Effects of air conditioning on sound propagation in a large

space
Yasutaka Ueda
Graduate School of Science and Technology, Kobe University, Rokkodai Nada Kobe 657, Japan and
Research and Development Center, Hazama Corporation, Karima Tsukuba 305, Japan

Yoichi Ando
Graduate School of Science and Technology, Kobe University, Rokkodai Nada Kobe 657, Japan

~Received 26 July 1996; revised 25 July 1997; accepted 30 July 1997!

Sound-pressure level ~SPL! fluctuation caused by air conditioners is measured using pure tones as
the sound source, in a gymnasium. When the air conditioner is turned on, the mean standard
deviation of SPL changes about 0.9 dB at 2 kHz, 2.5 dB at 4 kHz, 3.3 dB at 8 kHz, and 3.8 dB at
16 kHz, while that in the case with air conditioning off is less than 0.6 dB at all frequencies. The
cumulative density of mean-squared amplitude corresponds well with the gamma distribution.
Based on this statistical analysis, the SPL change is thought to be related to the impulse response
which is represented by the direct sound ~regular! and changing delay time of reflection sound
~irregular!. The cumulative density of mean-squared amplitude for this model using pure tones also
indicated a gamma distribution. It is found that the frequency characteristic and the cumulative
density of the SPL change have the same tendencies as the method of modulation for the delay time
of reflected sound. 1997 Acoustical Society of America. @S0001-4966~97!05011-X#
PACS numbers: 43.55.Br, 43.58.Fm @SLE#

INTRODUCTION

More and more buildings with large indoor spaces have

been built in recent years. The air conditioning systems used
in such buildings have also increased in size, leading to significant changes in the physical environment. There are
many studies18 on the effects of air turbulence on sound
transmission outdoors. In outdoor sound propagation in air
turbulence, sound-pressure level ~SPL! fluctuation and reduction are caused by phase changes and the interference of
waves. In this paper the effects of air currents produced by
an air conditioning system on sound transmission in a gymnasium are investigated.9 This investigation was done to assess how the dynamic environment, such as air currents produced by an air conditioning system, would influence sound
propagation in a large indoor space. It is confirmed that there
is a large fluctuation in the SPL for high-frequency ranges.
Based on statistical analysis of the SPL fluctuation, a sound
propagation model that incorporates a time variant system is
proposed.
I. SPL MEASUREMENT

SPL was measured in a gymnasium because the gymnasium allows for a large mean-free path, and its interior is
made with hard materials which lower the attenuation of reflected sounds ~1:89.4 m w: 67.6 m h: 18.0 m 5000 seats and
RT: 2.1 s at 500 Hz!. Table I shows the experimental conditions. The maximum air speed at the outlet duct was about 8
m/s. We could not measure the air speed and the direction
throughout the room, but the air speed did not exceed 0.5 m/s
in the audience area at 1.5 m above the floor level ~case 2!.
There are 36 outlet ducts with a diameter of 450 mm and 44
ducts with a diameter of 750 mm on the ceiling. The vertical
temperature difference from floor to ceiling did not exceed
2771

3 C in the condition of the air conditioner on or off. Temperature was uniform in the horizontal plane. A dodecahedron loudspeaker was used to produce pure tones of 0.5 and
1 kHz, and a plasma loudspeaker was used for 2-, 4-, 8-, and
16-kHz frequencies. Figure 1 shows the location of the
sound source and observation points. The height of the sound
source was 1.5 m above the floor, and observation points
were 1.2 m above the floor. The distance between the sound
source and the observation points was from 27 m to 57 m.
The SPL was recorded for about 1 min at each observation
point.

II. RESULTS OF MEASUREMENT

A. Fluctuation of SPL

Figure 2 shows the SPL fluctuation recorded at observation point K1 ~0.516 kHz!. Figure 3 shows the mean standard deviation of SPL which was measured at 11 observation
points. In case 1, the strong SPL change was not found at
any frequency or observation points. In case 2, strong SPL
fluctuations were observed at frequencies higher than 2 kHz.
As the frequency increases, SPL fluctuation became stronger.
These tendencies were apparent regardless of the distance
between the sound source and the observation point.
TABLE I. Experimental conditions.

Case

Air conditioner

Air speed in audience area

@m/s#

Noise criteria

1
2

off
on

'0.0
,0.5

25
40

J. Acoust. Soc. Am. 102 (5), Pt. 1, November 1997 0001-4966/97/102(5)/2771/5/$10.00 1997 Acoustical Society of America

2771

FIG. 1. Location of observation points. d: sound source, s: observation

points.

B. Statistical analysis of amplitude

When waves of differing amplitude and phase are synthesized irregularly, they can be characterized by a Rayleigh
or NakagamiRice distribution.1013 The measured amplitude was compared with a gamma distribution which includes Rayleigh, Gauss, and approximate NakagamiRice
distributions, to evaluate how reflected sounds are composed.
The general formula of this distribution is

FIG. 3. Mean standard deviation of SPL. d in case 1 and s in case 2. Top

and bottom bars show the maximum and minimum standard deviation of
SPL.

where p 2 @ 5m(l11) # is the mean value and s (5m Al11)

is the standard deviation.11 The parameter l represents the

degree of fluctuation. When l equals 1.0, this distribution is
equivalent to a Rayleigh distribution, and when l.1.0, it is
approximately equal to a NakagamiRice distribution. Since
the simple correlation coefficient between the distribution of
measured data and theory is more than 0.98, we chose to use
the values of l and m by changing the parameter l gradually
such that l50,1,2,... . Figure 4 shows an example of the cu-

FIG. 2. Sound pressure level at observation point K1. ~a! in case 1 and ~b!
in case 2.

FIG. 4. Cumulative density of the mean-squared pressure. Observation

point: K1 Frequency: 2 kHz, ~a! case 1; ~b!: case 2.

p ~ x; P 2 ! 5

2772

l11

1
2
x l e 2x /m ,
G ~ l11 !

~1!

J. Acoust. Soc. Am., Vol. 102, No. 5, Pt. 1, November 1997

Y. Ueda and Y. Ando: Effect of air conditioner on sound in room

2772

FIG. 6. A time-variant model of impulse response.

FIG. 5. Mean value of l at each frequency. h is the mean value of 1 in case
1 and j is in case 2. The top and bottom bars show the 61 standard
deviation.

mulative density of the measured values compared with this

gamma distribution. Figure 5 shows the mean values of parameter l at each point. In both cases, the measured values
agreed with this theoretical curve. The slope of the cumulative density curve for case 1 is steeper than that for case 2.
Because l is more than 0.0, the fluctuation of SPL must be
caused by the composition of regular waves with irregular
waves. As the frequency becomes higher, the value of l becomes smaller. It is considered that the ratio of irregular
components to the regular components differed in each case
~see Table II!.
III. TIME-VARIANT MODEL OF IMPULSE RESPONSE
A. Concept of the time-variant impulse response

In addition, t n (t) is defined by a n and b n (t). The representing of the delay time of reflections is introduced as
t n ~ t ! 5t n0 1 a n t n0 b ~ t ! 5t n0 l1 a n b n ~ t ! ,

~4!

where a n is the change width of delay time, b n (t) is change

the property of delay time, t n0 is the initial value for delay
time under the geometry ray theory, and ^ t n (t) & 5t n0 . By
substituting the above expression into formula ~2!, expression ~5! is obtained:
`

p~ t !5

n50

f ~ t ! * R n h n ~ t2t n0 l1 a n b n ~ t ! ! .

~5!

It is simply supposed that a n becomes larger as delay time

increases. Figure 6 shows the concept of this time-variant
impulse response model.

The sound pressure at the observation point is usually

represented by
B. Statistical character

p~ t !5

n50

f ~ t ! * R n h n ~ t2t n ! ,

~2!

where R n is the amplitude of each reflection ~n50 represents

the direct sound! and t n is the delay time of each reflection.14
The mechanism of SPL change is represented by the timevariant impulse response model in which the delay time of
reflection varies:

The statistical character of mean-squared pressure for

this model was examined as follows. When f (t) is equivalent
to sine wave, the sound pressure at a point can be represented
p~ t !5

(n

sin v t * R n h n ~ t2t n !

p~ t !5

n50

f ~ t ! * R n h n t2t n ~ t ! .

~3!

TABLE II. Relationship between statistical characteristics and the condition

of wave composition.

Wave composition
~a! Differing amplitude and
phase waves synthesized
~b! Steady waves synthesized
with above irregular waves

2773

Distribution of
amplitude

Distribution of
mean-squared
amplitude

Rayleigh distribution Gamma distribution

(l50.0)
NakagamiRician
distribution

Gamma distribution
(l.0.0)

J. Acoust. Soc. Am., Vol. 102, No. 5, Pt. 1, November 1997

(n R n sin~ v t2 v t n !

5 $ R 0 sin v t 0 1R 1 sin~ v t2 v t 1 ! 1R 2 sin~ v t2 v t 2 !

1 %
5sin v t ~ R 0 1R 1 cos v t 1 1R 2 cos v t 2 1 !
1cos v t ~ R 1 sin v t 1 1R 2 sin v t 2 1 ! .

~6!

When t n distributes according to formula ~4!, v t n are simply

denoted by a *
ne
Y. Ueda and Y. Ando: Effect of air conditioner on sound in room

2773

FIG. 7. Comparison of the SPL simulated by some modulation method of

delay time. K1, 2 kHz, ~a! Measured data; ~b! by 0.1-Hz octave band; ~c! by
0.2-Hz octave band; ~d! by 0.3-Hz octave band; ~e! by 0.4-Hz octave band;
~f! by 0.5-Hz octave band.

nl

p ~ t ! 5sin v t R 0 1R 1

1cos v t R 1

(i

n2

cos a i 1R 2

nl

(i

(j

cos a j 1

n2

sin a i 1R 2

(j

sin a j 1

FIG. 8. Comparison of the envelope frequency in SPL change. K1, 2 kHz,

~a! Measured data; ~b! by 0.1-Hz octave band; ~c! by 0.2-Hz octave band;
~d! by 0.3-Hz octave band; ~e! by 0.4-Hz octave band; ~f! by 0.5-Hz octave
band.

5sin v t ~ R 0 1r 1 1r 2 1 ! 1cos v t ~ s 1 1s 2 1 !
5sin v t ~ R 0 1r ! 1cos v t ~ s ! .

~7!

The distribution of a i is flat (0<a i <2 p ). However, it approaches normality as n1,n2,...`, due to the central limit
theorem. Consequently the distribution of each term including the trigonometric function becomes a normal distribution. The variables r 1 ,r 2 ,s 1 ,s 2 , etc., are normally distributed variables. Since the normal distribution reproduces
itself by composition, r and s are also normal variables. The
fundamental equation of NakagamiRice distribution is
p 25

1
2

@~ R 0 1r ! 2 1s 2 # .

lation methods in band waves. Here, octave-band noise was

chosen and examined. The fluctuation width ( a n ) was controlled so that it has the same standard deviation as that of
the measured data. Figure 7 shows the measured SPL and the
calculated SPL modulated by octave-band waves from 0.1 to
0.5 Hz ~at observation point K1, 2 kHz!. As the frequency of
the modulation became higher, the simulated SPL fluctuated.
Next the frequency envelope of SPL change was examined
~Fig. 8!. The frequency envelope of simulated SPL change
with a 0.2-Hz octave-band wave, resembled that of measured
data at this point. Figure 9 shows the SPL from 0.5 to 16
kHz, calculated by a 0.2-Hz octave-band wave.

~8!

Consequently, the distribution of this model can be applied

to a gamma distribution approximately.
IV. SOUND PRESSURE LEVEL (SPL) SIMULATION

SPL was simulated to confirm the appropriateness of

this time-variant model for the measured SPL change. The
initial values for delay time t n0 and amplitude R n of each
reflected sound were calculated at each observation point in
the gymnasium using the geometrical sound theory. The calculation conditions were that the delay time of reflections
was within 500 ms, and the reflection frequency was within
three times.
A. Examination of b t

Delay time change @ b (t) # can be modulated by random

noise, periodical waves ~like sine waves!, and band waves. It
can be considered that SPL calculated by random noise
would change randomly, and that by periodical waves would
change periodically. And there are also many types of modu2774

J. Acoust. Soc. Am., Vol. 102, No. 5, Pt. 1, November 1997

FIG. 9. Calculated SPL by using 0.2-Hz octave-band noise at receiving

point K1.
Y. Ueda and Y. Ando: Effect of air conditioner on sound in room

2774

A time-variant impulse response in which the delay time

of reflected sound changes over time was devised. Based on
the comparison of the simulated SPL, using some modulation of delay time, with the measured SPL, it is found that
SPL fluctuates more as the modulation frequency becomes
higher. From the analysis of envelope frequency in the SPL
change, the envelope of the measured data resembles that of
the 0.2-Hz octave-band wave. The SPL that was calculated
by using 0.2-Hz octave-band noise shows the same characteristics as the measured data, and the cumulative density of
the mean-squared pressure corresponded to the gamma distribution.
It is almost possible to confirm the appropriateness of
the time-variant model, but there is need for more physical
analysis. The authors also are proceeding with psychological
experiments that use this time-variant model.

FIG. 10. Cumulative density between the measured and calculated meansquared amplitude with the gamma distribution. K1, 2 kHz. s: calculated
mean-squared pressure; d: measured mean-squared pressure, the solid line
indicates the gamma distribution.

B. Statistical distribution of simulated SPL

The cumulative density of the mean-squared amplitude

of Fig. 9 was compared with the gamma distribution of Fig.
10. The cumulative density of the amplitude fits the gamma
distribution well, as does the measured data.
V. CONCLUSION

Analyses of SPL and the statistical distribution of amplitude for both measurement conditions ~air conditioning on
and off! gave the following conclusions.
A change in SPL was observed when the air conditioning was on. The influence is significant in the frequency
domain, above 2 kHz. As the frequency becomes higher, the
SPL fluctuates more. Since the distribution of amplitude
agrees with the gamma distribution, SPL change occurs as a
result of the combination of the direct wave ~regular! and the
changing delay time of reflected sound ~irregular!.

2775

J. Acoust. Soc. Am., Vol. 102, No. 5, Pt. 1, November 1997

G. A. Daigle, Correlation of the phase and amplitude fluctuation between direct and ground-reflected sound, J. Acoust. Soc. Am. 68, 297
302 ~1980!.
2
U. Ingard and G. C. Maling, Jr., On the effect of atmospheric turbulence
on sound propagation over ground, J. Acoust. Soc. Am. 35, 1056 ~1963!.
3
G. A. Daigle, J. E. Piercy, and T. F. W. Embleton, Effect of atmospheric
turbulence on the interference of sound wave near a hard boundary, J.
Acoust. Soc. Am. 64, 622630 ~1978!.
4
R. J. Thompson, Ray theory for an inhomogeneous moving medium, J.
Acoust. Soc. Am. 51, 1675 ~1972!.
5
R. J. Thompson, Ray-acoustic intensity in a moving medium 1, J.
Acoust. Soc. Am. 55, 729 ~1974!.
6
R. J. Thompson, Ray-acoustic intensity in a moving medium 2, stratified
medium, J. Acoust. Soc. Am. 55, 733 ~1974!.
7
H. Tachibana and K. Ishii, Scale model experiment of the effects of
wind on sound propagation, Inter-Noise 75, 623 ~1975!.
8
L. L. Beranek, Noise Reduction ~McGraw-Hill, New York, 1960!, p. 185.
9
Y. Ueda, S. Iwashita, and Y. Ando, Sound fluctuation due to the air
turbulence in a large indoor space, Proc. ICA 14, f8-3 ~1992!.
10
R. V. Waterhouse, Statistical properties of reverberant sound fields, J.
Acoust. Soc. Am. 43, 14361444 ~1968!.
11
L. Maisel, Probability, Statistics and Random Processes ~Simon and
Schuster, New York, 1971!.
12
N. Mohanty, Random Signals Estimation and Identification Analysis and
Applications ~Van Nostrand Reinhold, New York, 1986!, pp. 139141.
13
H. Cramer, Mathematical Methods of Statistics ~Princeton U.P., Princeton,
NJ, 1946!, pp. 126, 233236.
14
Y. Ando, Concert Hall Acoustic ~Springer-Verlag, Berlin, 1985!.

Y. Ueda and Y. Ando: Effect of air conditioner on sound in room

2775