Building and Environment 94 (2015) 752e763

Contents lists available at ScienceDirect

Building and Environment

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

Sound reflection measurements on noise barriers in critical conditions

M. Garai, P. Guidorzi*
Dept. of Industrial Engineering, School of Engineering and Architecture, 40136 Bologna, Italy

a r t i c l e i n f o a b s t r a c t

Article history: It is often necessary to check the intrinsic acoustic characteristics of installed noise barriers, like sound
Received 29 March 2015 reflection and airborne sound insulation, to verify their compliance to design specifications or their
Received in revised form quality after some years of life. These characteristics may be measured in-situ following CEN/TS 1793-5.
18 June 2015
These guidelines have been substantially improved in the frame of the European project QUIESST (2009
Accepted 22 June 2015
e2012), which is now under consideration by the relevant CEN working groups to produce new Euro-
Available online 25 June 2015
pean standards. The new method for measuring sound reflection specifies the usage of an electroacoustic
sound source and a microphone grid, in order to obtain a set of impulse responses; these are processed
Keywords:
Acoustic barriers
by means of improved algorithms to compute the required results. The impulse responses are acquired
In-situ measurements using MLS (Maximum Length Sequence) or ESS (Exponential Swept-Sine) as test signals. While the
Sound reflection acoustical characteristics of a noise barrier obtained using the two signals are generally equivalent, in
Sound insulation critical conditions e e.g. excessive background noise or local meteorological variability e some dis-
Impulse response crepancies may occur. Moreover, different type of background noise (broadband or impulsive) give
ESS different effects on the final result, using MLS or ESS test signals. This paper presents a series of ex-
periments designed to put in evidence the differences between Reflection Index measurements per-
formed in the mentioned critical conditions, according to the QUIESST guidelines, done using MLS or ESS
signals. The relative advantages and drawbacks are analysed and discussed in detail. Conclusions are
drawn on the selection of the best test signal for each situation.
© 2015 Elsevier Ltd. All rights reserved.

1. Introduction repeatability and reproducibility [8,9]. However, no explicit

The in-situ measurement method for sound reflection described
in CEN/TS 1793-5 [1] was firstly developed in the frame of the
ADRIENNE European project (1995e1997), borrowing from the
sound reflection method for measuring the sound absorption with
MLS (Maximum Length Sequence) signals implemented by Garai
some years before [2]. A different source and microphone set up
permits to measure also the airborne sound insulation [1,3]. This
kind of measurements is used to know the intrinsic acoustic char-
acteristics of an installed noise barrier and to verify their compli-
ance to design specifications after its installation aside of a road or a
railroad [4e6]. The same method can be used to check the quality of
the installed barriers after some years of life. In the frame of the
European project QUIESST (2009e2012) [7], working package 3, the
test method has been completely revised, increasing its robustness
and obtaining for the first time an objective evaluation of its
* Corresponding author.
E-mail addresses: massimo.garai@unibo.it (M. Garai), paolo.guidorzi@unibo.it
(P. Guidorzi).
guidelines exist about the selection of the test signal used to
measure the impulse responses (IRs) required for the computation
of the Reflection Index and Sound Insulation Index. Two test signals
are mentioned in the QUIESST reports [7]: MLS and ESS (Expo-
nential Swept-Sine), considered equivalent if meeting some general
criteria. In fact, the acoustical characteristics of a noise barrier ob-
tained using the two signals are generally equivalent in normal
conditions. Instead, in critical environmental conditions e like
excessive background noise or time variation of the system under
test due to local meteorological instability e some discrepancies
may occur between the acoustic characteristics of the same noise
barrier in the same conditions obtained using the two mentioned
signals. Therefore, it is necessary to investigate the problem more in
depth in order to understand how to select of the best test signal for
each situation. In the light of this, the paper presents a series of
Reflection Index measurements on a real scale noise barrier,
following CEN/TS 1793-5 and the QUIESST improved method [7,8].
The experiments are designed to put in evidence the differences
between Reflection Index measurements done using MLS or ESS
signals in critical conditions. The relative advantages and

http://dx.doi.org/10.1016/j.buildenv.2015.06.023
0360-1323/© 2015 Elsevier Ltd. All rights reserved.
 M. Garai, P. Guidorzi / Building and Environment 94 (2015) 752e763
drawbacks are analysed and discussed in detail. Finally, some
conclusions are drawn on the selection of the best test signal for
each situation. A complete description of the measurement method
can be found in Refs. [6e10].
2. Test signals
The test signal type (MLS or ESS) must be chosen to minimize
the possible troubles in the impulse response measurements.
When applied to in-situ barrier characterization, impulse response
measurements may be affected mainly by:
- time variance, due to wind and temperature changes during the
measurement session;
- distortion of the system (e.g. loudspeaker non-linearity);
- background noise, broadband or impulsive, coming from noise
sources in the surrounding of the measurement position or from
the measuring equipment itself.
In this regard, the output level of the power amplifier during the
measurement must be carefully selected. In fact, an excessively low
level will decrease the signal to noise ratio (SNR) of the measure-
ment and will increase the influence of time variance on the
recovered impulse response; an excessively high output level will
cause non-linearity of the loudspeaker and possibly other elements
of the measurement chain. Considering all the above, the most used
test signals are MLS and ESS, well known for their reliability even in
presence of a non-negligible background noise. On the other hand,
MLS and ESS, have different behaviours in relation to their noise
rejecting properties and their capability to withstand some degree
of distortion and time variance. Consequently the Reflection Index
values coming from impulse responses measured in the same
conditions using the two signals may show differences in case of
critical measurement conditions (when the system under test
doesn't comply the hypothesis of linear and time invariant system).
The use of a MLS (Maximum Length Sequence) signal for
measuring impulse responses is well established [11e13]. The sine
sweep signal is also widely used [14]. In particular, the ESS (Expo-
nential Sine Sweep) signal [18] has gained considerable interest
since Farina introduced it in 2000 [15] and refined it in 2007 [16]. If
compared to MLS, it reveals both advantages and drawbacks. The
main advantage of the ESS method is the separation of the linear
part of the measured impulse response of the system from most
part of the harmonic distortion, even if recent studies have shown
that some amount of odd orders distortion still remains, as Torras
et al. [17] formally proved in 2011. The separation of the most part
of distortion from the linear part allows having a much better signal
to noise ratio (SNR) than with MLS, because the impulse response is
free from the spurious peaks distributed on the time axis typically
caused by distortion when using MLS. On the contrary, using ESS,
the impulse response is recovered by means of an aperiodic linear
convolution, avoiding the time-aliasing problem of MLS. Moreover,
the use of an ESS measurement signal allows to describe easily the
nonlinearities of the measured system by means of the Volterra
model [21] and its simplified implementation (diagonal Volterra
model). The crest factor of about 3 dB of the ESS can be exploited
performing high power measurements in (steady) noisy conditions.
Typically, in similar conditions the ESS has a dynamic range of
about 15 dB higher than MLS.
Whereas stationary background noise can be somehow rejected
and compensated in different ways for both MLS and ESS methods
[20], impulsive noises can contaminate the data sampled using an
ESS signal, causing spurious effects on the deconvolved impulse
response in form of a frequency decreasing sweep [14,16]. Farina in
Ref. [16] proposed a possible workaround for correcting a
753
measurement corrupted by an impulsive noise, consisting in the
rejection of the portion of the corrupted sampled ESS through a
narrow-band filter, tuning the filter itself at the same frequency of
the ESS at the very instant the disturbance occurs. However this
procedure can be applied only if the sampled ESS is available and
not when a measurement system gives in output directly the
deconvolved impulse response. In addition, depending on the kind
and duration of the disturbance, the manual correction of the ESS
may not be possible.
3. Experimental results
Three series of measurements were executed in order to
compare the MLS and ESS signals applied to Reflection Index
measurement in critical conditions. The loudspeaker and micro-
phone grid used for the tests are visible in Fig. 1. In a first set of
measurements MLS, pink-filtered MLS and ESS signals, in all cases
single shot, were employed. The measurements were done at
different sound pressure levels and with a powerful fan, switched
on or off, placed near the microphones in order to test the resilience
of the measurement to time variance and stationary noise.
Following the same procedures, a second set of measurements
were performed using unfiltered MLS signal, with various average
options, and ESS. A third set of measurements were performed in
presence of two types of impulsive background noise.
3.1. Single shot measurements
A series of Reflection Index measurements were carried out on
the reflective and the absorbing surfaces of the noise barrier. Three
signal types were used for this comparison: MLS, pink-filtered MLS
and ESS, all of them 256 K samples long. No averages were done on
the measurements (it is worth remembering that averaging can
improve the MLS performance, while it is useless employing the
ESS). A broadband background noise, having an equivalent sound
pressure level of about 65 dB, coming from the heating system was
present during the measurements. Since the aim of this test was a
comparison of different signals in presence of background noise
and time variance, measurements at different sound pressure levels
and with a powerful fan, switched on or off, placed near the mi-
crophones were performed. The first measurement session was
done with the fan switched off (air speed nearly 0 m/s) and opti-
mizing the system levels so as to measure a Leq of about 80 dB at the
Fig. 1. Loudspeaker and microphone grid close to the sample noise barrier used for the
tests. Reflecting metal sheet on the right; absorptive melamine layer on the left.
754 M. Garai, P. Guidorzi / Building and Environment 94 (2015) 752e763

Fig. 2. Reflection Index measurement in stationary conditions; comparison among different signals. Sound absorbing surface (left) and sound reflecting surface (right).

microphones; the three signal types were calibrated in order to
have the same energy content. The same measurements were
repeated switching on the fan; this increased the overall sound
pressure level at the microphones of about 5 dB and induced a high
time variance to the system. The air speed at the microphones was
around 5 m/s. The measurements were repeated lowering the
overall output level in steps of 10 dB and keeping the fan turned on.
In each session, the measurement on the absorptive surface, on the
reflective surface and in the free field were performed. It must be
noted that the measurement on the highly reflective surface is
problematic, mainly because of stationary waves between the
barrier surface and the sound source and for computational rea-
sons. The stationary waves effect (partially or totally fixable by
using a reference measurement and correction procedure similar to
that recommended in ISO 13472-1 [19]) is visible as an oscillation of
the Reflection Index values around the unity value, but the aim of

Fig. 3. Reflection Index measurement in non-stationary conditions (fan switched on) and output level lowered of 10 dB. Sound absorbing surface (left) and sound reflecting surface
(right).

Fig. 4. Reflection Index measurement in non-stationary conditions (fan switched on) on the sound absorbing surface; comparison among different output levels (lowered in steps of
10 dB). MLS signal (left) and ESS (right).
 M. Garai, P. Guidorzi / Building and Environment 94 (2015) 752e763 755

Fig. 5. Reflection Index measurement in non-stationary conditions on the sound reflecting surface; comparison among different output levels (lowered in steps of 10 dB). MLS
signal (left) and ESS signal (right).

this work is the comparison of different measurement signals, so
this aspect has been disregarded. Actually, the sound reflection
measurement on a highly reflective surface can be considered the
“worst case”.
Fig. 2 shows the measurements with the fan switched off and
optimal output level settings (in front of the absorptive surface on
the left and in front of the reflective surface on the right): the
Reflection Index values measured with all three signals are almost
identical, both on the absorptive and reflective surface.
Fig. 3 shows an example of what happens when the system is
made time variant, switching on the fan, and the output level is
lowered of 10 dB. Keeping the average of the results obtained with
the fan switched off as reference it can be seen that on the
absorptive surface, the ESS and pink-filtered MLS measurements

Fig. 6. Comparison of the single number ratings DLRI measured on absorptive and reflective surfaces with different signals and environmental conditions. Absorptive surface
(upper); reflective surface (lower).
756 M. Garai, P. Guidorzi / Building and Environment 94 (2015) 752e763
are still valid, while the measurement done with the “plain” MLS
shows evident errors at low frequencies. On the reflective surface,
both the MLS measurements show errors at low frequencies, while
only the measurement done with ESS is almost invariant. This
confirms the better immunity of the ESS signal to the time variance
of the system.
Fig. 4 shows the Reflection Index measurements in non-
stationary conditions on the sound absorbing surface, lowering
the signal output level in steps of 10 dB, starting from the optimal
setting (labelled “-0dBFS”). When the MLS signal is employed (left),
both the measurements at
10 dB and at
20 dB show obvious
errors below 400 Hz. When the ESS signal is used (right), the
measurement at optimal level is quite correct, the one at
10 dB
has only some small deviations from reference while the mea-
surement at
20 dB shows an underestimation of Reflection Index
below 200 Hz, but a good agreement with the reference over
200 Hz. In all presented cases, a measurement in ideal conditions is
used as Reference; as shown for example in Fig. 2 (left), ideal
conditions measurements give almost perfect match between re-
sults obtained by means of MLS and ESS.
Fig. 5 shows the Reflection Index measurements in non-stationary
conditions on the sound reflecting surface (worst case), lowering the
signal output level in steps of 10 dB, starting from the optimal set-
tings. When the MLS signal is employed (left), the measurement
already shows errors (overestimation of Reflection Index at low fre-
quencies) at the optimal output level; the measurement at
10 dB
shows clearly wrong results between 160 and 315 Hz and the mea-
surement at
20 dB is totally unreliable. When the ESS signal is used
(right), the measurement at the optimal level is quite correct, since it
has only some small deviations from the reference. The measurement
Fig. 7. Reflection Index measurement in stationary conditions on the sound absorptive surface. MLS (single, 4 and 8 averages) and ESS signals. Output level: optimal (upper
left),
10 dB (upper right),
20 dB (lower left) and
30 dB (lower right).
at
10 dB shows a small underestimation of Reflection Index below
250 Hz; the measurement at
20 dB shows again errors below
400 Hz. ESS results are generally better than the MLS ones in the
same conditions.
Fig. 6 shows a comparison on the single number ratings DLRI of
all the Reflection Index measurements taken in this session. For the
sake of clarity, the values are not rounded to integers as prescribed
in Refs. [1] and [7]. In ideal conditions, all single number ratings are
very close, on both the absorptive and reflective surfaces. When the
fan is switched on, but still the output level is optimal, a good
agreement between the values can be still observed. In case of fan
turned on and output level reduced of
10 dB and
20 dB, a higher
variance from the optimal level value, kept as reference, can be
observed; again, the measurements done using the ESS are gener-
ally better.
3.2. Averaged measurements
A second series of Reflection Index measurements was done in
order to investigate the reliability of Reflection Index measure-
ments using averaged MLS and ESS signals in critical conditions.
Single shot (kept here for consistence with the other sections), 4
and 8 averages (flat) MLS measurements and single cycle ESS
measurements were carried out. The length of both signals was
increased to 512 K samples with the aim of further enhancing the
differences among the results.
The first measurement session was done with the fan switched
off.
Fig. 7 shows the Reflection Index measured using the mentioned
signals options on the absorptive surface, in stationary conditions.
 M. Garai, P. Guidorzi / Building and Environment 94 (2015) 752e763
The upper left plot shows the results obtained using the optimal
output level (the sound pressure levels of the signal and the
background noise were the same as before): a perfect match among
results gathered by means of MLS (single, 4 and 8 averages) and ESS
can be observed so this measurement was taken as reference (0 dB).
The upper right plot shows the measurements performed lowering
the output signal level of 10 dB. Apart some small discrepancies at
low frequencies, there is a good agreement among different mea-
surements. The lower left plot shows the measurements taken
at
20 dB output level ref. the optimal value: the single shot and
four-average MLS measurements now show evident errors at low
frequencies below 200 Hz while the eight-average MLS and ESS
measurements match the reference curve. Finally, the lower right
plot shows measurements taken at
30 dB: in this case also the
eight-average MLS and ESS measurements show clear deviations
from the reference curve at low frequency, less pronounced for the
ESS one.
Fig. 8 shows the Reflection Index measured as in the previous
case, but on the reflective surface, in stationary conditions. The
upper left plot shows the results obtained using the optimal output
level: again, a very good match among results gathered by means of
MLS and ESS can be observed, so this measurement was taken as
reference. The upper right plot shows the measurements per-
formed lowering the output signal level of 10 dB. Apart some small
discrepancies at low frequencies, there is a good agreement among
different measurements. The lower left plot shows the measure-
ments taken at
20 dB output level: the single shot MLS mea-
surement shows some visible deviations below 250 Hz while the
other measurements match well the reference curve. The lower
right plot shows measurements taken at
30 dB: in this case only
Fig. 8. Reflection Index measurement in stationary conditions on the sound reflective surface. MLS (single, 4 and 8 averages) and ESS signals. Output level: optimal (upper
left),
10 dB (upper right),
20 dB (lower left) and
30 dB (lower right).
757
the ESS measurement does not show large deviations from the
reference curve at low frequencies.
The second measurement session was done with the fan
switched on.
Fig. 9 shows the Reflection Index measurements done on the
absorptive surface of the barrier, in non-stationary conditions, with
the fan on as in the previous measurement session. The results
obtained using the optimal output level (upper left plot) show a
very good match among results, with some little discrepancies
below 160 Hz (only the ESS measurement matches perfectly the
reference). The same behaviour becomes more evident when
lowering the output signal level of 10 dB (upper right plot).
At
20 dB (lower left plot), the only Reflection Index measurement
matched with the reference is the ESS one. The lower right plot
shows the measurements taken at
30 dB: in this case also the ESS
measurement does show errors, more pronounced below 400 Hz.
Fig. 10 shows the Reflection Index measurements done on the
reflective surface in non-stationary conditions, with the fan on
(“worst case”). The Reflection Index obtained using the optimal
output level (upper left plot) show an almost perfect match among
results, and with the reference, for all measurement signals. The
already observed errors at low frequencies can be observed when
lowering the output signal level of 10 dB, 20 dB and 30 dB, but now
the 8-average MLS exhibits a better behaviour than the ESS
measurements.
Fig. 11 shows the single number ratings DLRI calculated in the
different cases described above and not rounded. Despite the de-
viations from the reference of some of the 1/3 octave Reflection
Index data, the single number ratings show a good reliability also in
critical conditions. The measurements on the absorptive surface in
758 M. Garai, P. Guidorzi / Building and Environment 94 (2015) 752e763
Fig. 9. Reflection Index measurement in non-stationary conditions (fan on) on the sound absorptive surface. MLS (single, 4 and 8 averages) and ESS signals. Output level: optimal
(upper left),
10 dB (upper right),
20 dB (lower left) and
30 dB (lower right).
stationary conditions done using the MLS and ESS signals (upper
left plot) show relative differences not greater than 0.25 dB. The
measurements on the absorptive surface in non-stationary condi-
tions e fan on e (upper right plot) show differences not greater
than 0.5 dB, with the only exception of the single shot MLS mea-
surement at
30 dB (deviation of about 2 dB). The measurements
on the reflective surface show even smaller differences: less than
0.2 dB for both the stationary and non-stationary measurements,
with the exception of the single shot non-stationary MLS case
at
30 dB, like on the absorptive surface.
3.3. Measurements in presence of an impulsive background noise
The comparison of measurements in critical conditions dis-
cussed in the previous sections were performed in presence of a
broadband stationary (constant average amplitude) disturbing
noise source (a fan) or generally considering the laboratory back-
ground and measurement chain noises, when the output signal was
kept much under the optimal level. Further investigation has been
carried out considering 2 different kinds of noises: the first one was
the noise of cars passing by (“soft” impulsive noise), and the other
one was the noise of a hammer hitting on concrete (“hard”
impulsive noise). The mentioned impulsive noises were played
during the measurements, performed in laboratory, using a sec-
ondary loudspeaker, in order to simulate ordinary in-situ envi-
ronmental conditions at the side of a road or on a building site.
Two series of measurements were performed, on laboratory, on
the absorbing and on the reflecting surface of the noise barrier,
using single shot MLS, 4 averages MLS and single shot ESS. The
measurements were performed at optimal output level and 30 dB
below the optimal output level, without disturbing background
noise (as reference), and with the presence of the soft and hard
disturbing noise, played from a secondary loudspeaker system, at
an average SPL level, measured at the microphone grid, equal, on
average, to the measurement signal SPL at the same point. The
length of the measurement signals (MLS and ESS) were 512 K
samples.
Fig. 12 (upper left) and (upper right) show Reflection index
measurements on the sound absorbing surface and different signal
level and type (MLS and ESS). In Fig. 12 (upper left) the measure-
ments performed with traffic noise disturbance “soft” type are
shown. It can be noted that all the measurements at optimal output
level have a quite good accordance with the reference curve, with
only some negligible differences at the lower frequency bands in
the single shot MLS case. Measurements performed in critical
conditions (30 dB below optimal level) show large errors, as ex-
pected: single shot MLS is the worst one and below the 400 Hz third
octave band is totally invalid. The two measurements obtained with
4 averages MLS and the single shot ESS show both errors on the
middle and low frequency bands, with different behaviour: the 4
averages MLS is better on lower frequencies, the ESS is better on the
middle frequencies. It is worth noting that, despite the similar
amount of errors (in different third octave bands) of the Reflection
Index curves obtained with these 2 measurements, the corre-
sponding SNR, shown in Fig. 13, would indicate a better result ex-
pected for the ESS case (10.15 dB for the ESS and 4.95 dB for the 4
averages MLS). Fig. 12 (upper right) shows same measurements
with “hard” type noise disturbance. In this case, the results ob-
tained at optimal output level, with both signals, are excellent and
comparable to the reference curve. For measurements performed
30 dB below optimal level, comments similar to the previous case
can be said: the single shot MLS measurement is markedly wrong at
 M. Garai, P. Guidorzi / Building and Environment 94 (2015) 752e763 759

Fig. 10. Reflection Index measurement in non-stationary conditions (fan on) on the sound reflective surface. MLS (single, 4 and 8 averages) and ESS signals. Output level: optimal
(upper left),
10 dB (upper right),
20 dB (lower left) and
30 dB (lower right).

lower frequencies, while the 4 averages MLS and ESS measure- ones in the “soft” type noise case. But in this case, the SNR values on
ments show some discrepancies with the reference, of similar the measured impulse responses show an opposite trend since the
global quantity, in different third octave bands and smaller than the 4 averages MLS has an SNR higher than ESS SNR (7.23 dB and

Fig. 11. Comparison of the single number ratings DLRI values measured on the absorptive surface in stationary conditions (upper left), on the absorptive surface in non-stationary
conditions e fan on e (upper right), on the reflective surface in stationary conditions (lower left), on the reflective surface in non-stationary conditions e fan on e (lower right).
760 M. Garai, P. Guidorzi / Building and Environment 94 (2015) 752e763

Fig. 12. Reflection Index measurements on the sound absorbing surface and different signal level and type (MLS and ESS). Traffic noise disturbance “soft” type (upper left), hammer
hitting on concrete “hard” type (upper right). Normalized Reflection Index measurement on sound reflecting surface and different signal level and type (MLS and ESS). Traffic noise
disturbance “soft” type (lower left), hammer hitting on concrete “hard” type (lower right).

Fig. 13. Average residual background noise (RMS power) computed on the Reflection Index measurements on absorbing surface presented in upper part of Fig. 12 (upper). Cor-
responding estimated SNR of the same measurements (lower).
 M. Garai, P. Guidorzi / Building and Environment 94 (2015) 752e763 761

Fig. 14. Average residual background noise (RMS power) computed on the normalized Reflection Index measurements on reflecting surface presented in lower part of Fig. 12
(upper). Corresponding estimated SNR of the same measurements (lower).

Fig. 15. Spectrograms of the free-field Impulse Responses (central microphone of the grid) measured in non-stationary conditions and with an output level
40 dB under the
optimal one. MLS single shot (upper left), MLS 4 averages (upper right), MLS 8 averages (lower left) and ESS (lower right).

Table 1
Suggested signal type for different kind of measurement conditions.

Ideal conditions Background broadband noise Background impulsive noise Wind

MLS Yes No No No
MLS (Averaged) Yes Yes Yes No
ESS Yes Yes No Yes
762 M. Garai, P. Guidorzi / Building and Environment 94 (2015) 752e763
2.38 dB respectively). This confirms the higher tolerance of MLS to
the “hard” impulsive disturbance, compared to ESS, but is not
directly connected to better results on the Reflection Index
coefficient.
Fig. 12 (lower left) and (lower right) show Reflection index
measurements on the sound reflecting surface and different levels
and measurement signals. The results have been normalized to a
reference curve, in order to highlight the errors. The measurements
on the reflective surface are also inherently more difficult from a
physical and computational point of view, so worst results are
generally expected, also without disturbance noise. In Fig. 12 (lower
left) the measurements performed with traffic noise disturbance
“soft” type are shown. The ESS measurement at optimal level was
taken as reference. The 4 averages MLS and single shot MLS mea-
surements at optimal level show some little discrepancies with
reference curve at low frequencies, but generally a good match can
be observed. The single shot MLS measurement 30 dB below
optimal level shows very large errors below 600 Hz. In this case
there is a difference between the 4 averages MLS and the ESS
measurements, since the latter is much better. In Fig. 12 (lower
right) the measurements performed with the “hard” disturbance
noise show better results in case of optimal output level for the 4
averages and single shot MLS cases. When the level is 30 dB below
optimal, the 4 averages and single shot MLS measurements show
both large errors, below 400 Hz. The ESS measurement, instead, has
a good agreement with the reference curve. On the SNR, shown in
Fig. 14, the same trend as in previous case can be observed.
3.4. Results and discussion
At a first glance, the measurements shown in Section 3.2, related
to stationary noise, seem to indicate that the ESS signal is recom-
mended for Reflection Index measurements in critical conditions,
but the results from the second session (fan on) show that the 8-
average MLS signal in some circumstances can give more reliable
results (see for example Fig. 10, lower right). Thus, the decision of
using the ESS or averaged MLS signal (and the number of averages)
is not trivial. As previously discussed, the ESS signal is generally
more reliable when the measured system is time variable (due for
example to wind) but is also much more sensible to background
noise (certainly present in case of wind). The MLS is more vulner-
able to time variance, and for this reason the number of averages
must be carefully chosen, but it has a strong immunity to back-
ground noise. The actual immunity to noise in different circum-
stances can be evaluated also from the spectrograms of the free-
field impulse responses, gathered from the central microphone of
the grid. Measurements taken at optimal output level and
at
40 dB (not discussed in previous sections), in stationary and
non-stationary conditions, from the second session of measure-
ments (Section 3.2) were studied.
Fig. 15 shows the spectrogram of the “worst case” free-field
Impulse Responses, measured in non-stationary conditions and
with an output level
40 dB under the optimal one. The first peak
of the impulse response is located around t ¼ 3 ms and the first
ground reflection is visible around t ¼ 12 ms. These peaks have
correspondence with the peaks of the free-field impulse responses
located around 4 ms and 13 ms respectively and are slightly shifted
in the time axis due to the width of the FFT analysis window of the
spectrogram. MLS single shot (upper left), MLS 4 averages (upper
right), MLS 8 averages (lower left) and ESS (lower right) measure-
ments are plotted. The measurement background noise can be
visually evaluated in the initial part of the time axis, before the
arrival of the first impulse response peak, from t ¼ 0 ms to about
t ¼ 3 ms, and in the later part of the impulse response; lighter
colour means louder sound pressure level (and higher background
noise). The averaging benefits on the MLS signal are evident. It can
be seen that the background noise of the MLS measurements
progressively decreases increasing the number of averages, while
the background noise of the ESS measurement is even higher than
the noise contained in the single shot MLS measurement. The
background noise of the 4- and 8-average MLS measurements in
non-stationary conditions and low output level is similar to the
background noise present in the corresponding MLS measurements
done in stationary conditions (not shown for space reasons) and
certainly better, from this point of view, than the ESS one. Data in
Fig. 10 lower right (actually measured at
30 dB below the optimal
level), support these conclusions: the 8-average MLS Reflection
Index measurement is closer to the reference curve than the ESS
measurement.
From the study of impulsive noise disturbance, presented in
Section 3.3, a partial preference for the ESS measurements seems to
be given, at least when the measurement is applied to Reflection
Index computation. But some cautions must be taken because of
the specific behaviour of the ESS when impulsive noise is present
during the measurement.
4. Conclusions
Reflection Index measurements done in non-ideal conditions on
a real scale noise barrier, with the surface treated to be very
absorptive or very reflective, generally show the advantage of using
an ESS signal, thanks to its higher dynamic than MLS, good resis-
tance to time variance due to air turbulence and separation of most
part of distortion. It has been shown, anyway, that in some cir-
cumstances (very low output measurement signal level, in order to
simulate high noise in the surroundings of the test location and
non-stationary conditions) the use of averaged MLS is preferred.
Moreover, when the measurements are done in presence of
short impulsive noises the use of averaged MLS signal is always
recommended, thanks to its properties of phase randomization of
any disturbing signals.
Table 1 shows a summary of rules for selecting the optimal test
signal.
When more than one kind of disturbances are present, the
measurement signal type selection must be evaluated case by case.
Future developments of this study will include an enhancement
of the method for estimating the Signal to Noise Ratio of the
measurement, computed from the impulse responses measured in
critical condition, improving the robustness of the SNR definition
used in Ref. [8] and an extension of the research to sound insulation
measurements.
References
[1] CEN/TS 1793e5, Road Traffic Noise Reducing Devices - Test Method for
Determining the Acoustic Performance - Part 5: Intrinsic Characteristics - in
Situ Values of Airborne Sound Reflection and Airborne Sound Insulation, 2003.
[2] M. Garai, Measurement of the sound-absorption coefficient in situ: the
reflection method using periodic pseudorandom sequences of maximum
length, Appl. Acoust. 39 (1993) 119e139.
[3] prEN 1793-6: Road traffic noise reducing devices - Test method for deter-
mining the acoustic performance - Part 6: intrinsic characteristics - In situ
values of airborne sound insulation under direct sound field conditions.
[4] M. Garai, P. Guidorzi, European methodology for testing the airborne sound
insulation characteristics of noise barriers in situ: experimental verification
and comparison with laboratory data, J. Acoust. Soc. Am. 108 (3) (2000)
1054e1067.
[5] G. Watts, P. Morgan, Measurement of airborne sound insulation of timber
noise barriers: comparison of in situ method CEN/TS 1793-5 with laboratory
method EN 1793-2, Appl. Acoust. 68 (2007) 421e436.
[6] M. Garai, P. Guidorzi, In situ measurements of the intrinsic characteristics of
the acoustic barriers installed along a new high speed railway line, Noise
Control Eng. J. 56 (5) (2008) 342e355. http://dx.doi.org/10.3397/1.2969244.
[7] http://www.quiesst.eu.
[8] P. Guidorzi, M. Garai, Advancements in sound reflection and airborne sound
 M. Garai, P. Guidorzi / Building and Environment 94 (2015) 752e763
insulation measurement on noise barriers, Open J. Acoust. 3 (2A) (2013)
25e38. http://dx.doi.org/10.4236/oja.2013.32A004.
[9] M. Garai, E. Schoen, G. Behler, B. Bragado, M. Chudalla, M. Conter, J. Defrance,
P. Demizieux, C. Glorieux, P. Guidorzi, Repeatability and Reproducibility of in
Situ Measurements of Sound Reflection and Airborne Sound Insulation Index
of Noise Barriers, 2014 to be published on Acta Acustica united with Acustica.
[10] P. Robinson, N. Xiang, On the subtraction method for in-situ reflection and
diffusion coefficient measurements, J. Acoust. Soc. Am. 127 (3) (March 2010).
EL99e10, http://dx.doi.org/10.1121/1.3299064.
[11] M.R. Schro € der, Integrated-impulse method measuring sound decay without
using impulses, J. Acoust. Soc. Am. 66 (2) (1979) 497e500.
[12] D.D. Rife, J. Vanderkooy, Transfer-function measurement with maximum
length sequences, J. Audio Eng. Soc. 37 (6) (1989) 419e444.
[13] G. Stan, J.J. Embrechts, D. Archambeau, Comparison of different impulse
response measurement techniques, JAES 50 (4) (2002 April) 249.
[14] S. Müller, P. Massarani, “Transfer-function measurement with sweeps”. Di-
rector's Cut including previously unreleased material and some corrections,
Orig. J. Audio Eng. Soc. 49 (2001) 443e471.
[15] A. Farina, Simultaneous Measurement of Impulse Response and Distortion
763
with a Swept-Sine Technique, presented at the 108th Convention of the Audio
Engineering Society, 2000. Feb, paper 5093.
[16] A. Farina, Advancements in Impulse Response Measurements by Sine Sweeps,
presented at the 122nd Convention of the Audio Engineering Society
Convention, 2007 May. paper 7121.
[17] A. Torras-Rosell, F. Jacobsen, A new interpretation of distortion artifacts in
sweep measurements, J. Audio Eng. Soc. 59 (2011) 283e289. May.
[18] K. Vetter, S. di Rosario, ExpoChirpToolbox: a Pure Data Implementation of ESS
Impulse Response Measurement, 2011. Pure Data Convention 2011, Weimer-
Berlin.
[19] ISO 13472e1, Acoustics d Measurement of Sound Absorption Properties of
Road Surfaces in Situ d Part 1: Extended Surface Method, 2002.
[20] M. Guski, M. Vorla €nder, Comparison of Noise Compensation Methods for
Room Acoustic Impulse Response Evaluations, Acta Acustica united with
Acustica, 100, 2014, pp. 320e327.
[21] A. Nov ak, L. Simon, F. Kadlec, P. Lotton, Nonlinear system identification using
exponential swept-sine signal, IEEE Trans. Instrum. Meas. 59 (8) (2010)
2220e2229.