Construction and Building Materials 140 (2017) 274–281

Contents lists available at ScienceDirect

Construction and Building Materials

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

Significance and reliability of absorption spectra of quiet pavements

Filippo G. Praticò, Rosario Fedele, Domenico Vizzari ⇑
DIIES Department, University Mediterranea of Reggio Calabria, Italy

h i g h l i g h t s g r a p h i c a l a b s t r a c t

Three methods for assessing the

acoustic absorption of a pavement
were compared.
KS: ISO 13472-2

New objective functions to fit
observed data and model were set up.

Boundary conditions were attentively KL: ISO 10534-2
investigated and experimental data
were used.

The precision and the accuracy of AD: ISO 13472-1
each device were studied.
Design Construction Noise performance

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

Article history: The objectives and scope of the study described into this paper were to compare different methods to
Received 18 November 2016 assess (measure or predict) the acoustic properties of a pavement in order to derive information about
Received in revised form 3 February 2017 use in practical applications. To this end, in-lab Kundt tube, on-site Kundt tube, Adrienne device and the-
Accepted 22 February 2017
oretical modelling were used. Based on measurements, absorption spectra were derived. Analyses and
Available online 2 March 2017
results focused also on composite indicators, first-pick frequencies and objective functions.
Based on results, when fitting experimental data through theoretical models the least square method is
Keywords:
unsatisfactory and alternative algorithms are proposed. The on-site Adrienne method provides a good
Acoustic absorption
Bituminous mixture
estimate of thickness, tortuosity, resistivity and maximum, while in-lab impedance tube, even if quite
Quiet pavement time consuming, seems to achieve the best accuracy. The on-site Kundt tube provides a satisfactory esti-
Quality assurance mate of porosity and further investigations and analyses are recommended to investigate its use also for
Quality control (QA/QC) porous surfaces. Future research will address two competing needs which emerged from the study: i) the
need for simplifying the procedures/methods; ii) the need for improving the quantity of information
gathered (e.g., texture), through the same device.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction
As is well known, traffic noise is one of the most important
environmental problems in Europe (see [1–5]). Power unit and
tyre-road contact are the key sources, and in the mid-to-high speed
range (above 40 km/h-80 km/h) rolling noise prevails [6]. Based on
the Weyl-Van Der Pol’s equation these phenomena generate
acoustical pressures, which, in turn, generate loudness ([7–10]).
⇑ Corresponding author.
E-mail address: domenico06@hotmail.it (D. Vizzari).
Tyre/road noise may vary >15 dB, based on tyre and pavement
type. To this end, noise reduction at the source can be more
cost-effective than treatments on the buildings or on the propaga-
tion path (e.g., noise barriers). Generation factors (pavement
texture, tyre type, see [11] and [12]) and acoustic absorption
govern rolling noise ([13,14]). Pavement characteristics
(volumetric and surface proprieties) affect generation, absorption
and propagation ([8,15–19]).
Based on the above, road agencies require a satisfactory level of
acoustic absorption, which can be assessed through in-lab (ISO
10534-2; [20]) and/or on-site (ISO 13472-1; ISO 13472-2) tests.

http://dx.doi.org/10.1016/j.conbuildmat.2017.02.130
0950-0618/Ó 2017 Elsevier Ltd. All rights reserved.
 F.G. Praticò et al. / Construction and Building Materials 140 (2017) 274–281
Furthermore, other relevant tests were set up in order to provide
accurate and high-speed methods and procedures in terms of
absorption properties ([21]). These tests have an outstanding
importance in worldwide quality assurance/quality control
(QA/QC, see [22–25]) because of noise-related health issues,
because of the growing diffusion of quiet pavements ([22,26–28]),
and because pavement acceptance procedures include adjusted
payment schedules ([22]). This notwithstanding, it may be
observed that: i) there is no common agreement on the best mea-
surement method; ii) different methods and devices can be used
and each one implies different issues, time consumption, traffic
restrictions, and results. In more detail, it is stated the inapplicabil-
ity of the ISO 13472-2 to porous asphalts and there is a certain lack
of attention on ‘‘true” values, i.e., about variability (precision) and
trueness (accuracy); iii) in both design and acceptance plans, the
consistency and dependency between material properties (e.g., air
voids content) and acoustic performance (as per measurements)
is a key-factor but is often neglected due to the complexity of the
involved relationships and algorithms and to the diverse cultural
backgrounds involved (acoustics versus civil engineering).
The facts above imply potential biases and uncertainties from a
practical and scientific standpoint.
2. Objectives
Based on the above, the objectives and scope of the study
described into this paper were the following:
1. to match and analyse three methods for measuring the acoustic
absorption of a pavement (Adrienne, ISO 13472-1, AD; in-lab
Kundt tube, ISO 10534-2, KL; on-site Kundt tube, ISO13472-2,
KS).
2. to compare measurements and expected results based on theo-
retical modelling. This implies: i) To study the objective (error)
function that allows better comparing experimental and theo-
retical spectra (least square method versus new recipes for opti-
misation), for the three different methods; ii) To fit the
theoretical model to the experimental data, without boundary
conditions (i.e., deriving tortuosity and resistivity, thickness,
and porosity without imposing ranges they belong to) or with
boundary conditions (using as input the real thickness and
the effective porosity and deriving the remaining two parame-
ters); iii) To study the relationships among the actual value of
thickness and porosity (i.e., based on experiments on cores)
and the corresponding values obtained from the optimisation
above; iv) consequently, to ascertain which method (out of
the three ones involved in the experiments) best fits the theo-
retical results in terms of sound absorption spectra (accuracy).
3. Design of experiments
Fig. 1 illustrates the flow chart of the experimental
investigation.
Task 1 focused on field investigations according to ISO 13472-1
and ISO 13472-2. A motorway, located in Southern Italy was inves-
tigated. Measurements were carried out on the hard shoulder lane
and on the inside lane, but not on the outside (overtaking) lane.
The surface layer main characteristics were the following: nominal
maximum aggregate size: 16 mm; asphalt binder content, percent
by total weight of mixture: 4.7%; air voids content: 20%; clogging
not noticed. For each device, operating time and performance were
analysed.
In Task 2, after core extraction (in the sections corresponding to
the quoted on-site measurements), cores were labelled and trans-
ported to the laboratory. There, each core was preliminary cleaned,
275
swept and controlled (for imperfections in the lower face). After-
wards it underwent the nondestructive tests (KS, ISO 10534-2;
effective porosity ASTM D6752-11; thickness EN 12697-36:2003).
Task 3 focused on data analysis (absorption spectra).
Finally, in Task 4 experimental spectra (from Tasks 1 and 2)
were compared with the ones derived from the theoretical model
(Task 3).
4. Modelling, implementation and preliminary analyses
As mentioned above, Task 1 and Task 2 provided experimental
data to Tasks 3 and 4.
In terms of theoretical modelling, note that the sound pressure
level depends on the acoustic absorption coefficient based on the
Van der Pol formula [8]. For the absorption coefficient, note that
the main parameters that define the acoustic coupling between
the two phases that comprise a porous material (i.e., porous
asphalt concrete) are: porosity (O, dimensionless), resistivity (Rs,
Ns/m4), tortuosity (q2, dimensionless), thickness (t, e.g., m), and
viscous and thermal factors. Porosity refers to connected voids
(O, dimensionless). Under several hypotheses it can be easily
related to air voids content. Resistivity can be derived from the air-
flow resistance (R, Nsm5). This latter is the ratio between the
pressure difference across a sample and the flow rate through
the sample. Tortuosity (which is dimensionless) refers to the
square of the ratio between pore lengths and sample length [31].
The parameters above (with the exception of the thickness) affect
the complex, dynamic density (qg, Ns2m4) and the bulk modulus
(kg, N/m2). The dynamic density and the bulk modulus (together
with the porosity of the material) determine the characteristic
impedance (Zc, Ns/m3) and the complex wave number (k, s/m).
In turn, the characteristic impedance and the thickness of the por-
ous layer determine the surface impedance (Z, Ns/m3). The acoustic
behaviour of a rigid-frame porous material (including its absorp-
tion coefficient) is completely characterized by its surface impe-
dance and its complex wave number:
jZ  qcj2
a0 ¼ 1  ð1Þ
jZ þ qcj2
where q is the air density and c is the speed of sound in air.
For the four most relevant parameters which impact the
absorption, note that (Table 1):

Thickness (t) is easy to measure. The higher the thickness is the
lower the frequency of the first maximum is. Absorption tends
to be lower and smoothed ([32]);

Porosity (O) is quite easy to measure (in the laboratory). The
higher the porosity is, the higher the absorption coefficient is.
Maximum frequency does not depend on porosity ([32]);

Resistivity (Rs) is quite difficult to measure. The higher the
resistivity, the lower the maxima, the smoother the curve
([32]);

Tortuosity (q2) is difficult to measure ([33–37]). The higher the
tortuosity is, the lower the frequency of maximum is. The
impact on the maximum value of absorption is usually quite
negligible ([32,38]).
Finally, note that, for the total resistance (Rt = Rs*t), the higher
the total resistance is, the higher the maxima are when Rt is lower
than about 100 Ns/m3. For Rt higher than about 100 Ns/m3, the
behaviour is opposite.
From a theoretical and computational standpoint, apart from
the overall complexity of models, it is very relevant to ‘‘confine”
both expected spectra and input values (O, q2, t, Rs). In this regard,
Porous European mixes (PEMs) and dense-graded friction courses
276 F.G. Praticò et al. / Construction and Building Materials 140 (2017) 274–281

Fig. 1. Main tasks. Notes. AD: Adrienne device (ISO 13472-1, see also [29]); KS: on site Kundt device (ISO 13472-2, method for reflective surfaces); KL: in-laboratory Kundt
device (ISO 10534-2, see also [30]).

Table 1
From design to physical parameters and how they affect the acoustic absorption of a pavement.

Design Physical parameter Main effect on a0(f) Easily Reference values

Aggregate gradation and Thickness (t, 0.01 m)
type, bitumen content
and type, compaction
process Porosity (O,%)
Resistivity (Rs, Ns/m4)
Tortuosity (q2)
Total resistance (Rt = Rs*t;
Ns/m3)
measurable
DGFC PEM
The higher the thickness is the lower the frequency of Y 2–4 4–6
the first maximum is. Absorption tends to be lower and
smoothed.
The higher the porosity is, the higher the absorption Y 4–8 16–30
coefficient is. Maximum frequency does not depend on
porosity.
The higher the resistivity, the lower the maxima, the N 0.6 M–30 M 1 k–60 k
smoother the curve.
The higher the tortuosity is, the lower the frequency of NA 1–10
maximum is. The impact on the maximum value of
absorption is usually quite negligible.
If 83 < Rt < 110, the maximum value of absorption (a0 See above See above
max) is 1; If Rt < 83, a0 max increases; If Rt > 100, a0 max
decreases ([32])

Fig. 2. PEMs and DFGCs envelopes of spectra based on different ranges of porosity, tortuosity, thickness and resistivity.

(DGFCs) represent the opposite ‘‘poles” which help split inputs/
output in different ranges. Table 1 lists: i) the main variables that
the civil engineer can handle in pavement construction (first col-
umn); ii) the four main inputs (physical parameters); iii) how they
affect a0; iv) how easy is to measure the four inputs; v) their refer-
ence values based on international literature and simulations. Note
that the choice of a given couple t, and X, based on experiments
(see Fig. 2), implies two (main) degrees of freedom, i.e. q2 and Rs.
This notwithstanding, q2 and Rs values cannot be varied outside
a reasonable range, which depends on mix typology and
physical-based boundaries (see Table 1 and [39]).
Based on the range of variation listed above, the envelopes of a0
can be derived as in Fig. 2 below.
The values of t, O, q2, Rs in Table 1 imply what follows: i) the
first point of maximum (frequency) is expected to vary from 400
to 1600 Hz, while the first maximum (absorption coefficient) is
expected to range from 0.5 to 1.0 for PEMs (Fig. 2A) and from
0.02 to 0.18 for DGFCs (Fig. 2B); ii) the product Rs·t = Rt (Ns/m3)
 F.G. Praticò et al. / Construction and Building Materials 140 (2017) 274–281
Fig. 3. Maximum values of a0 based on airflow resistance/resistivity (Rt, Rt), under given conditions.
plays a crucial role in terms of max a0 of a wearing course. For
example, if t = 0.03 m, O = 4–8%, q2 = 2–6 (DGFCs), then values of
max a0  0.2 can be obtained only for Rt > 20 KN/m3, i.e., for
Rs > 600 KN/m4 (see also Table 1). Fig. 3 illustrates this concept
in terms of Rt (Fig. 3A) and Rs (Fig. 3B).
Objective functions are needed to fit functions (hereafter ter-
med a0) to data (hereafter termed a0S). To this end the least square
method is usually used. In quality assurance/quality control of
quiet pavements, these functions are very relevant because they
allow verifying if data are significantly biased with respect to the
actual properties of a pavement. In the end, they may impact the
acceptance of the as-built pavement and have a great financial
effect.
For objective functions, note that when comparing theoretical
and experimental spectra, another issue refers to the method used
to fit data, because of the well-known tendency of the three mea-
surement methods not to provide significant outputs close to the
origin (frequencies lower than 400 Hz). For the given i-th fre-
quency, fi, (e.g. 800 Hz), the a0 (f) derived from four main parame-
ters (t, X, q2, Rs) may be compared with the a0S(f) observed
(measured, as mentioned above, e.g., through the KS). To this
end, in the following, these definitions apply (see Fig. 5):
i) a0S is the spectrum of the absorption coefficient as per mea-
surements carried out through a device (e.g., KL);
Fig. 4. Assessing data (device) consistency (close up of Task 4 and Task 3).
277
ii) a0(fi, t, O, q2, Rs) is the spectrum of the absorption coefficient
derived according to the theoretical model ([31]), through an
optimization process in which the four main variables were
set ‘‘free”, i.e., without constrains;
iii) a0tr = a0(fi, tex, Xex, q2ex, Rsex) is the spectrum of the absorption
coefficient derived according to the theoretical model ([31]),
through an optimization process in which two out of the four
main variables were constrained (t = tex; X = Xex, where the
subscript ex refers to the fact that cores were extracted and
used to this purpose).
When solved in terms of least square method, the objective
function of the over determined system (four or less unknowns
and much more than four equations) would be the following:
X 2
U1 ¼ min ½a0 ðf i ; t; X; q2 ; RsÞ  a0S ðf i Þ ð2Þ
i
Due to the fact that the theoretical model and the measure-
ments do not fit properly in given ranges, such as very low fre-
quencies, the following additional class of objective functions is
considered in this study:
n n
U2 ¼ minbða20  a20S Þ 1  ða20 þ a20S Þ 2 Þc ð3Þ
with n1 and n2 natural and even numbers. This formula builds on
the higher reliability and soundness of high values with respect to
low values. The sum of squared values corresponds to this rationale.
278 F.G. Praticò et al. / Construction and Building Materials 140 (2017) 274–281
Fig. 5. Device-based results (s) versus ‘‘true” results (ex) for inputs (Fig. 5A) and for outputs (Fig. 5B).
Furthermore, due to the complexity of the process and being the
shape of the absorption coefficient essentially governed by the first
maximum point (in terms of abscissa and ordinate), the following
objective function is set up:
0:5
U3 ¼ min½ða0maxS  a0maxex Þ2 þ ðnfmaxS  nfmaxex Þ2  ð4Þ
where the normalised values of frequency are obtained as follows:
f max  min f max
nfmax ¼ : ð5Þ
max f max  min f max
Note that nfmaxtr and nfmaxS range from 0 to 1, while the normal-
ization of the absorption coefficient is not required being this
parameter already in 0, 1. The rationale behind the objective func-
tion above (Eq. (4)) is a very simple Euclidean distance between the
data (subscript S) and the theoretical-based spectrum (subscript
ex). The normalization process (Eq. (5)) is needed to correct the
bias that higher/lower frequencies would introduce.
Apart from the particular form of the objective function, a visual
comparison (data versus calibrated model) is always needed. Note
that Eqs. (3)–(5) are new (herein proposed by the authors) while
Eqs. (1) and (2) are based on the literature.
Fig. 4 illustrates the procedures used to assess device consis-
tency. In terms of metric and accuracy, note that precision refers
to random errors while accuracy often refers to systematic errors
(i.e., statistical bias), which is the meaning herein intended. Note
that ISO 5725-1 refers to systematic errors in terms of trueness,
while associates the term accuracy to trueness and precision.
Assessing the accuracy of a measurement method /technique
(e.g., KS) is a very difficult task. Different devices may give different
spectra for the same pavement and considering a method better
than another one based on use or specifications, may be incoherent
and lead to wrong decisions in noise management. The following
tentative method is here proposed and applied (Fig. 4). By fitting
the experimental data without imposing an upper and lower spec-
ification limit to each of the four parameters (unknowns free, i.e., t,
q2, O, Rs free), a given spectrum of the absorption coefficient can be
obtained (corresponding to a given objective function / as per Eqs.
(2)–(4), see path A B C of Fig. 4). On the other hand, under the
hypothesis of X = Xex, t = tex (i.e., thickness and porosity experi-
mentally determined (cores extracted from the pavement section),
different values for the remaining unknowns (Rs, q2) and for the
objective function chosen (/, see path A, D, E of Fig. 4) are obtained.
If the values of q2 and Rs obtained through the optimization are
unrealistic or/and an excessive increase of the objective functions
(Eqs. (2)–(4)) is obtained, then the experimental data may be
wrong (e.g. the data are unrealistic because the device does not
work properly), and viceversa. And if a given device data imply a
bias in t or X (experimental data) then it is possible to conclude
that there is a given bias in terms of working principle (see path
F G H).
5. Results and observations
In terms of objective functions, the least square method (Eq. (2))
was compared to Eq. (3). It may be observed that the introduction
of an average-based multiplier (Eq. (3)) optimised the reliability of
results. Indeed, when using objective functions such as (Eq. (3)),
porosity (as a result of minimization process) approached the
experimental data (i.e., Oex). The lesson learned is as follows: i)
neglecting or undervaluing lower data (in practice, the ones which
refer to lower frequencies) through average-based objective func-
tions can help optimise data versus theoretical model fitting; ii)
to this end, having n1 = n2 = 2 (Eq. (3)) provides satisfactory results.
Figs. 5–8 refer to the analyses carried out in terms of method
accuracy and precision.
Fig. 5 refers to the absolute value of the difference between
‘‘true” (ex) values and ‘‘observed” (s) data (e.g., │Os–Oex│), as a
part of the process depicted in Fig. 4. Importantly, it is noted that:
i) X-axes refer to the differences difference between ‘‘true”
(ex) values and ‘‘observed” (s) data, for thickness, porosity,
tortuosity, resistivity, maximum absorption coefficient, and
peak frequency (frequencies were divided by 1000);
ii) Oex and tex were measured on cores, while q2ex and RSex were
derived based on data fitting (being thickness and porosity
constrained);
iii) OS, tS, q2S, RSS, a0S refer to the best (and free) data fitting;
iv) The plot depicted on the right (Fig. 5B) refers to the output in
terms of absolute maximum of the spectrum.
a0
f (Hz)
Fig. 6. Estimation of the first peak of the acoustic absorption: how the three devices
work.
 F.G. Praticò et al. / Construction and Building Materials 140 (2017) 274–281 279

KL

AD
ex
S KS

A B
Fig. 7. Comparing expected (ex) and observed (S) spectra in terms of maxima (Fig. 7A) and error increase (Fig. 7B).

Fig. 8. Main statistics for each range of frequency.

In other words, the subscript S refers to observed spectra
(‘‘measured”) while ex refers to measured (Oex, tex) or
‘‘quasi-measured” (q2ex, RSex) material characteristics .
For example, on average, the value of the difference between
the true value of the thickness (tex as per core height) and the thick-
ness value which best fit the obtained spectrum (tS) is 0.04 for AD,
0.19 for KS, and 0.13 for KL (Fig. 5A).
Based on results, it may be observed that:

KL tends to underestimate thickness, as well as KS and AD (in
fact texts > 0, Fig. 5A). On average, for thickness, AD and KL
have the best performance and KS has the worst performance
(the lower is the bar the better is the result).

Overall, for porosity, KL performs the best while AD the worst
(AD overestimates the porosity).

It may be observed that KS and AD do not ‘‘see” the right poros-
ity. In contrast, KL better ‘‘sees” t and X.

For tortuosity, note that AD performs better than KL and KS.

For resistivity, note that AD and KS perform worse than KL.

By referring to the detection of the (first) maximum point
(frequency and value, Fig. 5B), which has an outstanding impact
on all the remaining frequencies, note that the differences
among the three devices seem relevant only in terms of
frequency (KL and KS perform better than AD).

The frequency of maximum (maximum point) of AD is usually
lower than the corresponding value for KS and KL. On average
KL offers the best estimate of the point of maximum together
with KS. In more detail, for the first maximum (aomax), note that
AD and KL perform better than KS.
In summarising it is possible to observe that:

the in-lab procedure (KL) may entail: i) a destructive procedure,
being the core extracted from the pavement; ii) splitting the
core into single layers (i.e., porous asphalt concrete and binder
course); iii) the treatment of the bottom surface of the porous
asphalt concrete (in terms of milling and/or planning through
plastiline or similar materials; iv) The treatment of the lateral
surface in order to fill/smooth possible irregularities. This
notwithstanding, KL is quite accurate.

KS clearly diverges from the expected values in the low fre-
quency domain (f < 600 Hz). Its accuracy is neither the best
nor the worst. Test duration is low.

AD tends to underestimate thickness and overestimate porosity
(i.e., Xex < Xs, see Fig. 5A). It offers a satisfactory estimate of a0
(Fig. 5B), a wider range of frequency, and the test is very time
consuming.
Scatter plots in Fig. 6 refer to the maxima. Y-axis refers to the
maximum value of the absorption spectrum and x-axis refers to
the frequency (Hz). Maxima range from 0.6 to 0.8 and frequencies
from 700 to 1200 Hz. It points out that the assessment of the peak
frequency may be an issue.
Fig. 7A still refers to how the three devices estimate the first
peak of absorption (as in Fig. 6). Importantly, in this case, a 3-D
scatter plot (or bubble chart) is used, which illustrates the relation-
ship among the spectra obtained based on the measurements (S)
(‘‘filtered” using the theoretical model), the ones, true (ex), based
on experimental data for thickness and porosity, and method pre-
cision. Y-axis refers to the maximum value of the absorption spec-
trum (a0) and x-axis refers to the normalised frequency (nf, Hz/Hz,
equation 4). Consequently, both the variables range from 0 to 1.
The disks’ radii is scaled to the square root of the standard
280 F.G. Praticò et al. / Construction and Building Materials 140 (2017) 274–281
deviation and not to the standard deviation in the pursuit of avoid-
ing misinterpretations (indeed the human visual system naturally
experiences a disk’s size in terms of its area). The standard devia-
tion is derived from the one of a0 and the one of nf in terms of
square root of the sum of squares (due to their variation in the
same range). When data measured on cores (porosity and thick-
ness) are used in the theoretical model (ex), AD spectra undergo
appreciable changes in frequency. In contrast (towards higher fre-
quencies), KL and KS remain close to the observed spectrum (S). For
KL and KS, the bias in porosity and thickness does not imply appre-
ciable biases nor in terms of maximum value (y-axis) neither in
terms of frequency.
Finally, Fig. 7B illustrates the increase of errors (bubble area)
when true values of thickness (t) and porosity (O) are used. In more
detail, x-axis refers to thickness (0.1 m), y-axis refers to porosity,
while the area of the bubbles refers to the error. This latter relates
to the minimization process carried out in order to make the the-
oretical curve fit the data).
In summarising, it may be observed that while KS and AD move
from the left to the right part of the plot, KL moves from above to
below and its radius (error) undergoes the lowest variation among
the three devices under consideration.
Even if additional analyses are needed, this fact seems to imply
that for KL the appreciable error in porosity is balanced in terms of
variation in the remaining two input parameters (i.e., q2 and Rs).
This circumstance finally implies the lowest error and probably
this indicates the best accuracy.
For precision, note that the hypothesis which underlies this
analysis is the homogeneity of the surface. Fig. 8 illustrates the
main results (averages in Fig. 8A; coefficients of variation in
Fig. 8B). Note that X-axes refer to frequency range, while y-axes
report averages (dimensionless) and coefficient of variations
(dimensionless). These plots are very relevant for practical pur-
poses (QA/QC), because usually contract specifications define
thresholds per frequency range. When different frequencies are
considered, the following observations can be pointed out: i) KL
and KS usually yield values of absorption which are lower than
the ones of AD (Fig. 8A); ii) By referring to the highest and lowest
values of the absorption spectra (peaks), for each frequency, for
each device, note that all the devices yield an absolute maximum
at about 800-1250 Hz. It may be observed that the lower the fre-
quency is the higher the discrepancy among the three outputs
becomes (Fig. 8A); iii) The precision approaches lower values in
the range 400–630 Hz, while it increases in 800–1600 Hz. In more
detail, at 250 Hz (not represented in Fig. 8), the in-lab device (KL)
provides the lowest value (absorption coefficient about 0.1) while
the on-site Kundt tube KS provides the highest value (about 0.3).
Furthermore note that the standard deviation is the lowest (high-
est precision) for KS. In contrast, for AD, the lowest precision is
obtained. iv) Overall, although the standard deviation yields that
the three devices are quite comparable in terms of precision
(0.08–0.14, not represented in Fig. 8), in terms of coefficient of vari-
ation (Fig. 8B), in 400–630 Hz, KL (in-lab measurements) results
are less precise than the remaining two methods (KS, AD).
6. Conclusions and reccomendations
The following conclusions may be drawn:

The least square method overrates the importance of low fre-
quencies and originates an unsatisfactory fitting of the peak fre-
quency and value. Equations herein set up work properly
regardless of the type of device used and help provide informa-
tion about method trueness and accuracy.

AD builds on the assumption that the quantity of energy
reflected by the pavement surface depends on its acoustic
absorption. It performs measurements in 0–4000 Hz and it is
quite time-consuming. It has a coefficient of variation of
0.1–0.4 and a good estimate of thickness, tortuosity, resistivity
and maximum.

KS builds on the fact that pressures into the tube are affected by
the sample absorption and it is supposed to work properly only
for reflective surfaces (as per standard specifications). It pro-
vides a straightforward tool for practitioners, its coefficient of
variation ranges from 0.1 to 0.4 and its estimate of porosity is
better than the one of AD (even if worse than in the KL).

KL is based on the same principle used in KS and it is quite time
consuming (due to coring, transport, drying, and finally core fix-
ing in the tube). Several issues may arise in fixing the core
inside the tube, generating false responses. It has the worst
coefficient of variation (0.3–0.5), but builds on the best estimate
of porosity and resistivity. Even if further studies and analyses
are required, based on the analyses of errors above, it seems
to achieve the best accuracy.
Based on this study, recommendations are as follows:

In the pursuit of investigating the trueness of absorption spectra
in quality assurance/quality control process, attention must be
focused on the fitting of the ‘‘pick(s)” (maximum(a) point(s))
and the least square method must not be used.

It is suggested to have both on site and in-lab measurements.
The best accuracy is provided by the in-lab measurement and
this latter should be used to calibrate the acceptance plan based
on on-site measurements.

In terms of normative elements (requirements, recommenda-
tions and statements), it is advised to carry out further investi-
gations and analyses to study the possibility to use the on-site
Kundt tube also for porous surfaces.
References
[1] European Commission, Directive on environmental noise, 2002/49/CE, 25th
June 2002, DOCE L 89, 7/18/02.
[2] SMILE workshop, Guidelines for road traffic noise abatement, 2003. Berlin,
Germany, <http://www.smile-europe.org/>.
[3] E. Freitas, C. Mendonça, J.A. Santos, C. Murteira, J.P. Ferreira, Traffic noise
abatement: how different pavements, vehicle speeds and traffic densities
affect annoyance levels, Transp. Res. Part D 17 (2012) 321–326.
[4] F.G. Praticò, F. Anfosso-Lédée, Trends and Issues in Mitigating Traffic Noise
Through Quiet Pavements, SIIV. in: 5th International Congress – Sustainability
of Road Infrastructures, Procedia – Social and Behavioral Sciences, 2012.
E201994, 1887–0428, 1994.
[5] T. Kihlman, W. Kropp, W. Lang, Quieter cities of the future. Lessening the
severe health effects of traffic noise in cities by emission reductions.
International Council of Academies of Engineering and Technological
Sciences, CAETS, 2014.
[6] U. Sandberg, J.A. Ejsmont, Tyre/Road Noise Reference Book, Informex, Kisa,
Sweden, 2002.
[7] K. Attenborough, Acoustical characteristics of rigid fibrous absorbents and
granular materials, in: JASA 73 (1983).
[8] F.G. Praticò, Roads and loudness: a more comprehensive approach, J. Road
Mater. Pavement Des. 2 (4) (2001).
[9] M. Frías, J.M. Jiménez-Mateos, J. Pfretzschner, J. Olmeda, R.M. Rodríguez, M.I.
Sánchez de Rojas, Development of blended cement mortars with acoustic
properties using petroleum coke, Constr. Build. Mater. 25 (2011) 1086–1092.
[10] Jimenez-Espadafor et al., Optimal design of acoustic material from tire fluff,
Mater. Des. 32 (2011) 3608–3616.
[11] F.G. Praticò, R. Vaiana, A study on the relationship between mean texture
depth and mean profile depth of asphalt pavements, Constr. Build. Mater. 101
(30) (2015) 72–79.
[12] F.G. Praticò, R. Vaiana, T. Iuele, Macrotexture modeling and experimental
validation for pavement surface treatments, Constr. Build. Mater. 95 (2015)
658–666.
 F.G. Praticò et al. / Construction and Building Materials 140 (2017) 274–281
[13] J. Luong, M. Bueno, V.F. Vázquez, S.E. Paje, Ultrathin porous pavement made
with high viscosity asphalt rubber binder: a better acoustic absorption?, Appl
Acoust. 79 (2014) 117–123.
[14] G. Licitra, L. Teti, M. Cerchiai, A modified close proximity method to evaluate
the time trends of road pavements acoustical performances, Appl. Acoust. 76
(2014) 169–179.
[15] F.G. Praticò, A. Moro, R. Ammendola, Potential of fire extinguisher powder as a
filler in bi-tuminous mixes, J. Hazard. Mater. 173 (1–3) (2010) 605–613.
[16] F.G. Praticò, A. Moro, In-lab and on-site measurements of hot mix asphalt
density: convergence and divergence hypotheses, Constr. Build. Mater. 25 (2)
(2011) 1065–1071.
[17] F.G. Praticò, R. Vaiana, A study on volumetric versus surface properties of
wearing courses, Constr. Build. Mater. 38 (2013) 766–775.
[18] F.G. Praticò, R. Vaiana, Improving infrastructure sustainability in suburban and
urban areas: Is porous asphalt the right answer? And how?, WIT Trans Built
Environ. (2012).
[19] V.F. Vázquez, J. Luong, M. Bueno, F. Terán, S.E. Paje, Assessment of an action
against environmental noise: acoustic durability of pavement surface with
crumb rubber, Sci. Total Environ. 542 (2016) (2016) 223–230.
[20] F.G. Pratico, R. Vaiana, A. Moro, Dependence of volumetric parameters of hot-
mix asphalts on testing methods, J. Mater. Civil Eng. 26 (1) (2014)
45–53.
[21] M. Lia, W. van Keulenc, E. Tijsd, M. van de Vena, A. Molenaar, Sound absorption
measurement of road surface with in situ technology, Appl. Acoust. 88 (2015)
12–21.
[22] D.J. Hanson, R.S. James, C. NeSmith, Tire/Pavement Noise Study, NCAT Report
04–02, August 2004.
[23] G. Liao, M.S. Sakhaeifar, M. Heitzman, R. West, B. Waller, S. Wanga, W. Ding,
The effects of pavement surface characteristics on tire/pavement noise, Appl.
Acoust. 76 (2014) 14–23.
[24] T. Fujiwara, S. Meiarashi, Y. Namikawa, M. Hasebe, Reduction of equivalent
continuous A-weighted sound pressure levels by porous elastic road surfaces,
Appl. Acoust. 66 (7) (2005) 766–778.
[25] M. Kayhanian, P.T. Weiss, J.S. Gulliver, L. Khazanovich, The application of
permeable pavement with emphasis on successful design, water quality
benefits, and identification of knowledge and data gaps, A summary report
from the National Center for Sustainable Transportation, June 2015.
281
[26] W. Gardziejczyk, The effect of time on acoustic durability of low noise
pavements – the case studies in Poland, Transp. Res. Part D 44 (2016) 93–104.
[27] C. Sangiorgi, P. Tataranni, A. Simone, V. Vignali, C. Lantieri, G. Dondi,
Assessment of waste bleaching clay as alternative filler for the production of
porous asphalts, Constr. Build. Mater. 109 (2016) 1–7.
[28] K.P. Biligiri, Tyre/road noise damping characteristics using nomographs and
fundamental vibroacoustical relationships, Transp. Res. Part D 43 (2016) 82–
94.
[29] Niranjan Londhe, Mohan D. Rao, Jason R. Blough, Application of the ISO 13472–
1 in situ technique for measuring the acoustic absorption coefficient of grass
and artificial turf surfaces, Appl. Acoust. 70 (1) (2009) 129–141.
[30] M. Wolkesson, Evaluation of Impedance Tube Methods. A Two Microphone In-
situ Method for Road Surface and the Three Microphone Transfer Function
Method for Porous Materials, Department of Civil and Environmental
Engineering, division of Applied Acoustics, Vibroacoustics Group. Chalmers
University of Technology, Goteborg, Sweden, 2013.
[31] M.R. Stinson, Y. Champoux, Propagation of sound and the assignment of shape
factors in model porous materials having simple pore geometries, J. Acoust.
Soc. Am. 91 (2) (1992).
[32] Bereinger, M., Hamet, J.-F., Propriétés acoustiques des enrobés drainants. Le
phénomène d’absorption. Bullettin des liaison des ponts and chausses, 1990.
[33] J. Bear, Dynamics of Fluids in Porous Media, American Elsevier, New York,
1972.
[34] F.A.L. Dullien, Prediction of ‘‘tortuosity factors” from pore structure data, AIChE
J. 21 (1975) 820–822.
[35] L. Pisani, Simple expression for the tortuosity of porous media, Transp. Porous
Media 88 (2) (2011) 193–203.
[36] M.M. Ahmadi, S. Mohammadi, A.N. Hayati, Analytical derivation of tortuosity
and permeability of monosized spheres: a volume averaging approach, Phys.
Rev. E 83 (2 Pt 2) (2011) 026312.
[37] A.E. Khabbazi, J. Hinebaugh, A. Bazylak, Determining the impact of rectangular
grain aspect ratio on tortuosity–porosity correlations of two-dimensional
stochastically generated porous media, Sci. Bull. 61 (8) (2016) 601–611.
[38] F.G. Praticò, On the dependence of acoustic performance on pavement
characteristics, Transp. Res. Part D 29 (2014) 79–87.
[39] M.-J. Crocker, Springer Handbook of Acoustics, edited by Thomas Rossing,
2014.