Applied Acoustics 133 (2018) 123–132

Contents lists available at ScienceDirect

Applied Acoustics
journal homepage: www.elsevier.com/locate/apacoust

Time-evolution of sound levels around a roadside building T

a,⁎ a b b
Dietrich Heimann , Arthur Schady , Roland Schuster , Tobias Berkefeld
a
German Aerospace Center (DLR), Institute of Atmospheric Physics, Wessling-Oberpfaffenhofen, Germany
b
German Aerospace Center (DLR), Institute of Aerodynamics and Flow Technology, Göttingen, Germany

A R T I C L E I N F O A B S T R A C T

Keywords: The time evolution of the sound impact due to single cars driving along a road was evaluated at seven receivers
Sound propagation in front of and behind a road-side building. The sound impact was measured in the field and around a scale
Diffraction model of the building in a laboratory. In addition, two calculation methods were applied: a complex numerical
Building finite-difference time-domain model based on the Euler equation and a simple ray-based calculation according to
Measurement
the ISO 9613-2 standard. On the backside of the building where sound waves only arrive by diffraction, the
Simulation
differences between the four methods are disappointingly large. Some of the discrepancies can be explained by
diverse assumptions and boundary conditions which go along with the individual methods. However, a major
disagreement largely remains an open problem, namely the magnitude of the drop of the sound level in the
acoustical shadow of the building. It is fairly strong in the scale-model measurements and the Euler model
simulations on the one hand, but rather weak or even not existing in the field measurements and ISO 9613-2
calculations on the other hand.

1. Introduction
Road traffic noise is a major concern in many European countries
and worldwide. Most traffic-noise prediction procedures and national
standards only provide energy-equivalent average sound levels Leq on
the base of traffic parameters for roads and railway lines. However, the
assessment should also be interpretable with respect to possible noise
effects on human beings such as health impairments, sleeping dis-
turbances or annoyance. Conventional exposure-response relationships
rely on average sound levels (Miedema and Voss [1]; Miedema and
Oudshoorn [2]; Lercher et al. [3]). More recent studies also consider
event-related noise parameters which characterize the time evolution of
sound levels like the number of events, the maximumlevel, the noise
duration or the sound-level change rates (e.g. Basner et al. [4]). El-
menhorst et al. [5] found that the maximum level of an event and the
maximum level rise are highly significant indicators of sleep dis-
turbances caused by railway noise. For annoyance they found that the
Leq alone is not a significant predictor, while the number of events in
combination with non-acoustical parameters is significant.
If the vicinity of a road is free of obstacles, the temporal variation of
the sound level due to moving vehicles can be easily calculated using
analytical formulae (Heimann and Schady [6]). However, buildings and
other obstacles lead to reflections and shading. Some sound may pro-
pagate through gaps between the obstacles, either directly or by dif-
fraction. This geometry often results in a rather irregular time
behaviour of the sound level at receivers between and behind a road-
side development. The average sound level, maximum level and max-
imum change rate are affected by the buildings and thus differ from
those in unobstructed situations.
Diffraction is the key process that enables sound to propagate into
areas which are shaded by buildings or other obstacles. Diffraction
theory has been made progress during the 1970ies and 1980ies
(McNamara et al. [7]). Numerical propagation models that solve the
wave equations explicitly account for diffraction (e.g. propagation over
a barrier; Salomons [8]; Heimann and Blumrich [9]). However, ray-
based models often use simplified heuristic diffraction models. These
calculate the diffractive attenuation based on the difference between a
diffracted ray from source to receiver around and over an object and a
direct ray between source and receiver neglecting the object. These
models can also be described as detour models. A simple detour ap-
proach is implemented in the ISO 9613-2 standard [10] to calculate
sound levels in the acoustical shadow of screens and buildings. A re-
lated but slightly more complex diffraction model is described in the
Harmonoise project (Salomons et al. [11]). It is a detour model which
calculates the diffracted ray based on Fresnel zones. A more sophisti-
cated approach was introduced by Wei et al. [12].
Sound propagation over a single building was investigated by
Alberts II et al. [13]. In another paper Remillieux et al. [14] studied the
sound propagation at an isolated building not only in the field, but also
compared the results with numerical simulations. In both papers

⁎
Corresponding author.
E-mail address: d.heimann@dlr.de (D. Heimann).

https://doi.org/10.1016/j.apacoust.2017.12.011
Received 31 July 2017; Received in revised form 14 November 2017; Accepted 11 December 2017
0003-682X/ © 2017 Elsevier Ltd. All rights reserved.
D. Heimann et al.
artificial sources of impulsive broadband sound were used. However,
the time-history of sound levels according to moving sources (e.g. cars)
were not addressed.
The present study investigates a real situation with a roadside
building with four different methods: sound from moving sources is
studied by (1) field measurements, (2) laboratory scale-model mea-
surements, (3) advanced numerical sound propagation modelling, and
(4) engineering-type sound prediction. The resulting sound level var-
iations in space and time are compared and deviations are discussed. So
far, such a setting has not been in the focus of previous publications,
although dynamic simulations of traffic sound in built-up areas were
recently published. Most of these papers aim at probability statistics of
sound levels in urban areas with many buildings (e.g. Wang et al. [15],
Walker et al. [16]) and therefore do not really compare with the present
study. Hou et al. [17] discuss measured and simulated sound levels at a
single roadside building. However, the building in their study is of ra-
ther complex shape and only two of totally six measuring points are
located outdoors, one near the road and another one on the backside of
the building. Unfortunately, time-series are only shown in rather coarse
resolution (1 s) and for the point near the road and two indoor points.
Temporal variations of the sound level are due to fluctuating traffic
rather than due to shading by the building.
2. Case study
2.1. General situation
The study considers a single two-storey building with a saddle roof
A 96
ay
ilw
ra
an
rb
bu
su
68
20
St
DLR
0 100
124
Applied Acoustics 133 (2018) 123–132
located close to a two-lane state road (St 2068) about 20 km west of
Munich/Germany (Fig. 1). The building belongs to the Oberpfaffen-
hofen site of the German Aerospace Center (DLR) and is used as a
kindergarten. The detailed geometry of the building and its position
relative to the road is shown in Fig. 2. Photos are provided in Fig. 3.
They show that the building (A) is not symmetric. In particular, there is
a one-storeyed annex (B) and a small separate flat-roofed housing of a
natural-gas distributor (C) on the northern side. The nearest facade of
buildings outside the area shown in Fig. 2 is about 55 m to the southeast
(cf. Fig. 1).
The building complex A-B-C stands in a grass garden, but its im-
mediate surrounding is paved. Between building and road there is a
green strip with bushes and a few single trees. The road is 7 m wide and
paved with asphalt. The speed limit is 70 km/h. On the side of the
building, the road is accompanied by a paved pedestrian/bike way.
Except the state road, there are other potential sources of noise: a
suburban railway line in the west (closest distance 320 m), the federal
motorway A96 in the north (closest distance 620 m), and the Special
Airport Oberpfaffenhofen (ICAO-code: EDMO) in the east (closest dis-
tance to the runway 700 m).
2.2. Field measurements
Acoustical measurements were performed during the weekend
12–13 March 2016. The evaluation was restricted to selected drive-by
events of vehicles during the second half of the weekend nights (11/12,
12/13 and 13/14 March). Extraneous noise is minimized at this time.
The kindergarten is closed on weekends and the airport is not operating
Fig. 1. Topographical map showing the mea-
surement site (encircled) within the DLR
Oberpfaffenhofen research campus (enclosed by
dotted line) and its vicinity with the state road
A 96 to Munich St2068, the motorway A96, the suburban railway
and the airport runway (RWY).
2
/2
04
Y
RW
O
M
ED
N
200 300 400 500 m
© OpenTopoMap (CC-BY-SA)
D. Heimann et al. Applied Acoustics 133 (2018) 123–132

10 m
St 2068

bikeway
20 m
x

fence
+30 m

C
+20

P7
T
B P6

80 m
P5
+10

A 4
P2 P3 P
P1
0

y
10 20 30 m
-10
-20

50 m
Fig. 2. Sketch of the state road St2068, the building (A: main building, B: annex, C: gas
distributor housing) and the measurement positions (P1–P7). The hatched lines at
building “A” indicate ridge and overhang of the saddle roof. “T” is the automatic traffic
counter with its sensor range. The coordinate x points into the driving direction. The
coordinate y provides the normal distance from the center of the lane. The outer frame of
the sketch corresponds to the numerical model domain (see Section 2.4).

during night. Also train operation is ceased during the second half of
the night. The remaining nightly background noise varies between 38 Fig. 3. Photos of the building from the southeast (a), northeast (b), and north (c). Panel
and 42 dB(A). It mainly originates from the motorway or is wind-in- (c) shows the traffic counter mounted at the fence.
duced vegetation noise.
Four class-1 sound level meters were installed at seven positions Table 1
(P1–P7) in front of and behind the building (Fig. 2). Only at P1 and P3 Measuring positions (cf. Fig. 2) of the four sound meters. CET = Central European Time.
the devices were not moved and measurements are permanently
available. The other two devices were temporarily installed at the other 2016 Device 1 Device 2 Device 3 Device 4

positions so that always four positions were occupied at once (Table 1).
from to Norsonic 121 Topsonic TOPNMS15
The microphones are fixed 2 m above ground at P1, P4, and P7 and
1.5 m above ground at P2, P3, P5, and P6. The measurement position 11 March17, 12 March15, P2 P3 P1 P4
P1 is the nearest to the road and without obstacles towards the road CET CET
12 March17, 13 March14, P5 P3 P1 P4
(except the chain-link fence seen in Fig. 2c). P1 serves as a reference for
CET CET
the sound emission. 13 March16, 14 March07, P6 P3 P1 P7
All four sound meters recorded the A-weighted sound pressure level CET CET
with ‘fast’ time weighting. Two devices (Type 1: Norsonic 121) re-
corded the level in 1/8-s intervals including spectral analysis with 1/3-
octave resolution (8 Hz–16 kHz). The other two devices (Type 2: In addition to the sound also meteorological parameters (wind,
Topsonic TOPNMS15) recorded the level as energy-equivalent average temperature, relative humidity) were measured at position P4 and P7.
sound level for each 1-s interval together with the sound pressure at a The traffic was monitored by an automatic radar-based traffic counter.
rate of 44,100 s−1. For these sound meters the sound level in 1/8-s It was mounted at the fence between P1 and the road. The traffic
intervals and the frequency spectrum in 1/3-octave resolutions were counter recorded the time of the day, driving direction, speed and
derived by post-processing. length of each vehicle passing by. From the latter the vehicle type
(motorcycle, passenger car, delivery van, and truck) was derived using

125
D. Heimann et al.
a statistical categorization.
A selection of 13 drive-by events of northeastbound passenger cars
was chosen for further evaluation. To minimize the probability of ex-
traneous noise (rail, aircraft, birds, wind noise), the events were se-
lected such that all these events took place between 01 and 03 a.m.
Central European Time and in weak-wind situations. In addition, the
selection ensures that there are no other cars within a 5-min window
around each event so that the sound of a selected event is caused by just
one single car. The speed of the cars ranged between 62 and 77 km/h.
The mean speed of the 13 cars is 70.15 km/h. During the events the
wind speed (2 m above ground) varied between 1 and 3 m/s from
northeasterly directions (0° to 90°). The temperature ranged between 2
and 4 °C and the relative humidity between 75 and 90%.
Correspondingly, the adiabatic sound speed amounted to about
c = 330 m/s . According to the prevailing wind direction the sound of
the northeastward moving vehicles was subject to upwind propagation
during approach and downwind propagation during departure.
2.3. Scale-model measurements
To gain reference data of the diffraction problem of sound propa-
gation around and over a building an experimental study has been
carried out in the anechoic chamber of the DLR in Göttingen. The an-
echoic chamber is a Class 1 absorbent chamber according to ISO 3745
[18]. Its room dimensions are 7 × 7 × 6 m3 . Due to the limited space, a
1:25 scaled setup is used. The scale model was manufactured of wooden
multiplex panels with a thickness of 18 mm. The maximum height of
the saddle-roof building A is 0.28 m at the top and the height of the
annex building B is 0.18 m. The building models are placed in the
center of the room. In order to simulate a quasi-steady moved noise
source a traverse is mounted alongside the house model. A loudspeaker
is mounted on the traverse 2 mm above the ground, pointing down-
ward. During measurement, it is moved to 35 equally spaced positions
along the traverse. The spacing amounts to 0.208 m which corresponds
to 5.2 m in reality. At each position the loudspeaker is excited for 15 s
using a signal generator together with a power amplifier to generate a
broadband noise signal. An analogous bandpass filter is used with a
Butterworth filter function 48 dB/octave. The cut-off frequencies are set
to 2000 Hz and 60 kHz. Seven ¼″-microphones with a preamplifier are
mounted 6 cm above the ground. The microphone signals are recorded
with a 16-bit data acquisition system. The sampling rate is set to
200 kHz. Data acquisition and signal generator modules are placed
outside the anechoic chamber to prevent acoustical disturbance during
measurement. A reference microphone is placed on the traverse, which
is moved together with the loudspeaker. Its signal is used to cross-
correlate with the microphone array signals to filter out uncorrelated
noise. In Fig. 4 the experimental setup with the line of loudspeaker
positions and the locations of microphones is shown.
The measurement data processing is performed block-wise. All mi-
crophone signals are divided into blocks which have a length of 16,384
samples. The data in each block are weighted using the Hann window
function. The auto- and cross-spectral densities are calculated using
FFT. They are averaged over all blocks resulting from 15 s measure-
ment. Differences of the atmospheric absorption between ray tracer and
measurement due to the scaling are not taken into account.
As the experimental setup is scaled by 1:25, the spectrum of the
loudspeaker needs to be scaled as well. The non-scaled frequency range
of 100 Hz–2 kHz is scaled to a frequency range of 2.5–50 kHz. A com-
pression driver of the type BMS 4540-8 was chosen to satisfy this re-
quirement. It is further adjusted in two ways. Firstly, to reduce noise
emissions from the back of the compression driver, it is placed in an
acoustically sealed housing. The housing consists of several wooden
layers. In between these layers a heat conducting aluminium foil is
laminated to increase the cooling. Secondly, to approximate a point
source, the sound radiating surface of the compression driver is re-
duced. An adapter is produced and screwed onto the compression
126
Applied Acoustics 133 (2018) 123–132
driver. The adapter connects the compression driver with a thin pipe.
The sound leaves the acoustic pipe at the end on a diameter of 2.1 mm.
2.4. Euler model simulations
A three-dimensional finite difference time-domain (FDTD) model
based on the linearized Euler equation (Blumrich and Heimann [19])
was applied to the measured situation. The basic model equations for
sound propagation in a calm atmosphere provide the time derivatives
∂U 1
= − ∇p
∂t ρ0 (1)
∂p
= −κp0 ∇ ·U
∂t (2)
of the sound velocity vector U and the sound pressure p . ρ0 and p0 are
the atmospheric density and the atmospheric pressure, respectively.
κ = 1.4 is the ratio of the specific heat capacities of air at constant
pressure and constant volume. The equations are solved by time in-
tegration using a six-step 4th order Runge-Kutta scheme after Hu et al.
[20] in combination with an 11-point 6th order selective filter after
Bogey at al. [21]. Spatial derivatives are approximated by 7-point
central differences according to Bogey and Bailly [22]. Impedance
boundary conditions (Heutschi et al. [23]) are used at the ground and
advective boundary conditions (Bogey and Bailly [24]) are im-
plemented at the lateral and top limits of the model domain.
Simulations were performed in a 50 × 80 × 16 m3 large domain
(Fig. 2) with 0.1 m resolution corresponding to 64 million grid cells.
Sound propagation was calculated for 15 surface-based source points at
the center of the northeastbound lane of the road with an equal spacing
of 5 m in separate simulations. Each simulation was initialized with a
Gaussian sound pressure pulse. The amplitude of the pulse was set to an
arbitrary value as the results are subject to subsequent normalization.
The spectrum of the broadband pulse has an energy peak at 200 Hz. The
numerical differencing scheme allows to evaluate the spectrum down to
a wavelength of approx. five-times the grid spacing, i.e. up to a fre-
quency of about 650 Hz.
The building with all parts A, B, and C (Fig. 2) is represented by
setting U = 0 at the respective grid cells. This corresponds to totally
reflecting walls. The simulated sound pressure was stored at the grid
cells which correspond to the microphone positions P1 to P7. Here, the
sound level was determined during postprocessing.
The Euler model is capable of calculating the effects of diffraction
explicitly, i.e. without parameterization. This is demonstrated by
comparing its results with laboratory scale-model measurements and
analytical solutions based on diffraction theories as published by Kawai
[25]. Fig. 5a/b shows sufficient coincidence for the diffracted sound
above and behind two laboratory-scale house-shaped fully reflecting
barriers on non-reflecting ground. The results refer to a 500 Hz tone.
Scaled-up to the size of the real building A with a scaling factor of 5.68,
this corresponds to a frequency of 88 Hz.
2.5. Noise prediction
To compare the measurements data with noise mapping calcula-
tions, the noise calculation standard ISO 9613-2 was applied as it is
implemented in the CadnaA software [26]. For roads and other line-like
sources CadnaA does not only provide long-term sound levels such as
Leq, but also the time history of the sound pressure level according to a
single vehicle with specified sound emission travelling along that line at
a prescribed speed.
A monopole source of sound is placed in steps of 1.94 m along the
center line of the northeastbound lane at y= 0 . For a speed of 70 km/h
this corresponds to a time resolution of 0.1 s. The source height is set to
z= 0 , representing the rolling noise of the cars. As with the Euler si-
mulations the emission level was set to an arbitrary value. The ground
D. Heimann et al. Applied Acoustics 133 (2018) 123–132

Fig. 4. Layout of the scale-model measurements. The

loudspeaker was adjusted at 35 positions along the
dotted arrow to simulate the sound emission of a
driving road vehicle. The microphone positions corre-
spond to the positions P1–P7 in Fig. 2.

line of loudspeaker positions

microphone
positions
building A

building B
loudspeaker

microphone positions

is assumed to be non-reflecting. Air absorption is not considered. 3. Results

The buildings are implemented as obstacles and reflectors. Since
saddle roofs cannot be represented in this noise prediction procedure,
buildings and parts of buildings are assumed to be cubic blocks with flat
roofs of a height that corresponds to 80% of top heights of the real
saddle roofs (building parts A: 8 m, B: 6 m, C: 3 m).
3.1. Evaluation method
The time evolutions of A-weighted sound levels are compared at the
microphone positions P1 to P7. The origin of the time axis (t = 0 ) is

(a) -20
-25 -25
-20 -30 -20 -20 -25 -30

2 -15 -15
z in m

-10
-10 -10 -35
-35
-25
1 -10

-40
1.02 m

-5
-5
-10-5 0
-40
-5 0 S S
0
-2 -1 0 1 2 -2 -1 0 1 2
x in m x in m

(b) -20
-30 -15 -20 -25
-15 -25

-10 -30
2
z in m

-35
-35
/9
-5
1 -10
0.68 m

-40 0 -40
5
S S
-5
0 5
0
-2 -1 0 1 2 -2 -1 0 1 2
x in m x in m
Fig. 5. a. Comparison of the Euler model results (left) with calculated (right, solid isolines) and measured (right, broken isolines) sound levels for the propagation of 500 Hz sound from a
source (S) over a laboratory-scale flat-roof building. The right panel is reproduced from Fig. 13d in [25]. b. As Fig. 5a, but for a saddle-roof building. The right panel is reproduced from
Fig. 13 h in [25].

127
D. Heimann et al. Applied Acoustics 133 (2018) 123–132

Table 2
Overview of considered influences and physical processes in the diverse sound levels (F: field measurements, S: scale-model measurements, E: Euler model simulations, I: ISO 9613-2
calculations, A: analytical free-field solution).

F S E I A

Spectral range 8 Hz–16 kHz 100 Hz–2 kHz 8 Hz–630 Hz 63 Hz–8 kHz –
(1/3 octave bands)
Geometrical spreading Yes Yes Yes Yes Yes
Air absorption Yes No No No No
ground effects Yes No Simplified No No
Vegetation effects Yes No No No No
Refraction Yes No No No No
Diffraction Yes Yes Yes Simplified No
Doppler effect Yes No No No No
Building Yes Yes Yes Yes No
Reflectivity of building Unknown 1 1 0.8 –
Source directivity Yes No No No No
Background sound Yes Simplified Simplified Simplified Simplified

defined to be the time of maximum sound level at position P1. This time
corresponds to the source position with the smallest distance to P1
(xs = ys = 0). To ease the comparability and to eliminate the effect of
different source strengths all sound levels (field measurements, Euler
model, and scale model) are normalized to LA,0 = 73.27 dB at t = 0 and
P1. This value equals the observed energy-equivalent mean maximum
sound level at P1 averaged over all selected events. The Euler simula-
tions and the laboratory scale-model measurements were additionally
adapted to the spectrum for passenger cars as proposed by the European
project Harmonoise (Jonasson et al. [27]). All sound levels (except the
field measurements) were corrected with respect to a constant A-
weighted background sound level of LA,b = 40 dB.
The A-weighted sound levels (Euler model and scale model) at time
t and at position Pi (i = 2–7), LA,Pi (t ), are calculated from the un-
weighted 1/3-octave band sound levels Lj,Pi (t ) by using the A-weighted
source spectrum LA,j,s [27], where j with l ⩽ j ⩽ u is the index of the
available 1/3-octave bands of the respective data source according to
Table 2.
u (Lj,Pi (t ) + LA,j,s − Lj,P1 (t = 0))
⎡ ⎤
LA,Pi (t ) = 10lg ⎢∑ ⎜⎛10 10 ⎞
⎟ + C1
⎥ all selected driving-by events at each receiver position. Also the
⎣ j=l ⎝ ⎠ ⎦
+ LA,0−LA,P1 (t = 0)
with C1 = 10 LA,b /10 and LA,0 = 73.27dB.
The normalization and the adaption to a realistic frequency spec-
trum enable the comparability between the events and among the four
data sources with respect to the reference position P1. However, there
absorption, but comprise ground absorption. The comparability is also
hampered by the fact that the data sets do not cover the same frequency
range (see Table 2). An overlap is only given for the 1/3 octave bands
with the center frequencies 100–630 Hz. Nevertheless, the evaluation is
performed with all available frequency bands for each method. The role
of cut-off frequencies is discussed in Section 3.4. Finally, monopole
sources are assumed in the numerical and scale-model experiments,
while the real sources are multipoles with directivities. With regard to
the range of travel distances and emission angles, the error of the A-
weighted sound level due to missing air absorption is estimated to be
not much higher than 1 dB. Higher deviations can be expected from the
ignored refraction due to downwind and upwind propagation. An
overview of the considered environmental influences and physical
processes is given in Table 2.
3.2. Results of the field measurements
The results of the field measurements are illustrated in Fig. 6. The
graphs show the A-weighted energy-equivalent mean sound level over
variability of the events is indicated. It results from variations of the
(3)
70 P1 P2
60
50
40
70 P3 P4
A-weighted SPL in dB

are also facts that impede the comparability. First, the measurement 60
devices were not automatically synchronized. A time comparison was 50
made once a day and time was adjusted between the comparisons by
40
assuming a linear drift. This may result in an inaccuracy of the obser-
ving time of ± 0.5 s. Second, the time evolutions of the results of the 70 P5 P6
numerical simulation, the scale-model and the ISO 9613-2 noise pre- 60
diction were derived from the results for individual source points along 50
the road and the assumption of a constant vehicle speed of 70 km/h. 40
Hence, the sound level at the receiver points (P1 to P7) at a given point 70 P7 -5 0 +5
on the time axis is uniquely associated with a specific emission point on
60 t in s
the road, considering the straight-line travel time of sound. In reality,
50
sound from a certain source point on the road may arrive at different
times at a receiver because of different propagation paths (direct, re- 40
flected, diffracted). However, the error due to different travel times in -5 0 +5
the given geometry is estimated to be not larger than 0.15 s. Besides t in s
errors in the temporal coordination, there are further sources of po- Fig. 6. Time histories of A-weighted sound levels of driving cars measured at the posi-
tential deviations of sound levels. While the field measurements com- tions P1–P7. The solid line represents the energy-equivalent mean of all selected events.
prise all physical processes during propagation including refraction, air The grey shade indicates the range of variability of the events. The dotted curves show the
and ground absorption, the latter effects are not considered in the scale- analytical solution for free-field propagation. All sound levels are normalized to 73.27 dB
at P1 for t = 0 .
model results. The numerical results ignore refraction and air

128
D. Heimann et al. Applied Acoustics 133 (2018) 123–132

vehicle speed and meteorological influences, but also depends on the 70 P1 P2

number of recorded events at the individual receivers (P1 and P3: 13,
60
P2 and P5: 5, P4: 10, P6 and P7: 3).
In addition to the measurements, analytic solutions for free-field 50
propagation (no building, no absorption, no refraction) are shown for
better interpretation (cf. Heimann and Schady [6]). They were calcu- 70 P3 P4

A-weighted SPL in dB
lated by 60

ρ c 50
L (t ) = 10lg ⎧ 0 2 P [V 2A2 + yr2 ]−1 + 10 LB /10⎫
⎨ 2πp0 ⎬ (4)
⎩ ⎭ 70 P5 P6
with 60

V 2 (t −x /V )2 + yr2 50
xr r
A = t− −
V c -3 -2 -1 0 +1 +2 +3
70 P7
where P is the source power, V is the velocity of the vehicle, c is the t in s
60
sound speed. x r , yr are the receiver coordinates, and p0 = 2·10−5 Pa . Eq.
(4) considers the sound-wave travel time and thus results a slightly 50
asymmetric behaviour with respect to t = 0 . LB = 40 dB is the back- -3 -2 -1 0 +1 +2 +3
ground sound level. The source power was chosen such that
t in s
L (t = 0) = 73.27 dB at P1. The assumptions of the analytical solution
resemble those of the scale-model (no absorption in the air and at the Fig. 7. Time histories of A-weighted sound levels of driving cars valid at the positions
ground) except that effects of the building (reflection, shading, dif- P1–P7. Blue: field measurements (energy-equivalent average over all selected events);
green: scale-model results; brown: numeric Euler model results; red: ISO 9613-2 calcu-
fraction) are not considered in Eq. (4).
lations; dotted black: analytic free-field solution (Eq. (4)). All sound levels are normalized
The measured level vs. time curve at P1 is clearly asymmetric with to 73.27 dB at P1 for t = 0 . (For interpretation of the references to color in this figure
smaller levels during approach (t < 0 ) and higher levels during de- legend, the reader is referred to the web version of this article.)
parture (t > 0 ). Note that the asymmetry due to the travel time of the
sound is already included in the free-field solution of Eq. (4). It is much
positions. The scale-model measurements and the Euler model simula-
less than the observed one. Therefore, the most likely explanation is the
tions show rather strong drops of the sound level at P3 and P4. These
refraction effect due to the prevailing wind from north-east which
are only weakly developed in the field measurements and the ISO 9613-
causes upwind propagation with upward refraction during approach
2 calculations. At P5 and P6 the sudden peak in the field measurements
and downwind propagation with downward refraction during de-
could be reproduced by the Euler model and the ISO 9613-2 calcula-
parture. Note that the time axis covers ± 9 s which corresponds to
tion, but with smaller amplitude.
propagation paths from the 70 km/h fast cars to P1 of up to 175 m
A summary of the degree of agreement between the different sound-
length. Nearly the same asymmetry is observed at P2, but the sound
level time evolutions is presented in Fig. 8. It shows the concordance
from the departing cars is more damped and close to the analytic so-
correlation coefficient after Lin [28] for all combinations of the con-
lution, probably because of the vegetation in the slant propagation
sidered sound levels. This coefficient is a measure of both precision and
path. The maximum sound level at P2 of the field and scale-model
accuracy. The correlation analysis refers to the period
measurements is by about 3 dB higher than the maximum of the ana-
−1.5 s⩽ t ⩽ +2.5 s within which all sound levels are available. All
lytically calculated sound level. This is caused by the backward re-
correlations are rather high at positions which are not in the immediate
flection at the roadside facade of the building.
Behind the building (P3 to P7) the situation is more complex. The 1.0
field measurements result in sound levels which are generally lower 0.8
than the free-field solution. This is expected because of the shading by 0.6
0.4
the building and other propagation effects such as attenuation by the
0.2
ground and by vegetation. At P3 and P4 there are even weak dips in the 0.0
P1 P2
sound level curve at the time when the free-field propagation yields
maximum levels. Striking features are the short-time increases of the 0.8
sound level at P4 and P6. These are caused by reflection as it is dis- 0.6
0.4
cussed in the following. 0.2
0.0
P3 P4
3.3. Comparison of measured and calculated sound-level evolutions
0.8
0.6
Fig. 7 allows to compare five different time-series of sound levels at 0.4
each receiver position: the field measurements, the scale-model mea- 0.2
surements, the Euler model simulations, the ISO 9613-2 calculations 0.0
P5 P6
and the free-field solutions. At P1 and P2 the curves coincide quite well SF EF IF AF ES IS AS
0.8
except the strong asymmetry of the field measurements due to wind- 0.6
induced refraction. This process is not considered in all other curves. At 0.4
P2 all curves show higher levels than the free-field solution because of 0.2
0.0
the backward reflection by the building. However, the offset is dif- -0.2
P7
ferent. A possible cause is the degree of reflection of the building. It was SF EF IF AF ES IS AS
not determined for the real building and different assumptions were Fig. 8. Concordance correlation coefficient between pairs of A-weighted sound-level
met in the calculations (the reflectivity was set to 1 in the Euler si- time-series of the field measurements (F), the scale-model measurements (S), the Euler
mulation and 0.8 in the ISO 9613-2 calculation). model simulations (E), the ISO 9613-2 calculations (I), and the analytical free-field so-
lution (A) for all receiver positions P1–P7.
Larger deviations between the curves are found for all other receiver

129
D. Heimann et al. Applied Acoustics 133 (2018) 123–132

Table 3 0
S1 S2 S3 S4 S5 S6
Extreme change of the A-weighted sound level within 1 s at positions P1–P7 (F: field
10

y in m
measurements, S: scale-model measurements, E: Euler model simulations, I: ISO 9613-2
calculations, A: analytical free-field solution).

N
20 A
B C

Minimum change rate (drop) in dB/s

30 P3
F S E I A 60

A-weighted SPL in dB
P1 −6.1 −6.9 −7.4 −7.6 −7.4
P2 −7.2 −7.1 −6.4 −5.4 −4.7
55
P3 −2.6 −17.4 −12.1 −3.2 −2.9
P4 −2.6 −10.0 −6.7 −6.3 −2.5 50
P5 −4.4 −4.9 −2.0 −3.6 −2.8
P6 −6.0 −3.7 −5.8 −4.7 −3.0
P7 −4.2 −3.7 −9.1 −5.9 −2.4 45
Maximum change rate (rise) in dB/s
-60 -40 -20 0 20 40 60
F S E I A x in m
P1 7.8 8.1 7.2 8.6 8.0 Fig. 9. A-weighted sound levels (colors as in Fig. 7) at P3 as a function of the source
P2 11.2 7.8 5.2 6.8 5.0 position for a vehicle speed of 70 km/h. The source positions of distinctive features in the
P3 2.4 4.7 6.1 2.1 3.1 sound-level curves are indicated as S1–S6. Direct and diffracted (over the roof) rays are
P4 2.7 9.3 3.8 2.9 2.6 shown as long-dashed and short-dashed lines, respectively. (For interpretation of the
P5 11.2 10.0 9.6 4.8 3.1 references to color in this figure legend, the reader is referred to the web version of this
P6 13.1 9.7 4.5 6.4 3.3 article.)
P7 4.5 15.1 10.3 10.2 2.7

acoustical shadow of the building (P1, P2, and P7). At the shaded po-
sitions P3 to P6 the following is observed:
• the agreement between both types of measurements (field vs. scale
model “SF”) is rather poor.
• at P3 and P4 the Euler simulation agrees much better with the scale-
model measurements (“ES”) than with the field measurements
(“EF”).
• at P5 and P6 the rather simple ISO 9613-2 calculation agrees slightly
better (“IF”, ”IS”) with the measurements than the complex Euler
model (“EF”, “ES”). The free-field solution completely fails behind
the building as expected (“AF”, “AS”).
The discrepancies between the results also show up in the sound-
level rise and drop rates. As these are important in assessing human
effects of noise (cf. introduction), they are listed in Table 3. The sound-
level change rates deviate most at P3 and P4 (decrease rate over-
estimated by Euler simulations and scale-model measurements) and at
P7 (increase rate overestimated by Euler simulations, scale-model
measurements and ISO 9613-2 calculations).
3.4. Discussion of discrepancies and singularities
The field measurements deviate from the scale-model measurements
and Euler simulations at the backside of the building (positions P3 to
P6). In particular, the prominent drop of the sound level by 12–15 dB at
P3 which is present in the scale-model measurements and Euler simu-
lations, is not visible in the field measurements. The ISO 9613-2 cal-
culations do not show this feature either. Fig. 9 presents a close-up to
the situation at position P3. The time axis t is converted to the space
axis x along the road by x = Vt with the vehicle speed V = 70 km/
h = 19.4 m/s. So it is possible to coordinate features in the temporal
development of sound levels with the approximate position of the
sound emitting car.
Fig. 9 shows that P3 becomes shaded once the source passes S1. At
this instance the ISO 9613-2 sound level drops by about 4 dB. However,
neither the field measurement nor the scale-model measurement shows
this drop. The field measurement even shows a slight increase. As the
source passes S3 the sound towards P3 has to propagate around two
diffraction edges in the horizontal and vertical plane as well. At this
situation the scale-model result shows a significant drop by about
15 dB. Also the Euler model results in a decrease of nearly that amount,
but at a slower rate. The minimum sound level of the field measure-
ments, the scale-model measurements and the Euler model coincides
with the source position S4. From this position the diffracted propa-
gation paths over and around the building are longest. Once the car
reaches S5 the diffraction path over the low annex building B is pos-
sible. This seems to lead to the secondary maximum in the field mea-
surements, the scale-model measurements and the Euler model. The
features that are related with the source positions S3, S4 and S5 do not
show up in the ISO 9613-2 results. Presumably this is caused by the fact
that only the diffraction path over the building is considered and the
roof is assumed to be flat. The weak minima in the Euler and scale
model results associated with source position S6 cannot be satisfactorily
explained without additional experimentation. This minimum appears
neither in the field measurement nor in the ISO 9613-2 calculations.
In sum, it can be stated that the results at the receiver position P3 on
the shaded backside of the building are curious and difficult to under-
stand. In particular, the large difference in the decline of the sound level
between the field measurement on the one hand and the Euler and
scale-model results on the other hand is surprising. Some possible
reasons for the shallow curve of the field measurement at P3 were
considered. Local background noise (e.g. wind induced noise of nearby
vegetation) can be ruled out, because the curves level off at about 40 dB
for |t| > 6 s as they do at the other microphone positions. A possible
influence of reflections from the facades of neighbouring buildings in
the east (cf. Fig. 1) can be excluded, too. Test calculations with the ISO
9613-2 procedure suggest that such reflections would increase the
sound level at P3 by less than 1 dB. A certain sound transmission
through the building could be also a reason for the observed behaviour.
However, this could not be verified, but it is rather improbable because
of the solid masonry and sound-insulating windows at least on the road
side of building A. The sound meters were calibrated before and cross-
checked after the field experiment without abnormalities.
The differing spectral range of the methods (cf. Table 2) could be
also the origin of discrepancies. Therefore, the spectral behaviour at P3
is shown in Fig. 10 for the field and scale-model measurements and for
the Euler simulations. The figure shows the deviation from the spec-
trum near the source at P1 and t = 0 . The scale-model as well as the
Euler model shows a strong drop throughout the available frequency
range once the source passes S3. This drop slightly increases with
growing frequency as low-frequency sound is stronger diffracted than
high-frequency sound. By and large, the structures of the scale-model

130
D. Heimann et al. Applied Acoustics 133 (2018) 123–132

S1 S2 S3 S4 S5 S6 0
S1 S2 S3 S4
2000
(a) 10

y in m
frequency in Hz

1000
20 A
B C

500 30 P6

N
A-weighted SPL in dB
200 60

100 55
2000
(b)
50
frequency in Hz

1000
45
500
-60 -40 -20 0 20 40 60
x in m
200
Fig. 11. A-weighted sound levels (colors as in Fig. 7) at P6 as a function of the source
100 position for a vehicle speed of 70 km/h. The blue dashed-dotted curve shows the field
2000 measurements for 25–630 Hz (A-weighted and normalized as the other curves). The
(c) hatched red curve provides the results of ISO 9613-2 calculations without reflections at
frequency in Hz

1000
500
200
100
-3 -2 -1 0 1 2 3
time in s
Fig. 10. Deviation of the spectral distribution at P3 (in dB) from the spectrum at P1 and
t = 0 for (a) the field measurements, (b) the scale-model measurements and (c) the Euler
model simulations. The results of the Euler simulations are not available in the grey area.
For a better comparability with Fig. 9 the positions S1–S6 of a 70 km/h fast source are
indicated at the upper margin.
and Euler model results are similar. However, the field measurements
are very different. Frequencies below 200 Hz are generally less damped
than higher frequencies. This is almost independent of the source po-
sition. The slow decrease of the sound level between source positions S2
and S4 applies almost equally to all frequencies between 100 and
2000 Hz. From Fig. 9 it is evident that the different behaviour of the
broadband A-weighted sound levels shown in Fig. 8 is not caused by
differences in the spectral range of the respective data set. Note that the
Doppler effect is not considered in the scale-model and Euler model
results. For c = 330 m/s and V = 70 km/h the maximum Doppler shift
between arrival and departure amounts to 11.7 percent of the fre-
quency.
Another peculiarity of the results is the strong short-term peak of the
sound level in the field measurements at receiver positions P5 (at t = 0 )
and P6 (at t = −0.4 s ); see Fig. 6. This feature is closer inspected in
Fig. 11. Test calculations with the ISO 9613-2 procedure suggest that
the peak is caused by reflections at the southwestly facade of the small
building C as long as the source moves from S1 to S3. From this road
section sound rays reach the receiver P6 with one reflection at building
C and without being blocked by other parts of the building. In the ISO
9613-2 curves the peak disappears when the reflection at building C is
disregarded. The peak does not exist in the results of the scale-model
measurements because only the building parts A and B were considered
in the layout. The peak is also present in the results of the Euler model
simulations, but the amplitude is as small as in the ISO 9613-2 calcu-
lations. A spectral evaluation of the field measurements shows that the
strong amplitude of the peak results from high frequencies
(f > 630 Hz). If only the 1/3-octave bands 25–630 Hz are considered
the peak amplitude of the A-weighted sound level reduces from 9 to
4 dB.
the walls of building C. The source positions of distinctive features in the sound-level
curves are indicated as S1–S4. Direct, diffracted (over the roof) and reflected (at building
C) rays are shown as long-dashed, short-dashed and dot-dashed black lines, respectively.
(For interpretation of the references to color in this figure legend, the reader is referred to
the web version of this article.)
Another weak maximum of the sound level appears when the source
reaches S4. From here the sound can pass through a gap between
building parts B and C. This explains the sudden rise of the sound level
in the field measurements and most notably in the ISO 9613-2 calcu-
lations. The subsequent decrease of the level is a shadow effect of
building C. The ISO 9613-2 calculation without reflections at building C
also shows this decrease. Note that only the reflections at the wall of
building part C are switched off (totally absorbing walls), while
building part C still acts as an obstacle around and over which dif-
fraction is possible. The scale-model results do not show this feature,
possibly because building C was not modelled.
4. Conclusions
The investigation of road traffic noise in the vicinity of a building
with four different methods (field measurements, laboratory scale-
model measurements, numerical Euler model simulations, and standard
noise prediction calculations) reveal the complexity of the problem. The
real situation in the field is characterized by a structured building with
an annex building, roof overhangs, a chimney, and with different ma-
terials involved. The building is located on a partly paved, partly grass
covered ground and bush vegetation and a few trees between the house
and the road. Any model, be it a numerical one or a scale model, cannot
reproduce the full complexity of the reality, but it has to simplify the
situation depending on the specific abilities of the method. Moreover,
the models are only approximations of the physical processes which
determine the sound propagation from the source on the road to the
receivers around the building. Some processes are simplified, others are
neglected at all. Therefore, it is not surprising that the results of nu-
merical and physical models deviate from each other and from field
measurements.
Nevertheless, from the present study one has to conclude that the
differences between the applied methods can be rather grave. Most
notably there are two pairs of methods which among each other re-
semble with respect to their results: the numerical model (“E”) and the
scale model (“S”) on the one hand, and the field measurements (“F”)
and the rather simple ISO 9613-2 calculation method (“I”) on the other
hand. The setups (Table 2) for “E” and “S” are quite similar. This could
be one reason why the results of these methods look alike.
The reason for this behaviour could not be sufficiently elucidated in

131
D. Heimann et al.
this study. It is remarkable that the ISO 9613-2 standard reproduces the
in-field measured sound levels quite well although the buildings are
simplified (flat instead of saddle roof) and only diffraction over the
building is considered while lateral diffraction is omitted. Moreover,
the results do not generally confirm the statement in [12] that ISO
9613-2 overestimates the sound pressure level of diffracted sound. The
Euler simulation and the scale-model measurements generate similar
results although they also differ in their assumptions. Of course, the
pairwise agreements can be incidental to some degree. The disagree-
ment between the numerical Euler model and the laboratory scale-
model on the one side and the field measurements on the other side
might be an indication that either not modelled small-scale details of the
real setup (e.g. pillars, trees, fences, playground items in the garden)
play a role or the diffraction process is not properly reproduced in the
Euler model. The latter is not very likely as the comparison with the
results of [25] and the agreement with the scale model suggests.
So far, it has to be concluded from the present study that the cal-
culation of sound-level time histories according to moving road ve-
hicles, the prediction of maximum sound levels or maximum sound-
level change rates on the backside of a roadside building need parti-
cular caution. Especially, one has to be aware of effects by features
which are not or cannot be considered in the models because of their
inherent degree of approximation or missing information. For instance,
these could be effects of architectural details like avant-corps, niches, or
overhangs, the reflectivity of walls and windows, possible sound
transmission through the building (e.g. open windows), effects of ve-
getation and obstacles in the surroundings, and ground effects. Also
particularities of the sound sources (spectrum, directivity, elevated
partial sources) may lead to discrepancies. In the present study the
reasons for the deviations could not be completely identified.
Acknowledgements
Special thanks go to Klaus Ehrenfried who supported the study with
his acoustical know-how. We also thank Felix Karaman who designed
and scaled the indoor measurement models and helped during the
build-up of the indoor experiment and the measurement. Two anon-
ymous reviewers helped to improve the manuscript.
References
[1] Miedema HME, Vos H. Exposure-response relationships for transportation noise. J
Acoust Soc Am 1998;104:3432–45.
[2] Miedema HME, Oudshoorn C. Annoyance from transportation noise: relationships
with exposure metrics DNL and DENL and their confidence intervals. Environ
Health Perspect 2001;109:409–16.
[3] Lercher P, Brink M, Rudisser J, Van Renterghem T, Botteldooren D, Baulac M, et al.
The effects of railway noise on sleep medication intake: results from the ALPNAP-
study. Noise Health 2010;12:110–9. http://dx.doi.org/10.4103/1463-1741.63211.
[4] Basner M, Müller U, Elmenhorst EM. Single and combined effects of air, road and
rail traffic noise on sleep recuperation. Sleep 2011;34:11–23.
[5] Elmenhorst EM, Pennig S, Rolny V, Quehl J, Müller U, Maaß H, et al. Examining
132
Applied Acoustics 133 (2018) 123–132
nocturnal railway noise and aircraft noise in the field: sleep, psychomotor perfor-
mance, and annoyance. Sci Total Environ 2012;424:48–56. http://dx.doi.org/10.
1016/j.scitotenv.2012.02.024.
[6] Heimann D, Schady A. Derivation of sound-level characteristics to assess traffic
development scenarios. Appl Acoust 2016;103:1–9. http://dx.doi.org/10.1016/j.
apacoust.2015.10.003.
[7] McNamara DA, Pistorius CWI, Malherbe JAG. Introduction to the uniform geome-
trical theory of diffraction. Boston, MA, USA: Artech House; 1990.
[8] Salomons EM. Reduction of the performance of a noise screen due to screen-induced
wind-speed gradients. Numerical computations and wind tunnel experiments. J
Acoust Soc Am 1999;105:2287–93. http://dx.doi.org/10.1121/1.426835.
[9] Heimann D, Blumrich R. Time-domain simulations of sound propagation through
screen-induced turbulence. Appl Acoust 2004;65:561–82. http://dx.doi.org/10.
1016/j.apacoust.2003.09.007.
[10] ISO 9613-2. Acoustics - Attenuation of sound during propagation outdoors - Part 2:
General method of calculation. Berlin: Beuth Verlag; 1999.
[11] Salomons E, Van Maercke D, Defrance J, De Roo F. The Harmonoise sound pro-
pagation model. Acta Acustica United with Acustica 2011;97:62–74. http://dx.doi.
org/10.3813/AAA.918387.
[12] Wei W, Van Renterghem T, Botteldooren D. Simplified analytical model for sound
propagation at shielded urban locations involving multiple diffraction and reflec-
tions. J Acoust Soc Am 2015;138:2744–58. http://dx.doi.org/10.1121/1.4932585.
[13] Alberts II WCK, Noble JM, Coleman MA. Sound propagation in the vicinity of an
isolated building: an experimental investigation. J Acoust Soc Am
2008;124:733–42. http://dx.doi.org/10.1121/1.2945151.
[14] Remillieux MC, Corcoran JM, Haac TR, Burdisso RA, Svensson UP. Experimental
and numerical study on the propagation of impulsive sound around buildings. Appl
Acoust 2012;73:1029–44. http://dx.doi.org/10.1016/j.apacoust.2012.04.011.
[15] Wang H, Cai M, Luo W. Areawide dynamic traffic noise simulation in urban built-up
area using beam tracing approach. Sustain Cities Soc 2017;30:205–16. http://dx.
doi.org/10.1016/j.scs.2017.02.004.
[16] Walker ED, Hart JE, Koutrakis P, Cavallari JM, VoPham T, Luna M, et al. Spatial and
temporal determinants of A-weighted and frequency specific sound levels – an
elastic net approach. Environ Res 2017;159:491–9. http://dx.doi.org/10.1016/j.
envres.2017.08.034.
[17] Hou Q, Cai M, Wang H. Dynamic modeling of traffic noise in both indoor and
outdoor environments by using a ray tracing method. Build Environ
2017;121:225–37. http://dx.doi.org/10.1016/j.buildenv.2017.05.031.
[18] ISO 3745. Determination of sound power levels and sound energy levels of noise
sources using sound pressure - Precision methods for anechoic rooms and hemi-
anechoic rooms. Berlin: Beuth-Verlag; 2012.
[19] Blumrich R, Heimann D. A linearized Eulerian sound propagation model for studies
of complex meteorological effects. J Acoust Soc Am 2002;112:446–55.
[20] Hu FQ, Hussaini MY, Manthey JL. Low-dissipation and low-dispersion Runge-Kutta
schemes for computational acoustics. J Comp Phys 1996;124:177–91.
[21] Bogey C, de Cacqueray N, Bailly NC. A shock-capturing methodology based on
adaptive spatial filtering for high-order non-linear computations. J Comp Phys
2009;228:1447–65. http://dx.doi.org/10.1016/j.jcp.2008.10.042.
[22] Bogey C, Bailly NC. A family of low dispersive and low dissipative explicit schemes
for flow and noise computations. J Comp Phys 2004;194:194–214.
[23] Heutschi K, Horvath M, Hofmann J. Simulation of ground impedance in finite
difference time domain calculations of outdoor sound propagation. Acta Acustica
United with Acustica 2005;91:35–40.
[24] Bogey C, Bailly NC. Three-dimensional non-reflective boundary conditions for
acoustic simulations: far field formulation and validation test cases. J Comp Phys
2002;88:463–71.
[25] Kawai T. Sound diffraction by a many-sided barrier or pillar. J Sound Vib
1981;79:229–42.
[26] DataKustik; URL: < http://www.datakustik.de/ > [last opened 06 Nov 2017].
[27] Jonasson H, Sandberg U, van Blokland G, Watts G, Luminari M. Source modelling of
road vehicles. Harmonoise project, Technical Report No. HAR11TR-041210-SP10;
2004, 52p.
[28] Lin L. A concordance correlation coefficient to evaluate reproducibility. Biometrics
1989;45:255–68.