Acoustics and Electroacoustics Module I:

Fundamentals
Alex Heldring
March 4, 2020

1
Contents
1 Introduction to Acoustics 4
1.1 The wave equation . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Solutions to the wave equation . . . . . . . . . . . . . . . . . . . 6
1.2.1 Plane wave solution . . . . . . . . . . . . . . . . . . . . . 7
1.2.2 Spherical wave solution . . . . . . . . . . . . . . . . . . . 8
1.3 Characteristic Acoustic Impedance . . . . . . . . . . . . . . . . . 9
1.4 Energy and Power . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Frequency Spectrum 11
2.1 Frequency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Octaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 Sound Pressure Level 15

3.1 Effective Sound Pressure . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Spectral density . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3 Sound Pressure Level . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.4 Frequency Weighting . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.5 Sound Pressure Meter . . . . . . . . . . . . . . . . . . . . . . . . 21
3.6 Time averaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

4 Classification of Sounds 25
4.1 Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2 Spectrum of the human voice . . . . . . . . . . . . . . . . . . . . 26
4.3 Spectrograms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

5 Source Power Levels 32

5.1 Directivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.2 Line sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

6 Summation of Sound Fields 37

6.1 Incoherent sums . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
6.2 Coherent sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

7 Attenuation, Reflection and Diffraction 40

7.1 Attenuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
7.2 Reflection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
7.2.1 Plane waves . . . . . . . . . . . . . . . . . . . . . . . . . . 42
7.2.2 Spherical waves . . . . . . . . . . . . . . . . . . . . . . . . 44
7.3 Diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
7.3.1 Fresnel zones . . . . . . . . . . . . . . . . . . . . . . . . . 46

2
8 Low frequency theory: Acoustic Impedance 51
8.1 Mechanical Impedance . . . . . . . . . . . . . . . . . . . . . . . . 52
8.2 Acoustic Impedance . . . . . . . . . . . . . . . . . . . . . . . . . 53
8.3 Helmholtz resonators . . . . . . . . . . . . . . . . . . . . . . . . . 53
8.4 Acoustic Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

3
1 Introduction to Acoustics
1.1 The wave equation
Acoustics is the study of the physical phenomenon that local disturbances of the
pressure in a gas, fluid or solid medium, will propagate through that medium.
In the special case that the medium is the air that surrounds us at sea level, and
that the frequency content of the disturbance lies within the range that can be
detected by the human ear, this phenomenon is called sound. In this course we
shall be focusing on the latter case, that is, audible sound in the air at sea level,
without forgetting that it is just a small part of the general field of acoustics.
Starting with a homogeneous gas in equilibrium filling an infinite volume (no
boundaries anywhere) with a spatially and temporally constant static pressure,
density and temperature (eg. air at 105 Pa or approximately 1 Atmosphere, at
20o Celcius = 293o Kelvin and a density of about 1.2 kg/m3 ), we need to setup
the equations that describe the evolution of the pressure in time and space when
a forced disturbance takes place somewhere in this gas. In order to do this we
introduce another variable, the particle velocity → −u.
Like any gas, the air consists of a huge density of particles, mainly (>99%)
nitrogen and oxygen molecules, in total about 25×1015 molecules per mm3 .
These molecules bounce around and into each other continuously with an aver-
age speed that depends on (or rather, causes) the pressure and the temperature
of the gas. The pressure p is the average force working on a surface of one
square millimeter exerted by all the molecules that hit it, in other words, p is a
macroscopic quantity. For air at sea level the average speed of the individual
particles is about 500 m/s. But the vector average of the velocities of all the
particles in a cubic millimeter is zero if the gas is in equilibrium. The particle
velocity introduced above represents this vector average, over a small region in
space, but still containing millions of particles. It has nothing to do with the
speed of the individual molecules. Just like the pressure, the particle velocity →−
u
is also a macroscopic quantity. Unlike the pressure, which is a scalar quantity,
the particle velocity is obviously vectorial.
At any point inside a gas, the mechanical state is described by the scalar
pressure field p(~r, t) and the vector particle velocity field ~u(~r, t). The time
evolution of these fields is governed by two fundamental laws of physics. The
first one is the law of conservation of momentum (Newton’s second law):

∂→
−u
−∇p = ρ . (1)
∂t
The second equation is the law of conservation of mass:
∂ρ
−ρ∇ · →
−
u = . (2)
∂t
We now have two equations relating three variables, the third one being
the density which, like the pressure, is a scalar field ρ(~r, t). To obtain a third
equation we need to know that the propagation of sound in air is very nearly an

4
adiabatic process: The only transport of energy that takes place is an exchange
of kinetic energy due to the molecules elastically bouncing against each other.
Furthermore, the perturbations of p and ρ are very small in comparison to
their static background values. Then, there is a linear relationship between the
change of pressure and the change of density at all points:
∂p p0 ∂ρ
=γ (3)
∂t ρ0 ∂t
where p0 and ρ0 are the static pressure and density respectively and γ is a
dimensionless constant for a given gas (it is called the heat capacity ratio and
its value is about 1.4 for air)1 . The factor Ks = γp0 is called the adiabatic
bulk modulus of a gas, which is the inverse of the compressibility. A further
approximation that is perfectly valid in our case is to replace the density in (1)
and (2) by its static value ρ0 . Now, if we substitute (3) in (2) and then take
the divergence of (1) and the time derivative of (2), we can combine the two
resulting equations by eliminating the particle velocity, and obtain:

∂2p
c20 ∇2 p = , (4)
∂t2

where ∇2 = ∇ · ∇ is the Laplace operator ∂x2 + ∂y2 + ∂z2 and c0 = Ks /ρ0 ,

p

as we will see below, the speed of sound. Equation (4) is the Helmholtz wave
equation. If, instead of the above operations, we take the gradient of (1) and
the time derivative of (2), we find that →−
u obeys almost the same wave equation,
the only difference is that it becomes a vectorial instead of a scalar equation.
Because the wave equation only contains derivatives of the total pressure,
not the total pressure itself, in a homogeneous medium we can write the total
pressure ptot as the sum of the static background pressure p0 and the acoustic
pressure p = ptot −p0 which equals zero when the medium is in equilibrium. The
acoustic pressure is equal to the aforementioned perturbation of the pressure.
Table 1 illustrates the claim that the perturbation is indeed very small compared
to the background pressure. Also shown are some typical values of the maximum
particle velocity and the maximum particle displacement, which is again a
macroscopic quantity, representing the average distance, in the direction of → −
u,
of millions of particles, from a ficticious static equilibrium position.
The speed ofpsound, as we have seen above, depends on the type of gas, and
on the relation p0 /ρ0 . The air behaves very closely as an ideal gas, for which
the following equation of state (the ideal gas law) is valid:

p = ρRspec T (5)

in which Rspec is the specfic gas constant, whose value only depends on the
type of gas and T is the temperature in Kelvin. Eq. (5) shows that the speed of
sound in depends on the square root of the absolute temperature. In the air a
1 Eqs.(1) to (3) are called Euler’s equations of motion. They are approximations to the
Navier-Stokes equations of motion in fluids, only valid for small perturbations in an ideal gas.

5
  
~
|p|max (P a) |~u|max (m/s) δ  (m)
max
pain threshold 200 0,5 8 × 10−5
minimum audible sound 2 × 10−5 5 × 10−8 10−10

Table 1: Limits of pressure (p), particle velocity (~u) and particle displacement
~δ amplitudes for sound waves (at the reference frequency of 1kHz , see section
2). Compare with the static pressure in air at sea level of 105 P a and the average
speed of individual molecules of 500 m/s at T = 20o C.

Figure 1: The speed of sound in different media (in m/s) [1].

commonly used approximation, accurate for temperatures of practical interest

is given by
c0 = 331 + 0.6T [o C] m/s. (6)
In other media, the speed of sound can be very different. Without entering
into the physics of sound propagation in media other than near-ideal gases, Fig.
1 shows the speed of sound in a range of different media to give a global idea.
Note that in solid media and to a lesser degree in liquid media, the speed of
sound is generally much higher than in air.

1.2 Solutions to the wave equation

In order to solve (4) for any specific problem, we need to supply it with sources
and boundary conditions (restrictions imposed on p by obstacles, enclosures,
discontinuities in de medium). In general we cannot derive an exact analytical
solution for an arbitrary source and arbitrary boundary conditions. However,
since (4) is a linear equation (only sums of terms containing p to the power
one), we know that any sum of functions that are a solution of (4), is again a
solution of (4). Therefore, we can try to find simple homogeneous solutions (in
an infinite homogeneous medium) with a single, simple source, and then try to

6
construct the specific solution as a sum of these simple solutions such that the
sum fulfills the boundary conditions of the actual problem.
There are two simple homogeneous solutions that serve this purpose well.
The first one is the plane wave solution, which is exact when the source is
infinitely far away from the region under study, but it is often a very good
aproximation when the source is ’sufficiently far’. The second one is the spherical
wave solution, which is exact when the source is a point-source. In many cases
the source may be approximated by a single point source or a sum of point
sources. In some situations the plane wave model is more appropriate, in others
the spherical wave model.

1.2.1 Plane wave solution

If the source is far from the region where the solution for the pressure field is
sought, the orientation of the disturbance of the pressure that it causes will
be aproximately the same throughout this region. If we choose a coordinate
system aligned with this orientation, for example such that the disturbance is
only along the x-axis, then the wave equation becomes one-dimensional:

(c20 ∂x2 − ∂t2 )p = 0, (7)

which is solved by any function f (x, t) that can be written as

f (x, t) = f1 (c0 t − x) + f2 (c0 t + x). (8)

In (8), f1 is any function that, if it has a given profile along the x-axis at
time t = 0, then at t = 4t it has exactly the same profile, but shifted along
the x-axis over a distance 4x = c0 4t. The disturbance has moved, unaltered,
at a speed of c0 , in the positive x-direction. Likewise, f2 moves in the negative
x-direction. f1 and f2 are called acoustic plane waves, and we now see why c0
is called the speed of sound. Of course, for a single source we will only have one
term. We can choose the x-axis such that f = f1 . Then p(x, t) = p(c0 t − x).
Typically, sound sources emit transient signals: Initially the (acoustic) pres-
sure is zero, then the source starts emitting and finally it returns to silence. So,
at every point x, p is non-zero only during a finite period. However, every
function of a single variable, finite or infinite, can be Fourier transformed to
obtain a decomposition in a finite or infinite number of components that vary
sinusoidally with that variable. In the case of a plane wave:
X ωi
p(c0 t − x) = pi sin( (c0 t − x) + ϕi ) (9)
i
c0

where the sum is over all the different f requency components present in the
time signal p, each with its own radial frequency ωi , its own amplitude pi and
its own relative phase ϕi . If the time signal is transient, the frequency decom-
position is a continuous function of ω and the sum becomes an integral. If it
is periodic, there is a finite number of discrete frequency components. This de-
composition is very important in acoustics, for several reasons. Firstly, sound is

7
often studied as a function of frequency, because the human ear is very sensitive
to the frequency content of acoustic signals. Also, the interaction of sound with
matter is very frequency dependent. Furthermore, some important classes of
sound, for example music, are dominated by a few discrete frequencies or even a
single frequency. Such signals are called harmonic waves. Lastly, when a source
emits a signal with a constant frequency content, the pressure field around it
will reach steady state: it can be described as a function of frequency instead
of time.
We can simplify (9) by introducing the wave number k = ω/c0 , obtaining,
for each frequency component
p(x, t, ω) = pω sin(ωt − kx + ϕω ) (10)
or, equivalently, in complex notation:
p(x, t, ω) = Re{pω ej(ωt−kx+ϕω ) } = Re{Pω e−jkx ejωt } (11)
where Pω = pω e jϕω
is the complex pressure amplitude of the wave component,
called a phasor. Since we know the frequency ω of every component, we can
use the time-independent complex representation of the wave
p(x, ω) = Pω e−jkx , (12)
manipulate it (for example sum it to other waves) and then use (11) to retrieve
the true time dependent pressure.
The phasor representation of a plane wave is easily extended to arbitrary
directions of propagation as:
~→−
p(→
−
r , ω) = Pω e−j k· r (13)
→
− →
−
where k = k̂k, with k̂ a unit vector in the direction of propagation. k is called
the wave vector.

1.2.2 Spherical wave solution

When we have a single point-source (the disturbance is created in an infinitely
small region in the gas), the field is best described in spherical coordinates with
the source at the origin. Then the solution to the wave equation does not depend
on the spherical angles φ and θ, only on the radial coordinate r, the distance
from the origin. The solution for p is then
1
p(→
−r , t) = f (c0 t − r) (14)
r
which represents a wave propagating away from the source in all directions
(the ’inward’ wave that propagates towards the source is also a solution, but
of no practical interest. The factor 1/r follows from the maths, but it is also
physically necessary for conservation of energy, as we shall see later. In the
frequency domain, (14) becomes
−jkr
e
p(→
−
r , ω) = Pω . (15)
r

8
1.3 Characteristic Acoustic Impedance
As mentioned earlier, the wave equation can also be written in terms of the
particle velocity. In the case of a plane wave along the positive x-axis, the
solution
Pω −jkx
ux (x, ω) = e , uy = uz = 0, (16)
ρc0
compare (12), complies with the wave equation and with (1) and (2) as is easily
verified by taking the appropriate derivatives of (12) and (16). The relation p
/|→
−u | = ρc0 is found for plane waves propagating in arbitrary directions. For a
plane wave, the pressure and the particle velocity are in phase, and their ratio is
a medium-dependent constant . This constant is the characteristic impedance
Z0 = ρc0 of the medium, and it plays an important role in calculations of
the reflection and transmission of acoustic waves at the boundary between two
different media, among other things. For air at 1 atm and 20o C, the value of
the characteristic impedance is approximately 413 kg m−2 s−1 .
For a spherical wave, the solution for →
−
u is somewhat more complicated:
Pω e−jkr 1
ur (→
−
r , ω) = (1 + ), uφ = uθ = 0. (17)
Z0 r jkr
The extra factor appearing in (17) with respect to (15) is a near f ield effect. As
the wave propagates away from the source the ratio between p and → −
u converges
to Z0 .
If there is more than one source, the fields due to the different sources can be
summed, as noted before. However, the sum of the particle velocities is a vector
sum; if the two waves propagate in opposite directions, the vector components
have opposite sign and cancel out. So, the relation p /|→ −
u | = Z0 only holds for
f ree f ields (in an infinite homogeneous medium) due to a single source.

1.4 Energy and Power

Although the pressure field is necessary to obtain a complete description of the
propagation of sound in a given situation, in practise one is generally more
interested in the energy contained in the field and in particular the propagation
of energy in terms of a power density, in W atts/m2 . The acoustic energy density
at any point in the region under interest is the sum of two components, the
potential energy which depends on the acoustic pressure according to
p(t)2
wpot (t) = (18)
2ρc20
and the kinetic energy:
1 →
ρ |−
2
wkin (t) = u (t)| . (19)
2
For a plane wave (and the far field of a spherical wave), using p /|→
−
u | = ρc0 = Z0 ,
we get a total energy density of
p(t)2
w(t) = . (20)
Z0 c0

9
 The power density, or the amount of energy per square meter that flows
→
−
through any point, is a vector field called the acoustic intensity I which is
related to the energy density by the law of energy conservation:
→
− ∂w
∇· I =− . (21)
∂t
Being a vector, it cannot be computed from the scalar pressure alone, we also
need the particle velocity which, obviously, has the same direction:
→
−
I (t) = p(t)→
−
u (t). (22)

For a plane wave, the power density in the direction of the wave equals

−  p(t)2
→ 
 I (t) = W/m2 (23)
Z0

Although for a spherical wave the simple relation between p and → −

u is only
valid in the far field, it can be shown that (23) is true everywhere for a spherical
wave as well. Since p is proportional to 1/r for spherical waves, (23) shows
that the power density decreases with 1/r2 . This is necessary, because the total
power radiated by a point source is homogeneously distributed over a sphere of
radius r, which has a total surface of 4πr2 .

10
2 Frequency Spectrum
2.1 Frequency
In the previous section we introduced the concept of harmonic waves of a given
frequency, and we mentioned that any sound can be decomposed into harmonic
components by way of the Fourier transform. This allows to study the different
frequencies making up the sound separately. Each frequency component has
a fixed sinusiodal time dependence sin(ωt) where ω is the radial frequency, in
rad/s. At every point in space, after every time interval 4t = T = 2π/ω, the
pressure repeats itself. T is called the period of the harmonic wave. Its inverse

f = 1/T = ω/2π (24)

is the frequency of the wave, with dimension Hertz (Hz = s−1 ). For a propa-
gating wave along the x-axis, the spacial dependence at a fixed time is sin(kx).
So, moving along the x-axis, after a distance 4x = λ = 2π/k, the pressure also
repeats itself. λ is called the wave length. Since, as we saw before, ω = c0 k, we
can relate the frequency and the wave length of a harmonic wave through

f = c0 k/2π = c0 /λ. (25)

Acoustic waves in the air at sea level can theoretically have any frequency
between 0Hz and some high maximum (in the order of 10GHz 2 ), provided
there exists a source that generates those frequencies. However, we shall focus
on the frequency band between 20Hz and 20.000Hz, which corresponds to the
frequencies that the human ear can detect, the audible f requency band, also
called Sound. Acoustic waves outside this band are called infra- and ultra-
sound respectively.

2.2 Octaves
With (25) we can compute the limits of the audible band in terms of the wave
length (for c0 = 345 m/s): λmin ≈ 1, 7 cm, λmax ≈ 17 m. The wide range
of frequencies and wave lengths that are included in the audible band makes
it impractical to study sound phenomena frequency by frequency on a linear
scale. Fortunately, both the sensitivity of the human ear to frequency differences
and the frequency dependence of important acoustic phenomena such as sound
absorption, are logarithmic in nature; the higher the frequency, the less they
vary as a function of frequency. We can therefore sub-divide the full range into
sub-bands with a width that grows with the frequency. The basic unit for this
sub-division is the octave. A frequency band with a width of one octave is
any band that runs from a lower frequency fl to a higher frequency fh where
2 The mean f ree path of the molecules making up the air is about 70 nm, much longer than

the effective radius of the molecules (approximately 0,15 nm). Their average speed at room
temperature is approximately 500 m/s. The minimum period T cannot be shorter than the
average time between collisions of the molecules, or Tmin ≈10−10 s. This gives an indication
(order of magnitude) of the maximum frequency in air: 10GHz.

11
fh = 2fl . The band width fh − fl of an octave is equal to the lower frequency
fl . The center-frequency of an octave is defined as

(26)
p
fc = fl fh ,

the
√ geometric √ mean of the limiting frequencies. From (26) we see that fc =
2fl = fh / 2, so the octave is fully specified by its center frequency. The
limiting frequencies and the center frequencies of a sequence of octaves follow a
geometric series with common ratio 2 (sn = 2n s0 ). Observe that the geometric
series is exponential. Therefore its logarithm grows linearly and on a logarithmic
scale, the distance between center frequencies is constant.
If a more refined sub-division is needed, steps of a fraction of an octave are
used. A common choice are third-octaves (also written as 1/3-octaves). These
subdivide each octave into three parts that have equal width on the logarithmic
scale. The centre and limiting frequencies of the third-octaves follow the same
rules as the full octaves, to the power 1/3:
p 1/3
fh = 21/3 fl , fc = fl fh = 21/6 fl . (27)

Another common choice for even more frequency resolution are 12th-octaves.
Full octaves are observed to have a fixed relative band width
BWoct 1
= √ ≈ 70%. (28)
fc 2
whereas third-octaves have a relative band width of
BW1/3oct 1
= √
6
≈ 23%. (29)
fc 2
A logarithmic scale with less resolution that is sometimes encountered is
the decade scale. This scale is based on a common ratio of 10, so the center
and limiting frequencies are a factor 10 apart. A decade does not contain an
integer number of octaves, but, because 210/3 ≈ 10, one decade contains almost
10 third-octaves.
Like any logarithmic scale, the frequency scale is not complete without a
ref erence value that tells us where to start (on a linear scale the reference is
typically zero, but on a logarithmic scale, zero becomes minus infinity). The
reference value should somehow incorporate the range of the aubible band, 20 −
20.000Hz. It is not possible to fit an integer number of octaves in this range.
So a different approach has been taken. The octaves are identified by their
center frequencies, so it would be convenient to have these center frequencies at
’simple’ values. The most important region of the audible band is the region
around 1.000Hz because of the role it plays in human speech. So, 1.000Hz
is chosen as the reference (it doesn’t matter that it isn’t at the beginning).
Then, with ten octaves centered at 31.25Hz to 16.000Hz, we approximately
cover the entire audible band. This sub division has been standardized by the

12
International Organization for Standardization (ISO), as shown in Fig. 2. Note
that it does not follow the geometric series exactly; the lowest band is centered
at 31.5Hz instead of 31.25Hz; thanks to this small deviation, the third-octave
center frequencies span exactly a decade.

13
Figure 2: Standardized octave bands.

14
3 Sound Pressure Level
3.1 Effective Sound Pressure
Up to now we have considered the exact instantaneous sound pressure p(t) at a
position in space. We have also seen that within the frequency band of interest,
its value may oscillate with a period of 50 microseconds (for 20.000Hz). The
same happens with the transported power, which is generally the quantity of
interest. If we want to measure it accurately, we need an awful lot of time-
samples. In practise, we usually deal with signals that are stationary over much
larger time scales, meaning that their frequency spectrum varies much slower
than the instantaneous pressure. To characterize such signals, we would want to
average out the rapid oscillations. We cannot average the instantaneous pres-
sure directly, because it oscillates around zero, but we can average the squared
pressure. This leads to the concept of effective sound pressure (also known as
root-mean-square or RMS pressure), defined as
v
u T
u Z
u1
pef,T = t p(t)2 dt, (30)
T
0

where T is supposed to contain several full cycles of all frequencies present

in the signal. Eq. (30) gives the total effective pressure of the signal. Often
we are interested in the effective pressure per frequency band (octaves, third
octaves,..). Then we first have to pass p(t) through an appropriate filter. For a
harmonic signal with complex amplitude Pω , it is easily verified that
√
pef = 1/ 2 |Pω | . (31)

For a single plane or spherical wave, the time averaged intensity or power
density through the measured point in the direction of propagation equals, ac-
cording to (23):
2
−  pef
→ 
I = W/m2 . (32)
Z0
The time averaged power density is important, for instance if we want to know
the total power radiated by a source. If the source is stationary (the radiated
power is constant over some time interval), and it can be considered a point
source - which is always the case at a sufficiently long distance away from it, in
terms of the wave length (r/λ  1 or kr  1) -, we can measure pef at several
points around it, to cover a closed surface, preferably a sphere for computational
ease. Then we use (32) and
→
−
Z
Prad = I · dS, (33)
S

where S denotes the closed surface, to find the radiated power. If the source
is not a point source, we cannot know the direction of propagation of the wave

15
from the pressure alone. Then we also need to measure the effective particle
velocity, →
−
u ef , which is defined analogously to (30), and use the time averaged
→
−
equivalent of (22) to find I .
For general sound sources, the above procedure is the only way to determine
the radiated power.

3.2 Spectral density

Let us see what happens to the effective pressure of the sum of two pure tones
in steady state (T → ∞), at different frequencies:

ZT
1
p2ef,sum = lim (pω1 sin(ω1 t + ϕω1 ) + pω2 sin(ω2 t + ϕω2 ))2 dt (34)
T →∞ T
0

ZT
1
= p2ω1 sin(ω1 t)2 + p2ω2 sin(ω2 t)2 + 2pω1 pω2 sin(ω1 t) sin(ω2 t)dt. (35)
T
0

The relative phases do not affect the integral and the third term on the RHS
of (35) vanishes unless ω1 = ω2 . Another way of saying this is that different
frequency components are uncorrelated.
Eq. (35) for two different frequencies leads directly to:

p2ef,sum = p2ef,ω1 + p2ef,ω2 . (36)

The square of the effective pressure of the sum of different frequency compo-
nents is the sum of the square of the effective pressures of each component. In
other words, the total power of a signal is the sum of the power in each fre-
quency component. Eq. (36) is a version of Parseval’s teorem for two discrete
frequencies. For a continuous frequency spectrum, Parsevals teorem reads (in
terms of the frequency f instead of the radial frequency ω)
Z∞ Z∞
2 2
|p(t)| dt = |Pf (f )| df (37)
−∞ −∞

which implies that we can measure the power carried by a signal in any frequency
interval 4f and sum over all the intervals to obtain the total power. Thus, we
can characterize a signal by the power present in each octave, or third octave,
or even as a continuous spectral power density in terms of W/Hz.

3.3 Sound Pressure Level

The sound pressure is the fundamental quantity for the characterisation of sound
fields. However, as we have seen in Table 1, the values of practical interest

16
(globally between the weakest sound that the human ear can detect and the
sound pressure that causes unbearable pain and permanent damage to the ears),
vary by a factor of 10 million. The human ear does not perceive this variation
linearly, we never say that one sound is a million times as loud as another.
Our perception of the strength (volume, loudness) of a sound is more adjusted
to the logarithm of the sound pressure, similar to what we have seen with the
frequency. To simplify calculations, and to better reflect our perception, acoustic
engineering almost always uses this logarithmic scale. As we have seen before, a
logarithmic scale always needs a reference value. In the case of sound pressure,
this reference is chosen to be the (approximate) hearing threshold for a pure
tone of 1kHz, which equals
pref = 20µP a (38)
(this value was found through experiments on a large sample of healthy young
adult people). The Sound P ressure Level is then defined as
 
pef
Lp = 20 log dB SP L, (39)
pref

(adopting the convention that log without subscript always refers to log10 ).
The units of Lp are the decibel (dB). The extension SP L specifies that the
reference level (0 dB) equals 20 log(pref ). An important note: In acoustics, like
in all branches of engineering, decibels are always defined according to
 
P
10 log ,
Pref
where P denotes a measure of power. As we have seen, the power carried by
a sound wave is proportional to the square of the pressure. This is where the
leading factor 20 comes from in the definition of SP L in (39). Accordingly,
the intensity in a given direction through a point in a sound field can also be
expressed in decibels through
 →
−  
I 
LI = 10 log   dBI (40)
Iref

where the reference value of Iref = 10−12 W/m2 has been chosen such that for
a plane or spherical wave in air at sea level, the dB SP L value and the dBI
value are (very nearly) the same.

3.4 Frequency Weighting

With Eq. (39), we can characterize a sound field at any point in space by its total
sound pressure, or its SPL. However, as mentioned before, the human ear is more
sensitive to certain frequencies in the audible band than others, so two fields with
equal SPL can in fact be perceived as very different depending on their spectral
content. We could accomodate for this by reporting the SPL as a function

17
Figure 3: Typical Sound Pressure Levels [2].

18
 fc (Hz) 31,5 63 125 250 500 1k 2k 4k 8k 16k
A-weight (dB) -39,4 -26,2 -16,1 -8,6 -3,2 0 1,2 1,0 -1,1 -6,6

Table 2: A-weights of the standard octave bands

of frequency, as explained in section 3.2, but since the spectral sensitivity is

very similar for all humans, we can obtain a single value that captures the
perceived sound pressure level as the sum of the spectral components, assigning
a different weight to the different frequency components. On the basis of large
scale experiments, different weighting functions have been defined for different
purposes. The most important one is the so-called A − weighting, which is
almost universally used in acoustical engineering, for example in norms and
standards for noise control.
The A-weighting curve is shown in Fig. 4 together with some less important
weighting curves. The B weighting is no longer in use, the C weighting is
sometimes used for very high sound pressure levels as it turns out that the human
perception is non-linear with respect to power; at normal levels, the human is is
much more sensitive to higher frequencies, as reflected in the A weighting curve
whereas for very loud noise, this difference is much less pronounced. Sometimes
we are in fact interested in the true, unweighted level (all weights equal one).
This case is sometimes referred to as Z-weighting.
Often it is enough to measure the total sound pressure level for each octave
band or each third octave band and apply a discrete set of weights corresponding
to the average of the weighting function over the given frequency bands. Table
2 shows the discrete weights for the standard octave bands. Notice that all
weighting curves have the value one at the reference frequency of 1 kHz. As the
the human ear is most sensitive around 2 kHz, the A-weighting curve assumes
a maximum there, with a value greater than one. Below 500 Hz and above 10
kHz it drops off quickly.
In order to distinguish the weighted levels from from each other and from the
unweighted level, it is common to indicate the scale by reporting, for instance,
an A-weighted sound pressure level in dB(A). Also, the unweighted level is
usually indicated with the symbol Lp as above, while the weighted levels are
denoted with LA or LC .
Of course, the linear sum of the squared effective sound pressure in different
frequency bands does not translate to a sum of levels directly. To find the total
SPL of a signal for which (for example) the levels per octave are known, we need
to convert to linear scale, then sum the contributions, then reconvert to dB, as
in X
Lp,tot = 10 log( 10Lp,i /10 ) (41)
i

where the index i runs over all the relevant octave bands. The scale of (41) will
be dB SP L provided the levels per octave were also in dB SP L. If we need the

19
Figure 4: Different spectral weighting curves [3].

20
 Figure 5: Simplified diagram of a Sound Level Meter [4].

A-weighted level, we use

X
LA = 10 log( 10(Lp,i +Ai )/10 ) (42)
i

where Ai are the A-weights given in Table 2.

3.5 Sound Pressure Meter

We have seen in previous sections that the fundamental parameter for identi-
fying sound fields is the Sound Pressure (Level). The device used by sound
engineers to measure his parameter is the Sound Pressure Meter (SPM). SPMs
exist in many different forms, from sophisticated devices for official, legally
binding measurements to simple consumer gadgets (there are several applica-
tions that allow sound level measurements with modern smart-phones). Fig. 5
shows a generic scheme of the functionality of an SPM. A microphone converts
the instantaneous pressure p(t) into an electric signal V (t). This signal is then
either passed through a weighting network to apply A,C or Z weighting, or it is
separated into different frequency bands by filtering. Subsequently, after being
amplified, it enters a RMS detector which applies Eq. (30) to the time signal,
with a given time constant T . The resulting signal, which is now a sequence of
discrete values with a refresh rate T , is either displayed or sent to an external
storage device.

3.6 Time averaging

The time averaging operation to convert the instantaneous pressure into an
effective pressure as in (30) requires a choice of the time constant T . In order
to make sure that measurements are taken according to the same procedure
everywhere, the values of the time constant have been standardized. Sound
Level Meters usually offer two options, ’S’ for slow and ’F’ for fast. These are
defined in the international standard IEC 61672-1. For historical reasons they

21
include an exponential time-averaging factor:
v
u T
u Z
u1
pef,T (t) = t p2 (τ )e−(t−τ )/T dτ (43)
T
0

(the exponential averaging returns a ’smoothed’ response which facilitates

reading the result on an analogue needle-display). The time constant T in
the ’S’ mode equals 1 s. If more temporal resolution is necessary, because the
measured level varies at a time scale below one second, the ’F’ mode is chosen,
with a time constant of T = 125 ms.3 Apart from these two, many SPMs have
a ’peak’ mode. In this mode, the time constant is 50 µs, but typically only the
highest measured value over the full measurement time is retained. This value
is important in noise control because very high, short bursts of pressure can
be very damaging to the ears, even when they are too short to be consciously
noticed.
The result of a measurement in ’F’ or ’S’ mode is a time series of ’instan-
taneous’ sound pressure levels. However, we are often interested in an average
level on a much longer time scale. In noise control for example, it has long been
established, that (except in the case of very high ’peak’ levels), the quantity
that determines annoyance and, more importantly, long term damage to the
human hearing, is the total sound energy to which a person is exposed, i.e. the
sound pressure level (power), integrated over time. Even if the level varies over
the total duration of the measurement, the total energy will be adequately cap-
tured by the effective pressure as defined in (30). Hence, Sound Level Meters
generally offer the option, apart from the ’S’, ’F’ and ’peak’ modes to choose an
arbitrarily large time constant and measure a single average level for the corre-
sponding period. This value is defined as the equivalent level for the period T .
Combining (30) and (39), it is given by:
 T 
Z 2
1 p(t) 
Leq,T = 10 log  dt dB SP L. (44)
T p2ref
0

The value of T can vary from a few seconds, to measure the noise level due to
a continuously working machine, to 24 hours for the average noise level of city
traffic. Often a weighted equivalent level is required, for example LAeq,T , in
which case p(t) is subject to an A-weighting filter first.
Furthermore, modern professional Sound Pressure Meters have the option
’Short Leq ’, which measures a sequence of samples, typically each 125 ms, ac-
cording to (44), without exponential weighting. Storing these samples allows to
subsequently compute Leq,T for any T which is a multiple of 125 ms.
The equivalent level gives essential information on a sound field over a given
period in a single number. However, regularly we know that the sound pres-
sure level varies during the period of interest and we require some additional
3A third standard option, ’I’ for Impulse, is now obsolete.

22
information about this variation, without the need to know the actual detailed
time evolution of the signal. For this reason, professional SPMs usually generate
a number of additional standardized values, that may appear, for example, in
legal regulations for noise control. Apart from the maximum value during the
interval in ’S’ or ’F’ mode, which is often limited separately, below are a few of
the most commonly encountered ’complementary’ measures:
• Ld/n . Noise levels in public places often have different legal limits accord-
ing to the time of day. This can be incorporated by performing separate
measurements over different time frames, and comparing with separate
legal limits. Another approach often encountered is to measure the full
24 hours, and apply a correction factor, or penalty, to the result for the
more restrictive time frame. For example, at night-time, between 10PM
and 7AM, a penalty of 10 dB may be imposed. If the average 15h day-
time level is Ld and the 9h nighttime level is Ln , then, incorporating the
penalty in the total 24h equivalent level according to (44), yields
 
15 Ld /10 9 (Ln +10)/10
Ld/n = 10 log 10 + 10 . (45)
24 24

• LX , where X is an integer between 1 and 99, are so-called percentiles.

Common percentiles are L10 , L50 and L90 . They are defined as: the level
that is exceeded X% of the total duration of the measurement T . They
give a global idea of the variability of the level. If L10 is almost equal to
Leq,T , the noise is practically constant. If it is much higher, then there
are short intermittent bursts of noise in an otherwise quiet background.
• LN C , N oise Criteria Levels. Although A weighting yields a good mea-
sure of the pressure level adjusted to the sensitivity of the human ear,
it sums over all frequencies so it does not reveal any information on the
relative frequency content. It turns out that a pure tone or a narrow
band of noise around a specific frequency, in particular a high frequency,
can be much more distracting than homogeneous noise, even if the latter
has a higher LA . To take this into account, a mechanism has been in-
troduced, standardized by the ISO and other standardization institutes,
that does not sum the frequencial components, but it compares the entire
spectrum with a set of reference curves, as illustrated in Fig. 6. Then it
applies the following simple rule: LN C for the measurement in question
equals the value corresponding to the lowest curve that is higher than
the measurement everywhere. Noise criteria levels are typically used for
recommended noise limits in residential or public spaces. For example,
LN C in concert halls and theatres should not exceed 25 dB, in conference
halls 30 dB and in offices 35 dB.

23
Figure 6: Noise Criteria Curves according to ISO. The black curve is a mea-
sured spectrum that has a Noise Criteria Level of 45 dB (see text) [5].

24
4 Classification of Sounds
Different types of sound can be classified according to their spectral character-
istics, as shown in Fig. 7. A first sub-division is made between deterministic
sound and random noise according to whether the instantaneous pressure and
the corresponding power spectrum allow for an analytic description or not. In
Fig. 7 the first 4 wave forms (a) to (d) are deterministic, while (e) is random
noise. We discuss them in the order of the figure:
a) A pure tone. The pressure varies sinusoidally and the power spec-
trum has a single non zero peak at the corresponding frequency.
b) A complex sinusoid, a sum of a finite number of pure tones. The
first (lowest frequency) tone is called the fundamental frequency or
first harmonic. Often the higher frequencies are integer multiples
of the f undamental frequency, for example with most musical in-
struments, but also the human voice. If that is the case they are
called superior (second, third, etc.) harmonics. The figure shows
a combination of a fundamental frequency and its third harmonic.
Frequency components above the fundamental that are not an inte-
ger multiple are called overtones.
c) A general periodic wave form. If any continuous wave form repeats
itself after a given period, its spectrum will be discrete (frequency
components at finite intervals), although it can be infinite. The
lowest frequency present will be the period of the time signal. The
more rapid the change in the instantaneous pressure, the higher the
frequencies present. For a true block-function, with its infinitely fast
change of pressure, the frequency content goes to infinity.
d) A transitory (non periodic) signal will have a continuous spectrum,
i.e. a frequency density rather than discrete components. The
shorter the pulse width, the wider the frequency content.
e) Random noise. If the pressure at different time instances is uncorre-
lated, the signal is called random noise. Often, even if the pressure
as a function of time is unpredictable, the spectral content is ap-
proximately constant over time and can be measured. This is called
stationary random noise. When we measure the noise from indus-
trial machines, car engines, street noise, etcetera, we assume that
it is stationary. Examples of non-stationary noise are explosions, or
airplanes and missiles at take-off.

4.1 Noise
The spectral density of random noise can be any continuous function of the
frequency. However, certain simple functions are of particular interest because
they are good approximations of many naturally occuring noise phenomena, or

25
they can be generated by artificial sources for measuring purposes and they
are allow for simple analytical manipulation. These noise spectra are identified
with colors, a practise which started with the label white in analogy with optics
where white light has an approximately flat spectrum over the visible frequency
band.
Accordingly, white noise in acoustics has an approximately constant spectral
density over the audible band. Natural phenomena that are truly random, such
as the noise produced by a waterfall, are approximately white noise. Since in
acoustical engineering, measurements are often made on a logarithmic frequency
scale, most commonly in octaves, it is interesting to see how the power density
is distributed in terms of octaves. Observing that each octave spans twice the
band width of the previous one, and half that of the following one, we see that,
if the power density is constant in terms of W/m2 , the total power carried in
an octave goes up by a factor 2 or 3dB with each octave (see Fig. 8(a)). Since
the human perception of sound is closer to the logarithmic scale, white noise to
the human ear sounds as if the higher frequencies dominate.
In order to have a standardized noise spectrum that is better adjusted to
the human perception, and more appropriate for measurements per octave band,
pink noise has been introduced as a random noise that has an energy density
over the audible band according to (see Fig. 8(b))
dw
∼ 1/f. (46)
df
It is easily demonstrated that for pink noise, all octaves carry the same
power, as shown in Fig. 8. There are more standardized types of noise but
white and pink are the most important ones in acoustics.

4.2 Spectrum of the human voice

Fig. 9 shows the spectrum (averaged over many people) of human speech. Al-
though these results are for English, they are quite similar for other languages.
For normal speech, although there is a lot of energy in the lower frequencies,
with a maximum below 500 Hz, the voice has frequential components over the
entire audible band (the measurements in Fig. 9 do not include the frequency
bands below 125 Hz and above 8000 Hz because their role is insignificant for
speech transmission). In fact, the lower frequency bands contain the fundamen-
tal frequency and the first few superior harmonics of the vowels. The vowels do
not transmit a lot of information for understanding, they mainly serve to draw
the attention of a listener and to identify the speaker. Everybody has a unique
timbre (combination of overtones with varying amplitude) and different accents
reveal themselves in the spectrum of the vowels. The power of the vowels is
typically some 12 dB higher than that in the consonants (see Table 3). The
spectrum of the consonants is concentrated in the range from 1000 Hz to 4000
Hz. They are much more important for speech intelligibility (we shall get back
to this in Module II). Fig. 10 shows the relative importance of the different
frequency bands for correct understanding of a spoken message, As determined

26
Figure 7: Classification of sound according to spectral content. The images
on the left show the time dependent instantaneous pressure and on the right
the corresponding power spectrum. The 5 different wave forms are explained in
section 4.

27
Figure 8: Linear (above) and Octave (below) Spectra of White Noise (a) and
Pink Noise (b).

Avg. duration Dominant freq. relative Contribution to

(ms) content (Hz) SPL (dB) Intelligibility
Vowels 90 <500 0 Low
Consonants 20 1000-4000 -12 High

Table 3: Comparison of the characteristics of vowels and consonants in human

speech.

by measuring the level of understanding of several test subjects, while filtering

out the different octave bands.

4.3 Spectrograms
In practice, most sound phenomena are not steady state but transient; a source
starts emitting at some initial time, then it may or may not reach a stable state
where the frequency components remain constant, but eventually the source
stops and the sound dies out. Even so, we can choose a time-scale and deter-
mine the average level of all frequency bands that are present on this time scale.
The time scale, which is really the integration domain in (39), can be as short
as 50 µs as in the ’Peak’ measurements of a SLM, or it can be much longer.
Nevertheless, we will end up with several samples in time, each of which with
its own spectrum. This means that a complete measurement of an acoustic phe-

28
Figure 9: Average power spectrum of male and female speech (SPL at 1m) [6].

Figure 10: Relative importance of octave bands for speech intelligibility [6].

29
 Figure 11: Spectrogram Representation

nomenon is three-dimensional: There is a sound pressure level axis, a frequency

axis and a time axis. A full representation of a measurement on these three axes
is called a spectrogram. Fig. 11 shows an example where the three dimensions
are visualized in perspective. Often one is most interested in only two out of
the three components. Then, a projection of the spectrogram on one of the
three axial planes is enough. The names of these three planes are indicated in
the figure: The harmonic plane shows the (average) distribution of SPL over
the frequencies. The dynamic plane shows the evolution of the total SPL over
time, irrespective of the frequency content. It is the standard output of an SLM.
Lastly, the melody plane shows the evolution of the frequency content over time.
A more common way to visualize the three dimensional result is by using a
color- or grey- scale. Figs. 12 and 13 show two examples. The first one is a
measurement of a spoken phrase in English. The SPL level is represented by
the darkness of the graph at all time-frequency locations. The second one is
a recording of a short piece (6 sec) of violin play. The vertical scale is linear
frequency, the horizontal scale is time. The SPL is represented by a colorscale,
with yellow for the highest levels and black for the lowest. We clearly see the
many superior harmonics of each tone played by the violin.

30
Figure 12: Spectrogram of a spoken sentence. Darker lines have higher SPL
[7].

Figure 13: A spectrogram of a violin, with linear frequency on the vertical axis
and time on the horizontal axis. The coloring represents SPL [8].

31
Figure 14: Typical radiated sound powers (W) and sound power levels (dB
PWL) [9].

5 Source Power Levels

Although it is generally difficult to measure, with any source that produces an
approximately constant sound field over a given time period, we can associate a
given total radiated power. In Fig. 14 some typical radiated sound powers are
shown. Since, just like with typical sound pressure values, the range covering
sources often found in practice is very stretched, radiated sound power is usually
given on a logarithmic scale:
 
Prad
Lw = 10 log dB (P W L), (47)
Pref

where the reference value for the PWL (PoWerLevel) scale equals Pref =
10−12 W . The third column in Fig. 14 gives the PWL values corresponding
to the linear (Watt) values in the second column.
In section 1.4 we derived the formula (33) that gives the total radiated power
→
−
by a source from the intensity I on a closed surface surrounding the source. For
a point source (also called an omnidirectional source because it radiates equally
in all directions), this formula simplifies to the integration of a constant value
over a sphere of radius r:

Prad = 4πr2 I = 4πr2 p2ef /Z0 . (48)

Taking the logarithm of (48) and inserting the numerical values for Pref , pref
and Z0 for air under the usual typical circumstances, we obtain the following
useful relation between source power level and sound pressure level at a distance
r, for an omnidirectional source in free space:

SP L = P W L − 20 log r − 11 dB. (49)

32
Observe that according to (49), as we move away from the source, every time
we double the distance, we lose 6 dB of SPL.
The -11 dB term in (49) is due to the factor 10 log(4π) ≈ 11 from (48). This
is because the radiated power is distributed over a full sphere, which spans a
solid angle of 4π steradians. A source that is located very near an infinite,
flat and completely stiff surface (often a good aproximation of a loudspeaker
mounted very near a large wall), radiates all of its power into the half -space in
front of it. A half-space only spans a solid angle of 2π steradians. Thus, the -11
dB term now becomes a factor two or 3 dB smaller. Hence, formula (49) for a
source very near a large stiff surface becomes
SP L = P W L − 20 log r − 8 dB (half-space) (50)
We see that a loudspeaker radiating the same power but located near a wall will
produce 3 dB more SPL than if it were located in free space. This is simply
a consequence of the radiated power being forced to propagate into a smaller
portion of space than before. It is independent of the frequency content of the
signal. However, there is a second phenomenon associated with a source near
a wall. M irror Image theory tells us that the combination of a point source
and a rigid wall, can be modelled by replacing the wall with an image-source,
located at the same distance from the wall, on the other side, such that the
line connecting the two sources is perpendicular to the wall. This image source
radiates exactly the same signal as the original source, and the combination of
the two sources now radiates into free space. Now, if the distance between the
two sources is much smaller than the wave length, then wherever we measure
the total pressure, it will be twice as large as the pressure from the original
source alone. Hence, the radiated power will be four times larger. Of course,
since the wall is no longer there to confine the power to a half-space, half of the
radiated power is lost into the ’imaginary’ half-space behind the wall. But, we
still have twice more PWL than for the same source in free space. We see that
for low frequencies, we gain not 3 but 6 dB of SPL by placing the source near
a stiff wall. The gain in power as a function of the distance from the wall in
terms of frequency is shown in Fig. 15.
It seems curious that the interaction of a wall with a source can help the
source to radiate more power than in free space. The explanation is that the
radiation resistance seen by the source (more on this in module V on loud-
speakers) changes due to the presence of the wall. Likewise, two sources, located
closely together and radiating identical signals, will radiate four times as much
power in the low frequencies as the power radiated by either source on its own
(for example two loudspeakers side by side, driven by the same amplifier).
If a source is positioned near the intersection of two perpendicular reflecting
walls (on the floor against a wall, then the space into which it radiates spans
only π steradians:
SP L = P W L − 20 log r − 5 dB (edge), (51)
and for low frequencies, mirror image theory provides an additional 6 dB of
PWL with respect to free space. This is because the first reflecting surface is

33
Figure 15: Radiated power from a point source at a distance h from a reflecting
plane, relative to the radiated power in free space, as a function of the wave
number k [10].

replaced by one image source and the second reflecting surface is replaced by
two images, one for the original source and one for the image in the first surface.
Altogether, the equivalent in free space are four sources.
If a source is positioned near a corner where three surfaces intersect, we can
add yet another 3 dB to the SPL:
SP L = P W L − 20 log r − 2 dB (corner) (52)
and for low frequencies we have eight image sources. The SPL at low frequency
at a distance r is then 18 dB or a factor 64 higher than with the same source
in free space!

5.1 Directivity
When a source is not an omnidirectional (point-) source, the radiated field will
still be spherical at sufficient distance away from it (the f ar f ield). However
the power density will not necessarily be equal in all directions. The radiated
power may be concentrated in a single direction or in one particular plane. The
source will have a directivity pattern
I(θ, φ)
Q(θ, φ) = (53)
I0
where I0 is the average intensity. Clearly, for an omnidirectional source, Q(θ, φ)
equals one everywhere. Often one is not so much interested in the details on the

34
Figure 16: Relative Directivity of the human voice at different frequencies,
horizontal plane

directivity pattern, but only in the value at the maximum. When a source is said
to have a directivity, or a directivity f actor, of Q = 2, then this is assumed to
mean the maximum of (53). The directivity index is the logarithmic equivalent
of (53), defined as

DI = 10 log(Q) dB. (54)

Often, the directivity pattern is represented as a relative pattern, normalized
to the maximum. Also it is usually presented as one or more polar plots, showing
two dimensional cross-sections of the full 3D pattern.
As a general rule, sources can only be directive if their dimensions are large
compared to the wave length. Hence, the directivity is a function of the fre-
quency. Fig. 16 shows the (approximate) directivity of human speech in different
frequency bands. The source (the mouth) is small at most wave lengths in the
audible band, so the human voice is not very directive. The size of the human
head also plays a role. At the highest frequencies it is several wave lengths large
so it directs the sound propagation towards the frontal hemisphere.

5.2 Line sources

Sometimes rather than aproximating a source with a point source at a given
position in 3D space, it makes more sense to approximate it with an infinitely
thin but infinitely long straight line generating the same sound everywhere along
its axis. This is equivalent to reducing the situation to a 2-dimensional model,
since the sound field is identical in every plane perpendicular to the line source,
so we only need to study one such plane. Practical examples of situations that
allow for a line source approach are motorways, busy streets or long trains.

35
 Line sources are studied in cylindrical coordinates with the source at the
origin. The total radiated power is now expressed in W/m, i.e. the amount of
Watts radiated per meter along the length of the source. Rather than being
distributed over a sphere, it is distributed over a circle, so the intensity drops
not as 1/r2 but as 1/r. The intensity is the radiated power per unit length over
the circumference of the circle, so (if the line source is omnidirectional),

Prad
Ilinesource = W/m2 , (55)
2πr
which gives, in logarithmic scale

SP L = P W L − 10 log r − 8 dB (line source). (56)

36
6 Summation of Sound Fields
6.1 Incoherent sums
We have seen in section 3.2 that two pure tones of different frequency are un-
correlated or incoherent which means that the cross-term of the squared sum
of the pressures vanishes, such that the squared effective pressure of the sum is
the sum of the squared effective pressures. By extension this works for the sum
of any number of frequency components.
Two different sound waves with given spectral densities can also be inco-
herent, even if their spectra show overlap. In fact, practically all naturally
occurring sound phenomena are incoherent if they come from different sources,
the exception being natural sources producing a pure tone such as a tuning
fork. Hence, in the general case, if at some point in space, one source produces
a sound field pef 1 and another source produces pef 2 , then the total effective
pressure will follow from q
pef = p2ef 1 + p2ef 2 . (57)
In terms of Sound Pressure Levels this translates to
 
Ltot = 10 log 10L1 /10 + 10L2 /10 dB SP L, (58)

or in the general case of N incoherent sources:

N
!
X
Ltot = 10 log 10Li /10
dB SP L. (59)
i=1

Rearranging the equations, we can also subtract the contributions from dif-
ferent sources. This is often used to determine the level Ls due to a given source,
if we can only measure the level in the presence of some background noise. First
we measure the level LN of the background noise in the absence of the source
(eg. a noisy air-conditioner), and subsequently the total level Ltot . The level
due to the source is then estimated with

Ls = 10 log(10Ltot /10 − 10LN /10 ). (60)

6.2 Coherent sums

In practise, coherent sound fields are only encountered in the case of two sources
producing the same pure tone, or when the two sources, although spatially sep-
arated, actually produce the same signal, for example two loudspeakers driven
by the same amplifier, or one source and its reflection in a wall. When such
a situation occurs, it makes sense to study the total pressure frequency by fre-
quency. At every point in space, the total complex pressure is then the sum of
the phasors of the two fields, so the squared effective pressure equals (see Eq.
(31)):
1 2 1 2
p2ef = |Pω,tot | = |Pω,1 | + |Pω,2 | ej∆ϕ  (61)
2 2

37
where ∆ϕ is the position-dependent phase difference between the two phasors.
Observe that, depending on the position, the effective pressure assumes a value
between the sum and the difference of the two separate effective values. An
interference pattern is built up throughout space with minima and maxima. In
the special case that the two fields are of equal strength, (61) reduces to
1  2
p2ef = Pω,1 (1 + ej∆ϕ ) (62)
2
with minima of zero pressure when ∆ϕ equals π, 3π, .. and maxima of 2pef,1
when ∆ϕ equals 0, 2π, ..
If the two sources are point sources emitting identical signals, then the cor-
responding phasors are given by (15) with identical complex amplitudes p̂ω and
their sum is  −jkr1
e−jkr2

e
Pω,tot = Pω + . (63)
r1 r2
where r1,2 are the respective distances from the sources to the observation point.
If these distances are not too different, we can aproximate (63) by setting them
equal to an average distance r in the denominators of (63) and obtain

Pω
Pω,tot = (1 + e−jk∆r ) (64)
r
where ∆r = r2 − r1 is the path-length dif f erence of the two sources to the
observation point. We observe that the field is that of a single source, modulated
by the factor (1+e−jk∆r ) which depends on k∆r = 2π∆r/λ. Over the frequency
spectrum, this factor oscillates between the value two when ∆r = 0, λ, 2λ, ...
and zero when ∆r = λ/2, 3λ/2, ... The interference of the two sources acts as a
periodic filter. When the difference is a multiple of λ the two fields are in phase.
They interfere constructively. When the difference is a multiple of λ plus λ/2,
the interference is destructive and the two fields cancel eachother.
Figure 17 shows a typical situation giving rise to this effect. If the wall is
hard (no absorption) the amplitudes of the two signals will be practically equal
but some frequencies are filtered out while others are amplified, which leads to
a coloring of the signal. Fig. 18 shows the effect of the filter on a spectrally
flat signal, for some path-length differences. The linear scale representation
has inspired the name of this effect: the comb-f ilter ef f ect. Notice that the
maxima are at 6 dB. The pressure doubles (with repect to the single source),
so the intensity quadruples.

38
Figure 17: Interference between direct and reflected signal causes the comb
filter effect (see text).

Figure 18: Comb filter response due to interference between coherent sources
with different path-length differences (delays)

39
7 Attenuation, Reflection and Diffraction
Up to here we have exclusively treated sources in a homogeneous, lossless, infi-
nite medium (except for the reflection in a wall in the previous section). This
is often a good approximation leading to useful practical results in acoustic en-
gineering. However, equally often, we must deal with situations that cannot be
captured within this simplified model. In this section we discuss three different
phenomena that govern the behaviour of acoustic fields under more realistic cir-
cumstances; attenuation of waves in the medium due to absorption, ref lection
on walls and dif f raction of sound waves by solid obstacles.

7.1 Attenuation
In reality, no medium, including the air, is completely lossless. Some acoustic
energy gets converted into heat by complicated mechanisms at the molecular
level. For air, these losses are small so they can often be completely neglected,
leading to the lossless wave equation we’ve already seen. If they are not, they
can be incorporated in the free field solutions to the wave equation for the
intensity (plane wave, spherical wave), through an attenuation factor e−mx ,
where the attenuation coefficient m, in m−1 , is the sum of a number of rather
complicated energy absorption mechanisms, involving the molecular structure
of the medium, and its state (gas, liquid, solid). Even restricting ourselves to
air at sea level as usual, the value of m is highly frequency dependent and also
depends on the temperature and the relative humidity of the medium. Fig. 19
shows the dependence of m on the frequency and the relative humidity for a
temperature of 20ºC.
To give an idea of the consequence of air absorption in practical circum-
stances, if m = 0.0025m−1 , then the intensity of a wave decreases by about
5% or 0.2dB every 10m (remember the intensity is porportional to the pressure
squared). This may become an important factor outdoors or in very large en-
closures, like a concert hall, but in smaller enclosures it is generally negligible
compared with absorption by the walls. According to the figure, this is the
approximate value of m for 2kHz with a relative humidity above 30%4 . As a
rule of thumb, air absorption only becomes relevant above 2kHz and only in
large spaces.

7.2 Reflection
When a sound wave encounters a sudden change of medium, for example a
solid wall, or the ground beneath our feet, part of the power it carries may
penetrate into the wall, while the rest is reflected back into the space where
the wave originated. How much power is transmitted into the new medium
and how it will continue to propagate in there depends on the characteristics
of that medium, and also on the direction of the incident wave with respect to
the boundary between the two media. While this is of course a very important
4 In Barcelona, the yearly average relative humidity is 70% and it rarely drops below 65%

40
Figure 19: Attenuation coefficient for sound waves in air, sea level, 20ºC [11].

41
question in acoustics, we shall continue to restrict ourselves to the propagation
of sound in air, so our interest is mainly in what happens to the part of the
power that is reflected.

7.2.1 Plane waves

Although the word ref lection refers to the sound field re-emitted into the space
of origin by any boundary of any shape in general, when we use the word,
we usually mean specular reflection, which is reflection of plane waves by a
perfectly flat, infinite boundary between media. This seems very restrictive,
but in practice we can often assume that a general boundary is both flat and
without edges locally. We can then apply the rules of specular reflection to the
regions where these conditions hold.
The expressions for the transmitted and reflected waves, given an incident
wave, follow from the boundary conditions at the boundary between the media.
The acoustic boundary conditions are that both the pressure and the normal
component of the particle velocity are continuous across the boundary. Let us
consider the simplest case of a totally hard boundary, that does not move and
does not admit any transmitted wave. Since the wall does not move, there is
no condition on the pressure. The only boundary condition is that the normal
particle velocity has to vanish on the boundary.
Recall that a harmonic plane wave travelling in the direction kˆi is written as

~ →
− →
− Pω −j k~i ·→
−
pi = Pω e−j ki · r , u i = kˆi e r
(65)
Z0
We assume that the wall is in the plane z = 0 and the plane wave travels in the
positive z-direction. We can always choose the coordinate system such that the
wave travels in the zx-plane and we can set the phasor to Pω = 1, for simplicity:
 
kx
→
− 1 −j(kx x+kz z)
pi = e −j(kx x+kz z)
, ui = 0  e (66)
kZ0
kz

where kx2 + kz2 = k 2 . Now, a reasonable guess is that the reflected wave will also
be a plane wave, with the same frequency but not necessarily the same direction.
It should also have the same amplitude, because all the power is reflected back
into the medium of incidence:
 
mx
1 −j(mx x+my y+mz z)
pr = e−j(mx x+my y+mz z) , → −
u r =  my  e . (67)
kZ0
mz

Applying the boundary condition uz (z = 0) = 0 to the total acoustic field (the

sum of incident and reflected field) yields:

kz e−jkx x + mz e−j(mx x+my y) = 0, (68)

42
 Figure 20: Snell’s law for plane waves incident on a flat wall: θr = θi [12].

which must be true for all values of x and y. Hence mx = kx my = 0, and

mz = −kz . This corresponds to a reflected wave whose direction of propagation
makes the same angle with the normal vector n̂ of the wall as the incident wave,
and is in the plane defined by k~i and n̂ (see Fig. 20). This result is called Snell’s
law.
Generalizing to a plane wave according to (65), that is incident on a flat wall
with unit normal vector n̂, the reflected wave will obey

~ →
− →
− Pω −j k~r ·→
−
pr = RPω e−j kr · r , u i = Rkˆr e r
(69)
Z0

where k~r follows from

k~r = k~i − 2(k~i ·n̂)n̂. (70)
The complex ref lection coef f icient R is a frequency dependent parameter of
the material and the structure of the wall, as well as of the medium of the
incident wave and the angle of incidence.
If the wall is simply the boundary between two infinite media with charac-
teristic impedances Z1 and Z2 , then the reflection coefficient becomes
Z2 − Z1
R= (71)
Z2 + Z1

If the medium of the incident wave is air, then Z1 ≈ 413kgm−2 s−1 , as we have
seen in section 1.3. For a totally stiff or hard wall, Z2 = ∞ so R = 1. Conversely,
a completely soft wall has Z2 = 0 and thus R = −1. The third important case
is when Z2 = Z1 , then R = 0 and we have a perfect absorber5 . However, in
5 These values are idealized values never encountered in reality, However, any hard, heavy
solid material will approximate a totally hard wall. An approximately soft wall is harder to
find, because the characteristic impedance should be small relative to that of air, which is
already very small. The closest approximation to a perfect absorber can be found in sound
laboratories called anechoic chambers.

43
general, the wall is not an infinite medium but, for example, a brick wall, a
glass window, or a separation built up of multiple layers of material. Then, R
becomes complex (and hence frequency dependent), because of the interference
of multiple reflections inside the finite thickness layers.
As usual, we are often interested in the power carried by the reflected wave
rather than its exact pressure field. The power density of the plane wave equals
2
Ii = |pi | /Z0 , hence with (69) the power density of the reflected wave is
2
Ir = |R| Ii . (72)

The power that is ’lost’ (that does not return to the room where the incident
wave originates, irrespective of what does happen to it) is then equal to αIi
where
2
α = 1 − |R| (73)
is called the absorption coefficient of the wall.
Let’s examine the simple case of a plane wave in the air inciding perpen-
dicularly on a hard wall (R = 1). Then, the reflected wave will propagate
perpendicularly back into the air, due to (70), and its amplitude will be equal
to that of the incident wave. At the boundary, the pressure of the incident wave
and the reflected wave will be in phase and sum to twice the incident pressure.
The particle velocity of the reflected wave will be equal but opposite in sign to
that of the incident wave (because k~r = −k~i ). Hence the net particle velocity
will be zero, as it should be according to the boundary condition. As we move
away from the wall, the pressures will be more and more out of phase, until at
a distance d = λ/4, the phase difference will be π radians and the pressures of
the incident and reflected wave cancel each other out completely. At a distance
d = λ/2, the pressures are in phase again and add up to twice the original
pressure. This alternating of maxima (peaks) and zeros (nodes) continues peri-
odically from here on. Such a pattern is called a standing wave. The particle
velocity shows the same pattern, but with a zero on the wall and a maximum
at d = λ/4. If |R| < 1, a similar pattern appears, but the peaks are lower and
the nodes are not zero. Fig. 21 illustrates this case.

7.2.2 Spherical waves

If the incident wave is not plane but spherical, then the reflected wave will
obviously be much more complicated. However, the total field as a point in space
due to the incident and the reflected wave is rather easy to determine thanks
to a method that we have already mentioned in section 5, the mirror image
method: Let us assume that the wall is in the plane z = 0. The space above the
wall is homogeneously filled with air, and the incident field comes from a point
source at →−
r s = (xs , ys , zs ) above the wall. We wish to find the total field at a
receiver position →−
r r = (xr , yr , zr ). According to the mirror image method we
can replace the wall by an imaginary source at → −r 0 = (xs , ys , −zs ), or in general
at the point that is the perpendicular reflection of → −
r in the plane of the wall.
The field in the space above the wall is then the sum of the two sources radiating

44
Figure 21: Standing wave pattern due to a plane wave inciding perpendicularly
on a wall with a positive reflection coefficient smaller than one [13].

into infinite free space. If the wall has a reflection coeffcient unequal to one, we
multiply the field from the imaginary source with the reflection coefficient R of
the wall. The procedure is illustrated in Fig. 22
Observe that the mirror image method automatically complies with Snell’s
law, in the sense that, if we approximate the spherical waves both from the true
source and the imaginary source as locally plane around the point where the
incident wave hits the wall, then by geometrical construction the wave vectors
k~i and k~r comply with (70), that is, the angles of the incident and reflected
propagation directions with the surface normal vector are the same.

7.3 Diffraction
Dif f raction is the perturbation of the sound field in the medium in which
it propagates when it encounters an obstacle that is not flat and/or not of
infinite extent in all directions. Two other words for the same phenomenon
are scattering and dif f usion, although they usually refer to slightly different
situations: diffraction is typically used for the sound field penetrating the region
which is blocked from the direct view of the source (the ’shadow zone’) due to
the waves ’bending’ around the edges of the obstacle. Scattering is used to refer
to the perturbation of the sound field by the obstacle in all directions. Diffusion
is used when the scattering is so dominant and unstructured that the perturbed
field no longer has any dominant direction of propagation or net power transport
through any point in space.
An exact solution for the diffraction from an obstacle only exists if the obsta-
cle is a perfect sphere made of a homogeneous material. Good approximations
exist for simple geometries like arbitrarily shaped apertures in a flat hard wall

45
 Figure 22: Illustration of mirror image method in two dimensions [9].

(the Fresnel-Kirchoff diffraction formula), or the complementary problem of just

a flat hard obstacle in free space (by a theorem called Babinet’s principle, the so-
lution to the one problem can easily be converted into the solution to the other).
The diffracted field from a general shape and (combinations of) materials can
only be calculated numerically by computer simulations.
Diffraction is a consequence of the interference of the sound field scattered
from infinitesimallly small parts of the surface of the obstacles. It is therefore
highly frequency dependent. If an object is small compared to the wave length,
diffraction will be minimal and the incident field propagates around it practically
unaltered. See Fig. 23. As a consequence, an obstacle may be large enough
to shield off high frequency components but at the same time be practically
transparent to low frequencies.

7.3.1 Fresnel zones

A qualitative model developed by Fresnel in the 19th century (originally for
optical waves), still often used for estimating the effects of diffraction, subdivides
the space inbetween the source and the receiving point into Fresnel-zones. They
are often defined on a plane perpendicular to the line joining source and receiver,
as in Fig. 24, although they can be defined on any surface separating the source
and the receiver. First, the length of the straight path between the source and
the receiver is determined (d = p + q in the figure). Then, for every point in the
plane, the total path length from source to receiver passing through this point
is found and the difference with the direct line is determined (4 = 4p + 4q
in the figure). Now, the first Fresnel zone is the set of points with 4 < λ/2,
the second Fresnel zone has λ/2 < 4 < λ, etc. In the perpendicular plane, this
leads to concentric circles defining the separation of zones. In the figure, the

46
Figure 23: Diffraction by cylindrical obstacles that are much smaller than the
wave length and approximately of the size of one wave length.

47
 Figure 24: Fresnel zones [14].

first zone is the white disk in the centre, the second zone is the smallest black
ring, etc.
With this rule, different propagation paths through the same zone are never
more than π/2 radians out of phase with the ’central’ path (for the first zone
this is not the direct line but the paths through the points half-way between
the centre and the zone-edge), while propagation paths through adjacent zones
are on average π rad out of phase. So, on average, contributions from the same
zone interfere constructively (amplify eachother) while contributions one zone
apart tend to cancel eachother out. Fresnel constructed a screen covering the
even zones and showed that for light of the appropriate frequency, this screen
acted as a lens: a light beam from the source is focalized towards the receiving
point. The same can be done with sound.
In practise, Fresnel zone theory is mainly used to obtain a quick assessment
of the effect of an obstacle: If the objective is to create a shadow zone, then
a good ’rule of thumb’ is to ensure that a least the entire first Fresnel zone is
blocked. If on the other hand we want the effect of an obstacle to be minimized,
this theory learns us that it is not enough to ensure a ’Line of Sight’ from source
to receiver, but we should try to keep open the entire first Fresnel zone.
When (referring to Fig. 24), the distances p and q are large with respect to
the size of the obstacle (and therefore to the relevant Fresnel zone radius) the
approximate formula s
nλpq
rn = (74)
p+q
can be used to find the (outer) radius of the n-th Fresnel zone.
Although the Fresnel zone theory is very approximate, it is nevertheless
very useful because especially for relatively simple shaped obstacles (circular,

48
square) the attenuation due to blockage and the minimum path length difference
divided by the wave length are very much proportional in practical situations.
An obstacle can be assigned a Fresnel number
24
N= (75)
λ
(not necessarily an integer number), where 4 is the ’typical’ shortest path length
difference with the Line of Sight. For a perfectly circular obstacle, if it com-
pletely blocks the first Fresnel zone, then its Fresnel number equals N = 1.
As an example, we consider the very practical problem of finding the correct
height of a noise barrier between a motorway and a residential neighbourhood.
Considering the motoway as an infinite line source, and assuming that the noise
barrier will be much longer than its height, the problem is reduced to a two
dimensional problem. The geometry of the problem is sketched in Fig. 25. Fig.
26 shows the theoretical attenuation in dB as a function of the Fresnel number.
The graph is calculated by a somewhat more elaborate theory than the one
presented above, but based on the same concept of Fresnel zone diffraction.
It is used by the United States Federal Highway Administration to establish
noise barrier heights. A generally accepted rule for noise barriers is that the
attenuation should be at least -10dB at the lowest frequency with a significant
contribution to the noise. Observe that this is approximately achieved when
N = 1, i.e. the first Fresnel zone is entirely blocked.

49
Figure 25: Fresnel zone theory applied to Noise Barrier [15].

50
 Figure 26: Noise barrier attenuation as a function of Fresnel number [16].

8 Low frequency theory: Acoustic Impedance

We have seen that the local ratio between the pressure (a driving force) and
the particle velocity (the resulting displacement) in a free field gives rise to a
parameter characterizing the medium, the characteristic impendance (section
1.3). If the pressure and particle velocity are approximately constant in space
over a finite surface, we can calculate the ratio of the total force acting on that
surface and the resulting volume velocity U in m3 /s. The volume velocity is
defined as the particle velocity u in the direction perpendicular to the surface,
times the surface area S. From this ratio we can define the concept of Lumped
Acoustic Impedance, or simply acoustic impedance as a characteristic of such
a surface. Although this definition does not explicitly involve the frequency,
in practice the concept is typically applicable only for surfaces that are small
compared to the wave length, hence the name low frequency theory.
A related restriction is that a surface of constant pressure an particle ve-
locity is essentially only found in a plane wave, perpendicular to the direction
of propagation. Therefore the theory only applies to (approximately) normal
incidence of (approximately) plane waves. The two conditions mentioned above
are typically found in guided wave geometries, such as acoustic ducts (tubes),
or in transducers (microphones and loudspeakers). In both cases the theory is
an enormous aid in the design and analysis of the various elements.
By definition, impedance is a complex value, that only makes sense in the
frequency domain. The real part of the impedance is called the resistance, and
it determines the amount of power that is converted into a different form of
energy, usually thermal energy. From the acoustic view point, this power is lost
or dissipated. The imaginary part is called the reactance and it determines the
phase difference between the driving force and the resulting displacement. If

51
the reactance is positive it is called inductive, if it is negative, capacitive.

8.1 Mechanical Impedance

Acoustic impedance is very closely related to mechanical impedance, in fact
the two are practically interchangeable, and in order to introduce the acoustic
impedance, it makes sense to start with the mechanical equivalent (there is a
similar relation with electrical impedance which we’ll mention too in case the
reader is more familiar with that).
Mechanical impedance Zm is the reaction of a solid body to a force, in terms
of the ratio of the force F and the resulting velocity of the body v
F
Zm = . (76)
v
In its simplest form the force is just point-force acting upon a point of the struc-
ture. Part of the force will result in an acceleration of the body, and part will
result in a deformation of the body (stretching or compression). Another part
is transformed into heat through internal friction and friction with the environ-
ment. The different proportions are determined by the physical characteristics
of the body. The acceleration depends on the mass M of the body through
dv
F = Ma = M = jωvM, (77)
dt
so the inductive reactance due to the mass equals

Xm = jωM. (78)

The mass is analogous to the inductance of an electric coil in electric circuit

theory. The deformation depends on the stiffness (or spring constant) K of the
body through Hooke’s law:
Z
K
F = Kδ = K vdt = v (79)
jω
where δ is the change of length of the body in the direction of the force. Con-
sequently the capacitive reactance of the body equals
K
Xm = , (80)
jω
analogous to the electrical capacitance of an electrical condenser or capacitor.
These two characteristics of mechanical impedance of a solid body can be
represented as a mass hanging from a spring, as in Fig. 27. The total mechanical
impedance Zm is the sum of the two components above plus a resistance term
Rm that represents the losses though friction:
F K
Zm = = jωM + + Rm . (81)
v jω

52
 Figure 27: A mass M hanging from a spring with stiffness K.

8.2 Acoustic Impedance

The acoustic impedance of a surface S is defined as
p
Za = , (82)
U
the ratio between the pressure and the volume velocity (both assumed to be
constant over the surface). Notice that over the surface S, the total force by
the pressure equals F = Sp. At the same time, the volume velocity equals
U = Su. We can thus consider the air just in front of the surface as a single
homogeneous body with a constant velocity equal to v = u = U/S. So, if a flat
solid object, such as the diaphragm of a loudspeaker or a microphone with a
surface S vibrates in the air, it is loaded with a mechanical impedance equal to

Zm = Za S 2 . (83)

This shows that the analogy between mechanical and acoustical impedance is
not just an artefact. It can be used to derive a formulation for the total me-
chanical impedance of a microphone or loudspeaker, that includes the radiation
impedance of the sound that is dectected or produced respectively (Ultimately,
microphones and loudspeakers convert between electric and acoustic signals.
The analogy can be extended to electrical circuits as we have seen above; then
the entire system can be described in terms of either mechanical forces and ve-
locities or voltages and currents. We shall revisit this subject in Modules IV
and V).
The acoustic impedance can also be used to model acoustic ducts or tubes
guiding acoustic signals. Practical applications are air or heat exchangers in
buildings and vehicles, or mufflers on car exhaust pipes or guns, that must all
allow a maximum air flow, while attenuating the propagation of sound to a
maximum.

8.3 Helmholtz resonators

A special case are acoustic wave guide sections that are very short in comparison
with the wavelength. Consider a short very thin tube of cross-section S that

53
is open at both ends. If a low frequency wave incides on one end, since the
other end is open, the air column inside it will move as a whole rather than be
compressed. It therefore represents a mechanical reactance of

Xm = jωρ0 Slef (84)

(the mass is the volume times the density. The volume is of course Sl, where l
is the length of the tube, but we use an effective length lef ,slichtly longer than
l, because the moving air column extends a bit outside of the tube). Hence,
with (84), the acoustical reactance gives

Xa = jωρ0 lef /S. (85)

Now consider a hard cavity that is small with respect to the wave length,
with an opening of cross-section S. When a low frequency incident wave incides
upon the opening, the cavity is too small to allow wave propagation, so the air
inside the cavity will be uniformly compressed. The pressure p, and density ρ
in the cavity are related through equation (3):
ρ0
ρ= p. (86)
Ks
Where Ks is the adiabatic bulk modulus from section 1.1. The volume velocity
can be related to the density through continuity of mass: Considering that the
total mass of the air in the cavity M = V ρ, and that, the cavity having hard
walls, its volume V is a constant, we have (again for small relative changes of
ρ),
∂M
= jωρV = ρ0 U. (87)
∂t
Combining (86) and (87) to eliminate the density, we obtain for the impedance
of the cavity:
p Ks
Xa = = . (88)
U jωV
We see that a cavity at low frequency presents a capacitive reactance: The
air acts as a spring that is loaded by the incident pressure.
Finally, if we connect the small tube of (85) to the opening of the cavity of
(88), we obtain a configuration as in Fig. 28. The total acoustic impedance will
be the sum of the two. Using Ks = ρ0 c20 :
ρ0 c20
Za = jωρ0 lef /S + . (89)
jωV
It is easily verified that Za goes to zero when the frequency approaches
s
S
ω0 = c0 . (90)
V lef

This is the resonant frequency of the system of Fig. 28. Such a system is
called a Helmholtz resonator. It is the acoustic equivalent of the mechanical

54
Figure 28: Short open tube of length l and cross section S connected to a small
cavity of arbitrary shape and volume V .

resonator in Fig. 27. In both cases, the slightest driving force at the resonant
frequency will result in an arbitrarily large oscillation (in reality the impedance
will of course never be exactly zero because there will always be resistive losses).
Notice the similarity of the Helmhotlz resonator to an empty bottle. If we blow
across the neck of the empty bottle, our breath will cause random turbulences,
a small part of which is enough to excite the resonance and produce the typical
sound. Helmholtz resonators in various appearances are an important concept
in acoustics.

8.4 Acoustic Filters

In the previous section we saw that short segments of pipes act as an acoustic
mass if they are open-ended and as an acoustic spring if they are closed. When
they are neither open or closed but matched, that is, terminated without any
reflections, then the impedance in a cross-section of the pipe equals that of a
propagating plane wave:
Z0
Za = . (91)
S
Observe that the impedance changes when the cross-section area S changes.
Now, as an example, consider the sequence of segments depicted in Fig. 29.
The presence of the wide segment in the middle can be modeled a continuous
pipe of cross-section S in parallel with a closed cavity of volume V = L (S1 − S).
The reflection coefficient of this circuit is readily calculated as
1
R= 2c0 S
(92)
1+ jωV

2
which represents a low-pass filter with a cutoff-frequency (when |R| = 1/2)
of ωc = 2c0 S/V . A good reference with more examples of elementary acoustic
circuits is [17]. The analogy with electrical circuit theory is evident in the above.
As is the case with electrical circuits, when the length of the pipe segments is
longer than a fraction of the wavelength, the lumped element model ceases to be
valid and we have to allow for plane waves travelling up and down the segment.

55
Figure 29: Acoustic Low-pass filter made of a pipe of cross-section S with an
enlarged segment of length L and cross-section S1 .

References
[1] https://www.isover.co.za/acoustic
[2] https://ebom.com/2018/05/arduino-sound-level-meter/
[3] https://www.castlegroup.co.uk
[4] http://www.industrial-electronics.com/transformers_6c.html
[5] https://en.wikipedia.org/wiki/Noise_curve
[6] https://www.dpamicrophones.com/mic-university/facts-about-speech-
intelligibility
[7] http://home.cc.umanitoba.ca/~robh/archives/arc0705.html
[8] http://en.wikipedia.org/wiki/File:Spectrogram_of_violin.png
[9] https://www.researchgate.net/publication/
292150038_Fundamentals_of_Acoustics_and_Noise_Control
[10] J. Acoust. Soc. Am., Vol 29, p. 743 (1957)
[11] J. Acoust. Soc. Am., vol.40, pp. 148-159 (1966)
[12] http://dev.physicslab.org
[13] H. Kuttruff, Room Acoustics, 5th Ed. (2009)
[14] http://zoneplate.lbl.gov/theory
[15] http://acoustics.group.shef.ac.uk/asaproject.php
[16] https://rosap.ntl.bts.gov/view/dot/30259/dot_30259_DS1.pdf
[17] Fundamentals of Acoustics, 4th Ed. L. E. Kinsler ea.

56