Contemporary Communication Systems

Chapter 2

Review of Signals and

Linear Systems

M.F. Mesiya
9/18/2024 © 2012 by McGraw-Hill Education. This is proprietary material solely for authorized instructor use. Not authorized for sale or distribution in any 1
manner. This document may not be copied, scanned, duplicated, forwarded, distributed, or posted on a website, in whole or part.
Analog and Digital Signals
⚫ A continuous-time signal that assumes a continuum of
amplitude values between given maximum and minimum is
called an analog signal
⚫ Most signals we encounter in the real world are analog in
nature. Examples include speech, music, image, and video
signals
⚫ Digital signals, on the other hand, can change values at
discrete instants of time, assuming one of a finite number of
amplitude levels
Digital signals
Analog signal

9/18/2024 2
Deterministic and Random Signals
⚫ A deterministic signal x(t) is completely specified for each
value of time t – that is, its amplitude is known either
graphically or analytically for all values of t
⚫ An example is a simple sinusoidal waveform sin(4πt )

⚫ A random signal is not precisely known for each value of t −

it can only be specified in terms of probabilities
⚫ A very important class of signals that includes noise and all
information-carrying signals, such as speech and data

Sine wave Random noise

9/18/2024 3
Some Useful Basic Signals Triangular
pulse
Rectangular
pulse

- /2  t   / 2  2t
t 1 t 1 − , t  / 2
      
  0 otherwise   0,
Sinc  otherwise
pulse Unit
step

1 t0
sin( t ) u (t ) 
sinc(t ) 0 otherwise
9/18/2024 t 4
The Unit Impulse Signal (Delta Function)
⚫ The unit impulse signal (t) is defined by the equations
 (t ) = 0, t  0

   (t )dt = 1, for any real number   0
−

⚫ Thus the unit impulse signal is zero everywhere except at the

origin and it has unit area
⚫ The value of (t) at t = 0 is not defined. In particular, (0) ≠∞
⚫ A unit impulse signal can be viewed as a narrow pulse with
large amplitude and having a unit area
⚫ For example, it can be viewed as a limit of
the unity area rectangular pulse as its
width approaches zero and its amplitude
increases proportionately
9/18/2024 5
Important Properties of Delta Function
⚫ P1. x(t ) (t − to ) = x(to ) (t − to )

1
⚫ P2.  ( t ) =  (t )

⚫ P3. Sampling property
  
 −
x(t ) (t − to )dt =  x(to ) (t − to )dt = x(to )   (t − to )dt = x(to )
− −

⚫ P4. Convolution

x(t )   (t ) =  x( ) (t −  )dt = x(t )
−

⚫ The convolution of an arbitrary signal with the impulse

signal yields the signal itself
9/18/2024 6
 Frequency Domain Representation
⚫ Electrical signals used in communication systems are functions
of time, such as voltage and current, it is very useful to think
of signals in terms of their frequency content
⚫ Certain characteristics of signals are easier to analyze and
measure in the frequency domain.
⚫ Figure shows the time domain representation of a 10 Hz sine
wave embedded in noise

9/18/2024 7
Sine Wave Embedded in Noise
Time Domain Display

Frequency Domain Display

9/18/2024 8
Frequency Domain Representation (contd)
⚫ Difficult to identify a 10 Hz tone in the presence of
wideband (“white”) noise on an oscilloscope display
⚫ However, easy to identify 10 Hz tone in the frequency
domain using a spectrum analyzer display. Note that the
white noise forms the floor of the display
⚫ In more complex situations, the composite signal may consist
of hundreds of channels or carriers. An example is CATV
system where several hundred channels or signals may be
present

9/18/2024 9
Exponential Fourier Series (FS)
⚫ The Fourier series can be used to represent periodic signals in
the frequency domain
⚫ A periodic function xp(t) with fundamental period To can be
represented by an exponential Fourier series

x p (t ) =  n
C e
n =−
j 2 nf ot

where fo = 1/ To is called the fundamental frequency of the

periodic signal xp(t)
⚫ The FS coefficients Cn are given by
1
Cn =  x p (t )e − j 2nf o t dt
To To
⚫ Observe that the FS expands a periodic function as an infinite
sum of complex phasor signals
9/18/2024 10
Exponential FS (contd)
⚫ The term C0 represents the DC component of the signal
1
C0 =  x p (t )dt
To T o

⚫ n = 1 FS coefficients represent the fundamental frequency (fo)

component in the periodic signal xp(t)
⚫ n = 2, 3,… FS coefficients represent the harmonic (nfo)
components in the periodic signal xp(t)
⚫ Each phasor term in FS can be written as
j ( 2 nfot + C n )
C n e j 2 nfot = C n e

⚫ Plots of C n and C n versus discrete frequency values (n fo, n

= 0, 1, 2,…..) are called the magnitude and the phase line
spectra of the signal, respectively
9/18/2024 11
 Example: Rectangular Pulse Train
⚫ Determine the FS expansion of a periodic pulse train of
rectangular pulses

 ( t − nTo ) 
gTo (t ) =    
n =−   

⚫ Each pulse has unity amplitude and duration . The FS

coefficients are given by
To /2
1 1 1
 To 
− j 2 nf o t
Cn = g (t )e dt = e − j 2 nfot dt = − e − j 2 nfo − e j 2 nfo 
To To
T0 −To /2
j 2 nf oTo
 sin( nf o )  1
= = sinc(nf o ) Cn =  x p (t )e − j 2 nfot dt
To  nf o To To To

9/18/2024 12
Line Spectrum of a Rectangular Pulse Train

9/18/2024 13
Energy and Power Signals
⚫ Energy and power are useful parameters of a signal
⚫ The normalized energy of a signal x(t) is defined as the energy
dissipated by a voltage x(t) applied across a 1-ohm resistor (or
a current x(t) passing though a 1-ohm resistor)


2
Ex x(t ) dt
−

⚫ The energy of a signal is meaningful only if the integral value

is finite. Such signals are called energy signals
⚫ Example Energy of a rectangular pulse x(t ) = A(t / Tb )
 A, t  Tb / 2
x(t ) = 
0, otherwise
 Tb /2
Ex =  x(t ) dt =  A2 dt = A2Tb
2
− −Tb /2
9/18/2024 14
 Example: Energy of the Carrier Pulse
 A cos(2 f ct ) t  Tb / 2
x(t ) = 
0 otherwise
Tb / 2 2
 A Tb / 2
Ex =  x(t ) dt = A2  cos 2 (2 f ct )dt =  [1 + cos(4 f ct )]dt
2
− 2 −Tb / 2
−Tb / 2

A2 Tb / 2 A2 Tb / 2 A2Tb
=
2 −Tb / 2 dt + 2 −Tb / 2 cos(4 fct )dt = 2

⚫ The second integral is zero because carrier frequency fc >>

1/Tb has been assumed – true in practice
⚫ The energy content of the signal becomes infinite in the limit
as Tb → 
9/18/2024 15
Power Signals
⚫ The normalized average power of a signal x(t) is defined as
the power dissipated by a voltage x(t) applied across a 1-ohm
resistor (or a current x(t) passing though a 1-ohm resistor)
T /2
1 
 
2 2
Px lim x (t ) dt Ex x(t ) dt
T → T −
−T / 2

⚫ The normalized average power of a signal is meaningful only

if the limit exists (that is, finite). Such signals are called power
signals
⚫ For a periodic signal xp(t) with period To, the expression for
normalized power simplifies to
To / 2
1

2
P x p (t ) dt
To −To / 2
9/18/2024 16
 Example: Power of a Sinusoidal Signal
x p (t ) = A cos(2πfot +  )
2 To / 2
A
P=
To 
−To / 2
cos 2 (2 f ot +  )dt

A2 To / 2 To / 2 A2 A2
= [ dt +  cos(4 f ct + 2 )dt ] = +0 =
2To − To / 2 − To / 2 2 2

⚫ The second integral is zero because it evaluates the integrand

over two complete periods
⚫ A signal cannot be both power- and energy-type, because for
energy signals Px = 0 and Ex =  for power signals
⚫ A signal may be neither energy-type nor power-type

9/18/2024 17
Parseval’s Theorem
⚫ The normalized power Px of a periodic signal xp(t) is given by
1 2 1
Px =  x p (t ) dt =  x p (t ) x*p (t )dt
To To To To
⚫ Substituting the FS expansion for xp(t) yields
1   * − j 2 nfot 
Px =  x p (t )   Cn e dt
To To  n =− 
 1  * Average power in the
=    x p (t )e − j 2 nf o t
dt Cn frequency component at f = nfo
n =− 
 To To 
 
=  n n= 
* 2
C C Cn
n =− n =−
Average power of xp(t) = sum of the
average power of phasor components
9/18/2024 18
Bandwidth of a Signal
⚫ The bandwidth of a signal is a measure of the frequency
range that contains significant energy of the signal
⚫ The term significant here implies inclusion of those
frequencies that represent the signal with acceptable distortion
⚫ The latter is determined by the relevance in a given
application.
⚫ If the significant energy of the signal lies in the range of
frequencies f1 < f < f2, the bandwidth would be f2 − f1
⚫ There are many definitions of bandwidth depending on how
frequencies f1 and f2 are chosen
⚫ For example, if the frequencies f1 and f2 are chosen so that
99% of the power resides in the frequency band f1 < f < f2 ,
the quantity f2 − f1 is called the 99% power bandwidth
9/18/2024 19
99% Power Bandwidth Rectangular Pulse Train
⚫ We will assume To = 1 sec and / To = 0.5
1
f o = = 1 MHz
To
⚫ The FS coefficients of a rectangular pulse train are given by
A
Cn = sinc(nf o ) = 0.5 sinc(0.5n)
To
⚫ The normalized average power P is
 /2
A2
To / 2
1 1
 
2
P= x p (t ) dt = A dt =
2
= 0.5
To −To / 2
To − / 2
To

⚫ The power in various frequency components is given by

Cn = 0.25 sinc(0.5n)
2 2

9/18/2024 20
 Example: Power in Various Frequency Components
n Accumulated Power up
⚫ For example, power at Cn
to and including f = nfo
DC frequency
0 0.5 0.25
Co = 0.25 sinc(0) = 0.25
2 2
1 0.6366 0.4526
3 −0.212 0.4752
⚫ As the table shows, we 5 0.1273 0.4833
need to include 21 FS
7 −0.091 0.4874
coefficients to get 99%
9 0.0707 0.4899
power in the signal
11 −0.058 0.4916
⚫ Since each spectral
13 0.0490 0.4928
component is separated
15 −0.0424 0.4937
by 1 MHz, the 99%
power bandwidth of the 17 0.0374 0.4944
periodic pulse train is 19 -0.0335 0.4949
~21 MHz.
9/18/2024
21 0.0303 0.4954
21
Fourier Transform
⚫ Any continuous-time signal x(t) that has finite “energy”, i.e.,


 dt  
2
| x (t ) |
−

can be represented in the frequency domain via the Fourier

transform (FT)

X ( f ) =  x(t )e − j 2ft dt
−

⚫ In general, X(f) is a complex function of frequency f and can

be written as
X( f ) = X( f ) ej X(f )

where X ( f ) and X ( f ) are, respectively, called the

magnitude and the phase spectrum of the signal x(t)
9/18/2024 22
Fourier Transform (contd)
⚫ The signal x(t) can be recovered from its FT X(f) using the
inverse Fourier transform (IFT) formula

x(t ) =  X ( f )e j 2ft df
−


x(t ) ⎯⎯ → X ( f ) or X ( f ) = {x(t )} FT operation
−1
X ( f ) ⎯⎯→ x(t ) or x(t ) = −1{ X ( f )} Inverse FT operation

x(t ) ⎯ → X ( f ) FT or Inverse FT operation

9/18/2024 23
 FT of Rectangular Pulse A(t /  )
A

x(t ) = A(t /  )

− j f  j f 
  /2 e − e
X ( f ) = A  (t /  )e − j 2 ft dt = A e − j 2 ft dt = A
− − /2 − j 2 f
sin( f  )
=A = A sinc( f  )
f
Observe that the width
of the mainlobe of X(f)
increases as the pulse
width  narrows

9/18/2024 24
FT of Complex Exponential Signal e j 2 f t c

e j 2 fct ⎯

→ ( f − fc ) (*)
⚫ This can be verified by substituting in the inverse Fourier
transform formula as follows:
 
 df =   ( f − f c )e j 2f c t df = e j 2f c t
j 2ft
 ( f − f c )e
− −

⚫ The spectrum of a complex exponential signal contains

energy at only single frequency fc
⚫ (single tone)

⚫ Substituting fc = 0 into (*), we obtain the FT of a DC signal

as
1⎯
→ ( f )
9/18/2024 25
FT of Signum Signal sgn(t)
⚫ The signun signal sgn(t) can be expressed as
1, t0
sgn(t ) = 
−1, t0
− t

 e , t0
= lim   t
 →0

e , t0

⚫ The FT of sgn(t) is given by

 0
{sgn(t )} = lim  e e  t − j 2 ft
dt +  e − t e − j 2 ft dt 


 →0 
 − 0 

= lim  e
0
( − j 2  f ) t
dt +  e − ( + j 2 f )t dt 


 →0 
 − 0 
−4 j f 1
= lim 2 =
 →0  + 4 2 f 2 j f
9/18/2024 26
FT of Unit Step Signal
⚫ The unit step function u(t) can be expressed as
1 1
u (t ) = + sgn(t )
2 2
⚫ Taking the FT of both sides yields

1 1
U( f ) = ( f ) +
2 j 2 f

9/18/2024 27
FT of Unit Impulse Signal
x(t ) =  (t )

 
X ( f ) =   (t )e − j 2ft
dt =   (t )dt = 1
− −

⚫  the unit impulse signal contains all frequencies with equal

magnitudes as shown in the Figure

9/18/2024 28
Basic Fourier Transform Pairs

9/18/2024 29
Properties of Fourier Transform
⚫ There are a number of important properties of the Fourier
transform which are useful in the analysis and design of
communication and control systems

⚫ Linearity x(t ) ⎯→ X ( f )

y (t ) ⎯ →Y ( f )

x(t ) + y(t ) ⎯→


X ( f ) + Y ( f )
⚫ Taking the Fourier Transform of the left hand side yields

 x(t ) +  y (t ) ⎯→  [ x(t ) +  y (t )]e − j 2 ft dt

−
 
=   x(t )e − j 2 ft
dt +   y(t )e − j 2 ft dt
− −

=  X ( f ) + Y ( f )
9/18/2024 30
Properties of FT: Conjugate Symmetry
⚫ For real x(t),

X (− f ) =  x(t )e j 2ft dt = X * ( f )
−

⚫ Comparing magnitude and phase responses of both sides of

yields
X (− f ) = X ( f )

X (− f ) = − X ( f )
⚫ Thus X ( f ) and X ( f ) are even and odd functions of f,
respectively.

9/18/2024 31
Properties of FT
⚫ Time Shifting: x(t − to )  
⎯→ X ( f )e− j 2fto
⚫ This can be proved by using the inverse FT formula.
 
x(t − to ) =  X ( f )e j 2 f ( t −to )
df =  [ X ( f )e − j 2 fto ]e j 2 ft df = −1{ X ( f )e − j 2 fto }
− −

⚫ Note that the magnitude of the FT is unchanged by a time

− j 2 fto
shift. However, it introduces a linear phase shift of e

⚫ Frequency Translation: x(t )e j 2f c t  

⎯→ X ( f − fc )
⚫ Taking the Fourier transform of the left hand side yields
 
 dt =  x(t )e − j 2 ( f − fc )t dt = X ( f − f c )
j 2 f c t − j 2 ft
x(t )e e
− −

⚫  that multiplication of a signal x(t) by e j 2 fct translates its

frequency spectrum X(f) by the amount fc (to the right)
9/18/2024 32
Properties of FT: Convolution

x(t )  y(t ) ⎯→ X ( f )Y ( f )

⚫ From the definition of convolution operation


x(t )  y (t ) =  x( ) y (t −  )d
−

⚫ If we take the Fourier transform of the right hand side, and

exchange the order of integration, we get

  x( ) y (t −  )d e− j 2 ft dt =  d x( )   y (t −  )e − j 2 ft dt 
−  −  −  − 

= Y ( f ) x( )e− j 2 f  d = Y ( f ) X ( f )
−

⚫  the convolution operation in time-domain is equivalent to

multiplication in the frequency domain
9/18/2024 33
Properties of FT: Multiplication Property
⚫ This property is the dual of the convolution property. The
multiplication of two signals results in the convolution of their
spectra

x(t ) y(t ) ⎯ → X ( f ) Y ( f )

⚫ Modulation involves multiplication of a signal x(t) by a high-

frequency sinusoidal waveform. That is,
y (t ) = cos(2 f ct ) x(t )
⚫ Applying the multiplication property
1 1 
Y ( f ) =   ( f − fc ) +  ( f + fc )   X ( f )
2 2 
1 1
= X ( f − fc ) + X ( f + fc )
2 2 Spectrum X(f ) shifted
9/18/2024
by carrier frequency fc 34
Properties of FT: Time/Frequency Scaling
 1 f 
x(at ) ⎯→ X  
a a

⚫ Assume a > 0. Using the transformation of variables, u = at,

we have
 1  1  f 
x(at ) ⎯→ =  x(at )e
 − j 2 ft
dt =  x(u )e − j 2 ( f / a )u du = X  
− a − a a
⚫ Now with a < 0, substituting u = – |a|t yields

 1 
x(at ) ⎯→ =  x(− a t )e 
 − j 2 ft j 2 ( f / a ) u
dt = x(u )e du
− a −

1  f  1 f 
= X −  = X  
a  a a a
9/18/2024 35
Properties of Fourier Transform (contd)
⚫ The function x(at), for a > 0, is a time compressed (by a factor
a) version of x(t). On the other hand, a function X(f/a)
represents a function X(f) expanded by the same factor a.
⚫ The scaling property therefore states that compressing a signal
in time domain will stretch its Fourier transform. Similarly
stretching a time signal will compress its Fourier transform.
⚫ The result is intuitively satisfying since compression in time
by the factor a > 0 means that the function is varying rapidly
in time by the same amount
⚫ Consequently, the frequencies of its components will be
increased by the factor a. The converse can also be justified by
a similar argument.

9/18/2024 36
Properties of FT: Differentiation Property
d 
x(t ) ⎯→ j 2fX ( f )
dt
⚫ To prove this, we have
 
d d  d j 2 ft 
 df =  X ( f )  e
j 2 ft
x(t ) = X ( f )e df
dt dt − −  dt 

=   j 2 fX ( f )  e j 2 ft
df
−

⚫ That is, differentiation in time domain is equivalent to

multiplication by j2 f in the frequency domain

9/18/2024 37
Properties of FT: Parseval’s Relation
 
 −
x(t ) y * (t )dt =  X ( f )Y * ( f )df
−

⚫ To prove, substituting y *(t ) = − Y *( f )e − j 2 ft df and exchanging
the order of integration yields
 
 
dt
− −  −
− j 2 ft
x (t ) y *(t )dt = x (t ) Y *( f )e df


 
=  Y *( f )  x(t )e − j 2 ft  
dt df =  Y *( f ) X ( f ) df
−  −  −

⚫ We get well-known relationship for the energy of a signal in

time and frequency domains by letting y(t) = x(t) in Parseval’s
relation
 
E x =  x(t ) dt =  X ( f ) df
2 2
− −
(*)
9/18/2024 38
Fourier Transform Properties

9/18/2024 39
Fourier Transforms of Periodic Signals
⚫ The Fourier transform is strictly defined for finite energy
signals. However, we can extend its scope by allowing the FT
to include delta functions
⚫ Since a periodic signal can be expanded into exponential FS,
its FT can be obtained by taking the FT of the FS term by term
⚫ The FT expansion of a periodic function xp(t) is obtained as
  j 2 nf o t 

X p ( f ) =    Cn e   n
= C  e 
j 2 nf o t

 n =−  n =−


 n

j 2 nf ot
=  C  ( f − nf )
n o
x p (t ) = C e
n =− n =−

⚫  the FT of a periodic signal consists of impulses located at

harmonic frequencies of the signal. The weight of each
impulse equals FS coefficient, i.e., X p (nfo ) = Cn
9/18/2024 40
Fourier Transforms of Periodic Signals (contd)
⚫ Example: Fourier transform of a cosine wave
A j (2 fct + ) A − j (2 fct + )
x(t ) = A cos(2 f ct +  ) = e + e
2 2
⚫ Taking the FT of both sides and using Table, we get
A j A − j
X ( f ) = e  ( f − fc ) + e  ( f + fc )
2 2

⚫ Similarly, it can be shown that

A j A
A sin(2 f ct +  ) ⎯

→ e  ( f − f c ) − e − j  ( f + f c )
2j 2j

9/18/2024 41
FT of Periodic Impulse Train
⚫ The periodic impulse train with period To is given by

 p (t ) =   (t − nT )
n =−
o
FT also a periodic
⚫ The coefficient in its FS expansion is impulse train
To /2
1 1 1
Cn =   p (t )e − j 2 nfo t
dt =   (t )e − j 2 nf ot
dt = = f o
To To To −To /2
To

⚫ The FS can now be expressed as


 p (t ) = f o  e j 2 nf o t

n =−

⚫ Taking the FT of both sides


 p ( f ) =  p (t ) = f o   ( f − nf ) o
9/18/2024
n =− 42
Time-bandwidth Product
⚫ Recall the scaling property of the Fourier transform, which
states that the compression in the time domain is equivalent to
the expansion in the frequency domain, and vice versa
⚫ Thus, the frequency- and time-domain behaviors of a signal
are inversely related
⚫  a signal cannot be both time-limited and bandwidth-
limited
Time-limited Not Band-limited

A(t /  )
A

9/18/2024 43
Time-bandwidth Product (contd)
⚫ However, a signal can be “approximately” time-limited and
band-limited
⚫ That is, there exist numbers B > 0 and T > 0, such that |x(t)|
is small for |t|  T and |X(f)| is small for |f|  B
⚫ The product of a signal’s duration and its bandwidth is
constant.
⚫ Duration  Frequency Bandwidth = k

⚫ The constant k is determined by the precise definitions of

duration in the time domain and bandwidth in the frequency
domain
⚫ For a Gaussian pulse, if we use the RMS definitions of
duration and bandwidth of a signal, it can be shown that
1
f rms trms =
9/18/2024
4 44
Transmission of Signals Through LTI Systems
⚫ An Linear Time-invariant (LTI) system is completely
characterized in the time domain by its impulse response h(t)

y (t ) = x(t )  h(t ) =  x( )h(t −  )d
−

⚫ Applying the FT to both sides and using convolution property

Y ( f ) = X ( f )H ( f )

⚫  that the output of the system in the frequency domain is

given by multiplying the Fourier transform of the input by the
system frequency response H(f)

x (t ) LTI system y (t ) = x(t )  h(t )


X ( f ) h(t ) ⎯→ H ( f ) Y ( f ) = X ( f )H ( f )
9/18/2024 45
 Frequency Response of LTI Systems
⚫ H(f), in general, is a complex function of f . That is,

H ( f ) = H ( f ) e jH ( f )

where H ( f ) and H ( f ) are, respectively, called the

magnitude and the phase responses of the system
⚫ If h(t) is a real function, applying conjugation property of FT
H ( f ) = H (− f ) , H ( f ) = − H (− f )
even function of f odd function of f

⚫ In the frequency domain, the magnitude and the phase of the

system input and output are related by
Y( f ) = X( f )  H( f )
Y( f ) = X( f ) + H ( f )
9/18/2024 46
Example: First-order RC LP filter
⚫ The first-order RC LP filter is shown in Figure. The transfer
function of the network is given by R
Y( f ) 1/ j 2 fC 1
H( f ) = = =
X ( f ) R + 1/ j 2 fC 1 + j 2 fRC X ( f ) C Y( f )

1
=
1 + j ( f / f3dB )
1 −t / RC
3-dB cutoff
h(t ) = 2 f 3dB e −2 f3 dBt u (t ) = e u (t ) frequency
RC

9/18/2024 47
Distortionless Transmission
⚫ An LTI system is termed distortionless if it introduces the
same attenuation to all spectral components and offers linear
phase response over the frequency band of interest, that is,

 H o e− j 2 fto f1  f  f 2
H ideal ( f ) = 
0 otherwise

⚫ Substituting yields
Y ( f ) = X ( f ) H ideal ( f ) = H o X ( f )e− j 2 fto

⚫ Taking the inverse FT (IFT) of both sides, the output of a

distortionless LTI system due to an arbitrary input signal x(t) is
given by
y(t ) = H o x(t − to )
delayed and scaled replica of the input
9/18/2024 48
Group Delay
⚫ The group delay of an LTI system is defined as
1 d H( f )
g( f ) −
2 df
⚫ Represents time delay incurred by a spectral component at
frequency f as it passes through the LTI system
⚫ The phase response of an ideal filter is linear function of
frequency as given by
H ideal ( f ) = −2 fto , f1  f  f 2 , to is constant
⚫ For a linear phase LTI system, we obtain
1 −2 fto
g ( f ) = − = to = constant
2 df
⚫  all frequency components of the input signal undergo the
same time delay through the system  no distortion
9/18/2024 49
Ideal Filters
⚫ One key application of LTI systems is processing of signals in
order to enhance certain frequency components and to reject
certain others
⚫ For example, if a signal consists of a low-frequency
information-bearing message waveform and high-frequency
noise, we can employ a filter to reject the high frequencies
and thus remove the noise
⚫ An ideal filter designed to pass certain frequency components
should have a magnitude response that is constant and phase
response that is linear over these frequencies− called the
passband of the filter
⚫ The magnitude response of the ideal filter is zero over the
range of frequencies blocked by the filter − called the
stopband of the filter
9/18/2024 50
Ideal Filters (contd)

9/18/2024 51
Ideal Lowpass Filter
⚫ The magnitude response of an ideal lowpass (LP) filter is

 Ao −W  f  W
H LP ( f ) = 
0 otherwise

⚫ The passband of the filter is range of frequencies 0  f  W.

The range of frequencies f > W is the stopband of the filter
⚫ The frequency response of an ideal LP filter can now be
written as
H LP ( f ) = Ao( f / 2W )e− j 2fto

⚫ Taking the inverse FT yields

hLP (t ) = 2WAosinc[2W (t − to )]
9/18/2024 52
Ideal Highpass Filter
⚫ The magnitude response of an ideal highpass (HP) filter is

0 −W  f  W
H HP ( f ) = 
 Ao otherwise

⚫ The range of frequencies f  W is the stopband of the filter.

The range of frequencies f > W is the passband of the filter
⚫ The frequency response of an ideal HP filter can now be
written as
H HP ( f ) = Ao [1 − ( f / 2W )]e− j 2fto

⚫ Taking the inverse FT yields

H HP ( f ) = Ao 1 − ( f / 2W ) e− j 2 fto
9/18/2024 53
Ideal Bandpass Filter
⚫ The amplitude response of an ideal bandpass (BP) filter is
 Ao fc − W  f  fc + W
H BP ( f ) = 
0 otherwise

⚫ The range of frequencies fc − W  |f |  fc + W is the

passband of the filter. The range of frequencies |f | > fc + W
and |f | < fc − W are the stopband regions of the filter
⚫ The frequency response of an ideal BP filter can now be
written as
H BP ( f ) = H o ( f − f c ) + H o ( f + f c )
where
H o ( f ) = Ao( f / 2W )e− j 2fto
9/18/2024 54
Ideal Bandpass Filter (contd)
⚫ Ho(f) is a LP filter with impulse response

ho (t ) = 2WAosinc[2W (t − to )]
⚫ Since

H o ( f − f c ) ⎯ → ho (t )e j 2 fct

H o ( f + f c ) ⎯ → ho (t )e− j 2 fct

we can now write the impulse

response hBP(t) of the BP filter as
 e j 2 f c t + e − j 2 f c t 
hBP (t ) = 4WAosinc  2W (t − to )   
 2 
= 4WAosinc  2W (t − to )  cos(2 f ct )
9/18/2024 55
Ideal Bandstop Filter
⚫ The amplitude response of an ideal bandstop (BS) filter is
defined as
 Ao otherwise
H BS ( f ) = 
 0 fc − W  f  fc + W

⚫ The range of frequencies fc − W  |f |  fc + W is the

stopband of the filter. The range of frequencies |f | > fc + W
and |f | < fc − W are the passband regions of the filter.

9/18/2024 56
Power Density Spectrum
⚫ In the design of communication systems, we are interested in
power distribution of a power signal in the frequency domain
⚫ The problem in dealing with power signals in the frequency
domain is that their Fourier transform may not exist as they
have infinite energy
⚫ To overcome this problem, we define a new function xT(t) by
truncating x(t) outside the interval t > T/2

 x(t ) −T / 2  t  T / 2
xT (t ) = 
0 otherwise

⚫ xT(t) has finite energy as long as T is finite. Using Parseval’s

relation
 
ExT =  xT (t ) dt = 
2 2
X T ( f ) df
− −
9/18/2024 57
Power Density Spectrum (contd)
⚫ Since
T /2  
 x(t ) dt =  xT (t ) dt = 
2 2 2
X T ( f ) df
−T /2 − −

the average power of signal can be expressed as

T /2
1 1 
T → T  T → T −
Px = lim =
2 2
x (t ) dt lim X T ( f ) df
−T /2

⚫ Since x(t) is a power signal, the integral on the right hand side
exists in the limit as T →. Therefore, we can change the
order of integration and limit yielding
2 2
 XT ( f )  XT ( f )
Px = lim  df =  lim df
T → − T − T → T
9/18/2024 58
Power Spectral Density
⚫ The power spectral density (PSD) Gx ( f ) of power signal x(t)
is defined as
2
XT ( f )
Gx ( f ) lim
T → T

⚫ This allows us to express the normalized average power as


Px =  Gx ( f )df
−

⚫ Again Gx ( f o )f represents the power contained in 2 spectral

bands of width  f Hz centered at frequencies  fo
⚫ Thus Gx ( f ) may be interpreted as the power contained in
spectral components of x(t) centered at frequency f per Hz of
bandwidth. It is specified in units of W/Hz.
9/18/2024 59
PSD of Periodic Signals
⚫ For a periodic signal xp(t), the normalized average power is
given by

Px = 
2
Cn
n =−
2
⚫ Since Cn is power contained in the spectral component at f = nfo ,
the PSD Gx ( f ) of a periodic signal can be expressed as

Gx ( f ) = C  ( f − nf o )
2
n
n =−

⚫ Since FS coefficient Cn = X p (nf o ) , we can express the PSD as




2
Gx ( f ) = X p (nf o )  ( f − nf o )
n =−

9/18/2024 60
Time-average Autocorrelation Function
⚫ The time-average autocorrelation function of a power signal
x(t) is defined as
T /2
1
T → T 
Rx ( ) lim x(t ) x(t −  )dt
−T /2

⚫ The normalized average power Px of x(t) is related to Rx ( ) by

1 T /2
Px = lim  x(t ) dt = Rx (0)
2

T → T −T /2

⚫ It can be shown that the PSD of a power signal x(t) is the FT

of its time-average autocorrelation function

Gx ( f ) ⎯ → Rx ( )

9/18/2024 61
Response of LTI System: Deterministic Inputs
⚫ For a linear system with transfer function H(f), the output y(t)
in response to a deterministic input signal x(t) is given by
Y ( f ) = X ( f )H ( f )

Gy ( f ) = H ( f ) Gx ( f )
2
Gx ( f ) LTI System

X(f ) h (t ) ⎯ → H ( f ) Y( f ) = H( f )X ( f )

⚫ The PSD of a power signal y(t) can be written as

2
YT ( f )
Gy ( f ) = lim
T → T
⚫ Now
YT ( f ) = X T ( f ) H ( f )
9/18/2024 62
Response of LTI System (contd)
⚫ Substituting yields
2 2
X T ( f )H ( f ) XT ( f )
Gy ( f ) = lim = H ( f ) lim
2

T → T T → T
= H ( f ) Gx ( f )
2

⚫  that the output signal PSD in an LTI system depends on the

2
magnitude of H(f), and is given by H ( f ) times the input PSD
⚫ Taking inverse FT of both sides and applying the convolution
property, we obtain
Ry ( ) = h( )  h* (− )  Rx ( )

9/18/2024 63