4.

Lecture 4 – Frequency-Domain Analysis

Sinusoidal response. Arbitrary response. Ideal filters. Sampling.
Reconstruction. Aliasing. Amplitude modulation (DSB-SC). Demodulation.
Fourier series. Windowing. Multiplication and convolution..

Frequency-Domain Analysis
Since we can now represent signals in terms of a Fourier series (for periodic
signals) or a Fourier transform (for aperiodic signals), we seek a way to
We have a
describe a system in terms of frequency. That is, we seek a model of a linear, description of
signals in the
time-invariant system governed by continuous-time differential equations that frequency domain -
we need one for
expresses its behaviour with respect to frequency, rather than time. The systems
concept of a signal’s spectrum and a system’s frequency response will be seen
to be of fundamental importance in the frequency-domain characterisation of a
system.

The power of the frequency-domain approach will be seen as we are able to

determine a system’s output given almost any input. Fundamental signal
operations can also be explained easily – such as modulation / demodulation
and sampling / reconstruction – in the frequency domain that would otherwise
appear bewildering in the time domain.

Response to a Sinusoidal Input

We have already seen that the output of a LTI system is given by:
Starting with a

y(t ) = h(t ) ∗ x(t ) (4.1) convolution

description of a
system

if initial conditions are zero.

Suppose the input to the system is:

x(t ) = A cos(ω 0t + φ ) (4.2) We apply a sinusoid

Signal Theory (48541)

 4.2
We have already seen that this can be expressed (thanks to Euler) as:

A sinusoid is just a
(4.3)
A A
x(t ) = e jφ e jω 0t + e − jφ e − jω 0t
sum of two complex
conjugate counter-
rotating phasors 2 2
= Xe jω 0t + X *e − jω 0t
Where X is the phasor representing x(t ) . Inserting this into Eq. (4.1) gives:

∞
( )
y(t ) =  h(τ ) Xe jω0 (t −τ ) + X *e − jω0 (t −τ ) dτ
−∞
(4.4)

∞ ∞
=  h(τ )e − jω0τ Xe jω0t dτ +  h(τ )e jω0τ X *e − jω0t dτ
−∞ −∞

=  h(τ )e − jω0τ dτ  Xe jω0t +  h(τ )e jω0τ dτ  X *e − jω0t

∞ ∞

 −∞   −∞ 

This rather unwieldy expression can be simplified. First of all, if we take the
Fourier transform of the impulse response, we get:
The Fourier
∞
H (ω ) =  h(t )e − jωt dt
transform of the (4.5)
impulse response
appears in our −∞
analysis…

where obviously ω = 2πf . Now Eq. (4.4) can be written as:

y(t ) = H (ω0 ) Xe jω 0t + H (− ω0 ) X *e− jω 0t (4.6)

If h(t ) is real, then:

H (− ω ) = H * (ω ) (4.7)

which should be obvious by looking at the definition of the Fourier transform.

Now let:
…and relates the
output phasor with
the input phasor!
Y = H (ω0 ) X (4.8)

Signal Theory (48541)

 4.3
This equation is of fundamental importance! It says that the output phasor to a
system is equal to the input phasor to the system, scaled in magnitude and
changed in angle by an amount equal to H (ω 0 ) (a complex number). Also:

Y * = H * (ω0 ) X * = H (− ω0 ) X * (4.9)

Eq. (4.6) can now be written as:

y(t ) = Ye jω 0t + Y *e− jω 0t (4.10)

which is just another way of writing the sinusoid:

The magnitude and

y(t ) = A H (ω ) cos(ω t + φ + ∠H (ω
0 0 0 )) (4.11) phase of the input
sinusoid change –
according to the
Fourier transform of
the impulse
Hence the response resulting from the sinusoidal input x(t ) = A cos(ω 0 t + φ ) is response

also a sinusoid with the same frequency ω 0 , but with the amplitude scaled by

the factor H (ω 0 ) and the phase shifted by an amount ∠H (ω 0 ) .

The function H (ω ) is termed the frequency response. H (ω ) is called the Frequency,

magnitude and
magnitude response and ∠H (ω ) is called the phase response. Note that the phase response
defined
system impulse response and the frequency response form a Fourier transform
pair:
The impulse

h(t ) ⇔ H ( f ) (4.12) response and

frequency response
form a Fourier
transform pair
We now have an easy way of analysing systems with sinusoidal inputs: simply
determine H ( f ) and apply Y = H ( f 0 ) X .

There are two ways to get H ( f ) . We can find the system impulse response
Two ways to find the
h(t ) and take the Fourier transform, or we can find it directly from the frequency response
differential equation describing the system.

Signal Theory (48541)

 4.4

Example

For the simple RC circuit below, find the response to an arbitrary sinusoid,
assuming no stored energy in the system (zero initial conditions). This is
termed the steady-state response, since the input is assumed to be valid for all
time.
Finding the frequency
response of a simple
system
R

Vi C Vo

Figure 4.1

The input/output differential equation for the circuit is:

dvo (t ) 1 1 (4.13)
+ vo (t ) = vi (t )
dt RC RC
which is obtained by KVL. Since the input is a sinusoid, which is really just a
sum of conjugate complex exponentials, we know from Eq. (4.6) that the
output is Vo = H (ω 0 ) Ae j (ω0t +φ ) if the input is Vi = Ae j (ω0t +φ ) . Note that Vi and

Vo are complex numbers, and if the factor e jω0t were suppressed they would be

phasors. The differential equation (4.13) becomes:

d
dt
[ ]
H (ω 0 )Ae j (ω0t +φ ) +
1
RC
[ ]
H (ω 0 )Ae j (ω0t +φ ) =
1
RC
[
Ae j (ω0t +φ ) ] (4.14)

Signal Theory (48541)

 4.5
and thus:

1 1 (4.15)
jω 0 H (ω 0 ) Ae j (ω0t +φ ) + H (ω 0 ) Ae j (ω0t +φ ) = Ae j (ω0t +φ )
RC RC

Dividing both sides by Vi = Ae j (ω0t +φ ) gives:

(4.16)
1 1
jω 0 H (ω 0 ) + H (ω 0 ) =
RC RC
and therefore:

1 RC (4.17)
H (ω 0 ) =
jω 0 + 1 RC

which yields for an arbitrary frequency:

(4.18) Frequency response

1 RC
H (ω ) =
of a lowpass RC
circuit
jω + 1 RC

This is the frequency response for the simple RC circuit. As a check, we know
that the impulse response is:

h(t ) = 1 RC e −t RC (4.19) Impulse response of

a lowpass RC circuit

Using your standard transforms, show that the frequency response is the
Fourier transform of the impulse response.

The magnitude function is:

(4.20) Magnitude response

1 RC
H (ω ) = of a lowpass RC
circuit
ω 2 + (1 RC )2

Signal Theory (48541)

 4.6
and the phase function is:

∠H (ω ) = − tan −1 (ωRC )
Phase response of a
(4.21)
lowpass RC circuit

Plots of the magnitude and phase function are shown below:

A graph of the
frequency response –
in this case as
magnitude and phase |H (ω ) |
1

1 2

0
0 ω0 2ω 0 ω

0 ω0 2ω 0 ω
0º

-45 º

-90 º
H (ω )

Figure 4.2

The behaviour of the RC circuit is summarized by noting that it passes low

The system’s
behaviour described frequency signals without any significant attenuation and without producing
in terms of frequency
any significant phase shift. As the frequency increases, the attenuation and the
phase shift become larger. Finally as the frequency increases to ∞ , the RC

Signal Theory (48541)

 4.7
circuit completely “blocks” the sinusoidal input. As a result of this behaviour, Filter terminology
the circuit is an example of a lowpass filter. The frequency ω 0 = 1 RC is defined

termed the cutoff frequency. The bandwidth of the filter is also equal to ω 0 .

Response to an Arbitrary Input

If we can do one
It should now be obvious how we handle arbitrary inputs. sinusoid, we can do
an infinite number…
Periodic Inputs

For periodic inputs, we can express the input signal by a complex exponential
Fourier series:

∞ (4.22) which is just a

x(t ) = X
n = −∞
n e jnω o t Fourier series for a
periodic signal

It follows from the previous section that the output response resulting from the
complex exponential input X n e jnω0t is equal to H (nω 0 )X n e jnω0t . By linearity,

the response to the periodic input x(t ) is:

∞ (4.23)
y (t ) =  H (nω )X
n = −∞
0 n e jnω o t

Since the right-hand side is a complex exponential Fourier series, the output
y (t ) must be periodic, with fundamental frequency equal to that of the input,
ie. the output has the same period as the input.

It can be seen that the only thing we need to determine is new Fourier series The frequency
response simply
coefficients, given by: multiplies the input
Fourier series

Yn = H (nω 0 )X n
coefficients to
(4.24) produce the output
Fourier series
coefficients

Signal Theory (48541)

 4.8
The output magnitude spectrum is just:

Yn = H (nω 0 ) X n (4.25)
Don’t forget – the
frequency response is
just a frequency and the output phase spectrum is:
dependent complex
number

∠Yn = ∠H (nω 0 ) + ∠X n (4.26)

These relationships describe how the system “processes” the various complex
exponential components comprising the periodic input signal. In particular, Eq.
(4.25) determines if the system will pass or attenuate a given component of the
input. Eq. (4.26) determines the phase shift the system will give to a particular
component of the input.

Aperiodic Inputs

If we can do finite
sinusoids, we can do Taking the Fourier transform of both sides of the time domain input/output
infinitesimal sinusoids
too! relationship of an LTI system:
Start with the
convolution integral
again y(t ) = h(t ) ∗ x(t ) (4.27)

we get:

∞
Y( f ) =  [h(t ) ∗ x(t )]e − jωt dt
and transform to the (4.28)
frequency domain
−∞

Substituting the definition of convolution, we get:

Y ( f ) =   h(τ )x(t − τ )dτ e − jωt dt


∞ ∞ (4.29)

−∞ 
 −∞ 

This can be rewritten in the form:

Y ( f ) =  h(τ ) x(t − τ )e − jωt dt  dτ

∞ ∞ (4.30)

−∞  −∞ 

Signal Theory (48541)

 4.9
Using the change of variable λ = t − τ in the second integral gives:

Y ( f ) =  h(τ ) x(λ )e − jω (λ +τ )dλ  dτ

∞ ∞ (4.31)

−∞  −∞ 

Factoring out e − jωτ from the second integral, we can write:

Y ( f ) =   h(τ )e − jωτ dτ   x(λ )e − jωλ dλ 

∞ ∞ (4.32)

 −∞   −∞ 

which is:
Convolution in the

Y ( f ) = H ( f )X ( f ) (4.33) time-domain is
multiplication in the
frequency-domain

This is also a proof of the “convolution in time property” of Fourier

transforms.

Eq. (4.33) is the frequency-domain representation of the system given by The output spectrum
is obtained by
Eq. (4.27). It says that the spectrum of the output signal is equal to the product multiplying the input
spectrum by the
of the frequency response and the spectrum of the input signal. frequency response

The output magnitude spectrum is:

Y( f ) = H( f ) X ( f ) (4.34) The magnitude

spectrum is scaled

and the output phase spectrum is:

∠Y ( f ) = ∠H ( f ) + ∠X ( f ) (4.35) The phase spectrum

is added to

Note that the frequency domain description applies to all inputs that can be
Fourier transformed, including sinusoids if we allow impulses in the spectrum.
Periodic inputs are then a special case of Eq. (4.33).

By similar arguments together with the duality property of the Fourier Convolution in the
frequency-domain is
transform, it can be shown that convolution in the frequency-domain is multiplication in the
time-domain
equivalent to multiplication in the time-domain.

Signal Theory (48541)

 4.10

Ideal Filters

A first look at
Now that we have a feel for the frequency-domain description and behaviour
frequency-domain of a system, we will briefly examine a very important application of electronic
descriptions - filters
circuits – that of frequency selection, or filtering. Here we will examine ideal
filters – the topic of real filter design is rather involved.

Ideal filters pass sinusoids within a given frequency range, and reject
(completely attenuate) all other sinusoids. An example of an ideal lowpass
filter is shown below:

|H| Cutoff
1

ideal
Vi Vo
Pass Stop filter

0
0 ω0 ω

Figure 4.3

Filter types Other basic types of filters are highpass, bandpass and bandstop. All have
similar definitions as given in Figure 4.3. Frequencies that are passed are said
to be in the passband, while those that are rejected lie in the stopband. The
point where passband and stopband meet is called ω 0 , the cutoff frequency.

The term bandwidth as applied to a filter corresponds to the width of the

passband.

An ideal lowpass filter with a bandwidth of B Hz has a magnitude response:

 f  (4.36)
H ( f ) = Krect 
 2B 

Signal Theory (48541)

 4.11

Sampling
Sampling is one of the most important operations we can perform on a signal.
Sampling is one of
Samples can be quantized and then operated upon digitally (digital signal the most important
processing). Once processed, the samples are turned back into a continuous- things we can do to
a continuous-time
time waveform. (eg. CD, mobile phone!) Here we demonstrate how, if certain signal – because we
can then process it
parameters are right, a sampled signal can be reconstructed from its samples digitally
almost perfectly.

Ideal sampling involves multiplying a waveform by a train of impulses. The

weights of the impulses are the sample values to be used by a digital signal
processor (computer). An ideal sampler is shown below:

An ideal sampler
multiplies a signal
g (t ) g (t ) p (t ) = gs ( t ) by a uniform train of
impulses

p (t )

-2Ts -Ts 0 Ts 2Ts

Figure 4.4

∞
Let g( t ) be a time domain signal. If we multiply it by δ (t − kT )
k =−∞
s we get an

ideally sampled version:

∞ (4.49)
g s (t ) = g(t )⋅ δ ( t − kTs )
k =−∞

Signal Theory (48541)

 4.12
In the time domain, we have:

An ideal sampler
produces a train of
impulses - each g (t ) g s (t )
impulse is weighted B B
sampled signal
by the original signal g (t )
-2 /B -1 /B 0 1/B 2/B t -2 /B -1 /B 0 1/B 2/B t

p (t )

-2Ts - Ts 0 Ts 2Ts t

Figure 4.5

Taking the Fourier transform of both sides of Eq. (4.49):

∞ (4.50)
Gs ( f ) = G ( f ) ∗ f s  δ ( f − nf )
n = −∞
s

∞
= fs  G( f − nf )
n = −∞
s

Graphically, in the frequency domain:

A sampled signal’s
spectrum is a scaled
replica of the Gs ( f )
original, periodically G (f )
fs
repeated 1
- fs -B - fs - fs+B -B 0 B fs -B fs fs+B f
X (f )
-2B -B 0 B 2B f

P (f )

fs

-2 fs - fs 0 fs 2 fs f

Figure 4.6

Thus the Fourier transform of the sampled waveform is a scaled replica of the
original, periodically repeated along the frequency axis. Spacing between

Signal Theory (48541)

 4.13
repeats is equal to the sampling frequency (the inverse of the sampling
interval).

Similarly, if we sample a continuous Fourier transform G( f ) by multiplying it

∞
with δ ( f − nf ) and take the inverse Fourier transform we get:
n =−∞
s

∞ (4.51)
F [Gs ( f )] = g (t ) ∗ Ts
−1
δ (t − kT ) s
k = −∞

which is a periodic repeat of the inverse transform of the original signal.

Thus, sampling in the frequency domain results in periodicy in the time Sampling in one
domain implies
domain. We already know this! We know a periodic time domain signal can be periodicy in the
synthesised from sinusoids with frequencies nf 0 , ie. has a transform consisting other
of impulses at frequencies nf 0 .

We now see the general pattern: Sampling in one domain implies periodicy in
the other.

Reconstruction
If a sampled signal g s (t ) is applied to an ideal lowpass filter of bandwidth B,

the only component of the spectrum Gs ( f ) that is passed is just the original

spectrum G ( f ) .

We recover the
original spectrum by
lowpass filtering
gs (t ) g( t )
lowpass
filter

Figure 4.7

Signal Theory (48541)

 4.14

Hence the output of the filter is equal to g (t ) , which shows that the original
signal can be completely and exactly reconstructed from the sampled
waveform g s (t ) .

In the time domain:

A weighted train of
impulses turns back
into the original
signal after lowpass B g s (t ) g( t ) B
lowpass
filtering…
filter
-2 /B -1 /B 0 1 /B 2 /B t -2 /B -1 /B 0 1 /B 2 /B t

Figure 4.8

and in the frequency domain:

but it’s a much
clearer operation in
the frequency Gs ( f )
fs
domain!
G (f )
- fs -B - fs - fs+B -B 0 B fs -B fs fs+B f 1

-2B -B 0 B 2B f
lowpass
filter

H( f ) 1/ fs

-2 B -B 0 B 2B f

Figure 4.9

There are some limitations to perfect reconstruction though. One is that time-
We can’t sample
and reconstruct limited signals are not bandlimited (eg. rect function). Any time-limited signal
perfectly, but we can
get close! therefore cannot be perfectly reconstructed, since there is no sample rate high
enough to ensure repeats of the original spectrum do not overlap. However,
many signals are essentially bandlimited, which means spectral components
higher than, say B, do not make a significant contribution to either the shape or
energy of the signal.

Signal Theory (48541)

 4.15

Aliasing
We saw that sampling in one domain implies periodicy in the other. If the
We have to ensure
function being made periodic has an extent that is smaller than the period, there no spectral overlap
when sampling
will be no resulting overlap and hence it will be possible to recover the
continuous (unsampled) function by windowing out just one period from the
domain displaying periodicy.

Nyquist’s criterion is the formal expression of the above fact. It states:

Perfect reconstruction of a sampled signal is possible if the (4.52) Nyquists’s criterion

sampling rate is greater than twice the bandwidth of the
signal being sampled

f s > 2B

To avoid aliasing, we have to sample at a rate f s > 2 B . The frequency f s 2 is

Foldover frequency
called the spectral foldover frequency, and it is determined only by the selected defined

sample rate, and it may be selected independently of the characteristics of the

signal being sampled. The frequency B is termed the Nyquist frequency, and it Nyquist frequency
defined
is a function only of the signal and is independent of the selected sampling
rate. Do not confuse these two independent entities! The Nyquist frequency is a
lower bound for the foldover frequency in the sense that failure to select a
foldover frequency at or above the Nyquist frequency will result in spectral
aliasing and loss of the capability to reconstruct a continuous-time signal from
its samples without error. The Nyquist frequency for a signal which is not
bandlimited is infinity; that is, there is no finite sample rate that would permit
errorless reconstruction of the continuous-time signal from its samples.

Signal Theory (48541)

 4.16
To illustrate aliasing, consider the case where we have not selected the sample
rate higher than twice the bandwidth of a lowpass signal:
An illustration of
aliasing in the
frequency-domain
folded-over high-frequency
components
X s( f )

- fs - B 0 B fs f

Figure 4.10

If the sampled signal xs (t ) is lowpass filtered with cutoff frequency B, the

output spectrum of the filter will contain high-frequency components of x(t )

folded-over to low-frequency components.

How to avoid
To summarise – we can avoid aliasing by either:
aliasing

1. Selecting a sample rate higher than twice the bandwidth of the signal
(equivalent to saying that the foldover frequency is greater than the
bandwidth of the signal); or

2. By bandlimiting (using a filter) the signal so that its bandwidth is less than
half the sample rate.

Signal Theory (48541)

 4.17

Summary of the Sampling and Reconstruction Process

The sampling and

Time-Domain Frequency-Domain
reconstruction
g (t ) Sampling G( f ) process in both the
A time-domain and
C frequency-domain

0 t -B 0 B f

s(t ) S(f )
1 fs

0 Ts t -f s 0 fs f

g s (t ) Gs ( f )
Afs
C

0 t -f s -B 0 B fs f

Reconstruction

h (t ) H( f ) 1
1 fs

fs

0 Ts t - f s /2 0 f s /2 f

g r (t ) Gr ( f )
A
C

0 t -B 0 B f

Figure 4.11

Signal Theory (48541)

 4.18

Modulation
Another practical and very important application of Fourier analysis is when
we consider an operation called modulation.

Let x(t ) be a signal such as an audio signal that is to be transmitted through a

cable or the atmosphere. In amplitude modulation (AM), the signal modifies
(or modulates) the amplitude of a carrier sinusoid cos(ω c t ) . In one form of

AM transmission, the signal x(t ) and the carrier cos(ω c t ) are simply

multiplied together. The process is illustrated below:

Double side-band –
suppressed carrier
(DSB-SC) Signal multiplier
modulation
x( t ) y ( t ) = x ( t ) cos(2π f c t ) = modulated signal

cos(2π f c t )

Local
oscillator

Figure 4.12

The local oscillator in Figure 4.12 is a device that produces the sinusoidal
signal cos(ω c t ) . The multiplier is implemented with a non-linear device, and is

usually an integrated circuit at low frequencies.

By the multiplication property of Fourier transforms, the output spectrum is

obtained by convolving the spectrum of x(t ) with the spectrum of cos(2πf c t ) .

We now restate a very important property of convolution involving an impulse:

X ( f ) ∗ δ ( f − f0 ) = X ( f − f0 ) (4.53)

Signal Theory (48541)

 4.19
The output spectrum of the modulator is therefore:
DSB-SC up-
translates the
1
Y ( f ) = X ( f )∗ [δ ( f − fc ) + δ ( f + fc )] baseband spectrum
2
1
= [X ( f − fc ) + X ( f + fc )] (4.54)
2

The spectrum of the modulated signal is a replica of the signal spectrum but
“shifted up” in frequency. If the signal has a bandwidth equal to B then the
modulated signal spectrum has an upper sideband from f c to f c + B and a

lower sideband from f c − B to f c , and the process is therefore called double-

sideband transmission, or DSB transmission for short. An example of

modulation is given below in the time-domain:
DSB-SC in the time-
domain
x( t ) Signal multiplier y( t )
1 1
x( t ) y ( t ) = x ( t ) cos(2π f ct )

-2 -1 0 1 2 t modulated signal -2 -1 1 2 t
0
cos(2π f c t )

Local
oscillator

-2 -1 0 1 2 t

Figure 4.13

Signal Theory (48541)

 4.20
And in the frequency domain:
DSB-SC in the
frequency-domain
X (f )
1
X (f ) Y ( f ) = 1 [ X ( f − f c ) + X ( f − f c )]
2
-2 -1 0 1 2 f Y(f )
1/2
1
[δ ( f − fc ) + δ ( f + fc )] Local
2 - fc 0 fc f
oscillator
1/2

- fc 0 fc f

Figure 4.14

Modulation lets us The higher frequency range of the modulated signal makes it possible to
share the spectrum,
and achieves achieve good propagation in transmission through a cable or the atmosphere. It
practical also allows the “spectrum” to be shared by independent users – eg. radio, TV,
propagation
mobile phone etc.

Signal Theory (48541)

 4.21

Demodulation
The reconstruction of x(t ) from x(t )cos(ω c t ) is called demodulation. There are

many ways to demodulate a signal, here we will consider one common method
called synchronous or coherent demodulation.

Coherent
demodulation of a
x ( t ) cos(2π f c t ) x ( t ) cos2(2π f c t ) lowpass
x (t ) DSB-SC signal

filter

cos(2π f c t )

Local
oscillator

Figure 4.15

The first stage of the demodulation process involves applying the modulated Coherent
waveform x(t )cos(ω c t ) to a multiplier. The other signal applied to the demodulation
requires a carrier at
the receiver that is
multiplier is a local oscillator which is assumed to be synchronized with the synchronized with
carrier signal cos(ω c t ) , ie. there is no phase shift between the carrier and the the transmitter

signal generated by the local oscillator.

The output of the multiplier is:

1
[X ( f − f c ) + X ( f + f c )]∗ 1 [δ ( f − f c ) + δ ( f + f c )]
2 2
(4.55)
1 1
= X ( f ) + [ X ( f − 2 f c ) + X ( f + 2 f c )]
2 4

x(t ) can be “extracted” from the output of the multiplier by lowpass filtering
with a cutoff frequency of B Hz and a gain of 2.

Signal Theory (48541)

 4.22
Another way to think of this is in the time-domain:

1
x(t )cos 2 (2πf c t ) = x(t ) (1 + cos(4πf c t ))
2 (4.56)

Therefore, it is easy to see that lowpass filtering, with a gain of 2, will produce
x(t ) . An example of demodulation in the time-domain is given below:
Demodulation in the
time-domain
x ( t ) cos (2π f c t )
2

1
x ( t ) cos(2 π f c t )
-2 -1 0 1 2 t x( t )
1 1
x (t )
lowpass
-2 -1 0 1 2 t filter -2 -1 0 1 2 t

cos(2π f c t )

Local
oscillator

-2 -1 0 1 2 t

Figure 4.16

The operation of demodulation is best understood in the frequency domain:

Demodulation in the
frequency-domain
Y(f )=
1 1/2
[X ( f − f c ) + X ( f − f c )] 1/4 1/4
2
Y (f )
-2 f c 0 2 fc f
1/2
X (f )
x (t ) 1
lowpass
- fc 0 fc f
filter
-2 -1 0 1 2 f
1
[δ ( f − f c ) + δ ( f + f c )] H( f ) 2
2
1/2 Local
oscillator
-2 -1 0 1 2 f
- fc 0 fc f

Figure 4.17

Signal Theory (48541)

 4.23

Summary of DBS-SC Modulation and Demodulation

The DSB-SC
Time-Domain Frequency-Domain
modulation and
g (t ) Modulation G( f ) demodulation
C A process in both the
time-domain and
frequency-domain
0 t -B 0 B f

l (t ) L( f )
1 1/2

0
Tc t
-f c 0 fc f

gm(t ) Gm( f )
C A /2

t
0 -f c -B -f c +B f c -B f c +B f
-f c 0 fc

Demodulation
L( f )
1/2
l (t )
1

0 -f c 0 fc
Tc t f

Gm( f )
g i( t ) A /2
C
A /4 A /4

0 t -2 f c -f c -B B fc 2 fc f
0

H( f )
h (t )
2 2

0 T0 t - f0 0 f0 f

gd( t ) Gd( f )
C A

t -B 0 B f

Figure 4.18

Signal Theory (48541)

 4.24

Finding the Fourier Series of a Periodic Function From the

Fourier Transform of a Single Period

The quick way to It is usually easier to find the Fourier transform of a single period than
determine Fourier
performing the integration needed to find Fourier series coefficients (because
series coefficients
all the standard Fourier properties can be used). This method allows the
Fourier series coefficients to be determined directly from the Fourier
transform, provided the period is known. Don’t forget, only periodic functions
have Fourier series representation.

Suppose we draw one period of a periodic waveform:

A single period

g1(t)
1

0 T1 t

Figure 4.19

We can create the periodic version by convolving g 1 ( t ) with a train of unit

impulse functions with spacing equal to the period, T0 :

Convolved with a
uniform impulse
train
1

-2T0 -T0 0 T0 2T0 t

Figure 4.20

Signal Theory (48541)

 4.25
∞
That is, we need to convolve g 1 ( t ) with δ (t − kT ) .
k =−∞
0 Thus, g p ( t ) , the

periodic version is:

∞ (4.57)
g p ( t ) = g1 ( t )∗ δ (t − kT0 )
k =−∞

gives the periodic

waveform
gp( t )
1

-2T0 -T0 0 T1 T0 2T0 t

Figure 4.21

Using the convolution multiplication rule:

 ∞ 
[ ]
F g p (t ) = F [g1 (t )]⋅ F   δ (t − kT0 )
k =−∞  (4.58)
∞
= G1 ( f ) ⋅ f 0  δ ( f − nf )
n = −∞
0

In words, the Fourier transform of the periodic signal consists of impulses

located at harmonics of f 0 =1 T0 , whose weights are:

Fourier series
Gn = f 0 G1 ( nf 0 )
(4.59) coefficients from the
Fourier transform of
one period

These are the Fourier series coefficients.

In Figure 4.21 we have:

Gn = f 0 T1 sinc(nf 0 T1 )e − jπnf 0T1

(4.60)

Signal Theory (48541)

 4.26
Graphically, the operation indicated by Eq. (4.58) takes the original spectrum
and multiplies it by a train of impulses - effectively creating a weighted train of
impulses:
Sampling in the
frequency domain
produces a periodic G1( f )
waveform in the T1
time-domain

0
f1 2f1 f

f0

0 f 0 2f 0 f

Gp( f )
f 0 T1

-8 f0 -6 f0 6 f0 8 f0
-12 f0 -10 f0 -4 f0 -2 f0 0 2 f0 4 f0 10 f0 12 f0 f

Figure 4.22

According to Eq. (4.59), the Fourier series coefficients are just the weights of
the impulses in the spectrum of the periodic function. To get the nth Fourier
series coefficient, use the weight of the impulse located at nf 0 .

Remember that This is in perfect agreement with the concept of a continuous spectrum. Each
pairs of impulses in
a spectrum frequency has an infinitesimal amplitude sinusoid associated with it. If an
represent a sinusoid
in the time-domain impulse exists at a certain frequency, then there is a finite amplitude sinusoid at
that frequency.

Signal Theory (48541)

 4.27

Windowing in the Time Domain

Often we wish to deal with only a segment of a signal, say from t =0 to t = T .
Sometimes we have no choice, as this is the only part of the signal we have Some practical
effects of looking at
access to - our measuring instrument has restricted us to a “window” of signals over a finite
time
duration T beginning at t =0 . Outside this window the signal is forced to be
zero. How is the signal’s Fourier transform affected when the signal is viewed
through a window?

Windowing in the time domain is equivalent to multiplying the original signal Windowing defined
g( t ) by a function which is non-zero over the window interval and zero
elsewhere. So the Fourier transform of the windowed signal is the original
signal convolved with the Fourier transform of the window.

Example

Find the Fourier transform of sin( 2πt ) when it is viewed through a rectangular
window from 0 to 1 second:

A rectangular
window applied to a
sinusoid
1

0 1 t

Figure 4.23

The viewed signal is:

g w ( t ) = sin( 2πt ) rect( t − 0.5) (4.61)

Signal Theory (48541)

 4.28
The Fourier transform will be:

F[sin (2πt )]∗ F[rect(t − 0.5)]

− j j 
=  δ ( f − 1) + δ ( f + 1) ∗ sinc( f )e − jπf (4.62)
 2 2 
j j
= − sinc( f − 1)e − jπ ( f −1) + sinc( f + 1)e − jπ ( f +1)
2 2

Graphically, the magnitude spectrum of the windowed signal is:

Spectrum of a
rectangular
windowed sinusoid
1

-4 -3 -2 -1 0 1 2 3 4 f

1/2 90° 1/2 -90°

-1 0 1 f

G( f )
0.5

-4 -3 -2 -1 0 1 2 3 4 f

Figure 4.24

Signal Theory (48541)

 4.29
If the window were changed to 4 seconds, we would then have:
The longer we look,
the better the
spectrum
4

-4 -3 -2 -1 0 1 2 3 4 f

1/2 90° 1/2 -90°

-1 0 1 f

G( f )
2

-4 -3 -2 -1 0 1 2 3 4 f

Figure 4.25

Obviously, the longer the window, the more accurate the spectrum becomes.

Signal Theory (48541)

 4.30

Example

Find the Fourier transform of sinc( t ) when it is viewed through a window from
-2 to 2 seconds:

Viewing a sinc
function through a
rectangular window g (t )
w
1

-2 -1 0 1 2 t

Figure 4.26

We have:

g w (t ) =sinc(t )rect( t 4)
(4.63)

and:

F[g w (t )] = rect( f ) ∗ 4 sinc(4 f ) (4.64)

Signal Theory (48541)

 4.31
Graphically:

Ripples in the
spectrum caused by
a rectangular
1 window

-1 -0.5 0 0.5 1 f

-1 -0.5 0 0.5 1 f

-1 -0.5 0 0.5 1 f

Figure 4.27

We see that windowing in the time domain by T produces ripples in the

frequency domain with an approximate spacing of 2 T between peaks. (In a
magnitude spectrum, there would be a peak every 1 T ).

Signal Theory (48541)

4.32

Appendix: How to Use MATLAB® to Check Fourier Series

Coefficients
MATLAB® is a software package that is particularly suited to signal
processing. It has instructions that will work on vectors and matrices. A vector
can be set up which gives the samples of a signal. Provided the sample spacing
meets the Nyquist criterion the instruction G=fft(g) returns a vector
containing N times the Fourier series coefficients, where G(1) = N ⋅G0 is the DC
term, G(2) = N ⋅G1 , G(3) = N ⋅G2 etc. and G( N) = N ⋅G−1 , G( N-1) = N ⋅G− 2 etc.
where N is the size of the vector. G=ifft(g)does the inverse Fourier
transform.

Example

Suppose we want to find the Fourier series coefficients of g( t ) =cos( 2πt ) . Note
period = 1 s.

Step 1

Choose sample frequency - since highest frequency present is 1 Hz, choose

4 Hz (minimum is > 2 Hz).

Step 2

Take samples over one period starting at t =0 . Note N =4 .

g= 1[ 0 −1 0]
Step 3

Find G =fft(g) = 0 [ 2 0 ]
2 .

Hence G0 = 0 , G1 = 2 4 =1 2 , G−1 = 2 4 =1 2 . G±2 should be zero if the Nyquist

criterion is met. These are in fact the Fourier series coefficients of g( t ) .

Signal Theory (48541)

 4.33

Example

Find the Fourier series coefficients of a 50% duty cycle square wave.

Step 1

In this case the spectrum is ∞ so we can never choose f s high enough. There
is always some error. Suppose we choose 8 points of one cycle.

Step 2

[
g= 1 1 0.5 0 0 0 0.5 1]
Note: if samples occur on transitions, input half way point.

Step 3

G =fft(g) = 4 [ 2.4142 0 − 0.4142 0 0.4142 0 2.4142 ]

Therefore G1 = 2.4142 8 = 0.3015 . The true value is 0.3183. Using 16 points,

G1 = 0.3142 .

You should read Appendix A – The Fast Fourier Transform, and look at the
example MATLAB® code in the “FFT - Quick Reference Guide” for more
complicated and useful examples of setting up and using the FFT.

Signal Theory (48541)

4.34

Summary
• Sinusoidal inputs to linear time-invariant systems yield sinusoidal outputs.
The output sinusoid is related to the input sinusoid by a complex-valued
function known as the frequency response, H ( f ) .

• The frequency response of a system is just the Fourier transform of the

impulse response of the system. That is, the impulse response and the
frequency response form a Fourier transform pair: h(t ) ⇔ H ( f ) .

• The frequency response of a system can be obtained directly by performing

analysis in the frequency domain.

• The output of an LTI system due to any input signal is obtained most easily
by considering the spectrum: Y ( f ) = H ( f )X ( f ) . This expresses the
important property: convolution in the time-domain is equivalent to
multiplication in the frequency-domain.

• Filters are devices best thought about in the frequency-domain. They are
frequency selective devices, changing both the magnitude and phase of
frequency components of an input signal to produce an output signal.

• Linear phase is desirable in filters because it produces a constant delay,

thereby preserving waveshape.

• Sampling is the process of converting a continuous-time signal into a

discrete-time signal. It is achieved, in the ideal case, by multiplying the
signal by a train of impulses.

• Reconstruction is the process of converting signal samples back into a

continuous-time signal. It is achieved by passing the samples through a
lowpass filter.

• Aliasing is an effect of sampling where spectral overlap occurs, thus

destroying the ability to later reconstruct the signal. It is caused by not
meeting the Nyquist criterion: f s > 2 B .

Signal Theory (48541)

 4.35

• Modulation shifts a baseband spectrum to a higher frequency range. It is

achieved in many ways – the simplest being the multiplication of the signal
by a carrier sinusoid.

• Demodulation is the process of returning a modulated signal to the

baseband. Modulation and demodulation form the basis of modern
communication systems.

• Fourier series coefficients can be obtained from the Fourier transform of

one period of the signal by the formula: Gn = f 0 G1 (nf 0 ) .

• Using finite-duration signals in the time-domain is called windowing.

Windowing affects the spectrum of the original signal.

References
Haykin, S.: Communication Systems, John-Wiley & Sons, Inc., New York,
1994.

Kamen, E. & Heck, B.: Fundamentals of Signals and Systems using

MATLAB®, Prentice-Hall, 1997.

Lathi, B. P.: Modern Digital and Analog Communication Systems, Holt-

Saunders, Tokyo, 1983.

Signal Theory (48541)

4.36

Quick Check
Encircle the correct answer, cross out the wrong answers. [one or none correct]
1.
The convolution of x( t ) and y( t ) is given by:

∞ ∞ ∞
(a)  x(τ ) y( t − τ ) dτ (b)  x ( τ ) y ( τ − t ) dτ (c)  X ( λ )Y ( f − λ )dλ
−∞ −∞ −∞

2.
Vi Vo The Fourier transform
Vi Ideal Vo of the impulse response
0 0 t
t Filter
of the filter resembles:

|H( f )| h (t) |H( f )|

(a) f
(b) t (c) f

3.
The Fourier transform of one period of a periodic waveform is G(f). The
Fourier series coefficients, Gn , are given by:

(a) nf 0 G( f 0 ) (b) G(nf 0 ) (c) f 0G(nf 0 )

4.
x(t) y(t) The peak value of the
convolution is:
t t

(a) 9 (b) 4.5 (c) 6

5.
The scaling property of the Fourier transform is:

f (b) g( at ) ⇔
1
G( f ) (c) ag( t ) ⇔ aG( f )
(a) g( at ) ⇔ G 
 a a

Answers: 1. a 2. c 3. c 4. x 5. x

Signal Theory (48541)

 4.37

Exercises

1.
Write an expression for the time domain representation of the voltage signal
with double sided spectrum given below:

X( f )
2 30° 2 -30°

-1000 1000 f

What is the power of the signal?

2.
Use the integration and time shift rules to express the Fourier transform of the
pulse below as a sum of exponentials.

x(t)
2

-1 0 2 4 t

3.
Sketch the convolution of the two functions shown below.

x(t) y(t)
2
1

-1 0 4 t 0 0.5 t

Signal Theory (48541)

4.38

4.
Calculate the magnitude and phase of the 4 kHz component in the spectrum of
the periodic pulse train shown below. The pulse repetition rate is 1 kHz.

x(t)
10

0.4 0.6
0 0.1 0.5 0.9 1 t (ms)

-5

5.
By relating the triangular pulse shown below to the convolution of a pair of
identical rectangular pulses, deduce the Fourier transform of the triangular
pulse:

x(t)
A

−τ 0 τ t

6.
The pulse x( t ) = 2 Bsinc( 2 Bt ) rect( Bt 8) has ripples in the amplitude spectrum.

What is the spacing in frequency between positive peaks of the ripples?

7.
A signal is to be sampled with an ideal sampler operating at 8000 samples per
second. Assuming an ideal low pass antialiasing filter, how can the sampled
signal be reconstituted in its original form, and under what conditions?

8.
A train of impulses in one domain implies what in the other?

Signal Theory (48541)

 4.39

9.
The following table gives information about the Fourier series of a periodic
waveform, g( t ) , which has a period of 50 ms.
Table 1
Harmonic # Amplitude Phase (º)
0 1
1 3 -30
2 1 -30
3 1 -60
4 0.5 -90

(a) Give the frequencies of the fundamental and the 2nd harmonic. What is
the signal power, assuming that g( t ) is measured across 50 Ω ?

(b) Express the third harmonic as a pair of counter rotating phasors. What is
the value of the third harmonic at t = 20 ms ?

(c) The periodic waveform is passed through a filter with transfer function
H ( f ) as shown below.

|H( f )|
1

0.5

-100 -50 50 100 f (Hz)

H( f )
π

-100 -50 50 100 f (Hz)

−π

Draw up a table in the same form as Table 1 of the Fourier series of the
output waveform. Is there a DC component in the output of the
amplifier?

Signal Theory (48541)

4.40

10.
A signal, bandlimited to 1 kHz, is sampled by multiplying it with a rectangular
pulse train with repetition rate 4 kHz and pulse width 50 μs. Can the original
signal be recovered without distortion, and if so, how?

11.

Sketch the Fourier transform of the waveform g( t ) = (1 + cos( 2πt ) ) sin( 20πt ) .

12.
A scheme used in stereophonic FM broadcasting is shown below:

A
C
L Σ

R Σ
B
cos(2π f c t )

The input to the left channel (L) is a 1 kHz sinusoid, the input to the right
channel (R) is a 2 kHz sinusoid. Draw the spectrum of the signal at points A, B
and C if f c = 38 kHz .

13.
Draw a block diagram of a scheme that could be used to recover the left (L)
and right (R) signals of the system shown in Question 12 if it uses the signal at
C as the input.

14.
A signal is to be analysed to identify the relative amplitudes of components
which are known to exist at 9 kHz and 9.25 kHz. To do the analysis a digital
storage oscilloscope takes a record of length 2 ms and then computes the
Fourier series. The 18th harmonic thus computed can be non-zero even when
no 9 kHz component is present in the input signal. Explain.

Signal Theory (48541)

 4.41

15.
Use MATLAB® to determine the output of a simple RC lowpass filter
subjected to a square wave input given by:

∞
x(t ) =  rect(t − 2n)
n = −∞

for the cases: 1 RC = 1 , 1 RC = 10 , 1 RC = 100 . Plot the time domain from

t = −3 to 3 and Fourier series coefficients up to the 50th harmonic for each
case.

Signal Theory (48541)

 5A.1

Lecture 5A – Frequency Response

Frequency response function. Bode plots. Approximate Bode plots. Transfer
function synthesis. Digital filters.

Overview
An examination of a filter’s frequency response is useful in several respects. It
can help us determine things such as the bandwidth of the filter, how steep the
magnitude roll off is as well as its phase response. These parameters allow
engineers to determine the suitability of the filter in particular scenarios.

Despite all this, remember that the time- and frequency-domain are
inextricably related – we can’t alter the characteristics of one without affecting
the other. This will be demonstrated for a second-order system later.

Frequency Response Function

Recall that for a filter characterized by H (s ) , and for a sinusoidal input

x(t ) = A cos(ω 0 t ) , the steady-state response is:

yss (t ) = A H (ω 0 ) cos(ω 0 t + ∠H (ω 0 )) (5A.1)

where H (ω ) is the frequency response function, obtained by setting s = jω in

H (s ) . Thus, the system behaviour for sinusoidal inputs is completely specified

by the magnitude response H (ω ) and the phase response ∠H (ω ) .

The definition above is precisely how we determine the frequency response

experimentally – we input a sinusoid and, in the steady-state, measure the
magnitude and phase change at the output.

Signal Theory (48541)

 5A.2

Determining the Frequency Response from a Transfer Function

We can get the frequency response of a system by manipulating its transfer
function. Consider a simple first-order transfer function:

K
H (s ) =
s+ p (5A.2)

The sinusoidal steady state corresponds to s = jω . Therefore, Eq. (5A.2) is, for
the sinusoidal steady state:

K (5A.3)
H (ω ) =
jω + p

The complex function H (ω ) can also be written using a complex exponential

in terms of magnitude and phase:

H (ω ) = H (ω ) e j∠H (ω ) (5A.4a)

which is normally written in polar coordinates:

The transfer function

H (ω ) = H (ω ) ∠H (ω )
in terms of (5A.4b)
magnitude and
phase

We plot the magnitude and phase of H (ω ) as a function of ω or f . We use

both linear and logarithmic scales.

If the logarithm (base 10) of the magnitude is multiplied by 20, then we have
the gain of the transfer function in decibels (dB):

The magnitude of
the transfer function
in dB
H (ω ) dB = 20 log H (ω ) dB (5A.5)

A negative gain in decibels is referred to as attenuation. For example, -3 dB

gain is the same as 3 dB attenuation.

Signal Theory (48541)

 5A.3
The phase function is usually plotted in degrees.

For example, in Eq. (5A.2), let K = p = ω 0 so that:

1 (5A.6)
H (ω ) =
1 + jω ω 0

The magnitude function is found directly as:

1 (5A.7)
H (ω ) =
1 + (ω ω 0 )
2

and the phase is:

ω 
(5A.8)
∠H (ω ) = − tan   −1

 ω0 

Magnitude Responses
A magnitude response is the magnitude of the transfer function for a sinusoidal
steady-state input, plotted against the frequency of the input. Magnitude The magnitude
response is the
responses can be classified according to their particular properties. To look at magnitude of the
these properties, we will use linear magnitude versus linear frequency plots. transfer function in
the sinusoidal
For the simple first-order RC circuit that you are so familiar with, the steady state
magnitude function given by Eq. (5A.7) has three frequencies of special
interest corresponding to these values of H (ω ) :

H (0 ) = 1
1
H (ω 0 ) = ≈ 0.707
2
H (∞ ) → 0 (5A.9)

Signal Theory (48541)

 5A.4
The frequency ω 0 is known as the half-power frequency. The plot below

shows the complete magnitude response of H (ω ) as a function of ω , and the

circuit that produces it:

A simple lowpass
filter
|H|
1
R
1 2

Vi C Vo

0
0 ω0 2ω 0 ω

Figure 5A.1

An idealisation of the response in Figure 5A.1, known as a brick wall, and the
circuit that produces it are shown below:

An ideal lowpass
filter
|T| Cutoff
1

ideal
Vi Vo
Pass Stop filter
0
0 ω0 ω

Figure 5A.2

For the ideal filter, the output voltage remains fixed in amplitude until a critical
frequency is reached, called the cutoff frequency, ω 0 . At that frequency, and
for all higher frequencies, the output is zero. The range of frequencies with
Pass and stop output is called the passband; the range with no output is called the stopband.
bands defined
The obvious classification of the filter is a lowpass filter.

Signal Theory (48541)

 5A.5
Even though the response shown in the plot of Figure 5A.1 differs from the
ideal, it is still known as a lowpass filter, and, by convention, the half-power
frequency is taken as the cutoff frequency.

If the positions of the resistor and capacitor in the circuit of Figure 5A.1 are
interchanged, then the resulting circuit is:

Vi R Vo

Figure 5A.3

Show that the transfer function is:

s
H (s ) =
s + 1 RC (5A.10)

Letting 1 RC = ω 0 again, and with s = jω , we obtain:

jω ω 0
H (ω ) =
1 + jω ω 0 (5A.11)

The magnitude function of this equation, at the three frequencies given in Eq.
(5A.9), is:

H (0 ) = 0
1
H (ω 0 ) = ≈ 0.707
2
H (∞ ) → 1 (5A.12)

Signal Theory (48541)

 5A.6

The plot below shows the complete magnitude response of H (ω ) as a function

of ω , and the circuit that produces it:

A simple highpass
filter
|H|
1
C

1 2

Vi R Vo

0
0 ω0 2ω 0 3ω 0 ω

Figure 5A.4

This filter is classified as a highpass filter. The ideal brick wall highpass filter
is shown below:

An ideal highpass
filter
|T| Cutoff
1

ideal
Vi Vo
Stop Pass filter
0
0 ω0 ω

Figure 5A.5

The cutoff frequency is ω 0 , as it was for the lowpass filter.

Signal Theory (48541)

 5A.7

Phase Responses
Like magnitude responses, phase responses are only meaningful when we look Phase response is
at sinusoidal steady-state signals. A transfer function for a sinusoidal input is: obtained in the
sinusoidal steady
state

Y Y ∠θ
H (ω ) = = = H ∠θ
X X ∠0 (5A.13)

where the input is taken as the phase reference (zero phase).

For the bilinear transfer function:

jω + z
H (ω ) = K
jω + p (5A.14)

the phase is:

ω ω The phase of the

θ = ∠K + tan   − tan −1  
−1 bilinear transfer
 z  p (5A.15) function

We use the sign of this phase angle to classify systems. Those giving positive Lead and lag circuits
θ are known as lead systems, those giving negative θ as lag systems. defined

For the simple RC circuit of Figure 5A.5, for which H (ω ) is given by

Eq. (5A.6), we have:

ω
θ = − tan −1  
ω0 (5A.16)

Since θ is negative for all ω , the circuit is a lag circuit. When ω = ω 0 ,

θ = − tan −1 (1) = −45° .

Signal Theory (48541)

 5A.8
A complete plot of the phase response is shown below:

Lagging phase
response for a
simple lowpass filter 0 ω0 2ω 0 ω
0º

θ -45 º

-90 º

Figure 5A.6

For the circuit in Figure 5A.5, show that the phase is given by:

ω
θ = 90°− tan −1  
ω0 (5A.17)

The phase response has the same shape as Figure 5A.6 but is shifted upward
by 90° :
Leading phase
response for a
simple highpass
filter 90 º

θ 45 º

0º
0 ω0 2ω 0 ω

Figure 5A.7

The angle θ is positive for all ω , and so the circuit is a lead circuit.

Signal Theory (48541)

 5A.9

Frequency Response of a Lowpass Second-Order System

Starting from the usual definition of a lowpass second-order system transfer
function, we get the following frequency response function:

ω n2
H (ω ) =
ω n2 − ω 2 + j 2ζω nω (5A.18)

The magnitude is:

ω n2 The magnitude
H ( jω ) = response of a

(ω ) + (2ζω ω )
lowpass second
2
−ω 2
2 2 (5A.19) order transfer
n n function

and the phase is:

The phase response

 2ζω nω  of a lowpass second
θ = − tan −1  
2 
order transfer

 ωn −ω 
2 (5A.20) function

Signal Theory (48541)

 5A.10
The magnitude and phase functions are plotted below for ζ = 0.4 :

Typical magnitude
and phase ω r = ω n 1-2ζ 2
≈ωn
responses of a
lowpass second |H| 1
order transfer 2ζ
function 1

-40 dB / decade

0
0 ωn ω

0
All ζ
θ , degrees

-90
-180º asymptote
for all ζ

-180
0 1 ω

Figure 5A.8

For the magnitude function, from Eq. (5A.19) we see that:

H (0) = 1, H (ω n ) = 1 2ζ , H (∞ ) → 0 (5A.21)

and for large ω , the magnitude decreases at a rate of -40 dB per decade, which
is sometimes described as two-pole rolloff.

For the phase function, we see that:

θ (0) = 0°, θ (ω n ) = −90°, θ (∞ ) → −180° (5A.22)

Signal Theory (48541)

 5A.11

Visualization of the Frequency Response from a Pole-Zero Plot

The frequency response can be visualised in terms of the pole locations of the
transfer function. For example, for a second-order lowpass system:

ω n2 Standard form for a

H (s ) = 2 lowpass second

s + 2ζω n s + ω n2
order transfer
(5A.23) function

the poles are located on a circle of radius ω n and at an angle with respect to

the negative real axis of ψ = cos −1 (ζ ) . These complex conjugate pole locations
are shown below:

Pole locations for a

lowpass second
ω
j order transfer
function

p ωn

ψ
σ

p* - ωn

Figure 5A.9

In terms of the poles shown in Figure 5A.9, the transfer function is:

ω n2
Lowpass second

H (s ) =
order transfer

(s − p )(s − p ∗ )
function using pole
(5A.24) factors

Signal Theory (48541)

 5A.12
With s = jω , the two factors in this equation become:

Polar representation
of the pole factors jω − p = m1∠φ 1 and jω − p∗ = m2 ∠φ 2 (5A.25)

In terms of these quantities, the magnitude and phase are:

Magnitude function
1
H (ω ) =
written using the
polar representation
of the pole factors m1m2 (5A.26)

and:
Phase function
written using the
polar representation
of the pole factors
θ = −(φ 1 + φ 2 ) (5A.27)

Vectors representing Eq. (5A.25) are shown below:

Determining the
magnitude and jω jω jω
phase response
m1 jω 2
from the s plane
m1
p p jω n p
φ1
m1 φ1 φ1
jω 1

σ σ σ
m2 m2 m2

φ2 φ2 φ2
p* p* p*

θ
ω1 ωn ω2
0 ω
|H| 1
2ζ
1
-90 º

-180 º
0 ω1 ωn ω2 ω

Figure 5A.10

Signal Theory (48541)

 5A.13

Figure 5A.10 shows three different frequencies - one below ω n , one at ω n ,

and one above ω n . From this construction we can see that the short length of

m1 near the frequency ω n is the reason why the magnitude function reaches a

peak near ω n . These plots are useful in visualising the frequency response of

the circuit.

Bode Plots

Bode* plots are plots of the magnitude function H (ω ) dB = 20 log H (ω ) and the

phase function ∠H (ω ) , where the scale of the frequency variable (usually ω )

is logarithmic. The use of logarithmic scales has several desirable properties:

• we can approximate a frequency response with straight lines. This is called

an approximate Bode plot. The advantages of
using Bode plots

• the shape of a Bode plot is preserved if we decide to scale the frequency –

this makes design easy.

• we add and subtract individual factors in a transfer function, rather than

multiplying and dividing.

• the slope of all lines in a magnitude plot is ± 20n dB/decade , and

± n 45°/decade for phase plots, where n is any integer.

• by examining a few features of a Bode plot, we can readily determine the

transfer function (for simple systems).

We normally don’t deal with equations when drawing Bode plots – we rely on
our knowledge of the asymptotic approximations for the handful of factors that
go to make up a transfer function.

*
Dr. Hendrik Bode grew up in Urbana, Illinois, USA, where his name is pronounced boh-dee.
Purists insist on the original Dutch boh-dah. No one uses bohd.

Signal Theory (48541)

 5A.14

Approximating Bode Plots using Transfer Function Factors

The table below gives transfer function factors and their corresponding
magnitude asymptotic plots and phase linear approximations:

Magnitude Phase
Transfer
Function Asymptote Linear Approximation
Factor
H , dB ∠H , °

40
0
20
0 -90
K
-20 -180
-40
0.01ω n 0.1ω n ωn 10ω n 100ω n
0.01ω n 0.1ω n ωn 10ω n 100ω n

40
0
20
1 0 -90
(s ω n ) -20 -180
-40
0.01ω n 0.1ω n ωn 10ω n 100ω n
0.01ω n 0.1ω n ωn 10ω n 100ω n

40
0
20
1 0 -90
(s ω n + 1) -20 -180
-40
0.01ω n 0.1ω n ωn 10ω n 100ω n
0.01ω n 0.1ω n ωn 10ω n 100ω n

40
0
20
1
0 -90
s 2
2ζ 
 2 + s + 1 -20
 ωn ωn
-180

-40
0.01ω n 0.1ω n ωn 10ω n 100ω n
0.01ω n 0.1ω n ωn 10ω n 100ω n

The corresponding numerator factors are obtained by “mirroring” the above

plots about the 0 dB line and 0° line.

Signal Theory (48541)

 5A.15

Transfer Function Synthesis

One of the main reasons for using Bode plots is that we can synthesise a
desired frequency response by placing poles and zeros appropriately. This is
easy to do asymptotically, and the results can be checked using MATLAB®.

Example – Band-Enhancement Filter

The asymptotic Bode plot shown below is for a band-enhancement filter:

A band
enhancement filter
|H|, dB 20 dB

0 dB
2
10 10
3
10
4
10
5
ω rad/s (log scale)

Figure 5A.11

We wish to provide additional gain over a narrow band of frequencies, leaving

the gain at higher and lower frequencies unchanged. We wish to design a filter
to these specifications and the additional requirement that all capacitors have
the value C = 10 nF .

Signal Theory (48541)

 5A.16
The composite plot may be decomposed into four first-order factors as shown
below:

Decomposing a
Bode plot into first-
order factors |H|, dB 20 dB

0 dB
10 ω rad/s (log scale)
2 3 4 5
10 10 10

|H|, dB
1 4

ω rad/s (log scale)

2 3

Figure 5A.12

Those marked 1 and 4 represent zero factors, while those marked 2 and 3 are
pole factors. The pole-zero plot corresponding to these factors is shown below:

The pole-zero plot

corresponding to the
Bode plot jω

4 3 2 1

Figure 5A.13

Signal Theory (48541)

 5A.17
From the break frequencies given, we have:

H (ω ) =
(1 + jω 10 )(1 + jω 10 )
2 5

(1 + jω 10 )(1 + jω 10 )
3 4
(5A.28)

Substituting s for jω gives the transfer function:

H (s ) =
(s + 10 )(s + 10 )
2 5 The transfer function

(s + 10 )(s + 10 )
corresponding to the
3 4 Bode plot
(5A.29)

We next write H (s ) as a product of bilinear functions. The choice is arbitrary,

but one possibility is:

The transfer function

s + 10 s + 10
2 5
H (s ) = H 1 (s )H 2 (s ) =
as a cascade of
× bilinear functions
s + 10 3 s + 10 4 (5A.30)

For a circuit realisation of H 1 and H 2 we decide to use an inverting op-amp

circuit that implements a bilinear transfer function:

A realisation of the
specifications
-2 -3 -5 -4
10 10 10 10
Vi
Vo

1 1 1 1

Figure 5A.14

Signal Theory (48541)

 5A.18
To obtain realistic element values, we need to scale the components so that the
transfer function remains unaltered. This is accomplished with the equations:

Magnitude scaling is 1
required to get
realistic element
Cnew = Cold and Rnew = km Rold
values
km (5A.31)

Since the capacitors are to have the value 10 nF, this means k m = 108 . The
element values that result are shown below and the design is complete:

A realistic
implementation of
the specifications 1 MΩ 100 kΩ 1 kΩ 10 kΩ
Vi
Vo

10 nF 10 nF 10 nF 10 nF

Figure 5A.15

In this simple example, the response only required placement of the poles and
zeros on the real axis. However, complex pole-pair placement is not unusual in
design problems.

Signal Theory (48541)

 5A.19

Digital Filters
Digital filtering involves sampling, quantising and coding of the input analog
Digital filters use
signal (using an analog to digital converter, or ADC for short). Once we have analog filters
converted voltages to mere numbers, we are free to do any processing on them
that we desire. Usually, the signal’s spectrum is found using a fast Fourier
transform, or FFT. The spectrum can then be modified by scaling the
amplitudes and adjusting the phase of each sinusoid. An inverse FFT can then
be performed, and the processed numbers are converted back into analog form
(using a digital to analog converter, or DAC). In modern digital signal
processors, an operation corresponding to a fast convolution is also sometimes
employed – that is, the signal is convolved in the time-domain in real-time.

The components of a digital filter are shown below:

The components of
a digital filter
Analog Digital Analog
Vi ADC Signal DAC Vo
Filter Processor Filter

Figure 5A.16

The digital signal processor can be custom built digital circuitry, or it can be a
Digital filter
general purpose computer. There are many advantages of digitally processing advantages
analog signals:

1. A digital filter may be just a small part of a larger system, so it makes sense
to implement it in software rather than hardware.

2. The cost of digital implementation is often considerably lower than that of

its analog counterpart (and it is falling all the time).

3. The accuracy of a digital filter is dependent only on the processor word

length, the quantising error in the ADC and the sampling rate.

4. Digital filters are generally unaffected by such factors as component

accuracy, temperature stability, long-term drift, etc. that affect analog filter
circuits.

Signal Theory (48541)

 5A.20
5. Many circuit restrictions imposed by physical limitations of analog devices
can be circumvented in a digital processor.

6. Filters of high order can be realised directly and easily.

7. Digital filters can be modified easily by changing the algorithm of the

computer.

8. Digital filters can be designed that are always stable.

9. Filter responses can be made which always have linear phase (constant
delay), regardless of the magnitude response.
Digital filter
disadvantages Some disadvantages are:

1. Processor speed limits the frequency range over which digital filters can be
used (although this limit is continuously being pushed back with ever faster
processors).

2. Analog filters (and signal conditioning) are still necessary to convert the
analog signal to digital form and back again.

Summary
• A frequency response consists of two parts – a magnitude response and a
phase response. It tells us the change in the magnitude and phase of a
sinusoid at any frequency, in the steady-state.

• Bode plots are magnitude (dB) and phase responses drawn on a semi-log
scale, enabling the easy analysis or design of high-order systems.

References
Kamen, E. & Heck, B.: Fundamentals of Signals and Systems using
MATLAB®, Prentice-Hall, 1997.

Signal Theory (48541)

 5A.21

Exercises

1.
With respect to a reference frequency f 0 = 20 Hz , find the frequency which is

(a) 2 decades above f 0 and (b) 3 octaves below f 0 .

2.
Express the following magnitude ratios in dB: (a) 1, (b) 40, (c) 0.5

3.
Draw the approximate Bode plots (both magnitude and phase) for the transfer
functions shown.

4 1
(a) G (s ) = 10 (b) G (s ) = (c) G (s ) =
s 10s + 1

1
(d) G (s ) = (e) G (s ) = 5(s + 1) (f) G (s ) = 5(s − 1)
10s − 1

Note that the magnitude plots for the transfer functions (c) and (d); (e) and (f)
are the same. Why?

4.
Prove that if G (s ) has a single pole at s = −1 τ the asymptotes of the log
magnitude response versus log frequency intersect at ω = 1 τ .

Signal Theory (48541)

5A.22

5.
Make use of the property that the logarithm converts multiplication and
division into addition and subtraction, respectively, to draw the Bode plot for:

100(s + 1)
G (s ) =
s (0.01s + 1)

Use asymptotes for the magnitude response and a linear approximation for the
phase response.

6.
Draw the approximate Bode (magnitude and phase) for

100s 2
G (s ) =
s 2 + 2ζω n s + ω n2

when

ω n = 10 rads -1 and ζ = 0.3

7.
Given

20(6.66s + 1)
(a) G (s ) =
s (0.1s + 1) (94.6s + 1)
2

(b) G (s ) =
(
4 − s + 1 .5 + s 2 )
s 2 (10 − s )

Draw the approximate Bode plots and from these graphs find G and ∠G at

(i) ω = 0.1 rads -1 , (ii) ω = 1 rads -1 , (iii) ω = 10 rads -1 , (iv) ω = 100 rads -1 .

Signal Theory (48541)

 5A.23

8.
The experimental responses of two systems are given below. Plot the Bode
diagrams and identify the transfer functions.
(a) (b)
ω G1 ∠G1 ω G2 ∠G2
(rads-1 ) (dB) (°) (rads-1 ) (dB) (°)
0.1 40 -92 0.01 -26 87
0.2 34 -95 0.02 -20 84
0.5 25 -100 0.04 -14 79
1 20 -108 0.07 -10 70
2 14 -126 0.1 -7 61
3 10 -138 0.2 -3 46
5 2 -160 0.4 -1 29
10 -9 -190 0.7 -0.3 20
20 -23 -220 1 0 17
30 -32 -235 2 0 17
40 -40 -243 4 0 25
50 -46 -248 7 -2 36
100 -64 -258 10 -3 46
20 -7 64
40 -12 76
100 -20 84
500 -34 89
1000 -40 89

9.
Given G (s ) = K s (s + 5)

(a) Plot the closed loop frequency response of this system using unity feedback
when K = 1 . What is the –3 dB bandwidth of the system?

(b) Plot the closed loop frequency response when K is increased to K = 100 .
What is the effect on the frequency response?

10.
The following measurements were taken for an open-loop system:
(i) ω = ω1 , G = 6 dB , ∠G = 25°
(ii) ω = ω 2 , G = −18 dB , ∠G = 127°

Find G and ∠G at ω1 and ω 2 when the system is connected in a unity

feedback arrangement.

Signal Theory (48541)

5A.24

11.
An amplifier has the following frequency response. Find the transfer function.

40

30

20

10
|H(w)| dB

-10

-20

-30

-40
0 1 2 3 4 5
10 10 10 10 10 10
w (rad/s)

100

80

60

40

20
arg(H(w)) deg

-20

-40

-60

-80

-100
0 1 2 3 4 5
10 10 10 10 10 10
w (rad/s)

Signal Theory (48541)