J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 1 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 2
Fourier Series & Transform Summary Fourier Series & Transform Observations
Periodic signals can be represented as a sum of sinusoids
1
DTFS x[n] = X[k] ejko n X[k] = x[n]ejko n Nonperiodic signals can be represented as an integral of sinusoids
N
k=<N > n=<N >
DT signals
1
CTFS x(t) = X[k] e jko t
X[k] = x(t)e jko t
dt Can be represented by a finite frequency range
T T
k= Have transforms that are periodic
+
DTFT x[n] =
1
X(ej ) ejn d X(ej ) = x[n] ejn X[k] = X[k + N ] X(ej ) = X(ej(+2) )
2 2 n=
+ + This is a consequence of ejn = ej(2)n
1
CTFT x(t) = X(j) ejt d X(j) = x(t) ejt dt CT signals
2
Must be represented by an infinite frequency range, in general
Have transforms that are nonperiodic, in general
Each transform has a synthesis and analysis equation
X[k] = X[k + N ] X(j) = X(j( + o ))
Each transforms the signal (function of time) into a function of
frequency This is because ej1 t = ej2 t unless 1 = 2
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 3 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 4
Notes on Terminology Property 1: Linearity
Many of the transforms share the same properties FS
a1 X1 [k] + a2 X2 [k]
The following slides cover the primary properties of the transforms a1 x1 [n] + a2 x2 [n] FT
a1 X1 (ej ) + a2 X2 (ej )
I will use the word transform generally to refer to both the Fourier FS
series transforms and the Fourier transforms a1 X1 [k] + a2 X2 [k]
a1 x1 (t) + a2 x2 (t) FT
The book avoids this and uses the word representation a1 X1 (j) + a2 X2 (j)
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 5 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 6
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 7 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 8
Application Example 1: Workspace Application Example 2: Applying the Denition
Suppose you obtain an old recording of a song that you like that is
corrupted with a constant humming background noise. Is it possible to
design an LTI lter that will eliminate the humm? What
characteristics would you want the lter to have? How much would it
distort the original signal?
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 9 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 10
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 11 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 12
Example 2: Time Shift Example 2: Workspace
Derive one of the four time-shift relationships on the previous slide.
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 13 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 14
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 15 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 16
Application Example 3: Workspace Property 3: Frequency Shift
FS
ejks o n x[n] X[k ks ]
FT
ejs n x[n] X(ej(s ) )
FS
ejks o t x(t) X[k ks ]
FT
ejs t x(t) X (j( s ))
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 17 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 18
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 19 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 20
Application Example 4: Communications Channel Example 4: Workspace
Suppose you wish to transmit a stereo signal over a single telephone
line. You can model a telephone line as a lowpass lter with a 4 kHz
passband.
How can you combine the two signals into a single signal in such a
way that you can extract the original signals?
What must you compromise in combining the two signals into a
single signal?
What properties must the signals have in order to send them over
the telephone line without loss?
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 21 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 22
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 23 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 24
Example 4: Transform Symmetry Example 4: Workspace
Derive one of the four complex-conjugate symmetry relationships listed
on the previous slide given that the signal is real-valued.
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 25 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 26
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 27 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 28
Property 5: Transform Relationship to Signal Symmetry Example 5: Transform Relationship to Signal Symmetry
If the signal is real, then Derive one of the eight relationships listed on the previous slide given
FS FS that the signal is real-valued. Hint: use the complex-conjugate
Re{X[k]} Im{X[k]} symmetry of the transform.
xe [n] F T xo [n] FT
Re{X(ej )} Im{X(ej )}
FS FS
Re{X[k]} Im{X[k]}
xe (t) F T xo (t) FT
Re{X(j)} Im{X(j)}
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 29 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 30
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 31 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 32
Example 6: Transform Relationship to Signal Symmetry Example 6: Workspace
Derive one of the amplitude-phase forms listed on the previous slide
given that the signal is real-valued. Hint: use the complex-conjugate
symmetry of the transform.
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 33 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 34
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 35 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 36
Application Example 5: Polar versus Rectangular Application Example 5: Polar versus Rectangular
Use the fast Fourier transform to estimate the CTFT over a range of Linus: Philosophy of Wet Suckers
frequencies from 0 to half the sampling rate, fs /2, of a 60 ms segment 0.2
of speech.
0.15
What insights do you gain from studying the FT plots that you
0.1
did not gain from studying a time-domain plot?
0.05
What type of properties would a communications channel require
to transmit this signal? 0
What insights do you gain from studying the real, imaginary, 0.05
amplitude, and phase plots. 0.1
0.15
0.2
0.25
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 37 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 38
Application Example 5: Polar versus Rectangular Application Example 5: Polar versus Rectangular
15 20
FT Magnitude
10
15
Real FT
5
0 10
5
5
10
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500
200
FT Complex Phase (degrees)
20
100
Imaginary FT
10
0
0
100
10 200
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500
Frequency (Hz) Frequency (Hz)
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 39 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 40
Application Example 5: MATLAB Code set(h,LineWidth,0.6);
xlim([min(f) max(f)]);
ylim([1.05*min(real(X(k))) 1.05*max(real(X(k)))]);
function [] = Speech(); set(gca,XTick,[0:500:max(f)]);
close all; AxisLines;
box off;
[x,fs,nbits] = wavread(WetSuck.wav); ylabel(Real FT);
subplot(2,1,2);
k = round(fs*1.6):round(fs*1.66); % Look at only 0.5 s h = plot(f,imag(X(k)),r);
x = x(k); set(h,LineWidth,0.6);
nx = length(x); xlim([min(f) max(f)]);
ylim([1.05*min(imag(X(k))) 1.05*max(imag(X(k)))]);
figure; set(gca,XTick,[0:500:max(f)]);
FigureSet(1,LTX); AxisLines;
t = (k-1)/fs; box off;
h = plot(t,x,b); ylabel(Imaginary FT);
set(h,LineWidth,0.6); xlabel(Frequency (Hz));
xrng = max(x)-min(x); AxisSet(6);
xlim([min(t) max(t)]); print -depsc SpeechFTRectangular;
ylim([min(x)-0.01*xrng max(x)+0.01*xrng]);
AxisLines; figure;
xlabel(Time (sec)); FigureSet(1,LTX);
ylabel(); subplot(2,1,1);
title(Linus: Philosophy of Wet Suckers); h = plot(f,abs(X(k)),r);
box off; set(h,LineWidth,0.6);
AxisSet(8); xlim([min(f) max(f)]);
print -depsc Speech; ylim([0 1.05*max(abs(X(k)))]);
set(gca,XTick,[0:500:max(f)]);
X = fft(x,2^(max([12,nextpow2(nx)]))); box off;
nX = length(X); ylabel(FT Magnitude);
k = 1:floor((length(X)+1)/2); subplot(2,1,2);
f = (k-1)*(fs)./(nX+1); h = plot(f,angle(X(k))*180/pi,r);
set(h,LineWidth,0.6);
figure; xlim([min(f) max(f)]);
FigureSet(1,LTX); ylim([-200 200]);
subplot(2,1,1); set(gca,XTick,[0:500:max(f)]);
h = plot(f,real(X(k)),r); AxisLines;
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 41 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 42
box off;
ylabel(FT Complex Phase (degrees)); Amplitude/Phase versus Real/Imaginary Plots Continued
xlabel(Frequency (Hz));
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 43 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 44
Property 7: Convolution of Nonperiodic Signals Example 7: Nonperiodic Continuous-Time
Derive either the DT or CT relationship between the convolution of
x[n] h[n] = h[k]x[n k] = x[k]h[n k] two signals and their transforms
k= k= FT FT
x(t) h(t) X(j)H(j) or x[n] h[n] X(ej )H(ej )
x(t) h(t) = h( )x(t ) d = x( )h(t ) d
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 45 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 46
FT FT
x[n] h[n] X(ej )H(ej ) x(t) h(t) X(j)H(j)
This is very similar to what we used the Laplace transform for
Key dierences
Can be applied to two-sided signals
Cannot be applied to signals with innite energy
Works in discrete-time as well as continuous-time
Enables us to think of a system in terms of how it scales and
shifts the complex phase angle of each complex sinusoid
component of the signal
This is very useful if we have a good notion of which components
are present in a signal via the Fourier transform
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 47 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 48
Property 8: Convolution of Periodic Signals Example 8: Convolution of Periodic Signals
Derive either the DT or CT relationship between the convolution of
FS
x(t) z(t) = x( )z(t ) d x(t) z(t) T X[k]Z[k] two signals and their transforms
<T >
FS FS FS
x[n] z[n] = x[k]z[n k] x[n] z[n] N X[k]Z[k] x(t) h(t) T X[k]Z[k] x[n] h[n] N X[k]Z[k]
k=<N >
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 49 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 50
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 51 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 52
Property 10: Signal Multiplication Example 9: Signal Multiplication
FS Derive one of the four relationships on the previous slide.
x[n] z[n] X[k] Z[k]
FT 1
x[n] z[n] X(ej ) Z(ej )
2
FS
x(t) z(t) X[k] Z[k]
FT 1
x(t) z(t) X(j) Z(j)
2
Recall that convolution in the time domain is equivalent to
multiplication in the frequency domain
This shows that multiplication in the time domain is equivalent to
convolution in the frequency domain
Recall that the DT transforms are periodic and the CT transforms
are non periodic
Naturally, it works out that the DT transforms have circular
convolution and the CT transforms have ordinary convolution
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 53 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 54
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 55 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 56
Windowing Continued Example 5: Windowing
Your oscilloscopes in the lab must estimate the CT Fourier transform
x(t) |X(j)|
of a signal using a nite segment of the data. Calculate and plot the
CT Fourier transform of a windowed sinusoid with a fundamental
t
frequency of f0 = 10 Hz for a range of window lengths. Select the
w(t) |W (j)|
sinusoidal phase and window alignment such that the signal has even
symmetry (why?).
t
w(t) x(t) 1
2 |X(j) W (j)| How does the estimated spectrum compare to the true spectrum?
What happens as the window length decreases towards zero?
t
What happens as the window length increases towards innity?
Since only of a segment of the signal is available, the Fourier What would the transform of an ideal window be?
transform of the signal cannot be calculated exactly
What are the disadvantages of using a rectangular window,
Windowing is necessary to estimate the spectrum from a segment w(t) = pT (t)?
of the signal with nite duration
In the frequency domain, this lters or blurs the spectrum
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 57 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 58
Window w(t)
0.5
0 |t| > T
0
sin(T ) T
P (j) = 2T = 2T sinc 0.5
T
1
Windowed Signal y(t) = w(t) x (t)
5 4 3 2 1 0 1 2 3 4 5
s
1
0.5
0
0.5
1
5 4 3 2 1 0 1 2 3 4 5
Time (s)
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 59 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 60
Example 10: Window and Windowed Segment Transforms Example 10: Window and Windowed Segment
Pulse Spectral Amplitude fo= 10.0 Hz T=0.100 s Window fo= 10.0 Hz T=0.500 s
10 1
Window w(t)
0.5
W(j)
5 0
0.5
0
1
s
1
4 0.5
Y(j)
2 0
0.5
0
1
15 10 5 0 5 10 15 5 4 3 2 1 0 1 2 3 4 5
Frequency (rad/s) Time (s)
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 61 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 62
Example 10: Window and Windowed Segment Transforms Example 10: Window and Windowed Segment
Pulse Spectral Amplitude fo= 10.0 Hz T=0.500 s Window fo= 10.0 Hz T=1.000 s
10 1
Window w(t)
0.5
W(j)
5 0
0.5
0
1
Windowed Signal y(t) = w(t) x (t)
15 10 5 0 5 10 15 5 4 3 2 1 0 1 2 3 4 5
s
1
4 0.5
Y(j)
2 0
0.5
0
1
15 10 5 0 5 10 15 5 4 3 2 1 0 1 2 3 4 5
Frequency (rad/s) Time (s)
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 63 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 64
Example 10: Window and Windowed Segment Transforms Example 10: Window and Windowed Segment
Pulse Spectral Amplitude fo= 10.0 Hz T=1.000 s Window fo= 10.0 Hz T=2.000 s
10 1
Window w(t)
0.5
W(j)
5 0
0.5
0
1
s
1
4 0.5
Y(j)
2 0
0.5
0
1
15 10 5 0 5 10 15 5 4 3 2 1 0 1 2 3 4 5
Frequency (rad/s) Time (s)
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 65 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 66
Example 10: Window and Windowed Segment Transforms Example 10: Window and Windowed Segment
Pulse Spectral Amplitude fo= 10.0 Hz T=2.000 s Window fo= 10.0 Hz T=5.000 s
10 1
Window w(t)
0.5
W(j)
5 0
0.5
0
1
Windowed Signal y(t) = w(t) x (t)
15 10 5 0 5 10 15 5 4 3 2 1 0 1 2 3 4 5
s
1
4 0.5
Y(j)
2 0
0.5
0
1
15 10 5 0 5 10 15 5 4 3 2 1 0 1 2 3 4 5
Frequency (rad/s) Time (s)
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 67 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 68
Example 10: Window and Windowed Segment Transforms Example 10: MATLAB Code
15 10 5 0 5 10 15
4
Y(j)
15 10 5 0 5 10 15
Frequency (rad/s)
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 69 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 70
Example 10: MATLAB Code Continued Property 11: Time & Frequency Scaling
FT
x[an] X(ej/a )
FS
x(at) X[k]
FT 1 j
x(at) X
|a| a
Scaling a signal in time inversely scales the spectrum by the same
factor
Scaling a periodic signal shifts the spacing of the harmonic
components, but the amplitudes are the same!
In the DT case, a must be an integer (the signal is not dened at
non-integer samples)
This is equivalent to sampling only 1 out of every a values of x[n]
DT periodic signals are tricky because the fundamental period of
x[an] depends on how a and N are related
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 71 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 72
Example 11: Time & Frequency Scaling Example 11: Workspace
Derive one of the transform relationships on the previous slide.
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 73 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 74
1 2
2
Derive one of the four relationships on the previous slide.
DTFS |x[n]| = |X[k]|
N
n=<N > k=<N >
1
DTFT
2
|x[n]| = X(ej ) 2 d
n=
2 <2>
1 2 2
CTFS |x(t)| dt = |X[k]|
T <T > k=
2 1 2
CTFT |x(t)| dt = |X(j)| d
2
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 75 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 76
Example 12: Workspace Property 12: Parsevals Theorem Comments
Thus the magnitude of the spectrum squared is called
The energy spectral density for energy (nonperiodic) signals
The power spectral density for power (periodic) signals
This is why the squared magnitude is often plotted versus
frequency instead of just the magnitude
This enables us to talk about what fraction of signal energy is
contained within certain frequency ranges
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 77 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 78
1.6 1.61 1.62 1.63 1.64 1.65 1.66 1.67 1.68 1.69
Time (sec)
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 79 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 80
Application Example 6: Parsevals Theorem Application Example 6: MATLAB Code
400
300
PSD
200
100
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500
100
Net Spectral Energy (%)
90
80
70
60
50
40
30
20
10
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500
Frequency (Hz)
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 81 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 82
FT FT
x(t) X(j) X(jt) 2x(t)
FS FS 1
x[n] X[k] X[n] x[k]
N
We have already seen many times there are equivalent
relationships between the time and frequency domain
FT FT
x(t to ) ejto X(j) ejo t x(t) X (j( o ))
FT FT
(t) 1 1 2()
FT FT
pT (t) 2T sinc( T
) 2W sinc( tW
) 2 pW ()
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 83 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 84
Example 13: Duality Example 13: Workspace
Derive one of the four relationships on the previous slide.
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 85 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 86
J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 87 J. McNames Portland State University ECE 223 Fourier Properties Ver. 1.11 88