You are on page 1of 24

SkyGuard Enterprises

253 Humpty Way


Springfield, NM 38492
October 15, 2013
To:

Fred Armton, Vice President of Engineering Development

From:

Ali Alhamaly

Subject:

Experimental Investigation of the Feasibility of Using Acoustic Waves to


Determine the Location of a Sound Source

Summary and Introduction


As requested by the memo sent to us on August 27, 2013, we performed several tests
to investigate the effectiveness of using acoustics waves to find the location of a sound
source. Based on our analysis, we came up with a calibrated algorithm that is both accurate
and precise and is able to find the location of a sound source based on acoustics signals of
three microphones. The algorithm is believed to be applicable for scaled-up applications with
some modifications. In addition, we have conducted a simple study to decide on the best
method to find the speed of sound. The study shows that the theoretical calculation of speed
of sound is more accurate and also more practical.
The motivation behind this experimental work is to investigate the feasibility of using
acoustics measurements to detect the nuclear weapons testing in North Korea. The United
States Air Force (USAF) has proposed an idea of using high altitude microphones to locate
the position of the nuclear testing. USAF has hired us to make investigations on small scale
setup to assess the feasibility of the proposed idea.
The main experiments that we have conducted are the speed of sound measurement
and the determination of the location of a sound source using small microphones that are
fraction of the size that will be used by USAF. The speed of sound measurement utilizes two
small microphones that are mounted at a certain distant apart. In the location of a sound
source experiment, three microphones are used to detect the location of a sound source that is
mounted on a wooden panel with the measuring microphones. In both experiments, the
results can be obtained by analyzing the acoustic signals of the small microphones involved
in each experiment.
The main result of our experimental investigation is the algorithm for finding the
location of a sound source. The algorithm utilizes the acoustic signals of the measuring
microphones along with simple mathematical relations that relate the location of the
microphones through the speed of sound in the medium to the actual location of the sound
source. In addition, speed of sound determination and analyzing acoustic signature of an
aircraft are part of the results of this investigation.
This report presents the detailed experimental approach that was used to measure the
speed of sound and to conduct the sound source location experiment. The report starts by
describing the experimental setup for all the experiments that were done including the test
apparatus and the data acquisition system. Next, the report discusses the different methods of
calculating the speed of sound with the uncertainty analysis for each method. The next
section presents our proposed algorithm and describes how it works and how its
implemented. The section also presents the results of the algorithm and discusses the overall
1

accuracy. In addition, full uncertainty analysis is provided for all the results of the algorithm.
Next, we present the results of frequency decomposition of a sound signal. Finally, the report
discusses the extension of our algorithm into scaled-up applications.

Experimental Setup
This section presents the experimental apparatus and the test procedures that were
used to determine the location of a sound source. The section discusses the data acquisition
system, the speed of sound experiment and the microphone arrays experiment.
Experiment Apparatus. The experiment that we have conducted is divided into two

main sections: the determination of the speed of sound and the determination of a sound
source spatial location. The basic idea behind the two experiments is the transmission and
receiving of acoustic waves. Several microphones were used to receive the sound waves
generated by different sources. The sound wave output signals from the microphones were
connected to the National Instruments SC-2345 signal conditioning card. The output of the
SC-2345 was connected to a data acquisition device in which the analog signals from the
microphones were sampled at 15 kHz. The digitized sampled data were recorded and stored
by using LabVIEW.
The speed of sound experiment consists of two microphones mounted on a PVC
pipe and located 84 inches away from each other. The two microphones are connected to the
DAQ system in order to record their output signals for later data processing. Figure 1 shows a
picture of the two microphones with the PVC pipe.

Figure 1. Speed of sound test apparatus

The sound source location experiment consists of 8 x 4 wooden panel in which


several small speakers are mounted behind it. In addition, 15 known location microphones
are also mounted on the wooden panel. The microphones are installed in three sets in which
each set consists of five microphones. The output signal of the microphones can be connected
to the DAQ card to record the signals in LabVIEW. The maximum number of microphones
that can be connected at the same time is three microphones. Figure 2 shows a picture of the
wooden panel with the location of the 15 microphones.

Figure 2. Wooden panel with the mounted microphones for the sound source location
experiment
Test Procedures. One of the objectives of the current investigation is to measure the
speed of sound. The speed of sound was measured utilizing the speed of sound apparatus
shown in figure 1. The procedure for the test was fairly simple. A sound perturbation inside
the PVC pipe was produced by hitting the beginning the pipe with a wooden block. The hit
cause a sound wave to propagate through the pipe. The sound wave is captured by the two
microphones and the signal from the two microphones is recorded as well in LabVIEW. This
test was repeated for five different times. The signal analysis of this experiment and the
calculation of the speed of sound will be discussed in the next section of this report.
Additional to measure the speed of sound as discussed in the previous paragraph, the
speed of sound can be calculated theoretically by knowing the temperature of the air in
location were the speed of sound is needed to be determined. For this reason, we recorded the
temperature of the lab room using a thermometer. Calculation of the theoretic speed of sound
along with a discussion of which method is preferred to obtain the speed of sound is included
in the next section of this report.
The second experiment that we conducted was the determination of a sound source
location. The procedure of this experiment is as follows. From the different 15 microphones
that we have on the wooden panel (figure 2), we chose a set of three microphones (one
microphone from each section). These three microphones were then connected to the DAQ
card to record their output signals. A small speaker mounted behind the wooden panel was
used to produce a sound wave and the three microphones set to record that sound wave. This
procedure was repeated again for five different times and all the data from the output signals
were record and stored by LabVIEW. Once we finished recording the five trials, a new set of
three different microphones were used and the experiment was repeated again for the new
microphones set. The two different microphone combinations that we have used were:
(1,7,11) and (3,6,13).
The location of the small speaker was known for the first week of the experiment in
which we used the known location to calibrate our algorithm that finds the location of the
speaker. In the second week of the experiment, the location of the speaker was unknown and
it was our job to locate the speaker location using our calibrated algorithm.

Speed of Sound Calculations


This section presents the calculations of the speed of sound using both the theoretical
and experimental methods. The section includes discussion about the use of the two
microphone signals to determine the speed of sound. In addition, this section presents the
calculation of the random and systematic uncertainties associated with measuring the speed
of sound for both methods. The section concludes with our recommendation of which method
should be used to find the speed of sound.
Experimental Determination of the Speed of Sound. The experimental technique
of measuring the speed of sound relies on the simple concept of measuring the time
difference of arrival between two microphones. The time difference of arrival can be obtained
easily by analyzing the output acoustic signals of the two microphones that were recorded
during the speed of sound experiment. Once the time difference between the two
microphones is determined then the speed of sound can be calculated by dividing the distance
between the two microphones by the time difference of arrival. Mathematically the speed of
sound is given by

(1)

Where d is the distance apart between the two microphones and t is the time difference of
arrival between the acoustic signals between the two microphones. The distance between the
two microphones was 84 inches.
As mentioned earlier, the time difference in equation 1 can be obtained by looking at
the acoustic signals form the two microphones and what we expect to see is a two very
similar signals that are shifted slightly in time from each other and hence the time difference
between the two signals can be found by comparing two similar pattern between the two
waves like a valley for instance. Its known that a sound wave propagating through a medium
creates a low pressure region upstream of the wave front, and hence whenever we notice a
drop in the signal amplitude for any microphone we can tell that the sound wave just passed
by that particular microphone. This observation is used to find the sound wave time of arrival
for each microphone. An important comment here is that knowing the exact time of arrival
from the signal is a little bit tricky because its hard to know what value of drop in magnitude
should be used to identify the arrival of the sound wave because its dependent on the actual
response of each microphone. For this reason we used the first valley in the signal for each
microphone to find the time difference of arrival and this should be sufficient to find the time
difference because we noticed the same valley pattern for both microphones. Figure 3 shows
the output signal of the two microphones from the speed of sound experiment. The figure
shows that the two signals are similar and that they are shifted by some time, this time
difference is found by locating the time of first valley in the both signals as marked in the
figure.

Figure 3. Near and far microphone output signals for speed of sound experiment.

The speed of sound experiment was repeated for five different times as mentioned
earlier in the test procedures. The results of the five trials are summarized in table 1. Table 1
shows the time difference between the two signals for each trail along with the calculated
speed of sound using equation 1. Its worth to notice that the five different signals that we
obtained are very similar and they show the same pattern as figure 3. For this reason, the
signals for the other four trials are not shown and only the time difference results are given
within table 1. The results of table 1 indicate that the mean value of the speed of sound is
13830 in/s. Sample calculation of the results shown in table 1 is included in appendix A.
Table1. Main results of the speed of sound experiment for the five different trials

Trial
#
1
2
3
4
5

Time difference
t (s)
0.00607
0.00612
0.006048
0.0060699
0.0060609

Speed of
sound c (in/s)
13838.55
13725.49
13888.89
13838.78
13859.09

Theoretical Determination of the Speed of Sound. The speed of sound for an ideal
gas is derived from the assumptions that the pressure and temperature variations of the gas
across the sound wave are vanishingly small because the sound wave is assumed to be
infinitesimally thin and has infinitesimal strength. These assumptions lead the sound wave to
be an isentropic perturbation process. The speed of sound theoretic relation with the above
mentioned assumptions for ideal gas is given by

(2)

Where c is the speed of sound, is the specific heat ratio and for the case of air within the
temperature range that we conducted our experiment its equal to1.4, R is the gas constant
and for air its 287 J/kg K and T is the absolute temperature of the air. From equation 2 it can
be seen that the only parameter that need to be measured is the temperature of the room in
5

which the experiment is held at. The temperature of the room in which we performed the
experiment was 25 C which corresponds to 298 K. this leads using equation 2 to a speed of
sound of 13623 in/s. sample calculation of using equation 2 to find the speed of sound can be
found in appendix A.
Uncertainty Calculations of the Experimental Speed of Sound. Due to the

limitation in accuracy of the instruments that we used in the lab to measure the speed of
sound, there is definitely uncertainties associated with the final reported numbers. These
uncertainties are divided into random and systematic uncertainties. The random uncertainty is
a measure of the precision of the data collected, where the systematic uncertainty is a
measure of the accuracy of the measurement.
The measurement uncertainty of the measured speed of sound comes from the fact
that we had uncertainties in the measuring instruments. From equation 1, it can be seen that
the speed of sound depends on the measurements of the distance and time. The uncertainty in
the measured distance between the two speakers was 1/32 inches. The uncertainty of the
measured time in our instrumentation was 1/30000 seconds. This can is found from the
sampling frequency that we used. The sampling frequency that we used was 15 kHz, which
means that there is a time difference of 1/15000 seconds between each time measurement in
the DAQ system we used. Since the uncertainty in the measurement is half the resolution of
the instrument, hence the time uncertainty will be .5*(1/15000) which is 1/30000 seconds as
mentioned above. From these uncertainties, we need to find the total propagated error in the
measurement of the speed of sound. In order to do this we need to use the error propagation
formula which is given by
= (

Where uc is the measurement uncertainty of the speed of sound,

(3)

is the derivative of the

speed of sound with respect to an independent variable (in our case the variables are either
the distance or the time) and is the measurement uncertainty of the independent variable.
As mentioned above ud = 1/32 inches and ut = 1/30000 seconds. Equation 3 applied
specifically to the speed of sound formula (equation 1) yields the following expression
2

= ( ) + 2 ((
)
)2

(4)

We can see from equation 4 that the uncertainty of the speed of sound will have different
value for each trail that we conducted. This is because of the term in the equation in
which it varies from trail to trail. However, this variation is very small and any trial
calculation will yield almost the same value of the uncertainty. Table 2 shows the calculation
of the uncertainty in the speed of sound for each trial. Sample calculation of the uncertainty is
provided in appendix A.

Table2. Measurement uncertainty of the speed of sound for the five different trials

Trial #
1
2
3
4
5

Speed of sound
uncertainty uc (in/s)
107.59
105.84
108.37
107.59
107.91

As can be seen from table 2, all the uncertainties are very close to each other as expected. In
order to be conservative, we decided to take the largest value as the representative
measurement uncertainty, and hence the measurement uncertainty in the speed of sound is
108.4 in/s.
The random uncertainty in the measurement of the speed of sound can be found using
the standard error. The standard error is given by
=

(5)

Where is the standard error in the speed of sound measurement, is the standard deviation
of the speed of sound measurements and N is the number of trials. The standard error in our
measurement of speed of sound is 27.8 in/s. sample calculation of the standard error can be
found in appendix A.
The mean value of the speed of sound from our measurement is 13830 in/s. this value
is accompanied by measurement uncertainty of 108.37 in/s and random uncertainty of 27.8
in/s. the low value of the random uncertainty indicates high precision in the measurement of
the speed of sound. The measurement uncertainty is a reflection of the accuracy of the
instruments that we have used to measure the speed of sound and the value 108.37 in/s can be
improved by increasing the accuracy of the used instrumentation.

Uncertainty Calculations of the Theoretical Speed of Sound . The uncertainty


calculation of the theoretical speed of sound follows exactly same procedure as the
uncertainty in the experimental speed of sound. The only difference is the parameters that
lead to uncertainty in the speed of sound. To find the measurement uncertainty, equation 3 is
applied again but this time to equation 2. The expression of the measurement uncertainty is

(6)

uT is the measurement uncertainty of the thermometer we used and it has a value of 1K. In
equation 6 we assumed that the measurement uncertainty in is very negligible. This is
justifiable because has a constant value over the range of temperatures that we performed
the experiment. This means that the derivative of with respect to temperature is almost zero
and hence we can ignore its uncertainty.
The measured temperature was 298 K, = 1.4 and R = 287 J/kg K. these values give
measurement uncertainty in the theoretic speed of sound of 23 in/s. using equation 2 the
theoretic speed of sound is 13623 in/s. During the experiment that we have conducted, we
monitored the thermometer to see if the temperature changed from 298 K, but the
7

temperature remained constant at 298 K for the whole experiment. This means that the value
of the theoretic speed of sound didnt change and remained at 13623 in/s. this also means that
the random error in the theoretic speed of sound is zero since the value of the speed of sound
for all trials is the same. the final value of the theoretic speed of sound is (13623 23) in/s
From the uncertainty discussion for both the experimental and theoretical speed of
sound, we can see that the theoretical approach in calculating the speed of sound has less
uncertainty compared with the experimental approach. Because of this result, we recommend
that the speed of sound should be calculated using the theoretical approach since it has higher
accuracy than the experimental approach. In addition, the theoretical approach is very easy to
implement and only requires a temperature sensor to find the speed of sound which makes the
theoretical calculation more practical. From now and on, we will be using the speed of sound
determined by equation 2 throughout this report especially in the sound source location
calculations.

Determination of a Sound Source Location Using Acoustic Waves


This section presents the algorithm that we came up with to determine the location of
a sound source based on acoustic signals of three microphones. The section presents the
results from the algorithm for both the known source calibration experiment and the unknown
source location experiment. The section also discusses the uncertainties associated with both
experiments.
General Algorithm. Our algorithm to locate the sound source is based on analyzing
the acoustic signals from the three receiving microphones. The algorithm utilizes the idea that
the sound wave reaches the different microphones at different times due to the spatial
location differences between the three microphones.
Let the location of the three microphones in the plane be represented by the points P1,
P2 and P3 and let the location of the sound source be represented by the point Ps. The
coordinates of the microphones are given by P1 at (x1,y1), P2 at (x2,y2) and P3 at (x3,y3).
The coordinate of the unknown sound source is Ps at (xs,ys). From these locations we can
write three equations describing the distance between each microphone and the unknown
sound source, the three equations are given simply by the Euclidean distance between two
points in the plane

(1 ) 2 + (1 ) 2 = 1

(7)

(2 ) 2 + (2 ) 2 = 2

(8)

(3 ) 2 + (3 ) 2 = 3

(9)

In these equations R1, R2 and R3 correspond to the distances between the sound source and
the first, second and third microphone respectively. Examining equations 7-9, we can see that
they contain five unknowns (xs,ys,R1,R2 and R3) and hence the set of equations are not
closed and they dont have a unique solution. So, what we need to do here is to find relations
between the different distances from information that we have, namely from the acoustic
signals of the three microphones. Its also true that the right hand side of equations 7-9 can be
expressed as

1 = (1 0)

(10)

2 = (2 0)

(11)

3 = (3 0)

(12)

Where c is the speed of sound, t0 is the time that the sound source emits the sound wave and
t1, t2 and t3 are the times that microphones 1, 2 and 3 received the sound wave that was
emitted by the sound source. Substituting equations 10-12 into 7-9 gives
(1 ) 2 + (1 ) 2 = (1 0)

(13)

(2 ) 2 + (2 ) 2 = (2 0)

(14)

(3 ) 2 + (3 ) 2 = (3 0)

(15)

What is unique about equations 13-15 is that they contain only three unknowns, namely xs,
ys and t0. The speed of sound can be found using temperature measurements as was
described in the speed of sound calculations section. The three times t1, t2 and t3 can be
found by analyzing the output signals from the three microphones. As described in the speed
of sound calculations section, the time of arrival of the sound wave can be known by locating
the time of the first drop in the amplitude of the output time series signal. It was also
explained in the same section that finding the time of arrival from the time series signal is
tricky and has large uncertainty due to the unknown precise criteria for choosing the location
of first drop. The same argument holds for this situation as well and hence the time of arrival
for each signal is hard to know. For this reason, the times t1, t2 and t3 will be modified by
adding a constant time to each one them. This modification allows us to choose a time on the
output signal of the microphones that is easily recognizable and can be repeatedly chosen
from a different set of measurements. This modification in the timing can be justified by the
following argument. In equations 13-15, the right hand side is always a function of the
difference between the receiving time and the transmitting time. Hence, if we add a constant
time to both the reviving time (t1 for example) and the transmitting time t0, then the
difference t1-t0 will still be the same and hence for equations 13-15 the right hand side will
always have the same value whether we added a constant time or not. From this we can see
that the values of xs and ys will be the same with or without the constant addition. It is worth
to notice here that this modification will alter the value of t0 for its true value and will make
it larger than reality, but in our case we are not interested in the value of t0 and we are only
concerned with the values of xs and ys because they give the location of the unknown source
directly. Its important in this modification that the added time constant be the same for the
three signals, otherwise this method wont work and will yield to wrong results. The way that
we insured that the added time is the same for all the three signals is by choosing a distinctive
feature in the signal that shows up in all the microphone signals. Figure 4 shows a plot of the
time series signal for the three microphones and points out the time locations that we chose
for each of the three microphones.

Figure 4. Output signals form the three microphones.

The values of the three times that we get from the output signals are given as inputs to
equations 13-15, where t1 is the time for the right microphone, t2 is the time for the middle
microphone and t3 is the time for the left microphone. With these inputs, equations 13-15 are
ready to be solved for the values of xs and ys. Since equations 13-15 are nonlinear, then an
analytical solution to these equations is hard to obtain. For this reason, equations 13-15 are
solved using a nonlinear numeric solver in MATLAB.
Known Speaker Location Results. The first part of the determination of a sound

source expriment was to test the algorithm with known source location to see how accurate
the algorithm is in finding the source location. The source location was at (47,38) inches. As
described in the test producers section, we used two different microphone sets (1,7,11) and
(3,6,13) to test our algorithm and for each microphone set we repeated the expriment for 5
different times to assess the precision of our algorithm. Table 3 shows the results of the
algorithm for the two set of microphones and all the trials. While processing the 5 different
trials for both microphone sets, we noticed a bad output signal for each set of microphones.
The results of these trials are not included because they are outlier due to the odd behavior of
the signal for that particular trial. Hence table 3 shows only 4 trials of the data instead of 5.
Table3. Results of the location of the sound source

Trial
#
1
2
3
4

Set 1
microphones xs
(in)
47.1
47.1
47.5
47.6

Set 1
microphones ys
(in)
38.4
38.4
36.2
36.2

10

Set 2
microphones
xs (in)
47.7
47.1
47.1
47.7

Set 2
microphones
ys (in)
36.0
36.5
36.5
36.0

As can be seen from the results in table 3, the algorithm shows a good accuracy in
finding the actual location of the sound source for both microphone sets with set 1 (1,7,11)
being advantageous on set 2 (3,6,13). In addition, the algorithm shows a consistency in the
results from trial to trial.
Uncertainty Calculations of the Known Speaker Location . The measurement

uncertainty of the calculated position of the sound source comes from the uncertainty of
speed of sound and the uncertainty in the measured time. Equations 13-15 can be seen as a
general function in which it takes four inputs (c,t1,t2 and t3) and gives two outputs (xs and
ys). Since each input has uncertainty in it, then we would expect that these uncertainties
propagate through the output results. Lets denote our general function that is represented by
the solution of equations 13-15 by f(c,t1,t2,t3) where f is a two dimensional vector valued
function since it gives two outputs for the same set of inputs, then the uncertainty in f due to
uncertainties in its inputs argument is given by the general equation
= (

(16)

Where uxi is (uc , ut1, ut2, ut3). From the uncertainty calculations of the speed of sound, we got
that uc = 23 in/s and ut = 1/30000 seconds. In order to use equation 16 to get the measurement
uncertainty in the calculated position of the sound source, the derivative of f with respect to
each input variables need to be found. Since f is not known analytically, then these derivative
need to be estimated numerically. The method we chose to estimate the derivative is the
sequential perturbation method. The sequential perturbation method estimates the change in
multivariate functions by perturbing each independent input variable by its own uncertainty.

For instance, this method estimates the term ( ) (which is the change of f due to the

uncertainty in the speed of sound) in equation 16 by

) is
1 1

Similarly, the term (

(17)

estimated by

)
1 1

( +,1,2,3) ( ,1,2,3)

( ,1+1,2,3) (,1 1,2,3)


2

(18)

Equations 17 and 18 can be applied to the rest of the variables in the function f, and by doing
that the terms in equation 16 can be all estimated and hence the measurement uncertainty in
the position of the sound source can be obtained. The sequential perturbation calculations is
obtained using a simple code in MATLAB that we developed, this code utilizes the nonlinear
numerical solver function provided by MATLAB. A copy of the code is included in
Appendix B. Table 4 shows the results of measurement uncertainty calculations for the
location of the sound source for both microphone sets and for each trial. uxs is the
measurement uncertainty in the x position of the source and uys is the measurement
uncertainty in the y position of the source.

11

Table4. Measurement uncertainty in the known location sound source using perturbation
method

Trial
#
1
2
3
4

Set 1
microphones ux s
(in)
0.358
0.358
0.351
0.351

Set 1
microphones uys
(in)
0.682
0.683
0.642
0.641

Set 2
microphones
ux s (in)
0.352
0.353
0.353
0.352

Set 2
microphones
uys (in)
0.632
0.641
0.641
0.632

The random uncertainty in the results can be found using equation 5 which is the standard
error in the measurement. Table 5 summarizes the random uncertainty calculations of the
results from table 3. Table 5 shows the mean value of the position of the sound source from
the different trials reported in table 3 along with the standard error in each calculated mean
value.
Table5. Uncertainty summary from the results in table 3

Mean
value
Standard
error

Set 1
microphones xs
(in)

Set 1
microphones ys
(in)

Set 2
microphones
xs (in)

Set 2
microphones
ys (in)

47.35

37.37

47.43

36.29

.1434

0.6243

.1679

0.1274

We would like to summarize the uncertainty calculations by finding the random, systematic
and measurement uncertainties in the radius instead of the x and y coordinate. The radius of
the sound source location is given by
= () 2 + () 2
Using equation 19 and the results of table 3, we can get the calculated radius of the sound
source location. Table 6 summarizes the results

Table6. The radius of the sound source

Trial
#
1
2
3
4

Set 1
microphones Rs
(in)
60.8
60.8
59.8
59.8

12

Set 2
microphones Rs
(in)
59.8
59.6
59.6
59.8

(19)

The random uncertainty in the results of table 6 can be found using equation 5. The
measurement uncertainty of the radius can be found by using equation 16 applied to equation
19. The measurement uncertainty expression is given by

)
( ) 2+()2

= (

+(


( ) 2+( )

)
2

(20)

Where uRs is the measurement uncertainty in the radius. The values in the right hand side of
equation 20 are obtained from table 4 for the trial that has the largest measurement
uncertainty. The systematic uncertainty of the radius is defined as the absolute value mean
value of the radius minus the true value. The true value of the radius for the sound source is
60.44 in. table 7 summarizes the uncertainty calculation for the radius.

Table7. Uncertainty summary of the sound source radius

Mean value
Standard
error
Measurement
uncertainty
Systematic
uncertainty

Set 1
microphones
(in)
60.32

Set 2
microphones
(in)
59.72

.27

0.06

.51

.48

.11

.72

At the end of this section, we want to show that our choice of the speed of sound method
which is the theoretical calculation is justified by comparing the results of the radius using
the experimental speed of sound and the theoretical speed of sound. Table 8 shows the
comparison in radius for set 1 microphone using both calculation methods.

Table8. Comparison of speed of sound calculation method

Trial
#
1
2
3
4
Mean
value

Theoretical
speed of sound
Rs (in)
60.8
60.8
59.8
59.8
60.32

Experimental
speed of sound
Rs (in)
59.7
60.7
59.7
59.7
60

From table 8, we can see that the radius results obtained using the theoretical speed of sound
are more accurate that the results obtained by experimental speed of sound. For this reason

13

and for the reasons discussed earlier in the uncertainty of the speed of sound, we recommend
using the theoretical speed of sound.
Unknown Speaker Location Results. The second part of the determination of a sound
source expriment was to use the calibrated algorithm with known source location to locate the
position of the unknown speaker. The experiment was done identical to the known speaker
expriment in terms of the number of trials and the microphone sets that was used. Similar to
the known speaker data, we have found bad signals while processing the output microphone
signals and hence the results of only four trials are shown because the fifth trials led to outlier
result. Table 9 shows the results of the algorithm for four trials and both microphone sets.
Table9. Results of the position of the unknown location sound source

Trial
#
1
2
3
4

Set 1
microphones xs
(in)
60.26
59.87
63.06
59.65

Set 1
microphones ys
(in)
36.01
35.61
32.29
36.70

Set 2
microphones
xs (in)
60.17
60.17
59.59
63.27

Set 2
microphones
ys (in)
35.84
35.84
36.54
31.16

Uncertainty Calculations of the Unknown Speaker Location . Following the same

calculation methods as in the known speaker experiment, table 10 summarizes the


measurement uncertainty calculations.
Table10. Measurement uncertainty in the unknown location sound source using perturbation method

Trial
#
1
2
3
4

Set 1
microphones ux s
(in)
0.376
0.375
0.422
0.373

Set 1
microphones uys
(in)
0.677
0.666
0.670
0.682

Set 2
microphones
ux s (in)
0.387
0.387
0.382
0.432

Set 2
microphones
uys (in)
0.658
0.658
0.660
0.671

The random uncertainty calculations are done identically to what has been done in the known
speaker location. Table 11 summarizes the random uncertainty calculations.
Table11. Random uncertainty summary from the results in table 9

Mean
value
Standard
error

Set 1
microphones xs
(in)

Set 1
microphones ys
(in)

Set 2
microphones
xs (in)

Set 2
microphones
ys (in)

60.71

35.16

60.81

34.85

.7946

0.6243

.9797

1.24

14

Table 12 shows the radius of the unknown location sound source


Table12. The radius of the unknown location sound source

Trial
#
1
2
3
4

Set 1
microphones Rs
(in)
70.1
69.6
70.8
70.0

Set 2
microphones Rs
(in)
70.0
70.0
69.9
70.5

Following the same procedure as in the known sound location, Table 13 summarizes the
uncertainty calculation for the radius.
Table13. Uncertainty summary of the unknown location sound source radius

Mean value
Standard
error
Measurement
uncertainty

Set 1
microphones
(in)
70.85

Set 2
microphones
(in)
70.37

.51

0.26

.48

.49

Based in the calculation of the systematic uncertainty for the known location sound source, it
can be seen that set 1 microphones are more accurate than set 2 microphones. For this reason
the final value of the location of the unknown sound source will be taken as the mean value in
x and y coordinates for set 1 results. With this, the final answer for the unknown location is
(60.71 , 35.16) inches. Note that this value has random and measurement uncertainties
associated with it as discussed above.
Identification of aircraft type using acoustic signatures
This section presents the simple method that was used to identify the type of the
aircraft based on the acoustic signature of a sound signal recorded by microphone.
Frequency Decomposition of a Sound Signal. Any sound signal can be think of as
a combination of several frequency components. Aircrafts generate sound waves with
specific dominant frequencies that can be related directly to the aircraft itself. The main
sound frequencies that an aircraft generates are the operating frequency and the blade pass
frequency. The operating frequency is the propeller rotation speed, where the blade pass
frequency is the operating frequency multiply by the number of the blades on the propeller.
From this, we can see that if we analyze the frequency components of a sound signal and we
find both the operating frequency and the blade passing frequency for a certain aircraft, then

15

we can conclude that the source of the sound signal is that particular aircraft. In order to
obtain the frequency spectrum of the sound signal that was given to us, we need to use the
Fourier transform of the signal time series. To be more precise, we need to use the discrete
Fourier transform since the sound signal that we have is discrete in length. The way that we
performed the calculations of the discrete Fourier transform is using LabVIEW, in which the
transform is done using the fast Fourier transform (FFT) of the sound signal. Figure 5 shows
the frequency spectrum of the sound signal.

Figure 5. Frequency spectrum of the aircraft sound signal.

From figure 5, we can see the dominant frequencies in the sound signal which are represented
by the peaks in frequency spectrum. From the dominant frequency components we can
identify the type of aircraft and its component. Using the information provided to us about the
operating speeds and characteristics of various aircrafts and their corresponding components,
we can say that the sound signal represent the following components: USSR Antonvov An-2
Colt Engine, Weldon Fuel Pump and Airborne Fuel Pump. The USSR Antonvov engine
operates at 6900 rpm and has four propeller blades on it. This means that the sound signal
should have frequencies of 115 and 460 Hz. looking at figure 5 we can see the frequencies
114 and 459.9 Hz in the spectrum which means that Antonvov engine was part of the sound
signal. The Weldon Fuel Pump has operating frequency of 13 Hz and blade pass frequency of
65 Hz. these two frequencies can be seen in figure 5 approximately at 12 and 64 Hz. the
Airborne Fuel Pump has operating frequency of 30 Hz and blade pass frequency of 120 Hz.
from Figure 5 we can see these frequencies presents as 29.99 Hz and 120 Hz.
Discussion about Using the Location Algorithm in Scaled-up Applications
This section presents a discussion about using our algorithm in detecting the location
of nuclear weapons tests.
Recommendations in Scaled-up Applications. The main purpose of this

experimental investigation is to determine the feasibility of using the acoustics waves to find
the location of nuclear weapons testing. From our small-scale laboratory setup, we have
16

demonstrated in the previous sections the feasibility of our approach to find the location of a
sound source. Furthermore, we have shown that the method is both accurate and precise (the
accuracy is less than 1 inch). That being said, we believe that more investigation is needed
before the method be completely suitable for scaled up application. In the following
paragraph, we discuss some of the areas that need carful investigation before calling the
algorithm suitable.
With the small-scale setup, we have seen how accurate the algorithm is, but it is hard
to estimate the accuracy in scaled-up application because we cannot just simply assume that
the accuracy scale linearly. This means that the ratio of how off the algorithm was from the
true value compared to the total area covered by the microphones is probably not constant,
and hence we cannot scale the ratio to find what the accuracy would be in the case of scaledup applications. Another issue with scaled-up application is the variation of the temperature
between the sound source location and the microphone reviving location. This temperature
variation will lead to different sound speed between the source and reviving station which
will ultimately leads to inaccuracy in the location detection. For this reason, the sensitivity of
the speed of sound on the results needs to be investigated to determine how to account for the
variation in the sound speed. Another issue that needs to be addressed is the attenuation and
distortion of the sound wave through large distances, which is a factor we didnt investigate
due to the small-scale setup that we have. Its necessarily to know the properties of the wave
form received at large distance from the source location. One last comment that needs to be
considered for scaled-up application is the need to use four microphones instead of three.
This is necessary because the sound source and the reviving microphones lay in different
vertical planes, and hence a fourth microphone is needed to capture the out of plane distance.
With carful investigation of the above mentioned factors, our algorithm presents a fast
way to determine the location of a sound source. The advantage of our method is that it can
be fully automated and hence instant feedback can be obtained easily.
We want also to mention the fact that our algorithm can be used with some
modifications in other applications such as in exploration geophysics, in which sound waves
reflected from different surfaces from the earth ground can be analyzed to determine the
surface location.

Conclusions
The main focus of this experimental work was to investigate the effectiveness of
using acoustics waves to find the location of a sound source. Several testing have been
conducted in order to assess the feasibility of the proposed method. It was shown that the
acoustics waves can be used effectively to find the location of a sound source.
Speed of sound was determined experimentally and theoretically. The experimental
mean value of the speed of sound was 13830 in/s while the theoretical calculation of the
speed of sound indicated a value of 13623 in/s. uncertainty study has been done on both the
experimental and theoretical speed of sound and we showed that the theoretical speed of
sound has less measurement uncertainty than the experimental speed of sound. In addition, it
was shown that the accuracy of determining the sound source location is higher when the
theoretic speed of sound is used. For these two reasons, we recommend using the theoretical
speed of sound on the experimental approach.
We presented an algorithm in which it uses the acoustic signals of several
microphones to detect the location of a sound source. We showed that the algorithm has
location error of less than 1 inch based on a known source calibration study. The algorithm
was used to determine unknown speaker location and the result of the location was found to
17

be (60.71,35.16) inches. The extension of our algorithm to scaled-up applications was


discussed and it was recommended that more investigation is needed before applying this
algorithm into scaled-up application. Some of the issues we have identified with scaled-up
applications include: accuracy estimation, variation of the speed of sound between the source
and microphone and the need of four microphones instead of only three. Once these are
resolved and accounted for, then our algorithm should be a fast way to determine the location
of a sound source. In addition, the algorithm has the advantage of being able to be fully
automated by clever signal processing techniques.
Finally, we identified aircraft type and its component using the acoustic signature of a
recorded sound of an aircraft. The aircraft was found to be the USSR Antonvov. The
frequency decomposition of the acoustic signal indicated also a Weldon fuel pump and
Airborne fuel pump.
Attachments:
Appendix A: Sample calculations
Appendix B: MATLAB codes

18

Appendix A: Sample Calculations

This appendix presents sample calculations for the speed of sound using both the
experimental and theoretical techniques. In addition, the appendix provides sample
calculations for the measurement and random uncertainties of the speed of sound.
A-1: Experimental Speed of Sound Calculations
This sample calculation is based on the numbers in the first trial of table 1. The speed
of sound is given by
=

84()
0.00607(s)

= 13838.55

(A-1)

A-2: Theoretical Speed of Sound Calculations


The theoretical speed of sound is given by

= = 1.4 287 (
) (25 + 273)()

39.3701 ()
1 ()

= 13623

(A-2)

A-3: Measurement Uncertainty of the Experimental Speed of Sound


Calculations
This sample calculation is based on the numbers in the first trial of table 1. The
measurement uncertainty is given by
2

= ( ) + 2 ((
) =
)2

(A-3)

2
2
1
1
)()
84() (
) ()
30000
) + 2(
) = 107.59 /
( 32
(0.00607(s)) 2
0.00607(s)

A-4: Standard Error of the Experimental Speed of Sound Calculations


This sample calculation is based on the numbers in table 1. The standard error is given
by
1

=
( ) 2
(1)

(A-4)

(13838.55 + 13725.49 + 13888.89 + 13838 .78 + 13859.09) ( )


= 13830
=
5

19

[(13838.55 13830)2 + (13725.49 13830)2 + ( 13888.89 13830)2 + ( 13838.78 13830)2 + ( 13859.09 13830)2] = 27.8
4 5

A-5: Measurement Uncertainty of the Theoretical Speed of Sound Calculations


The measurement uncertainty is given by
=

1.4287 (

2( 25+273)( )

1( )

20

39.3701 ( )
1 ( )

= 22.85

(A-5)

Appendix B: MATLAB Codes

This appendix presents the MATLAB code that was used in the location finding
algorithm. In addition, this appendix presents the MATLAB code for doing the sequential
perturbation.

clear all
clc
close all
% comb1 11,7,1
% comb2 13,6,3
% comb3 15,9,5
% 1 = right
% 2 = middle
% 3 = left
% load the sound signals
load acoustic.mat
X = zeros(5,2);
Y = zeros(5,2);
uX = zeros(5,2);
uY = zeros(5,2);
%%
data = comb2run4;
t =data(:,1);
spanplus = .005;
spanminus = .01;
%% this removes the unwanted signal and define new time and amplitude
vectors

diffe1 = diff(data(:,2))./diff(data(:,1));
[C,idx] = max(diffe1);
idx2 = find(t>t(idx)+spanplus, 1, 'first');
idx1 = find(t>t(idx)-spanminus, 1, 'first');
sig1 = data(idx1:idx2,2);
tnew1 = data(idx1:idx2,1);
diffe2 = diff(data(:,4))./diff(data(:,1));
[C,idx] = max(diffe2);
idx2 = find(t>t(idx)+spanplus, 1, 'first');
idx1 = find(t>t(idx)-spanminus, 1, 'first');
sig2 = data(idx1:idx2,4);
tnew2 = data(idx1:idx2,1);
diffe3 = diff(data(:,6))./diff(data(:,1));
[C,idx] = max(diffe3);
idx2 = find(t>t(idx)+spanplus, 1, 'first');
idx1 = find(t>t(idx)-spanminus, 1, 'first');
sig3 = data(idx1:idx2,6);
tnew3 = data(idx1:idx2,1);

21

%% plot the three signals to choose the time location


figure
hold on
plot(sig1,'-xr')
plot(sig2,'-xb')
plot(sig3,'-xk')
legend('right =1','middle =2','left =3')
%%
% the
in.t1
in.t2
in.t3

time values for each of the microphone


= tnew1(159);
= tnew2(159);
= tnew3(160);

% speed of sound in/s


in.c = sqrt(1.4*287*(25+273))*39.3701;
in.comb = 2;
% random uncertainty in time
in.ut = 1/30000 ;
% random uncertainty in speed of sound
in.uc = 22.86;
% function that calculates the location of the speaker and perform
sequential perturbation to find random uncertainty in the found location
[x,deltaR] = locaation(in)

Locaation function
function [x,deltaR]

= locaation(in)

options = optimset('Display','off');

comb = in.comb;
% the function takes the three times + the speed of sound to find the
% speaker location
t1 = in.t1;
t2 = in.t2;
t3 = in.t3;
c = in.c;
ut = in.ut;
uc = in.uc;
if comb ==1

% comb 1
x0=[47;38;1.71];
% 3 equations to solve for the location
F = @(x)[sqrt((89-x(1))^2 + ((39+(3/8))-x(2))^2)-c*(t1-x(3));...
sqrt((53-x(1))^2 + (8-x(2))^2)-c*(t2-x(3));...
sqrt((9-x(1))^2 + ((21)-x(2))^2)-c*(t3-x(3))];
x = fsolve(F,x0,options);
% sequential perturbation calculations
% t1 perturbation
t1 = in.t1 + ut;
x0=[47;38;1.71];
22

F = @(y)[sqrt((89-y(1))^2 + ((39+(3/8))-y(2))^2)-c*(t1-y(3));...
sqrt((53-y(1))^2 + (8-y(2))^2)-c*(t2-y(3));...
sqrt((9-y(1))^2 + ((21)-y(2))^2)-c*(t3-y(3))];
y= fsolve(F,x0,options);
deltaR1P = y;
t1 = in.t1 - ut;
x0=[47;38;1.71];
F = @(y)[sqrt((89-y(1))^2 + ((39+(3/8))-y(2))^2)-c*(t1-y(3));...
sqrt((53-y(1))^2 + (8-y(2))^2)-c*(t2-y(3));...
sqrt((9-y(1))^2 + ((21)-y(2))^2)-c*(t3-y(3))];
y= fsolve(F,x0,options);
deltaR1N = y;
deltaR1 = .5*(deltaR1P - deltaR1N);
t1=in.t1;
% t2 perturbation
t2 = in.t2 + ut;
x0=[47;38;1.71];
F = @(y)[sqrt((89-y(1))^2 + ((39+(3/8))-y(2))^2)-c*(t1-y(3));...
sqrt((53-y(1))^2 + (8-y(2))^2)-c*(t2-y(3));...
sqrt((9-y(1))^2 + ((21)-y(2))^2)-c*(t3-y(3))];
y= fsolve(F,x0,options);
deltaR2P = y;
t2 = in.t2 - ut;
x0=[47;38;1.71];
F = @(y)[sqrt((89-y(1))^2 + ((39+(3/8))-y(2))^2)-c*(t1-y(3));...
sqrt((53-y(1))^2 + (8-y(2))^2)-c*(t2-y(3));...
sqrt((9-y(1))^2 + ((21)-y(2))^2)-c*(t3-y(3))];
y= fsolve(F,x0,options);
deltaR2N = y;
deltaR2 = .5*(deltaR2P - deltaR2N);
t2=in.t2;
% t3 perturbation
t3 = in.t3 + ut;
x0=[47;38;1.71];
F = @(y)[sqrt((89-y(1))^2 + ((39+(3/8))-y(2))^2)-c*(t1-y(3));...
sqrt((53-y(1))^2 + (8-y(2))^2)-c*(t2-y(3));...
sqrt((9-y(1))^2 + ((21)-y(2))^2)-c*(t3-y(3))];
y= fsolve(F,x0,options);
deltaR3P = y;
t3 = in.t3 - ut;
x0=[47;38;1.71];
F = @(y)[sqrt((89-y(1))^2 + ((39+(3/8))-y(2))^2)-c*(t1-y(3));...
sqrt((53-y(1))^2 + (8-y(2))^2)-c*(t2-y(3));...
sqrt((9-y(1))^2 + ((21)-y(2))^2)-c*(t3-y(3))];
y= fsolve(F,x0,options);
deltaR3N = y;
deltaR3 = .5*(deltaR3P - deltaR3N);
t3=in.t3;
% c perturbation
c = in.c + uc;
x0=[47;38;1.71];
F = @(y)[sqrt((89-y(1))^2 + ((39+(3/8))-y(2))^2)-c*(t1-y(3));...
sqrt((53-y(1))^2 + (8-y(2))^2)-c*(t2-y(3));...
sqrt((9-y(1))^2 + ((21)-y(2))^2)-c*(t3-y(3))];
y= fsolve(F,x0,options);
deltaR4P = y;
c = in.c - uc;
x0=[47;38;1.71];
23

F = @(y)[sqrt((89-y(1))^2 + ((39+(3/8))-y(2))^2)-c*(t1-y(3));...
sqrt((53-y(1))^2 + (8-y(2))^2)-c*(t2-y(3));...
sqrt((9-y(1))^2 + ((21)-y(2))^2)-c*(t3-y(3))];
y= fsolve(F,x0,options);
deltaR4N = y;
deltaR4 = .5*(deltaR4P - deltaR4N);
c=in.c;
deltaR = (deltaR1.^2 + deltaR2.^2 + deltaR3.^2 + deltaR4.^2).^.5;

24

You might also like