Introduction to

Telecommunication

The Impulse Response and

Transfer Function of Linear Systems

Asst. Prof. Dr. Ertuğrul SAATÇI

Overview
 In the previous week the we have examined the Fourier
Transform applied to energy signals.

 Today we examine the Fourier Transform applied to periodic

signals (i.e., power signals).

 Also, today we will apply the Fourier Transform to linear time

invariant systems.

 Specifically, we will show that an LTI system can be

characterized in the time domain by the impulse response,
 but an even more powerful way of characterizing an LTI
system is in the frequency domain through the use of the
transfer function.

 Reading
 2.5, 2.6

1
Lecture Objectives
 To show how the delta function allows the Fourier Transform to
be applied to power signals (i.e., periodic signals).

 To describe and show how to use a system’s impulse response

and transfer function.

Fourier Transform of Complex Exponentials

 Using the Fourier Transform of a constant and the frequency
shift property we can find the FT of a complex exponential:
z  t  1e j 2 f0t
 This is just a frequency shift of a constant, thus

Z  f    f  f 0 
 This results in

e j 2 f 0 t    f  f 0 
1 1
sin  2 f 0t     f  f0     f  f0 
2j 2j
1 1
cos  2 f 0t     f  f 0     f  f 0 
2 2

2
Fourier Transform of Periodic Signals
 Using this same approach we can develop the Fourier
Transform for any periodic signal.

 Specifically, we can define a Fourier Series for any periodic

signal (periodic with the period T0) that is valid over all time.

 Using the linearity property of the Fourier Transform we can

write the FT of any periodic signal:

x  t    ck e jk 2 f0t
k 

    

 
X  f   F   ck e j 2 kf0t    ck F e j 2 kf0t   ck   f  kf 0 
 k   k  k 

TW= T0 /2
To A=1

Example 3.1 -TW /2 TW /2

 Find the Fourier Transform of a 50% duty cycle square wave

with period To, amplitude 1 and average value 1/2.
The signal can be written as

 t  nT0 
x  t    rect  
n   T0 / 2 
 Clearly this signal is not integrable and thus does not have a
Fourier Transform in the strict sense.
 However, we can write this in terms of its Fourier Series
coefficients as  
1  k  j 2 kf0 t
x  t    ck e jk 2 f0 t   sinc   e ,
k  2 k  2
1 k
where ck  f 0Twsinc  kf 0Tw   sinc  
2 2

3
Example 3.1 – cont.
 Writing the signal in terms of its Fourier Series coefficients:

1  k
x t   
2 k 
sinc   e j 2 kf0t
2
 Using the linearity property and the FT for a complex
exponential:

 
F e j 2 f 0 t    f  f 0 
1  k  1  k

X  f   F   sinc   e j 2 kf0t    sinc   F e j 2 kf0t
2 2

 2 k   2 k 
1  k
  sinc     f  kf 0 
2 k  2
Note that since the signal is periodic the spectrum
(i.e., Fourier Transform) is discrete.

Example 3.1 – cont.

 Plotting the spectrum:

T0=0.1 sec
f0=10 Hz

4
Why study LTI systems?
 Several parts of a communication system can be modeled as an
LTI system including
● Filters
● Equalizers
● The channel
● Pulse shaping

Linear Time Invariant Systems

 A system is linear when superposition holds:

y (t )    a1 x1 (t )  a2 x2 (t )   a1  x1 (t )   a2   x2 (t ) 

 A system is time invariant if for a delayed input x(t-to) the

output is simply delayed by the same amount y(t-to).
 In other words the shape of the output is the same no
matter when the input is applied.

5
Impulse Response

 When the input to a LTI system is an impulse, we term the

output the impulse response of the system.

Impulse Response – cont.

 An LTI system may be characterized by its impulse response.
● If we represent the output and input of an LTI system by
y(t) and x(t), respectively we say that
 the impulse response h(t) is the output y(t) when
x(t)=δ(t).

 The impulse response is useful because it can be used to find

the output of a system
 when the input is not an impulse.

6
 Impulse Response – cont.
 Let us approximate a general input x(t) using a series of
impulses

x(t )   x  nt   (t  nt ) t
n 0

x  nt 

t
 Since the system is LTI we can approximate the output as

y (t )   x  nt   h(t  nt )  t
n0

Impulse Response – cont.

7
Impulse Response – cont.
 Now, if we let ∆t go to zero we get:

y (t )   x   h  t    d


Also,

y (t )   h   x  t    d

 This says that the output of an LTI system is the convolution of
the input and the impulse response.
 Thus we can find the output for any input if we know the
impulse response.

Impulse Response
Interpretation
 The current output of an LTI system is a weighted integral over
the past history of the input signal,
 weighted according to the impulse response of the system.
 
y (t )   x   h  t    d   h   x  t    d
 

 From the convolution integral we can see that the current

output is determined by the current input and the past inputs.
 The weighting of the past inputs is equal to the impulse
response.

 Thus, the impulse response acts as the memory of the system.

 What if the impulse response were an impulse?

8
Impulse Response in the Frequency
Domain
 Recall the Fourier Transform Property (convolution):

w1 (t ) * w2 (t )  W1 ( f )W2 ( f )

 A complicated operation in the time domain can be reduced to

a simple operation in the frequency domain.
 We can now more simply find the system output.

y (t )  x(t ) * h(t )  Y ( f )  X ( f ) H ( f )

Transfer Function
 Rewriting
Y( f )
H( f ) 
X(f )
H(f) is termed the Transfer Function of the system.

 Thus, we can find the transfer function of an LTI system by

 dividing the output signal’s Fourier Transform by the
Fourier Transform of the input.

9
Transfer Function – cont.
Alternative Definition
 Consider a LTI system characterized by impulse response h(t)
with input x(t)

x  t   e j 2 f 0 t
 The output can be determined as

y  t    x   h  t    d


  h   x  t    d



 h   e
j 2 f 0  t  
 d


Transfer Function – cont.

Alternative Definition

 Continuing
y (t )   h   e j 2 f0  t   d



 h   e
j 2 f 0 t  j 2 f 0
e d


 e j 2 f 0 t H  f 0 
 Thus, when the input to a LTI system is a complex sinusoid of
frequency fo,
 the output is also a complex sinusoid of frequency fo
weighted by the transfer function at that frequency, H(fo).

10
 Transfer Function – cont.
Alternative Definition
 In general, a signal x(t) can be expressed in terms of the
inverse Fourier Transform

x(t )   X  f  e j 2 f t df

which can be written in the limiting form

x(t )  lim
f  0
 X  k f  e
k 
j 2 k f t
f
Thus, from the previous development we can write
 
y (t )  lim
f  0

k 
H  k f  X  k f  e j 2 k f t f  H  f  X  f e
   
j 2 f t
df

Y f 

Thus Y ( f ) H  f  X  f 

Magnitude / Phase Response

 The transfer function H(f) is a characteristic property of an LTI
system.

 We are interested both in the amplitude response or magnitude

response |H(f)| and the phase response angle(H(f)).

 If the impulse response is real, then the magnitude response is

even
H ( f )  H ( f )
and the phase response is odd.
H ( f )  H ( f )

If h  t  is real, then H ( f )  H * (  f ).

11
 Magnitude of H(f)
Proof
 Taking the magnitude

  
H( f )   h  t  e  j 2 f t dt   h  t  cos  2 f t  dt  j  h  t  sin  2 f t  dt
  

2 2
   
   h  t  cos  2 f t  dt     h  t  sin  2 f t  dt 
     
  

 h t  e  h  t  cos  2 f t  dt  j  h  t  sin  2 f t  dt
 j 2   f  t
H ( f )  dt 
  

 The values for f and –f are the same.

Angle of H(f)
Proof
 Now, we can take the angle
 
H ( f )     h  t  e  j 2 f t dt 
  
 

    h  t  cos  2 f t  dt  j  h  t  sin  2 f t  dt 
   
     
 
 h  t  sin  2 f t  dt    h  t  sin  2 f t  dt 
 tan 1      tan 1   
   
  h  t  cos  2 f t  dt    h  t  cos  2 f t  dt 
     
where we assumed a real signal h(t) to make the first step.

12
 Angle of H(f) – cont.
Proof
 Substituting for –f we have
 
H ( f )     h  t  e   dt 
 j 2  f t

  
 

    h  t  cos  2 f t  dt  j  h  t  sin  2 f t  dt 
   
  
  h  t  sin  2 f t  dt 
 tan 1     H ( f )
 
  h  t  cos  2 f t  dt 
  

Example 3.2
 Consider a system with impulse response
h  t  e 5t u  t 
Determine the output when the input is

x  t  cos  200 t   cos  20 t 

What does the system do to the input?
 We can solve this in the time domain or in the frequency
domain. In the time domain we have
 
y  t    x   h  t    d    cos  200   cos  20   e u  t    d
5 t  

 

This doesn’t appear to be a fun convolution! (Although

manageable).

13
Example 3.2 – cont.
Frequency domain
 Let’s try this in the frequency domain instead.

In the frequency domain, we have

1
Hf
5  j 2 f
1
X  f     f  100     f  100     f  10     f  10  
2
Y f  H f X f 

1 1 1
e  at u (t )  , cos  2 f 0 t     f  f 0     f  f 0 
a  j 2 f 2 2

Example 3.2 – cont.

Frequency domain
 Calculating the output spectrum
1
Hf
5  j 2 f
1
Y  f    H 100    f  100   H  100    f  100 
2
 H 10    f  10   H  10    f  10  
1 1 1
Yf     f  100     f  100 
2  5  j 200 5  j 200
1 1 
   f  10     f  10  
5  j 20 5  j 20 

14
Example 3.2 – cont.
Frequency domain

1 1 1
Yf     f  100     f  100 
2  5  j 200 5  j 200
1 1 
   f  10     f  10  
5  j 20 5  j 20 
1  5  j 200 5  j 200
Yf     f  100     f  100 
2  25   200  2
25   200 
2

5  j 20 5  j 20 
   f  10     f  10  
25   20  25   20 
2 2


Example 3.2 – cont.

Frequency domain
 Collecting cosine and sine terms

1 5
Yf     f  100     f  100  
2  25   200 2 
200 1
   f  100     f  100  
25   200 
2
j
5
   f  10     f  10  
25   20 
2

20 1 
   f  10     f  10  

25   20 
2
j 

15
 Example 3.2 – cont.
Time domain
 In the time domain we have
 5 200
y t    cos  200 t   sin  200 t 
 25   200  25   200 
2 2

5 20 
 cos  20 t   sin  20 t  
25   20  25   20 
2 2

Now
a b 1
cos  2 f 0t   2 sin  2 f 0t   cos  2 f 0t   
a b
2 2
a b 2
a  b2
2

b
  tan 1  
a

Example 3.2 – cont.

Time domain
 Finally, we have
1 1
y t   cos  200 t  1   cos  20 t   2 
25   200  25   20 
2 2

 How do we interpret this?

 The input was a pair of sinusoids with frequencies f1=100
and f2=10.
 The output was also a pair of sinusoids with frequencies
f1=100 and f2=10.
 However, the two sinusoids are both attenuated with the first
being attenuated more.
 Additionally, both were delayed in time (i.e., phase shifted) –
θ1=1.56 radians, θ2=1.49 radians (both nearly /2).

16
 Example 3.2 – cont.
Transfer Function of the System

Magnitude Response Phase Response

1 1  2 f 
Hf  , Hf  , H  f    tan 1  
5  j 2 f 25   2 f 
2
 5 

Summary
 The Dirac delta function (or impulse) can be used to define the
Fourier Transform for periodic signals.
 When the Fourier Series combined with the Dirac delta
function, the Fourier Series coefficients allow us to determine
the Fourier Transform.

 LTI Systems can be characterized in the time domain by their

impulse response.

 LTI Systems can also be characterized in the frequency domain

by their transfer function
 which is the Fourier Transform of the impulse response.

 The transfer function tells us how input frequencies will be

attenuated (from the magnitude response) and how they will
be phase shifted (by the phase response).

17
Introduction to
Telecommunication

Filtering

Asst. Prof. Dr. Ertuğrul SAATÇI

Overview
 In the previous classes we have discussed using the Fourier
Transform to characterize systems.

 In this lecture we investigate an important subsystem of

communications systems: filters.

 It is important to view filters in the frequency domain in order

to more fully understand their characteristics.

 Reading
 2.6, 2.7

18
Filters
 A filter is a system which passes certain frequencies and rejects
other frequencies.

 Types of filters
• Lowpass filter
• Highpass filter
• Bandpass filter
• Bandstop filter

 Ideal filter
• An ideal filter is one which perfectly passes frequencies in a
certain range (termed the pass band) and perfectly rejects
frequencies in another range termed the stop band.
• An ideal filter doesn’t distort the signal in the pass band.

The Ideal Lowpass Filter

 The ideal lowpass filter passes without distortion all frequencies
inside the pass band of the filter (i.e., frequencies below
frequency B) and completely rejects all frequencies inside the
stop band.
 j 2  f t0
e f B  f   j 2  f t0
 Frequency response Hf  rect   e
 0 f  B   2 B time
delay
magnitude
response

19
The Ideal Lowpass Filter
Magnitude / Phase Responses

 f   j 2  f t0  f 
H  f   rect  e  rect  
 2B   2B 
  f   j 2  f t0 
arg  H  f    arg  rect  e 
  2B  
arg cos  2 f t0   j sin  2 f t0   f B
 
 0 f B
 1   sin  2 f t0  
 tan   f B  2 f t0
 f B
  cos  2 f t0   
f B
  0
 0 f B

The Ideal Lowpass Filter

Linear Phase
 What does it mean to have a linear phase response?

 f 
H  f   rect   H  f   H  f  e  j 2  f t0
 2B  
 A  f  e  j 2  f t0
arg  H  f    2 f t0

 Linear phase corresponds to all frequencies being delayed by

the same amount.
 In other words, there is no phase distortion imparted to the
input signal.

20
 The Ideal Lowpass Filter
Impulse response
 The impulse response of the ideal LPF can be found by taking
the inverse Fourier Transform of the ideal LPF frequency
response.

h  t   F 1  H  f 
2B=5, to=2
5sinc(5(t-2))
  f   j 2  f t0 
 F 1 rect  e 
  2B  
 2 Bsinc  2 B  t  t0  

1  f 
sinc  2Wt   rect  ,
2W  2W 
x  t  t0   e  j 2  f t0
Xf

The Ideal Lowpass Filter

Causality
 A causal system is a system that does not respond before the
input is applied.

 The impulse response is the response of a system to an

impulse applied at time t=0.

 A system whose impulse response is nonzero for t<0 is thus

noncausal.
 The ideal lowpass filter is thus noncausal and for that
reason is not physically realizable.
 We can attempt to approximate the ideal LPF by
introducing a delay into our system.

21
The Ideal Lowpass Filter
Causal LPF
 We can make the filter causal by simply truncating the impulse
response before t=0.

2B=5, to=0
5sinc(5t)u(t)

 The resulting filter is far from ideal.

The Ideal Lowpass Filter

Causal LPF
 A second option is to delay the impulse response and truncate it.
 Delay = 2

2B=5, to=2
5sinc(5(t-2))u(t)

 This makes the system closer to ideal, but requires a delay which
some applications may not tolerate.

22
The Ideal Lowpass Filter
Causal LPF
 A second option is to delay the impulse response and truncate it.
 Delay = 5

2B=5, to=5
5sinc(5(t-5))u(t)

 Larger delay leads to better approximation of the ideal LPF.

The Ideal Lowpass Filter

Causal LPF
 Consider a RC filter.
 Causal, exponential impulse response.
 Non-ideal LPF.

h  t   5 e 5 t Hf 
1
1  j  f / 2.5 

23
 The Ideal Lowpass Filter
Pulse Response of Ideal LPF
 Let us apply a rectangular pulse x(t)=rect(t/T) to an ideal
lowpass filter (impulse response h(t)=2B sinc(2Bt) ).

y t   x t   h t 

 
T /2
  rect   2 Bsinc  2 B  t     d   2 Bsinc  2 B  t     d
 T   T /2
T /2
sin  2 B  t     d
  2B d ,   2 B  t       d
T / 2
 2B t   2 B
2 B  t T /2 
1 sin   

 
2 B  t T /2 

d

The Ideal Lowpass Filter

Pulse Response of Ideal LPF – cont.
 Let us apply a rectangular pulse x(t)=rect(t/T) to an ideal
lowpass filter (impulse response h(t)=2B sinc(2Bt) ).
2 B  t T /2 
1 sin   
y t    d
 2 B  t T /2 

2 B  t T /2  2 B  t T / 2 
1 sin    1 sin   

 
0

d 
 
0

d

 Si  2 B  t  T / 2    Si  2 B  t  T / 2    , where
1


x
sin   
Si  x    d  is the Sine integral.
0


24
Time-Bandwidth Product
 The product of the signal’s duration and its bandwidth is always
a constant.
B T =  Bandwidth  ×  Duration  = constant
 Whatever definition we use for the bandwidth of a signal, the
time-bandwidth product remains constant over certain classes
of pulse signals.

Bandwidth for lowpass Bandwidth for bandpass

system system

Example 3.3
Pulse Response of Ideal LPF
 Consider a time-domain square pulse of width 1 second which
is passed through a filter with a bandwidth of 5Hz (BT=5).
 The bandwidth restriction does not cause a substantial change
in the pulse shape.

Time Domain
Frequency Domain

B=5Hz

25
Example 3.3 – cont.
Pulse Response of Ideal LPF
 Now consider a filter with a bandwidth of 3Hz (BT=3).
 The bandwidth restriction still does not cause a substantial
change in the pulse shape.
 Ringing occurs near sharp transitions.

Frequency Domain Time Domain

B=3Hz

Example 3.3 – cont.

Pulse Response of Ideal LPF
 Now consider a filter with a bandwidth of 2Hz (BT=2).
 The bandwidth restriction now begins to cause a more
substantial change in the pulse shape.
 Sharp transitions cannot occur.

Frequency Domain Time Domain

B=2Hz

26
Example 3.3 – cont.
Pulse Response of Ideal LPF
 Now consider a filter with a bandwidth of 1Hz (BT=1).
 The bandwidth restriction now changes the shape considerably.

Frequency Domain Time Domain

B=1Hz

Example 3.3 – cont.

Pulse Response of Ideal LPF
 Finally, consider a filter with a bandwidth of 100Hz (BT=100).
 The pulse shape is nearly unchanged with the exception of
Gibbs phenomenon.

27
The Ideal Bandpass Filter
 The ideal bandpass filter can be written as
 Ae  j 2 f t0 fL  f  fH
Hf 
 0 else
  f  f0   f  f 0    j 2  f t0
 A  rect    rect   e
  f   f   
  timedelay
magnitude
response

fL = lower frequency limit

fH = upper frequency limit
∆f = fH–fL
fo = (fH+fL)/2

The Ideal Bandpass Filter – cont.

 The impulse response of the ideal bandpass filter
h  t   F 1  H  f 
   f  f0   f  f 0    j 2 f t0 
 F 1  A  rect    rect   e 
   f   f   
   f  f 0   j 2  f t0  f  f 0   j 2 f t0  
 F 1  A  rect  e  rect  e 
   f   f   
 A  f sinc  f  t  t0   e 0  0   f sinc  f  t  t0   e
 j 2 f 0  t  t0 
j 2 f t  t


 Af sinc  f  t  t0    e 0  0   e
 j 2  f 0  t  t0 
j 2 f t  t


 2 Af sinc  f  t  t0   cos  2 f 0  t  t0  

28
The Ideal Bandpass Filter – cont.
 Ideal BPF is also noncausal.
 Realistic bandpass filters will be causal and not have a perfect
frequency response.

fL=7.5Hz, fH=12.5Hz
∆f=5Hz, fo=10Hz

Example 3.4
Noise
 A major use of filters is the elimination of noise.

 Noise typically has a much larger bandwidth than the signal of

interest.
 Filtering the received signal with a bandpass or lowpass filter
can reduce the amount of noise.

 We typically consider additive noise where the received signal

r(t) is equal to the desired signal x(t) plus noise n(t).
 r(t)=x(t)+n(t)

29
Example 3.4 – cont.
Noise
 Consider a square pulse with duration one second that is
received with the addition of noise.
 The ratio of the received desired signal power to the noise
power, signal-to-noise ratio or SNR, is 2.

Desired signal

x(t)

Received signal

r(t)=x(t)+n(t)

Example 3.4 – cont.

Noise
 If we filter the signal with an ideal LPF with bandwidth B=2Hz,
we know that we will introduce some distortion to the desired
signal, but we can also eliminate much of the noise.

Desired signal

Filter

Desired Noise
signal PSD
Received signal

Frequency Domain Time Domain

30
Example 3.4 – cont.
Noise
 We know that increasing the bandwidth to B=3Hz will reduce
the amount of distortion to the original signal.
 However, it also lets more noise in.

Desired signal

Filter

Desired Noise
signal PSD Received signal

Frequency Domain Time Domain

Summary
 An important function in communication systems is to filter the
input signal.
 Filtering helps eliminate noise in the system but can also
distort the desired signal.
 Increasing the value of the time-bandwidth product BT tends to
reduce the rise time and decay time of the filter pulse response
and helps to preserve the pulse shape.
 Increasing the bandwidth of the filter also allows more noise
into the system.
 Ideal filters are noncausal.
• Causal filters can approximate the ideal filter using delay.
• More delay can allow better approximation.
 Filters are built in the analog domain using resistors, capacitors
and inductors.
 They are often implemented digitally using simple tapped-
delay lines in DSP.

31