On the ability of consumer electronics microphones for environmental noise monitoring.
ABSTRACT The massive production of microphones for consumer electronics, and the shift from dedicated processing hardware to PC-based systems, opens the way to build affordable, extensive noise measurement networks. Applications include e.g. noise limit and urban soundscape monitoring, and validation of calculated noise maps. Microphones are the critical components of such a network. Therefore, in a first step, some basic characteristics of 8 microphones, distributed over a wide range of price classes, were measured in a standardized way in an anechoic chamber. In a next step, a thorough evaluation was made of the ability of these microphones to be used for environmental noise monitoring. This was done during a continuous, half-year lasting outdoor experiment, characterized by a wide variety of meteorological conditions. While some microphones failed during the course of this test, it was shown that it is possible to identify cheap microphones that highly correlate to the reference microphone during the full test period. When the deviations are expressed in total A-weighted (road traffic) noise levels, values of less than 1 dBA are obtained, in excess to the deviation amongst reference microphones themselves.
- SourceAvailable from: Weigang Wei[Show abstract] [Hide abstract]
ABSTRACT: Surveys show that inhabitants of dwellings exposed to high noise levels benefit from having access to a quiet side. However, current practice in noise prediction often underestimates the noise levels at a shielded façade. Multiple reflections between façades in street canyons and inner yards are commonly neglected and façades are approximated as perfectly flat surfaces yielding only specular reflection. In addition, sources at distances much larger than normally taken into account in noise maps might still contribute significantly. Since one of the main reasons for this is computational burden, an efficient engineering model for the diffraction of the sound over the roof tops is proposed, which considers multiple reflections, variation in building height, canyon width, façade roughness and different roof shapes. The model is fitted on an extensive set of full-wave numerical calculations of canyon-to-canyon sound propagation with configurations matching the distribution of streets and building geometries in a typical historically grown European city. This model allows calculating the background noise in the shielded areas of a city, which could then efficiently be used to improve existing noise mapping calculations. The model was validated by comparison to long-term measurements at 9 building façades whereof 3 were at inner yards in the city of Ghent, Belgium. At shielded façades, a strong improvement in prediction accuracy is obtained.Acta Acustica united with Acustica 12/2014; 100(6). · 0.71 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: To assess the overall noise exposure of the population in Flanders and in particular the exposure to street traffic noise, a measurement campaign was set up involving 250 randomly selected households in Flanders. Measurements were conducted first in 1996 and were repeated twice afterwards. This unique longitudinal noise monitoring exercise revealed that although the traffic intensity has grown over this period, noise exposure on average hardly changed. Small trends in exposure distribution and in statistical noise levels do nevertheless occur but they are marginally significant. Contrasting these measurements with status reporting based on noise maps, prediction of population exposure, and noise annoyance surveys shows that although all of these methodologies have their merits, they cannot be readily compared. In particular the difference between estimated trends in percentage of highly annoyed inhabitants based on noise level measurements and observed trends in reported noise annoyance, is striking.
- [Show abstract] [Hide abstract]
ABSTRACT: Several studies show that a significant portion of daily air pollution exposure, in particular black carbon (BC), occurs during transport. In a previous work, a model for the in-traffic exposure of bicyclists to BC was proposed based on spectral evaluation of mobile noise measurements and validated with BC measurements in Ghent, Belgium. In this paper, applicability of this model in a different cultural context with a totally different traffic and mobility situation is presented. In addition, a similar modeling approach is tested for particle number (PN) concentration. Indirectly assessing BC and PN exposure through a model based on noise measurements is advantageous because of the availability of very affordable noise monitoring devices. Our previous work showed that a model including specific spectral components of the noise that relate to engine and rolling emission and basic meteorological data, could be quite accurate. Moreover, including a background concentration adjustment improved the model considerably. To explore whether this model could also be used in a different context, with or without tuning of the model parameters, a study was conducted in Bangalore, India. Noise measurement equipment, data storage, data processing, continent, country, measurement operators, vehicle fleet, driving behavior, biking facilities, background concentration, and meteorology are all very different from the first measurement campaign in Belgium. More than 24 h of combined in-traffic noise, BC, and PN measurements were collected. It was shown that the noise-based BC exposure model gives good predictions in Bangalore and that the same approach is also successful for PN. Cross validation of the model parameters was used to compare factors that impact exposure across study sites. A pooled model (combining the measurements of the two locations) results in a correlation of 0.84 when fitting the total trip exposure in Bangalore. Estimating particulate matter exposure with traffic noise measurements was thus shown to be a valid approach across countries and cultures.Environment International. 01/2015; 74:89-98.
On the ability of consumer electronics microphones for environmental noise
Timothy Van Renterghem,*aPieter Thomas,aFrederico Dominguez,bSamuel Dauwe,aAbdellah Touhafi,b
Bart Dhoedtaand Dick Botteldoorena
Received 1st October 2010, Accepted 18th November 2010
The massive production of microphones for consumer electronics, and the shift from dedicated
processing hardware to PC-based systems, opens the way to build affordable, extensive noise
measurement networks. Applications include e.g. noise limit and urban soundscape monitoring, and
validation of calculated noise maps. Microphones are the critical components of such a network.
Therefore, in a first step, some basic characteristics of 8 microphones, distributed over a wide range of
price classes, were measured in a standardized way in an anechoic chamber. In a next step, a thorough
evaluation was madeof the ability ofthese microphones to beused forenvironmental noise monitoring.
This was done during a continuous, half-year lasting outdoor experiment, characterized by a wide
variety of meteorological conditions. While some microphones failed during the course of this test, it
was shown that it is possible to identify cheap microphones that highly correlate to the reference
microphone during the full test period. When the deviations are expressed in total A-weighted (road
traffic) noise levels, values of less than 1 dBA are obtained, in excess to the deviation amongst reference
Noise annoyance is a major environmental problem in urbanized
regions. Exposure to traffic noise is associated with a wide range
of negative effects on human health and well-being. It was esti-
mated that outside their homes, near 44% of the European
population (in the year 2000) was exposed to road traffic noise
levels above the World Health Organization’s threshold for onset
of negative health effects.1Examples of the adverse effects of
exposure to traffic noise are not only annoyance,2but also sleep
disturbance,3,4negative impacts on cognitive functioning (espe-
cially in children)5and the contribution to cardiovascular
The European Environmental Noise Directive8obliges each
member stateto makenoise maps of, amongst others, their major
highways and highly populated agglomerations. A noise map is
most often a calculation exercise, showing an estimation of long-
term averaged noise levels with a fine spatial resolution. Based on
such maps, action plans have to be proposed for problem areas.
However, producing accurate city noise maps is a hard task.
The complexity of the sound propagation problem in a densely
build-up environment is high.9Typically, geometrical acoustics
However, even in a single street, a large number of multiple
aGhent University, Department of Information Technology (INTEC),
Sint-Pietersnieuwstraat 41, B-9000 Gent, Belgium. E-mail: timothy.van.
bErasmushogeschool Brussel, Department of Industrial Sciences, Belgium/
Vrije Universiteit Brussel, Department of Electronics and Informatics,
Noise pollution is an increasingly growing threat for the well-being and the public health in industrialized countries. Although the
large advances made in predicting tools for noise exposure assessment during the last decades, the full complexity of the sound
propagation problem, together with an accurate representation of the distribution of noise sources, is most often not sufficiently
captured in an urban environment. Therefore, measurements are still an important tool for assessing the public’s exposure to noise.
The work presented in this paper shows that the large cost for extended noise monitoring networks can be strongly reduced by using
microphones appearing in consumer electronics devices. It was shown that it is possible to identify such microphones that result in
only small level differences compared to reference equipment, making them useful in many environmental noise monitoring
544 | J. Environ. Monit., 2011, 13, 544–552 This journal is ª The Royal Society of Chemistry 2011
Dynamic Article LinksC
Cite this: J. Environ. Monit., 2011, 13, 544
reflections between fac ¸ades (bordering the street) is needed,10
leading to very long computing times. In practice, the maximum
number of reflections taken into account is most often set to
a fixed number to limit computing times; this decision is usually
not based on accuracy considerations. At shielded locations,
predicting correct levels with a noise mapping calculation
method is even more problematic.11Besides computational and
propagation related issues, a good estimation of the relevant
noise sources and their spatial and temporal distribution is
needed given the fact that the acoustic environment is strongly
source-driven. Most often, noise maps highly rely on the output
of traffic models, inducing additional uncertainties.
Taking these problems into account, validation of such city
noise maps with measurements seems necessary. Although the
technology exists for noise measurement networks, their appli-
cation is very limited by the high cost of logging units and sensors
(microphones) found on the (commercial) market nowadays.
Two recent evolutions could lead to an affordable noise
monitoring network. Nowadays, microphones appear a lot in
consumer electronics (like mobile phones, laptops, portable
digital music players, etc.) and hearing aids. Due to mass
production, such devices come at a (very) low price. The
microphone technology of these cheaper devices is nevertheless
very similar to the technology of high-quality measurement
A second interesting evolution is that the processing and
logging of the raw signal produced by the microphone capsule
are shifted from dedicated hardware to PC-based systems. Of
special interest are the so-called Single Board Computers (SBCs).
Such devices can be seen as stripped-down integrated PCs, with
all basic functionalities, but with a more limited computational
performance. When equipping these with a sound card and
a network card, they are well-suited as nodes in a noise
measurement network. Furthermore, such SBCs use low-power
processing units, making them suitable for networked low-power
The price difference between cheaper approaches and dedi-
cated measurement microphones and logging hardware is huge,
and can easily exceed a factor of 100. In this paper, it is studied to
what extent such cheaper noise measurement systems can be used
for environmental noise monitoring. The studies presented in ref.
12 and 13 have similar interests in affordable noise monitoring
In this paper, results of the detailed testing of such cheap
microphones are presented. In a first step, the performance of
SBCs as logging units and microphones of different price classes
is checked in an anechoic chamber. The main focus in this paper
is on a half-year lasting outdoor test near a busy road. The noise
levels obtained by the cheaper variants were compared with
In this introduction, the validation of city noise maps is pre-
sented as a useful environmental application of an extensive
noise measurement network. It is clear that applications are
manifold, and could range from noise monitoring near industrial
facilities to prevent neighbourhood complaints in an early stage
to e.g. community-based noise monitoring near airports.
Selecting microphones and logging unit
Eight on-shelf and off-shelf microphones, distributed over a wide
range of price classes, were tested. An overview of some basic
characteristics is given in Table 1. All of these, except for one (the
MEMS microphone, see further), use pre-polarized condenser
microphone technology (also called electret microphones).
Professional measurement microphones can be categorized
into different accuracy classes, according to some preset norms.14
Types 0, I, and II are usually distinguished. With increasing type
number, accuracy goals become less strict. These goals deal e.g.
with a change in sensitivity in function of angle of incidence on
the microphone membrane, or the maximum change in observed
output after 1 hour in a constant sound field. Two professional
Table 1Product details, prices and measured noise floors of the 8 selected microphones
(dB re 1 V Pa?1)
at 1 kHz
?45 dB20 Hz to
20 Hz to
100 Hz to
40 Hz to
100 Hz to
20 Hz to
3.5 Hz to
6.3 Hz to
?68 dB3 V
?40 dB 30 V
?40 dB 50 V
?32 dB 30 V
?26 dB300 V
REF1 Electret 1/200
?26 dB2000 V
?26 dB 2000 V
aFollowing product sheets.
This journal is ª The Royal Society of Chemistry 2011J. Environ. Monit., 2011, 13, 544–552 | 545
high-quality (type I) measurement microphones were included in
the test (further indicated by REF1 and REF2). These will serve
as reference equipment producing the ground truth noise level. A
professional low-noise pre-amplifier with an Integrated Circuit
Piezoelectric (ICP) feeding is used to complete the measurement
chain. For operation outdoors, a professional outdoor protec-
tion unit (including windscreen and rain protection) is used.
Next, a type II microphone was added to the test as well (indi-
cated further byTYPEII). Themicrophone capsule wasdelivered
integrated with a pre-amplifier and an outdoor protection unit.
ICP feeding was needed here as well.
Next, 5 non-dedicated measurement microphones were
considered. A main advantage of these devices is that pre-
amplification is not needed. To make such microphones opera-
tional, a small RC-circuit was built. In this way, the output
voltage of a PC sound card could be used (‘‘line powering’’). As
the sensitivity of these microphones differs, an adequate ampli-
fication factor for the sound card was set to select an operational
amplitude range. Three microphones fell in the price range from
30 to 50 Euro; these are two electret microphones (ELECTRET3
and ELECTRET4) and one MEMS microphone. A MEMS
(micro-electrical–mechanical system) microphone is a recent type
of microphone technology.15
membrane is etched directly on the chip itself. It has a similar
working mechanism as a common electret microphone. Finally,
two very cheap electret microphones were selected (ELEC-
TRET1 and ELECTRET2) of only a few Euros. For these 5
devices, self-fabricated rain-caps and windscreens were made.
It is clear that the prices shown in Table 1 are indicative, and
are subject to (often rapid) market evolutions. Note that only
microphone capsules and pre-amplifiers (where needed) are
taken into account. For microphone TYPEII, the outdoor unit is
included in the price. Furthermore, the prices given for the
professional measurement microphones contain research and
development costs. Without taking these aspects into account,
the cost ratio between the cheapest and most expensive micro-
phone exceeds roughly 1000.
The logging and processing of the raw microphone data were
performed with a SBC. The choice of the specific SBC was
a compromise between its price, robustness (e.g. the absence of
moving hardware parts like fans and hard disk drives), energy
consumption, and past experience with this type of system board.
The SBC has a 500 MHz AMD Geode processor, and was
equipped with 256 MB DDR DRAM. An audio-card was placed
on the board with a signal-to-noise ratio exceeding 100 dB (18 bit
resolution), and delivers a microphone feeding voltage measured
at 1.5 V. The total energy consumption when performing noise
measurements is near 5 W. The full cost of the SBC is about 100
Euro. For the processing of the raw microphone signal, the
Euterpe software platform16
Windows XP operating system was used. Since this processing is
rather computational intensive, each microphone needs its own
Here, the pressure sensitive
running under a Microsoft
The specifications provided by the microphone product sheets
cannot be easily used for inter-comparison and relevant infor-
mation is often lacking, mainly for the very cheap variants.
Furthermore, the full measurement chain (including pre-ampli-
fiers where needed, ICP feeding, RC-circuit, the sound card of
the SBC, weather protection, etc.) determines the behaviour.
Therefore, in a first step, some basic characteristics are measured
in a standardized way in a full anechoic chamber. In all cases,
normal incident sound on the microphone membrane is consid-
ered. Of main interest are the noise floor (lowest sound pressure
level that can be measured, which is limited by instrumentation
noise in the circuit), saturation level (highest sound pressure level
that can be measured, limited by the maximum movement of the
membrane), and flatness of the frequency response and linearity
(similarity of the sensitivity at different sound frequencies and
sound pressure levels).
Calibration was performed in various ways. The measurement
microphones could be directly calibrated by putting a 1/200pis-
tonphone (Svantek SV30A, operating at 94 dB, producing a pure
tone of 1 kHz) on the microphone capsule. Microphones
ELECTRET4 and MEMS1 were designed in such a way that the
pressure sensitive membrane is located at the end of a rigid 1/200
cylinder, at its centre. In this way, a standard pistonphone can
still be used for calibration. ELECTRET1, ELECTRET2, and
ELECTRET3 had another design. For these, a free field cali-
bration was performed in a full anechoic chamber. These
microphones were placed directly beside a reference microphone,
and an intense 1 kHz pure tone was produced by the loudspeaker
in front of them. The exact level measured at the reference
microphone (which was calibrated in advance) was then used for
the microphone to be calibrated. For similarity, a level near
94 dB was used as well.
In a first step, the quality of the SBC and integrated sound card
was checked. Two identical type I reference microphones (Bruel
and Kjaer type 4189 microphone capsule) and pre-amplifiers
were placed directly beside each other, close to a loudspeaker in
the anechoic chamber. One reference microphone was connected
to a dedicated noise measurement hardware system (Bruel and
Kjaer PULSE software system, with front-end type 3560C), the
other one to the SBC (with an additional ICP feeding). Both
logging systems were put outside the anechoic chamber to
prevent noise generated by instrumentation fans. Both pink noise
and a 1 kHz pure tone were produced by the loudspeaker, at
various intensities. The only difference that could be observed
was a very small increase in the noise floor in the measurement
based on the SBC (for a 1 kHz pure tone, there was an increase
from 11 dB to 13 dB). Since these are extremely low levels, the
quality of the SBC was considered to be very good.
For the detailed testing, each microphone under test was
connected to a SBC. The amplification factor of the sound card
determines the dynamic rangeof the measurements. In a first test,
this factor was set as high as possible, without increasing the
noise floor relative to lower amplification values. This resulted in
fractions near 0.2 to 0.3 of the maximum possible amplification.
The reference noise level in the indoor test (REF0) is in all cases
provided by the reference microphone (Bruel and Kjaer type
4189 microphone capsule) connected to the dedicated noise
measurement hardware system (Bruel and Kjaer PULSE soft-
ware system, with front-end type 3560C).
Test results are shown in Table 1 and Fig. 1. Saturation near
100 dB was not observed for any of the microphones. As for the
546 | J. Environ. Monit., 2011, 13, 544–552This journal is ª The Royal Society of Chemistry 2011
microphones could be measured. For a 1 kHz pure tone, the
noise floors of the measurement microphones (REF1, REF2 and
TYPEII) are smaller than or equal to 15 dB. The much cheaper
MEMSmicrophone has a noise floor ofonly 23 dB. Forthe other
electret microphones, noise floors are significantly higher.
ELECTRET1, ELECTRET3 and ELECTRET4 have noise
floors between 32 and 36 dB. ELECTRET2 has a noise floor
exceeding 40 dB for the 1 kHz pure tone.
The frequency-dependent microphone sensitivity at 70 dBA
(measuredat REF0) isshown in Fig. 1. Pinknoise wasemitted by
the loudspeaker over the full audible frequency range. The level
difference at the 1 kHz 1/3 octave band between REF0 and the
tested microphones is used as a constant factor to correct other 1/
3 octave band levels. REF1 and REF2 have an almost flat
response up to 10 kHz. The TYPEII microphone gives a devia-
tion of a few dB at 10 kHz. ELECTRET1 and ELECTRET4
have a reasonably flat response up to a few kHz. At 10 kHz,
a deviation above 10 dB is measured for both. The MEMS
microphone, ELECTRET2 and ELECTRET3 show strong
deviations from flatness over the sound frequency range
Total sound pressure levels ranging from 50 dBA till 90 dBA
were considered for assessing linearity in the frequency response.
A linear response means that the deviations from a flat frequency
response are independent of the total sound pressure level at the
microphone. Highly linear behaviour is found for REF1, REF2,
and TYPEII over the full audible range (not shown). For
MEMS1, highly linear behaviour is observed up to 10 kHz. For
ELECTRET1, ELECTRET3, and ELECTRET4, linearity of the
frequency response is limited to 3–4 kHz. ELECTRET2 has
a non-linear behaviour over the full frequency range.
In general, it can be concluded that both the frequency
response and noise floor are related to the price of the micro-
phone. With increasing cost, the noise floor decreases and the
frequency response becomes more flat and linear. The non-flat
frequency response can be corrected for to some extent in the
processing software assuming that it is stable over time.
Combining several microphones at the same measurement node,
while applying cross-correlation techniques, is a possibility to
reduce the noise floor. However, such operations complicate the
signal processing in the data acquisition equipment too much for
Weather resistance and wearing can only be realistically tested
outdoors. Therefore, a half-year lasting outdoor test was set up.
A sufficiently long test period is needed to assess the microphone
performance in various weather conditions, and to check
cumulative weather effects. The experiment started at December,
21 in 2009 and ended at June, 30 in 2010.
All microphones were attached next to each other on a 2 m
wide horizontal bar on the roof of the Zuiderpoort-building in
the city of Ghent (Belgium), with direct view towards a busy
viaduct (at about 150 m, at an almost equal height as the roof
level and parallel to the test bar). The average microphone height
was 1.7 m relative to roof level. The hourly equivalent total
sound pressure levels are at most days limited to 65 dBA during
daytime, and drop to 50 dBA during the calmest hours at night.
A typical hourly equivalent frequency spectrum during morning
rush hour is depicted in Fig. 2.
microphones were placed in between them. The use of two refer-
ence microphones will provide certainty on the correct level, and
will also show the measurement difference over time that occurs
even when using best available techniques to measure sound
pressure levels. Since both reference microphones are located at
the ends of the bar, the maximum difference in sound pressure
level by the non-coinciding location is accounted for as well.
In front of the test bar, an outdoor loudspeaker (Bose Free-
Space 360P series II) was placed. At fixed times (twice a day, at
10.00 h and 22.00 h) 10 s pink noise events were emitted, with
sufficient energy in the 1/3 octave bands from 100 Hz to 10 kHz,
relative to the environmental noise levels. This is done since the
dominant traffic noise at the test location is rather limited in
frequency content, with small day-by-day variability. In this way,
the variation in spectral sensitivity over time could be checked
over a broader frequency range. The distance between the
loudspeaker and the test bar was 3.5 m.
anechoic chamber for pink noise with a total sound pressure level of 70
dBA (measured at REF0), relative to REF0. The level difference at the
1 kHz 1/3 octave band between REF0 and the tested microphones is used
as a constant factor to correct other 1/3 octave band values.
Frequency response of the tested microphones as measured in an
morning rush hour.
Typical hourly equivalent spectrum measured at REF1 during
This journal is ª The Royal Society of Chemistry 2011 J. Environ. Monit., 2011, 13, 544–552 | 547
On-site meteorological data was measured. The local wind
speed and air temperature were recorded with sensors, placed at
both ends of the test bar (see Fig. 3). Other relevant meteoro-
logical parameters like relative humidity and rainfall intensity
were obtained from a meteorological observation station (above
roof level) at 1.3 km from the test bar. Hourly averaged
meteorological data are available during the full monitoring
period. The winter period was characterized by long freezing
periods, and mostly high relative humidity values. The minimum
air temperature recorded near the microphones was ?10?C; the
maximum air temperature exceeded 30?C. Wind speeds were
mostly limited, and were at maximum 6 m s?1. The cumulative
rainfall intensity during the test period was 136 mm.
The basic logging consisted of 1 s equivalent sound pressure
levels, expressed in 1/3 octave bands. Each microphone was
connected to its own SBC. All SBCs were connected in a small
computer network. The clocks of the SBCs were synchronized by
a network time protocol (ntp) server.
The calibration values as obtained from the indoor test were
used, since exactly the same measurement chains were applied for
each microphone. It was chosen not to have (hard) calibration
moments during the experiment (see further for more details on
Results of the outdoor test were analysed bymeans of correlation
analysis and by calculating the long-term average level difference
between the microphones under test and one of the reference
microphones (REF1). Similar measures were used in ref. 17 for
the comparison of noise dosimeters. Furthermore, the influence
of meteorological parameters and the temporal frequency-
dependent variability are further explored.
Linear correlation analysis between the synchronized time series
of two microphones shows how well the course over time is
followed. The correlation between each microphone under test
and REF1 is calculated, for each hour separately during the test
period. The hourly correlation coefficient (R2) over time for total
A-weighted sound pressure levels is depicted in Fig. 4 for each
microphone separately. These same data are shown in a more
condensed way by means of the cumulative distribution curves in
Fig. 5. The results in Fig. 4 show that the R2-values can change
significantly from hour to hour. Also when comparing both
reference microphones (REF1 and REF2), correlation coeffi-
cients deviate from the optimal value of 1. Fig. 4 shows e.g. that
ELECTRET2 and ELECTRET3 failed in February. ELEC-
TRET1 failed at the end of June. During a sufficiently long
period, these microphones did not show any correlation with the
reference microphone. These microphones were removed from
the test at the end of month where failure was observed.
All microphones, except for ELECTRET2 and ELECTRET3,
have a limited fraction of data at low correlation classes. The
second reference microphone, REF2, has only 2.5% of data with
a correlation coefficient R2lower than 0.9. At this same corre-
lation value, ELECTRET4, TYPEII, ELECTRET1, and
MEMS1 have a cumulative fraction of 9.5%, 15.8%, 16.9%, and
Although the large difference in cost, ELECTRET4 shows
better hour-by-hour correlation with REF1 than TYPEII.
ELECTRET1 has a similar steep slope in the cumulative distri-
bution curve starting from R2¼ 0.9 as also observed in the case
of ELECTRET4 or TYPEII; the higher fraction at very low
correlation values is caused by the period with microphone
failure near the end of the outdoor experiment.
Deviations in measured overall sound levels
A more practical quality measure is a long-term averaged esti-
mation of the error in dBA when using a particular type of
microphone. Although all devices were calibrated, the measured
sound pressure levels in the first hours of operation outdoors
were set equal to REF1. In this way, the difference in location of
the microphones on the test bar, and the influence by the pres-
ence of the other microphones (e.g. reflection and scattering of
sound waves on their housings) will be levelled out. This virtual
calibration can be done in different ways. In a first approach
(approach a), only the total A-weighted level difference is
adjusted for. In the second approach (approach b), the different
frequency responses are taken into account. A correction factor
Photograph of test bar showing the tested microphones and some
microphone and REF1 during the full monitoring period.
Hour-by-hour correlation coefficients R2between each tested
548 | J. Environ. Monit., 2011, 13, 544–552 This journal is ª The Royal Society of Chemistry 2011
is then calculated for each 1/3 octave band separately. This
additional virtual calibration was not performed during the
correlation analysis since a fixed offset in level between two
microphones will not influence the correlation coefficients. In
practice, regular calibration is common in long-term noise
monitoring. Therefore, the effect of on-site (virtual) calibration
at the beginning of each month on the global error is assessed as
Results for the globally averaged error (based on hourly
equivalenttotal A-weighted sound pressure levels) for each tested
microphone, relative to REF1, are given in Table 2. The values in
between brackets are the standard deviations. The graphs in
Fig. 6 show the evolution over time by means of monthly aver-
aged errors for the single and month-by-month virtual calibra-
tions, combined with approaches a and b.
For the one-time on-site calibration at the beginning of the
experiment, the global error obtained in dBA over the full
monitoring period at REF2, relative to REF1, amounts to
0.7 dBA. When applying a month-by-month calibration, this
error reduces to 0.5 dBA. These will be the minimum errors that
can be expected with the cheaper microphones. Even with ‘‘best
available techniques’’, there is still a non-negligible variability in
the measured noise levels. The TYPEII microphone gives
a global error of 1.0 dBA. Monthly calibration leads to a slight
increase in the error. ELECTRET4 and ELECTRET1 give
similar errors for the one-time calibration of 1.6 dBA. For
multiple calibrations, this deviation relative to the reference
microphone reduces to respectively 1.3 and 1.5 dBA. Such
deviations are still acceptable, certainly in light of the variation
observed between reference microphones. The other micro-
phones in the test resulted in much higher errors. MEMS1,
ELECTRET2 and ELECTRET3 give 2.8 dBA, 4.3 dBA, and 6.6
dBA, respectively (single calibration, approach a). The MEMS1
microphone leads to much higher errors in the case of a month-
by-month calibration. The reason for this will be explained when
discussing meteorological effects on microphone performance
(see next section). The behaviour of ELECTRET3 is hardly
affected by month-by-month calibration, while for ELEC-
TRET2 an improvement is observed relative to a single calibra-
tion in time. For MEMS1, ELECTRET2, and ELECTRET3,
small differences between applying approach a and approach
b are observed, in contrast to the other microphones in the test.
For these 3 microphones, the frequency response differs much
more from flatness than for the others. However, approach
b does not decrease the overall error, relative to approach a. A
main cause for this is the presence of a very dominant traffic
noise source at our location, characterized by limited variation in
frequency content over time. At locations with a variety of other
noise sources, it is expected that approach b will lead to an
improvement, relative to approach a.
When looking at the evolution of monthly averaged errors for
REF2, there is an increasing trend. In the beginning of the test,
errors were very minor (<0.3 dBA), but they exceed 1 dBA
starting from June 2010. Month-by-month calibration seems to
temper this increase to a limited extent only.
(near 1 dBA) during the monitoring period. ELECTRET4 had
a very limited deviation relative to REF1 in the first 3 months of
the experiment (0.5dBA) andstayedbetween 1and2dBAduring
the rest of the experiment. Month-by-month calibration seems
especially interesting to limit errors in the second half of the
monitoring period and could result in a decrease in the error of
near 0.5 dBA relative to the one-moment calibration at the
beginning of the experiment. The deviation produced by ELEC-
TRET1 stayed constant near 1 dBA in the first few months of the
experiment, but gradually increased starting from March on. In
June, errors exceeded 2 dBA in all calibration approaches
considered, caused by microphone failure.
ELECTRET2 and ELECTRET3 do not seem to be suited for
long-term outdoor noise monitoring. While in December 2009
between each tested microphone and REF1.
Cumulative distribution curves of the correlation coefficients R2
end of June 2010), relative to REF1. Results are shown for a single (virtual) calibration at the beginning of the experiment, and for a monthly (virtual)
calibration. Approach a corrects for total sound pressure level only, approach b takes into account the possible non-flat (location-dependent) frequency
response. The values in between brackets are the standard deviations
Averaged, hourly equivalent total sound pressure level in dBA during the full monitoring period (from the middle of December 2009 until the
Single calibration in time Month-by-month calibration
Approach a (dBA) Approach b (dBA) Approach a (dBA)Approach b (dBA)
This journal is ª The Royal Society of Chemistry 2011 J. Environ. Monit., 2011, 13, 544–552 | 549
and January 2010 a constant error was observed, it increased
tremendously during February (which is, however, the month of
the failures, while data of the full month are included here). Even
the error in the beginning of the experiment was high, especially
in the case of ELECTRET3.
The MEMS1 microphone shows a completely different course
of the error over time. While in the first half of the experiment
very high errors were observed, an error of only 1 dBA (relative
to REF1) remained near the end of the experiment.
Influence of meteorological parameters
In Fig. 7, scatter plots between the error relative to REF1 and the
measured on-site air temperature are presented for some of the
tested microphones. Microphones that failed during the moni-
toring period are not considered here. Results for the one-
moment calibration and approach b are used. In Fig. 8 and 9, the
influence of relative humidity and local wind speed is shown.
The hourly equivalent level difference between the reference
microphones REF1 and REF2 shows no dependence on the
meteorological parameters measured. The difference between
microphone TYPEII and REF1 shows only a very small
dependence on temperature and relative humidity. ELECTRET4
is strongly (negatively) correlated with temperature, and posi-
tively correlated with relative humidity. Note, however, that at
the test location, air temperature and relative humidity are
strongly correlated as well. With increasing air temperature,
relative humidity decreases. MEMS1 shows inconsistent behav-
iour at air temperatures below 20?C, and similarly at high
relative humidity values, while a very limited error is found at
higher temperatures. This is consistent with the trend towards
smaller errors starting from May 2010, as shown in Fig. 6. If
periods with inconsistent behaviour occur during the moments of
(monthly) calibration, high errors in the rest of the month can be
expected. This seems to bethe reason for the high errors observed
in the case of month-by-month calibration of the MEMS1
microphone (see Table 2).
As an example, the effect of applying a simple temperature
correction for ELECTRET4 on the global error is assessed. The
best-fitted, linear regression curve reads C ¼ ?0.174T + 0.113,
pressure levels. Single and monthly calibrations are considered, combined with approaches a and b.
Month-by-month evolution of the deviation of the tested microphones, relative to REF1, for monthly averaged, hourly equivalent total sound
between the selected microphones and REF1, for the single moment
calibration using approach b. The full monitoring period is considered.
Scatter plots between air temperature T and the difference
550 | J. Environ. Monit., 2011, 13, 544–552 This journal is ª The Royal Society of Chemistry 2011
where C is the correction factor to be subtracted from the
ELECTRET4 data when considering total hourly equivalent
sound pressure levels in dBA, and T is the on-site air temperature
expressed in?C. The correlation coefficient R of this regression
line equals ?0.79. Note that the intercept depends on the mete-
orological conditions at the moment of calibration. Since air
temperature and relative humidity are also correlated at the test
site (R ¼ ?0.73), correcting for a single parameter was found to
be sufficient. As a result, the global error (single calibration,
approach b) over the full monitoring period was reduced from
1.6 dBA (with a standard deviation of 1.5 dBA, see Table 2) to
0.8 dBA (with a standard deviation of 0.8 dBA). The availability
of air temperature data could lead to smaller deviations relative
to reference equipment. Note that air temperature sensors are
typically very cheap. In Fig. 10, the effect of applying the
temperature correction on the evolution of the hourly error
relative to REF1 over the full monitoring period is shown. In the
uncorrected case, a clear drift in the data is observed when per-
forming a single virtual calibration at the beginning of the
experiment. By applying the air temperature correction, this
long-term drift is removed.
In Fig. 11, a detail of the first ten days of May, 2010 is depicted,
together with the air temperature measurements in this same
period. In the uncorrected case, it can be observed that at night
decreases. During the daytime, this error increases because of the
higher air temperatures. This is consistent with the fact that the
calibration was performed at a moment where the air temperature
was near 0?C. It seems that the magnitude of the correlation
the corrected case, day–night variation is still visible in the results.
However, the average error has become very small.
Starting from about 3 m s?1, the error obtained by the TYPEII
microphone strongly increases with increasing wind speed. At
5 m s?1, an error of near10 dBA is observed. A plausible reason is
the limited quality of the weather protection unit which was
delivered with this microphone. For the other microphones, no
trend between the on-site wind speed at microphone height and
the error relative to REF1 was observed. The rather broad data
spread at e.g. ELECTRET4 is caused by the coincidence of wind
speed with more influencing meteorological parameters like
temperature and relative humidity.
See caption of Fig. 7, but now for relative humidity, indicated by
wind speed, indicated by u.
See caption of Fig. 7, but now for the magnitude of the on-site
ELECTRET4, relative to REF1, for total noise levels during the full
monitoring period. The uncorrected case and the air temperature
corrected case are presented. In both cases, approach b is used.
Evolution over time of the hourly equivalent errors by using
May, 2010 is shown. In the lower panel, the evolution of the on-site
hourly averaged air temperature during these days is depicted.
See caption of Fig. 10, but now a detail in the first ten days of
This journal is ª The Royal Society of Chemistry 2011 J. Environ. Monit., 2011, 13, 544–552 | 551
Variability of spectral sensitivity over time
The temporal variability of the spectral deviation at all micro-
phones, relative to the reference microphone (REF1), is shown in
Fig. 12. This figure is assembled by considering the measure-
ments during the pink noise events that were emitted twice a day
(at 10.00 h and 22.00 h) by the outdoor loudspeaker placed in
front of the test bar. The noise levels produced by the loud-
speaker were typically 10 dB higher than the environmental noise
levels in the 1/3 octave bands ranging from 100 Hz to 10 kHz.
Measurements are referred to REF1 to account for possible
changes in the sound produced by the loudspeaker during the
ELECTRET4 shows a very constant performance over the
frequency range considered. The standard deviations are limited
and rather frequency-independent. The magnitude of the stan-
dard deviation is somewhat higher than the one observed at the
TYPEII microphone. For the latter, however, the frequency
response is more flat at higher frequencies, as can be seen in
Fig. 1. At most 1/3 octave bands, the TYPEII microphone shows
a slightly increased variation over time compared to REF2.
ELECTRET1 shows constant but much higher standard devia-
tions when compared e.g. to ELECTRET4. The MEMS1
microphone is characterized by a strong variability, which seems
to be most pronounced below 1 kHz. Note that for microphones
with a low variability in the frequency response, a non-flat
spectral response can be more accurately calibrated out.
Conclusions and discussion
In this study, it is assessed to what extent cheap microphones,
appearing in consumer electronics, can be used for environ-
mental noise monitoring. The long-term outdoor test showed
that it is possible to identify microphones that only resulted in
a small additional averaged error (limited to 1 dBA), relative to
the differences occurring between reference microphones them-
selves (measured at 0.5–0.7 dBA). This additional (but limited)
error must further be seen in the viewpoint of the very large
increase in cost of reference equipment relative to the cheaper
In this study, it was shown as an example that air temperature
correction could further reduce the difference in measured sound
pressure levels relative to the reference microphone. Air
temperature sensors are very cheap so the cost of the measure-
ment node will only increase to a very limited extent.
Replacing microphones that failed after a given period (e.g.
after a few months) is another viable option. The latter is still
much more economic given the huge difference in prices
compared to dedicated measurement microphones. Even for the
latter, human intervention is needed (for calibration checking)
and replacing the cheap sensors could become a part of normal
operation. A quick and reliable identification of deviant sensor
behaviour is then of course important in an extended micro-
phone network. Algorithms will be developed to perform this
This research is part of the IDEA (Intelligent, Distributed
Environmental Assessment) project, a 4-year strategic basic
research project, financially supported by the IWT-Vlaanderen
(Flemish Agency for Innovation by Science and Technology).
1 L. den Boer and A. Schroten, Traffic Noise Reduction in Europe:
Health Effects, Social Costs and Technical and Policy Options to
Reduce Road and Rail Traffic Noise, Report of CE Delft,
Commissioned by the European Federation for Transport and
Environment (T&E), Brussels, 2007.
2 D. Ouis, Journal of Environmental Psychology, 2001, 21, 101–120.
3 B. Griefahn, A. Marks and S. Robens, J. Sound Vib., 2006, 295, 129–
4 B. Griefahn, P. Br€ ode, A. Marks and M. Basner, Sleep, 2008, 31, 569–
5 C. Clark, R. Martinez, E. van Kempen, A. Tamuno, J. Head,
W. Hugh, M. Davies, M. Haines, I. Lopez Barrio, M. Matheson
and S. Stansfeld, Am. J. Epidemiol., 2005, 163, 27–37.
6 W. Babisch, Noise and Health, 2008, 10, 27–33.
7 L. Barregard, E. Bonde and E.€Ohrstr€ om, Occup. Environ. Med.,
2009, 66, 410–415.
8 Directive 2002/49/EC of the European Parliament and Council of 25
9 T. VanRenterghem,E. SalomonsandD.Botteldooren,Appl.Acoust.,
2006, 67, 487–510.
10 E. Salomons, H. Polinder, W. Lohman, H. Zhou, H. Borst and
H. Miedema, J. Acoust. Soc. Am., 2010, 126, 2340–2349.
11 T. Kihlman, M.€Ogren and W. Kropp, Proceedings of the 31st
International Congress and Exhibition on Noise Control Engineering,
12 R. Barhama, M. Goldsmith, M. Chan, D. Simmons, L. Trowsdale
and S. Bull, Proceedings of 8th European Conference on Noise
Control, Edinburgh, 2009.
13 M. Bell and F. Galatioto, Proceedings of 8th European Conference on
Noise Control, Edinburgh, 2009.
14 Standard of the International Electrotechnical Commission IEC
15 P. Scheeper, B. Nordstrand, J. Gullov, B. Liu, T. Clausen, L. Midjord
and T. Storgaard-Larsen, J. Microelectromech. Syst., 2003, 12, 880–
16 A. Touhafi, B. Coudron, G. Braeckman, M. Raadschelders and
J. Wambacq, Proceedings of 17th International Congress on
Acoustics, Rome, 2001.
17 F. Huang, T. Shih, J. Lee, T. Wang and P. Wang, J. Environ. Monit.,
2010, 12, 748–758.
for individual 1/3 octave bands ranging from 100 Hz to 10 kHz, over the
full monitoring period. The pink noise events emitted twice a day by the
outdoor loudspeaker are used to assemble this figure.
Outdoor level variability at all microphones, relative to REF1,
552 | J. Environ. Monit., 2011, 13, 544–552 This journal is ª The Royal Society of Chemistry 2011