Fourier Analysis

C. Blair, R. Braddell, T. Harris

25th October 2007
Abstract The objective of this experiment was to analyse periodic functions using Fourier
analysis. Sine and rectangular continuous waves and pulses were generated using various
methods and the Fourier transform of each waveform was determined and compared with
the theoretical prediction. The method was used to investigate human voices and also to
understand Heisenbergs Uncertainty Principle.
1 Introduction
Fourier analysis of a periodic waveform relies on our ability to write any periodic function (a
function which repeats itself after some p: f(x + np) = f(x), where n is some integer) as
a Fourier series, that is, an innite sum of simple trigonometric functions, ie cos and sin, as
follows:
f(x) = a
0
+

n=1
(a
n
cos(nx) + b
n
sin(nx))
where the coecients a
0
, a
n
and b
n
are given by the Euler formulas:
a
0
=
1
p
_
p/2
p/2
f(x)dx
a
n
=
2
p
_
p/2
p/2
f(x) cos
_
2n
p
x
_
dx
b
n
=
2
p
_
p/2
p/2
f(x) sin
_
2n
p
x
_
dx
These are derived by noting the orthogonality of trigonometric functions and that the
average value of cos and sin over one period is zero, and integrating the expression for the
Fourier series over one period.
1
The coecients a
n
and b
n
tell us directly how much of a certain frequency is present in
the Fourier series. This is very useful in determining how well a certain frequency correlates
to the original periodic function.
The Fourier series is useful for analysing functions which are periodic over a discrete
interval, however we must introduce the Fourier integral when dealing with functions which
range over the entire x-axis. This Fourier integral allows us to examine the Fourier transform
for all functions. The Fourier transform is an integral transform which takes a function of
one variable and transforms it into a function of a dierent variable, in this case the original
periodic function in the time domain (along the x-axis) is transformed into a function in
the frequency domain. The Fourier transform is analogous to the coecients in the Fourier
series and tells us how strongly particular frequencies correlate with our function. Thus the
complex form of the Fourier transform is denoted by x(f).
x(f) =
1

2
_

x(t)e
i2ft
dt
So the Fourier transform is essentially a convenient method of determining the frequencies
which are most prominent in a particular function, and x(f) gives us a useful frequency
spectrum corresponding to the signal x(t). There are many applications for this in areas
such as digital signal processing. Below is an example of the sine wave and its Fourier
transform:
Figure 1: Sine wave & its Fourier transform
In an experimental setting Discrete Fourier Analysis is used which involves reconstructing
a signal by collecting discrete data. This must be used due to the nite nature of the sampling
rate of an signal. An algorithm called the Fast Fourier Transform is widely used to reduce
the complexity in calculating the Fourier Transform, and in particular is used in the this
experiment by the CASSY Lab software.
2 Experimental Method
The appropriate adjustments in the CASSY Lab programme were made before the equipment
was used. This involved selecting the appropriate voltage range, and ensuring that CASSY
Lab was using the Fast Fourier Transform. Before each part of the experiment was carried
out the sampling rate of the apparatus was adjusted so that aliasing of the input signal did
not occur. Aliasing is a phenomenon which occurs when the sampling rate of an input is too
2
low for the Fast Fourier Transform to work correctly. If the sampling rate is too low then
the programme will erroneously try to reconstruct a wave which is not present in the input
signal. The sampling rate must satisfy f
input
>
f
sampling
2
for aliasing to be avoided.
2.1 Using the CASSY Lab Programme
A number of sine waves of varying frequency were generated using an oscillator. The wave-
forms were recorded using the CASSY Lab programme and the Fast Fourier Transform of
the input signals was determined using the software. A square wave of about 800Hz was
generated using the oscillator, and the resulting Fourier Transform was examined.
2.2 Using a microphone to detect sound waves
In this part of the experiment the oscillator was connected to a speaker to generate sound
waves and the microphone was connected as the CASSY Lab input.
A square wave was generated in air the input signal was recorded and its Fourier transform
was examined.
The upper/lower thresholds of human hearing were determined by increasing/decreasing
the frequency of the sound wave until the sound could no longer be heard.
Next, two tuning forks, f
a
=304.4Hz and f
b
=320.0Hz were used to create a sounds wave
with a beat frequency. Again the sound wave was recorded and its Fourier transform was
examined.
Lastly in this section, dierent voices were analysed by recording the sound of dierent
vowels of men and womens voices and examining the Fourier transform of the recording.
2.3 Eect of increasing the frequency of a sine wave on the Fourier trans-
form
A sine wave was generated using the oscillator, and the change in the Fourier transform of
the wave was observed as the frequency of the wave was varied.
2.4 Rectangular pulse
A single rectangular pulse was created by setting the oscillator to direct current, removing
one of the sensor inputs, starting to record the input and then briey the disconnected
input was reconnected (for less than a second). The frequencies of the rst three minima
were determined from the Fourier transform as well as the amplitude corresponding to the
maximum near zero.
2.5 The frequency spectrum of a repetitive pulse
The function generator was set to generate a repetitive pulse. The recordings Fourier
transform was analysed.
A sine wave pulse was also generated in a similar manner to the square wave pulse, and
its Fourier transform was identied.
3
2.6 Uncertainty in a particles position and momentum
In the nal part, the uncertainty in a particles position and momentum was investigated
using the given Fourier transform of a particles position.
3 Results & Analysis
3.1 Sine wave & its Fourier transform
The sine wave function can be written in terms of exponentials with imaginary exponents
using De Moivres theorem. Thus the Fourier transform of the continuous sine wave becomes:
x(f) =
1

2
_

e
i2f
0
t
e
i2f
0
t
2i
.e
i2ft
dt
=
1
2i

2
((f f
0
) (f + f
0
))
where (x) is the Dirac delta function, dened as (0) = and (x) = 0 where x = 0.
Thus, the Fourier transform is simply a peak at f
0
, as you would expect as the sine wave
consists only of one frequency.
The Fourier transform of a 10Hz 1Hz sine wave was determined using the CASSY Lab
FFT.
(a) Sine Wave (b) Fourier transform of sine wave
Figure 2: Sine wave & its Fourier transform
The square wave is dened by the function
f(x) =
_
L if L < x < 0
L if 0 < x < L
To prove only the odd harmonics are present in the Fourier series of the square wave,
we examine the b
n
coecients of the Fourier series. Because we know f(x) is odd, ie
4
f(x) = f(x) then it is true that only the odd terms the sine terms of the Fourier
series are non-zero.
b
n
=
1
L
_
L
L
f(x) sin
_
n
L
x
_
dx
=
1
L
__
0
L
f(x) sin
_
n
L
x
_
dx +
_
L
0
f(x) sin
_
n
L
x
_
dx
_
=
1
L
__
0
L
Lsin
_
n
L
x
_
dx +
_
L
0
Lsin
_
n
L
x
_
dx
_
=
2L
n
(1 cos(n))
And
1 cos(n) =
_
2 for odd n
0 for even n
Thus
b
1
=
2L

, b
2
= 0, b
3
=
2L
3
, . . .
Only the odd harmonics in the sine coecient in the Fourier series are non-zero.
3.2 Sound waves in air
The spectrum of the square wave in air was less distinct from the previous square wave
spectrum due to the nature of pressure waves traveling in air.
The upper and lower thresholds of hearing were determined to be 1.7610
4
Hz 250Hz
1.75 10
2
Hz 25Hz
The beat frequency of the tuning forks was measured to be 20Hz which corresponds
to the dierence in frequencies of the tuning forks, f
a
f
b
. The beat frequency spectrum
is comparable to the spectrum for the coupled pendulum. Also the damping eect on the
amplitude of the beat frequency was determined to be decaying e
At
The Fourier transform of men and womens voices were analysed and although no quan-
titative result was obtained, it was observed that the fundamental frequency for a mans
voice was lower than a womans for a vowel sound.
3.3 Aliasing of sine wave pulse
As the frequency of the sine wave generated by the oscillator was increased it was noted that
the waveform was aliased after the frequency was increased beyond the threshold dened by
the relation f
input
>
f
sampling
2
. The table below shows how the waveform became aliased
after the input waves frequency was greater than 250Hz:
5
Input (Hz) FFT of Input (Hz) Aliased/Not Aliased
100 93.59 Not Aliased
150 138.10 Not Aliased
200 181.20 Not Aliased
250 219.72 Not Aliased
300 147.9 Aliased
350 193.22 Aliased
400 235.65 Aliased
Table 1: Aliased sine wave
3.4 Single rectangular pulse
The single rectangular pulse is given by the formula:
f(x) =
_
V if |x| <
L
2
0 if |x| >
L
2
The Fourier transform is given by the Fourier cosine transform as it is known that the
rectangular wave function is an even one:
x(f) =
_
2

_
L/2
0
V cos(2fx)dx
= V
_
2

sin(fL)
2f
= V
1

2
sin(fL)
f
noting that from the denition of f(x) that it is zero everywhere outside of L/2. Thus
the minima will occur where sin(fL) = 0, or everywhere where fL = n or f = n/L.
In other words the frequencies at which the minima occur should appear where they are
integer multiples of the frequency at the rst minimum. The rst, second and third minima
were measured to be 2.20 0.25Hz, 4.10 0.25Hz and 6.35 0.25Hz respectively, thus
corresponding with the theory.
To predict the value of the rst maximum it is necessary to use the approximation
sin/ 1 as tends toward zero. This allows us to make the approximation A
1

V L

2
.
Then the values of the subsequent maxima were predicted in a similar manner to the minima
with modifying sin(fL) = 1 which implies that they will occur where fL = n/2 or
f = n/2L where n = 3, 5, . . .. We can rewrite the formula in terms of the value of the rst
maximum as follows:
A
n
=
V

2
.
1

_
n
2L
_ =
2
n
A
1
This formula accurately predicts the values of the second and third maxima 1.8V0.1V and
1.0V0.1V, where the rst is 8.9V0.1V.
6
(a) Rectangular Pulse (b) Fourier transform of Rectangular Pulse
Figure 3: Rectangular pulse & its Fourier transform
The Fourier transform of a single rectangular pulse is equivalent to the intensity prole
of a diraction pattern. This can be understood by examining the analogy between the
components of the frequency of the rectangular wave pulse which are odd harmonics, which
is similar to the analysis of the interference fringes which occur when light passes through
an aperture.
Varying the length of the rectangular pulse revealed that the longer the pulse then the
smaller the spacing between the maxima, this is similar to the reduction of the slit width of
a single slit diraction pattern.
3.5 Repetitive square and sine wave pulses
The square wave pulse is dened by
f(x) =
_
L if |x| <
L
2
0 if |x| >
L
2
This is identical to the rectangular pulse except V = L in this case. The analysis is also
similar of the repetitive square pulse. The results below correspond accurately with the
theory.
Minima Frequency (Hz)
1 1.10 0.1
2 2.1 0.1
3 3.1 0.1
Table 2: Minima
7
Maxima Amplitude (V)
1 4.1 0.1
2 1.1 0.1
3 0.6 0.1
Table 3: Maxima
The sine wave pulse is given by:
f(x) =
_
V cos(x) if L < x <
L
2
0 if |x| > L
The only non-zero Fourier coecients are the even cosine terms, which leaves:
a
n
=
_
L
L
V cos(x) cos(nx)dx
Which can be reduced a simpler integral by using trigonometric substitutions and com-
puted to give:
a
n
= V L
_
sin(L( + n))
L( + n)
+
sin(L( n))
(L( n))
_
= V L(sinc(L( + n)) + sinc(L( n)))
The Fourier transform of the sine wave pulse is as expected a peak at the frequency of
the sine wave (4000Hz in this case), this is similar to the Fourier transform of the continuous
sine wave as investigated earlier.
Figure 4: Fourier transform of sine wave pulse
8
3.6 Heisenbergs Uncertainty Principle
Momentum is dened as p = k in quantum mechanics. By Schrodingers wave equation
the particles state can be written as e
i
p

x
. Thus, the Fourier transform of the particles
position is its momentum. Now, we know that the Fourier transform cannot be computed
exactly unless the it is taken over the whole domain, which is theoretically impossible to do.
Thus the particles position and momentum cannot be determined simultaneously, which is
a statement of Heisenburgs Uncertainty Principle.
4 Conclusion
The theory of the Fourier analysis to decompose a periodic function into its consistent sine
and cosine waves was investigated and validated by using the CASSY Lab equipment and
software to analyse certain periodic waveforms generated using various methods. The Fourier
transform of the recorded waves was compared with the theoretical prediction and shown to
be in agreement.
References
[1] Kreyszig, E. 2006, Advanced Mathematics Engineering, 9th edn, Wiley
[2] Hecht, E. 2002, Optics, 4th edn, Addison Wesley
[3] Fourier Transform, available at: http://www.wikipedia.org/Fourier transform
9