1.

2 Signals
ISVR6136 Fundamentals of Acoustics
Contents
1. What is a signal?
2. How big is a signal?
3. Adding signals
4. Sinusoids
5. Introduction to Python (Jupyter)
6. Signals in Python (Jupyter)
1.2.1 What is a signal?
Wave motion
Pressure signal – a function of time

pressure

𝑝 𝑡
time
Loud tone at 1000 Hz in still air
Quantity Ambient value Fluctuation Comments
Displacement ±0.4 × 10−6 m Vector
Velocity ±2.5 × 10−3 m s −1 Vector
Acceleration ±15 m s −2 Vector
Density 1.2 kg m−3 ±8.5 × 10−6 kg m−3 Scalar
Pressure 105 Pa ±1 Pa Scalar
Temperature 20 ℃ = 293 K ±0.9 × 10−3 K Scalar
Barely perceptible tone at 1000 Hz in
still air
Quantity Ambient value Fluctuation Comments
Displacement ±7.7 × 10−12 m Vector
Velocity ±49 × 10−9 m s −1 Vector
Acceleration ±0.3 × 10−3 m s −2 Vector
Density 1.2 kg m−3 ±0.17 × 10−9 kg m−3 Scalar
Pressure 105 Pa ±20 × 10−6 Pa Scalar
Temperature 20 ℃ = 293 K ±18 × 10−9 K Scalar
Pressure signals
Pressure signals zoomed in
 Signal 𝑥(𝑡)

Deterministic Stochastic
e.g.
𝑥(𝑡) known Pr 𝑥 𝑡 = 𝑥0 known

Periodic Stationary
Non-
Aperiodic
𝑥 𝑡 + 𝑇 = 𝑥(𝑡) Pr 𝑥 𝑡1 = 𝑥0 = Pr 𝑥 𝑡2 = 𝑥0 stationary
Periodic signal
Period 𝑇

Period 𝑇

Frequency 𝑓 = 1Τ𝑇
Units: hertz (Hz)

Period 𝑇
Sampled signals
1.2.2 How big is a signal?
Signal statistics – random noise
Signal statistics – Cymbals clash
Signal statistics – bassoon clip
Signal ‘power’

𝑥 𝑡

𝑥2 𝑡
Average signal ‘power’ (mean square
value)

𝑡0 𝑡0 + Δ𝑡
𝑥2 𝑡
Equal areas
RMS – Root-Mean-Square value
RMS value

𝑥𝑅𝑀𝑆
Worked example 1: what is the RMS value
of a sine signal?

𝑥 2 𝑡 = 𝐴2 sin2 𝜔𝑡 = 12𝐴2 1 − cos 2𝜔𝑡

We perceive loudness logarithmically

1000 watts
100
10 watts
watts

Loudness difference Same loudness difference

Powers of ten and base ten logarithms

1000 = 103 log10 1000 =3

100 = 102 log10 100 =2
10 = 101 log10 10 =1
1 = 100 log10 1 =0
0.1 = 10−1 log10 0.1 = −1
0.01 = 10−2 log10 0.01 = −2
Reminder of log properties
ℓ = log10 𝑥 ⟺ 10ℓ = 𝑥

log10 𝑥 + log10 𝑦 = log10 𝑥𝑦

𝑥
log10 𝑥 − log10 𝑦 = log10
𝑦

𝑛 𝑛
1
log10 𝑥 = 𝑛 log10 𝑥 , log10 𝑥 = log10 𝑥
𝑛
log 𝑏 𝑥
log10 𝑥 = , log10 2 ≅ 0.3
log 𝑏 10
Level
• If signal 𝑥𝐵 𝑡 has ten times the average signal ‘power’ of signal
𝑥𝐴 𝑡 then the level of signal B is one bel higher than that of
signal A
• The number of bels between them can be found by

• The number of decibels between them is 10𝑁

Sound Pressure Level (SPL)
• SPL is given relative to a standard RMS reference pressure (in air)

𝑝𝑟𝑒𝑓 = 20 × 10−6 Pa

• This is the threshold of hearing for a pure tone at 1000 Hz presented to a

large sample of otologically sound subjects aged between 18 and 30
• The SPL of a pressure signal with RMS pressure 𝑝𝑅𝑀𝑆 is:

𝑝2 𝑡 𝑝𝑅𝑀𝑆
𝐿𝑝 = 10 log10 2 = 20 log10
𝑝𝑟𝑒𝑓 𝑝𝑟𝑒𝑓

2
𝑝2 𝑡 = 𝑝𝑟𝑒𝑓 × 10𝐿𝑝 ∕10 , 𝑝𝑅𝑀𝑆 = 𝑝𝑟𝑒𝑓 × 10𝐿𝑝 Τ20
Worked example 2: what is the SPL of a
1 kHz tone whose amplitude is 1 pascal?
𝑝 𝑡 = 𝐴 cos 𝜔𝑡 + 𝜙 , 𝐴 = 1 Pa, 𝜔 = 2000𝜋 rad s−1, 𝜙 =?

1
𝑝𝑅𝑀𝑆 = ≅ 0.707 Pa, 𝑝𝑟𝑒𝑓 = 20 × 10−6 Pa
2

𝑝𝑅𝑀𝑆 1 105
𝐿𝑝 = 20 log10 = 20 log10 = 20 log10
𝑝𝑟𝑒𝑓 2 × 20 × 10−6 2×2

= 100 − 20 log10 2 2 = 100 − 20 log10 23Τ2 = 100 − 32 × 20 log10 2

≅ 100 − 30 × 0.3 = 100 − 9 = 91 dB

• 𝜔 and 𝜙 don’t come into the calculation but noise measurements are often
weighted by frequency to account for perception (see lab)
1.2.3 Adding signals
Adding sounds
 Adding signals

𝑝1 𝑡

+
𝑝2 𝑡

=
𝑝2 𝑡 + 𝑝2 𝑡
Adding decibels
• Let 𝑥 𝑡 be a pressure signal with SPL 𝐿𝑥 dB
• Let 𝑦 𝑡 is a pressure signal with SPL 𝐿𝑦 dB
• What is 𝐿𝑥+𝑦 , the SPL of 𝑥 𝑡 + 𝑦 𝑡 ?
Adding RMS values
• If 𝑥 𝑡 has RMS value 𝑥𝑅𝑀𝑆 and 𝑦 𝑡 has 𝑦𝑅𝑀𝑆 what is the RMS value
of 𝑥 𝑡 + 𝑦 𝑡 ?
 Special case 1: Two identical signals
• Let 𝑦 𝑡 = 𝑥 𝑡 so 𝑥 𝑡 + 𝑦 𝑡 = 2𝑥 𝑡
• Example: a mono recording when a second loudspeaker is plugged in
• What is the difference in SPL between 𝐿𝑥 and 𝐿2𝑥?

2𝑥𝑅𝑀𝑆 𝑥𝑅𝑀𝑆
𝐿2𝑥 − 𝐿𝑥 = 20 log10 − 20 log10
𝑝𝑟𝑒𝑓 𝑝𝑟𝑒𝑓

2𝑥𝑅𝑀𝑆 𝑝𝑟𝑒𝑓
= 20 log10 × = 20 log10 2 ≅ 20 × 0.3 = 6 dB
𝑝𝑟𝑒𝑓 𝑥𝑅𝑀𝑆
Special case 2: Two random signals
• Let 𝑥 𝑡 and 𝑦 𝑡 be stochastic signals such that:
• They have the same SPL, so 𝐿𝑥 = 𝐿𝑦 and therefore 𝑥 2 𝑡 = 𝑦 2 𝑡
• They are completely uncorrelated so 𝑥 𝑡 𝑦 𝑡 = 0
• Example: two vacuum cleaners at the same distance
• What is the difference in SPL between 𝐿𝑥 and 𝐿𝑥+𝑦 ?
Special case 3: Two random signals
(cont’d)

= 20 log10 21Τ2

= 12 × 20 log10 2

≅ 10 × 0.3 = 3 dB
General case: many random signals
• Uncorrelated sounds 𝑥1 𝑡 , 𝑥2 𝑡 , …, 𝑥𝑁 (𝑡) have SPLs 𝐿1, 𝐿2, …, 𝐿𝑁
• What is 𝐿Σ, the SPL of 𝑥Σ 𝑡 = 𝑥1 𝑡 + 𝑥2 𝑡 + ⋯ + 𝑥𝑁 𝑡 ?

2
𝑥Σ2 = 𝑥1 𝑡 + 𝑥2 𝑡 + ⋯ + 𝑥𝑁 𝑡 = 𝑥12 𝑡 + 𝑥22 𝑡 + ⋯ + 𝑥𝑁
2
𝑡
2 2 2
= 𝑝𝑟𝑒𝑓 10𝐿1∕10 + 𝑝𝑟𝑒𝑓 10𝐿2∕10 + ⋯ + 𝑝𝑟𝑒𝑓 10𝐿𝑁 ∕10
2
= 𝑝𝑟𝑒𝑓 10𝐿1∕10 + 10𝐿2∕10 + ⋯ + 10𝐿𝑁 ∕10
 Regulation
Engineering Hearing protection
Why measure sound? Annoyance
...and more

Throughout a space:
Physics pressure waves
What is sound?
At a point: a
pressure signal

RMS pressure in dB
Maths relative to a
Quantify ‘size’ reference pressure
1.2.4 Sinusoids
Sinusoids

A sinusoid is any signal 𝑥 𝑡 of the form

𝑥 𝑡 = 𝐴 cos 𝜔𝑡 + 𝜙

where:
• 𝐴 is its amplitude (pascals, volts, etc depending on the signal)
• 𝜔 = 2𝜋𝑓 is its angular frequency (radians per second)
• 𝜙 is its phase (radians)
• It has a period 𝑇 = 1Τ𝑓 = 2𝜋Τ𝜔
(Make your own pictures for this section using Python)
Adding similar sinusoids
• Interesting fact about sinusoids No. 1:

“The sum of two sinusoids of the same frequency is a sinusoid

with that same frequency”

𝐴1 cos 𝜔𝑡 + 𝜙1 + 𝐴2 cos 𝜔𝑡 + 𝜙2 = 𝐴3 cos 𝜔𝑡 + 𝜙3

Three ways of representing sinusoids
1. The ‘amplitude and phase’ representation: 𝑥 𝑡 = 𝐴 cos 𝜔𝑡 + 𝜙
2. The ‘cos and sin’ representation: 𝑥 𝑡 = 𝑎 cos 𝜔𝑡 + 𝑏 sin 𝜔𝑡
3. The ‘complex exponential’ representation: 𝑥 𝑡 = ℜ 𝐶e ሚ j𝜔𝑡

• How are they related?

ሚ
• Polar representation: 𝐶 = 𝐶 e = 𝐶ሚ cos ∠𝐶ሚ + j sin ∠𝐶ሚ
ሚ ሚ j∠𝐶

• Euler’s formula gives ej𝜔𝑡 = cos 𝜔𝑡 + j sin 𝜔𝑡

ሚ j𝜔𝑡 = 𝐶ሚ cos ∠𝐶ሚ + j sin ∠𝐶ሚ cos 𝜔𝑡 + j sin 𝜔𝑡 , so…
• So 𝐶e
 Taking the real part
ሚ j𝜔𝑡 = 𝐶ሚ ℜ cos ∠𝐶ሚ + j sin ∠𝐶ሚ cos 𝜔𝑡 + j sin 𝜔𝑡
ℜ 𝐶e

= 𝐶ሚ cos ∠𝐶ሚ cos 𝜔𝑡 − sin ∠𝐶ሚ sin 𝜔𝑡

= 𝐶ሚ cos 𝜔𝑡 + ∠𝐶ሚ

• So 𝑎 = 𝐶ሚ cos ∠𝐶ሚ and 𝑏 = − 𝐶ሚ sin ∠𝐶ሚ in the cos & sin representation
• And 𝐴 = 𝐶ሚ and 𝜙 = ∠𝐶ሚ in the amplitude & phase representation
• Going the other way, 𝐶ሚ = A𝑒 j𝜙 = 𝑎 − j𝑏
Adding different sinusoids
• Interesting fact about sinusoids No. 2:

“The sum of two sinusoids of different frequencies is periodic if and

only if the ratio of their frequencies is that of two integers”

𝑥 𝑡 = 𝐴1 cos 𝜔1𝑡 + 𝜙1 + 𝐴2 cos 𝜔2𝑡 + 𝜙2

• If
𝜔1 𝑓1 𝑇2 𝑚
= = =
𝜔2 𝑓2 𝑇1 𝑛

and 𝑚 and 𝑛 are integers, then 𝑥 𝑡 + 𝑚𝑇1 + 𝑛𝑇2 = 𝑥 𝑡

 Sums of sinusoids
• Interesting fact about sinusoids No. 3:

“The sum of any number of sinusoids whose frequencies are all

integer multiples of one frequency is periodic
with the same period as a sinusoid of that frequency”
• If
∞
𝑥 𝑡 =෍ 𝐴𝑛 cos 𝑛𝜔𝑡 + 𝜙𝑛
𝑛=1
then
𝑥 𝑡+𝑇 = 𝑥 𝑡 , 𝑇 = 2𝜋Τ𝜔

(a consequence of interesting fact No.2)

Fourier series
• Interesting (amazing) fact about sinusoids No. 4:

“The converse is true – every periodic signal can be written as a

sum of sinusoids whose frequencies are integer multiples!”

• If 𝑥 𝑡 + 𝑇 = 𝑥 𝑡 then there exist 𝐴1 , 𝐴2 , … and 𝜙1 , 𝜙2 , … such that

𝐴0 ∞
𝑥 𝑡 = + ෍ 𝐴𝑛 cos 𝑛𝜔𝑡 + 𝜙𝑛
2 𝑛=1
where 𝜔 = 2𝜋 Τ𝑇
Fourier series – three representations
• Since 𝐴𝑛 cos 𝑛𝜔𝑡 + 𝜙𝑛 is a sinusoid, which can be written
three ways, we have

𝐴0 ∞
𝑥 𝑡 = + ෍ 𝐴𝑛 cos 𝑛𝜔𝑡 + 𝜙𝑛
2 𝑛=1

𝑎0 ∞
= +෍ 𝑎𝑛 cos 𝑛𝜔𝑡 + 𝑏𝑛 sin 𝑛𝜔𝑡
2 𝑛=1
∞
= 𝐶0 + ෍ ℜ 𝐶ሚ𝑛 ej𝑛𝜔𝑡
𝑛=1
Negative frequency!
• We can write ej𝑛𝜔𝑡 = cos 𝑛𝜔𝑡 + j sin 𝑛𝜔𝑡; what is e−j𝑛𝜔𝑡 ?

e−jn𝜔𝑡 = cos −𝑛𝜔𝑡 + j sin −𝑛𝜔𝑡 = cos 𝑛𝜔𝑡 − j sin 𝑛𝜔𝑡

• So
ej𝑛𝜔𝑡 + e−j𝑛𝜔𝑡 = 2 cos 𝑛𝜔𝑡
and
𝐶ሚ𝑛 ej𝑛𝜔𝑡 + 𝐶ሚ𝑛∗ e−j𝑛𝜔𝑡 = 2 𝐶ሚ𝑛 cos 𝑛𝜔𝑡 + ∠𝐶ሚ
∗
(𝐶ሚ𝑛∗ is the complex conjugate of 𝐶ሚ𝑛 , and e−j𝑛𝜔𝑡 = ej𝑛𝜔𝑡 )
Fourier series – best representation!
• If we set 𝑐ǁ𝑛 = 12𝐴𝑛 ej𝜙𝑛 = 12 𝑎𝑛 − j𝑏𝑛 and 𝑐ǁ−𝑛 = 𝑐ǁ𝑛∗ then we can
write

(even though RHS has no ℜ ⋅ , the imaginary parts cancel out so

LHS is real)
Finding the coefficients
• The 𝑐ǁ𝑛 for a particular 𝑥 𝑡 with period 𝑇 can be found from

(I won’t be asking you to calculate these, though other modules

might)
Fourier transform
• In the limit as 𝑇 → ∞ the Fourier series turns into an integral

ℱ[⋅] is the Fourier Transform (FT) operator

Power spectral density
• Stochastic signals don’t have Fourier transforms
• Stationary stochastic signals have autocorrelation functions (as
do deterministic functions)

𝑅𝑥𝑥 𝜏 = E 𝑥 𝑡 𝑥 𝑡 + 𝜏 =𝑥 𝑡 𝑥 𝑡+𝜏

• The Power Spectral Density (PSD) 𝑆𝑥𝑥 𝜔 = ℱ 𝑅𝑥𝑥 𝜏 provides

the frequency content of 𝑥 𝑡
• Comparable to ℱ 𝑥 𝑡 2 for a deterministic signal
• If 𝑥 𝑡 is a pressure signal with units Pa then 𝑆𝑥𝑥 has units
Pa2 Hz −1
1.2 Signals - conclusions
• SPLs vary logarithmically with amplitude
• When adding SPLs you need to know if the sounds are
correlated
• Representing sinusoids as complex exponentials makes things
easier
• Every signal consists of sinusoids added together