MIT OpenCourseWare http://ocw.mit.

edu

16.36 Communication Systems Engineering

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.

Lecture 2: The Sampling Theorem

Eytan Modiano Slide 1

Sampling

Given a continuous time waveform, can we represent it using discrete samples?
How often should we sample? Can we reproduce the original waveform?

Eytan Modiano
Slide 2

The Fourier Transform

Frequency representation of signals

X( f ) =

!#

x(t)e ! j2"ft dt

x(t) =

!#

X( f )e j 2 "ft df

Notation: X(f) = F[x(t)] X(t) = F-1 [X(f)] x(t) X(f)

Eytan Modiano Slide 3

Unit impulse (t)

! (t) = 0, "t # 0

&

% $%

! (t)=1

$ ! (t)x(t) = x(0)
"# #

! (t)

F[! (t)]
1

$ ! (t " % )x(% ) = x( t)
"#

F[! (t)] =

"$

! (t)e " j 2#ft dt = e 0 = 1

! (t) & 1

Eytan Modiano Slide 4

Rectangle pulse

!(t )
| t |< 1 / 2 " 1 $ !(t) = #1 / 2 | t |= 1 / 2 $ 0 otherwise %
1/2 1

1/2

F[!(t)] = ) !(t)e& j 2 ' ft dt = )

1/2 &1/2

e& j 2 ' ft dt

e& j' f & e j' f Sin(' f ) = = Sinc( f ) & j2' f 'f

Eytan Modiano Slide 5

Properties of the Fourier transform

Linearity
x1(t) <=> X1(f), x2(t) <=>X2(f) => x1(t) + x2(t) <=> X1(f) + X2(f)

Duality
X(f) <=> x(t) => x(f) <=> X(-t) and x(-f)<=> X(t)

Time-shifting: x(t-) <=> X(f)e-j2f Scaling: F[(x(at)] = 1/|a| X(f/a) Convolution: x(t) <=> X(f), y(t) <=> Y(f) then,
F[x(t)*y(t)] = X(f)Y(f) Convolution in time corresponds to multiplication in frequency and vice versa
x(t) * y(t ) =

$ x(t ! ") y(" )d"

Eytan Modiano
Slide 6

Fourier transform properties (Modulation)

x(t) e j2!fo t " X( f # fo ) e jx + e # jx Now, cos( x) = 2 x(t)e j 2 !f o t + x( t)e # j 2!fo t x(t) cos( 2!fo t) = 2 Hence, x( t) cos(2!fot ) " X( f # fo ) + X( f + fo ) 2

Example: x(t)= sinc(t), F[sinc(t)] = (f) Y(t) = sinc(t)cos(2fot) <=> ((f-fo)+(f+fo))/2

1/2

Eytan Modiano Slide 7

-fo

+fo

More properties

Power content of signal

"

!"

| x( t) | dt =
2

"

!"

| X( f ) |2 df

Autocorrelation

Rx (! ) =

"#

x(t )x * (t " ! )dt

Rx (! ) %| X( f ) |2

Sampling

x(to ) = x(t )! (t " to )

x(t)

n= "#

$ ! (t " nt ) = sampled version of x(t)

1 F[ ! (t " nto )] = to n= "#

$ !( f " t

n
o

)]

Eytan Modiano
Slide 8

The Sampling Theorem

Band-limited signal
Bandwidth < W
X( f ) X( f ) = 0, for all f , | f | ! W

-w

Sampling Theorem: If we sample the signal at intervals Ts where Ts 1/ 2W then signal can be completely reconstructed from its samples using the formula
x(t) =

n =!"

# 2W T x( nT )sin c[2W (t ! nT )]
' ' s s s

"

Where,

W $ W' $

1 !W Ts

1 WithTs = = >x(t )= 2W
"

n= !"

"

x( nTs )sin c[(

t ! n )] Ts

Eytan Modiano Slide 9

x(t) =

n =!"

# x( 2W )sin c[2W( t ! 2W )]
n n

Proof

x! (t) = x(t) $ ! (t " nTs )

X! ( f ) = X( f ) * F[ $! (t " nTs )]
n= "#

1 F[ $! (t " nTs )] = Ts n= "#

n= "#

$ !(f "

n ) Ts

1 # n X! ( f ) = $ X( f " Ts ) Ts n="#

The Fourier transform of the sampled signal is a replication of the Fourier transform of the original separated by 1/Ts intervals

-2/Ts
Eytan Modiano
Slide 10

-1/Ts

-w

1/Ts

2/Ts

Proof, continued

If 1/Ts > 2W then the replicas of X(f) will not overlap and can be recovered How can we reconstruct the original signal?
Low pass lter the sampled signal

Ideal low pass lter is rectangular

H( f ) = Ts!(

f ) 2W

Now the recovered signal after low pass ltering f X( f ) = X! ( f )Ts"( ) 2W f x(t) = F #1[X! ( f )Ts "( )] 2W $ t x(t) = % x(nTs )Sinc( # n) Ts n =#$

Eytan Modiano
Slide 11

Notes about Sampling Theorem

When sampling at rate 2W the reconstruction lter must be a rectangular pulse

In practice we can sample at a rate somewhat greater than 2W which makes reconstruction lters that are easier to realize Given any set of arbitrary sample points that are 1/2W apart, can construct a continuous time signal band-limited to W Sampling using impulses is also not practical
Narrow pulses are difcult to implement In practice, sampling is done using small rectangular pulses or zero-order-hold

Eytan Modiano
Slide 12

Zero-Order Hold

A form of interpolation The sampled signal holds its value until the next sample time
sampled values held values

In principle, zero-order hold can be realized with a cascade of an impulse train sampling and an LTI system with rectangular impulse response
p(t) =
n = "#

$ ! (t " nT )
s

x(t)
Eytan Modiano Slide 13

x! (t) = x(t) $ ! (t " nTs )

h0 (t)
1

x0 (t)

x0 (t)
T

Reconstruction from zero-order hold

$ ! (t " nT )
s #

p(t) =

H(f)
h0 (t)
hr (t) x0 (t)
T

n = "#

x(t)

x! (t)
1

x(t)

We know from the sampling theorem that in order to reconstruct x(t) from the impulse train samples on the left (x(t)) the lter on the right (H(f)) must be an ideal rectangular lter f H ( f ) = Ts !( ) = Ts !(Ts f ) Hr ( f ) 2W
H ( f ) = Ts !( f $ f ) = H 0 ( f )H r ( f ) & Ts !( )e j# fT # f 2W & 2W % ( Hr ( f ) = sin(# fT ) " j# fT sin(# fT ) & H0( f ) = e ( ) #f & '

-w

Eytan Modiano
Slide 14

Aliasing

Sampling theorem requires that the signal be sampled at a frequency greater than twice its bandwidth When sampling at a frequency less than 2W, the replicas of the frequency spectrum overlap and cannot be separated using a low pass lter This is referred to as aliasing
Higher frequencies are reected only lower frequencies Signal cannot be recovered

The term aliasing refers to the fact that the higher frequency signals become indistinguishable from the lower frequency ones

Eytan Modiano
Slide 15