiserslautern [Germany] | T +49 631 30322 100
Detroit, MI [United States] | T +1 248 952 5664
Torino [Italy] | T +39 011 6671 615