Vous êtes sur la page 1sur 5

Acoustic Instruments and Measurements

July 2014, Argentina

MICROFONE CALIBRATION BY FUNCTION TRANSFER


COMPARATION METHOD
NICOLAS DI DIEGO
ANDRES MITRE
ANDRES E. LARRAHONA
EZEQUIEL ALALA
JULIAN A. TINAO
UNTREF. Universidad Nacional Tres de Febrero, Buenos Aires, Argentina.
d.juliant@gmail.com
an.dres.l@hotmail.com
mitreandi@gmail.com
Abstract A microphone calibration curve is measured using a comparative method based on the transfer function. A
Behringuer measurement microphone is calibrated using an Earthworks M50 measurement microphone. The method
gave repeatable results and proved great immunity to external noise and not anechoic conditions. A dBSPL level
calibration in Smaart software is made using Earthworks M50 as a reference. Also an impulse response of the
Behringer microphone is found using convolution methods assuming the Earthworks M50 impulse response as a perfect
Dirac for reference.

1. INTRODUCTION
It is known that every microphone has
imperfections. Even the most expensive measurement
microphones will introduce error to the measurement.
An expensive measurement microphone will come
with a measured correction curve and sensitivity, this
curve is measured using advanced equipment [1]. A
correction curve is always useful as it will contribute to
the truth of the measure, but sometimes this curve is
not proportionated by the manufacturer. The purpose of
this investigation is to develop a method to find a
correction curve for an unknown microphone using
another certified microphone as a reference.
Also its possible to measure dBSPL with a
standard sound card and a microphone if it is known
the relation between dBSPL and dBFS on the software.
This relation depends on the microphone and the sound
card sensitivity. A level calibration was made using the
Earthworks M50 and its known sensitivity as a
reference.

an anechoic chamber. The room has good isolation but


there was music and speech in the contiguous rooms
that were perceived. These conditions are far from
optimal for a calibration measurement, but thanks to
the method used, the effects of these negative factors
were diminished.
3. INSTRUMENTS
The following instruments were used:

Microphone Behringer ECM 8000


Microphone Earthworks M50
Roland Quad Capture sound card.
Notebook
2 Microphone Stands
Measuring tape
Milivoltimeter
Dynaudio BM6A

4. CALIBRATION CURVE

2. MEASUREMENT CONDITIONS

4.1. CALIBRATION CURVE MEASUREMENT

The measure had place in the laboratory room. It is


a room relatively well acoustically damped, but its not

The calibration curve was measured using transfer


function from Smaart software. The two microphones

were connected to the Roland Quad Capture sound


card one in each input channel. A speaker was
connected to the output of the sound card. The measure
was made using the pink noise generated from the
Smaart itself. The speaker (Dynaudio BM6A) was
placed on the floor pointing upwards, and the two
microphones were pointed at approximately 70cm, in a
symmetric and coincident position as shown in Figure
1.

Figure 1: Speaker and microphones setup.


This setup was found to have a lot of advantages.
First, its immunity to background noise, due to the
nature of transfer function. Transfer function uses one
channel as reference, and calculates the difference that
exists with the other channel. The Earthworks mic was
connected to the reference channel, and the Behriguer
to the measure channel. This way, the background
noise and even the sound reflections are perceived by
the microphones as the same as the measure signal, as
function transfer calculates the difference of anything
that is being received by the mics, that is exactly the
same, at least until a hi frequency limit, were the size
of the mics become comparable to the wave lengths.
For that reason, the distance between mics was
minimized. It was found that the size of the mics set
the maximum frequency that the method can measure
with good certainty. As down in the results is shown a
good certainty exists until 10kHz approximately, this
corresponds with wave lengths of 34mm. At these
wave lengths each capsule itself generates diffraction
which modifies the signal that each mics receive and
the measurement becomes highly dependent on the
mics position introducing uncertainty.
All this was possible to appreciate thanks to the
real-time nature of the method.

For minimizing the uncertainty the position of the


mics was carefully achieved as shown in Figure 2. It
was possible to align the true acoustic center of the
mics thanks to the delay finder of Smaart software
assuming that when the delay between channels is 0
the mics are acoustically aligned. Due to the
differences of each mic capsule this position differed a
just a bit more than 1 millimeter from the metal top of
the mic.

Figure 2: Carefully placed microphones.


With the measurement repetition was found that the
mic placement become important only for frequencies
above approximately 9kHz where the uncertainty was
introduced. The method showed a really good
immunity to position misplacement from 8kHz and
below.
The Smaart setup was as follow :
Bit depth : 16bit
Sample rate : 48000Hz
FFT : MTW
Average : 16
Post Magnitude smooth : 1/6oct
The average was set to 16 just for precaution but
the measure showed the same stable results with
inferior averages. The magnitude smooth was selected
to optimize the repeatability of the results. It was
possible to check it on real time. It was found that
without magnitude smooth the results were in some
way less repeatable. Magnitude smooth averages
adjacent
frequency
magnitudes
concealing
measurement error [2]. As shown in Figure 3, with less
magnitude smooth the results from each measurement

contain information that cannot be taken as true due to


the lack of the repeatability. As shown in Figure 4 with
a 1/6oct of magnitude smooth, the results were highly
repeatable.

Figure 5: Results for 4 measurements with


completely different positions.

Figure 3: Four different measurements with 1/48oct


Magnitude smooth.

Below 35Hz the deviation of the results increase,


this is due to the lack of signal in that band of
frequency. The frequency response of the speaker was
measured, and showed almost no energy below 35Hz,
on the other extreme, it had good energy above 20kHz.
The measurement was restricted to 20kHz to be sure of
being out of the limitation of sampling rate.
4.2. CALIBRATION CURVE RESULTS
The four measurements were averaged using
Smaart Trace Average option. Figure 6 show the
average result curve.

Figure 4: The same four measurements of Figure 3


but now with 1/6oct magnitude smooth.
The function transfer method itself already has
averaging, because of that the result of one measure
was extremely stable. The exact formula that Smaart
uses for coherence function is unknown, but in some
way it expresses the certainty of the measurements.
The coherence is shown in Figure 4 and Figure 5 as the
red line on top. The coherence in the measurements
were >98%, 100% almost the full spectrum with
exception of the range over 9kHz and below 35Hz that
showed values >98%. For that reason, acquiring more
than one measure in the same conditions, which only
involved turn on-off the system, or taking another
instant snapshot was nonsense as it gave us almost
exactly the same results.
Therefore, four different measurements were done,
disconnecting the setup, moving the speaker to a
complete different position, and making another
similar mic arrangement every time at different
distances, 70cm, 75cm, 65cm, 68cm, from speaker to
microphones.
Figure 5 show the results of the four measurements.
As it can be seen in Figure 5 the method proved to be
highly repeatable from 8kHz and below, and showed
limitations for acquiring the same standard deviation
results over 10kHz.

Figure 6: Average result of the four measurements


in Smaart.
The four measurements were exported to excel and
a standard deviation is calculated. Figure 7 show the
standard deviation of the curve for only the 1/3oct
frequency values and the correction value for each
frequency.
Figure 8 show the average result calibration curve
and its standard deviation.

Figure 8: Correction curve (red) and its standard


deviation (green).

Figure 9: Sensitivity calibration in Smaart.


The sensitivity calibration offset in Smaart for the
Quad Capture sound card with minimum sensitivity
was : 157.41dB.
6. IMPULSE RESPONSE MEASSURMENT

Figure 7: 1/3oct values of correction curve and its


standard deviation.
The average correction curve was inverted in
Smaart software and exported to ASCII. A text archive
was created as this is the standard correction curve
format used on most measuring softwares.

5. SENSITIVITY CALIBRATION
The same microphone and speaker placement used
for the correction curve was used for sensitivity
calibration. A milivoltimeter was connected directly on
the output of the microphone. A 1kHz sine wave was
reproduced. The speaker level was such that the
milivoltimeter measure 1mV. This level assures
94dBSPL according to Earthworks sensitivity. Then in
Smaart software a sensitivity calibration was made,
introducing the 94dBSPL as a reference, using the
Behringer mic that was in the same position as
Earthworks mic. It is assumed both mics receive the
same level due to sharing the same position. The gain
in the sound card was set at minimum position to
ensure repeatability of the measurement.
Figure 9 show the sensitivity calibration screen in
Smaart software, and the calibration parameter
measured for the Behringer ECM 8000.

Using the same placement setup mentioned a test


signal was recorded with both microphones at the same
time. The signal was a logarithmic sine sweep of 15
seconds with frequency varying from 20 to 35kHz.
Using Farinas convolution method [3], the
recorded signal was convoluted with the inverse of the
logarithmic sine sweep to find an impulse response.
Four measurements were made. A previously
developed matlab code exposed in In situ
measurement of absortion of a material [4] was used
for the multiple convolutions.
The impulse response measured is a summation of
the speaker impulse response, the room impulse
response, the sound card impulse response and the
microphone impulse response [5].
The impulse response captured with the Earthworks
mic was phase inverted and summed with the
Behringer one. This is expected to cancel out the room,
sound card, and speaker impulse response, as due to
the nature of the method, all these impulse responses
are equally received by both mics. The resulting
impulse response should be the difference between the
Behringer impulse response and the Earthworks
impulse response. The impulse response of the
Earthworks is unknown, but if we assume that it is far
more ideal than the Behringer one. Then this result can
be assumed to be close to the Behringer microphone
impulse response.
The process was made in Nuendo for every pair of
the four measurements instances. Despite each impulse
response of each instance of measurement were
slightly different, the result of the summing with the
phase inverted impulse response of the Eartworks mic
result in four impulse responses that were quite similar.

Figure 10 show two of the Behringer impulse


response calculated. It can be seen the similarity
between them.

8. REFERENCES
[1] Tipton R. DIY Microphone Calibration. Audio
Press. USA, 2009.
[2] Smaart Live User Manual, Rational Acoustics,
2010.
[3] Farina A., Advancements in impulse response
measurements by sine sweeps. AES Convention
122nd Paper 2007.
[4] Abdala E., Tinao J., Larrahona A., Di Diego N.,
Mitre A. In situ measurement of absortion of a
material, Acoustical Instrument and Measurement
Paper, UNTREF, 2014.
[5] Proakis, G., J., Manolakis, G., D. Digital Signal
Processing. Prentice Hall. 2007.

Figure 10: Two measurements of Behringer impulse


response.
7. CONCLUSIONS
Comparison
between
two
measurement
microphones in a new way was made.
A method to obtain results with very small
variations measurement to measurement was used
This is a robust method of comparison because only
requires taking a good signal to noise relation because
the microphone is located almost in the same physical
location and both microphones are taking the same
signal, in addition to this a relatively high averaging
time is used. This allowed measurement at levels of
error and inaccuracy very low regardless of the
acoustic characteristics of the room, which are the
same in both microphones and when the transfer
function is performed, all that information in another
acoustic measurement system would be a problem,
disappears. The method proved to be highly repeatable
below 9kHz. Above that the diffraction from both mics
introduces error in the measurement.
The results obtained with the impulse responses of
both microphones were analyzed and checked, which
confirmed the validity of the results obtained by the
transfer function. Is pending for further work to
analyze the calculated impulse responses in matlab to
find the standard deviation of the measurement.

Vous aimerez peut-être aussi