TE312: Introduction to

Digital Telecommunications

PART II
BASEBAND DIGITAL
TRANSMISSION

Lecture #8
Optimum Digital Receivers
Introduction

Points to be discussed in this lecture

Model of a Binary Digital Communication
System.

Geometric Representation of Signals.

Optimum Receiver Design

Implementation of Optimum Receivers.


Introduction

Reading Assignment

Simon Haykin, Digital Communications,
J ohn Wiley & Sons, Inc., 1988, Chapter 3,
Sec. 3.1~3.8.

Simon Haykin, Communication Systems, 4
th

Ed., J ohn Wiley & Sons, Inc., 2001, Chapter
5.
Model of a Binary Digital Communication
System


{ }
1 2
, b b
{ }
1 2
s , s


Modulator
Vector
Transmitter
Data
Source
( ) ( ) { }
1 2
, s t s t






Transmitter


( )
r t


Vector
Receiver
Detector
Estimate
r




Waveform
Channel

b
Noise
( )
w t

Receiver


Model of a Binary Digital Communication
System

Binary Data Source:

Data source emits a bit , 1,2
i
b i = at every
seconds where
b
T
1 2
bit 0, bit 1 b b = = .

is the bit duration (in Sec.) and
b
T
b
T 1/
b
R = is the
transmission bit rate (in bits per sec.).

Data source is characterized by the a priori
probability for
i
p , 1, 2.
i
b i =

[ ] ( ) [ ] ( )
1 1 2 2 1 2
bit 0 , bit 1 , 1.0 p P b p P b p p = = + =
Model of a Binary Digital Communication
System

Transmitter

Modulator maps the bit into one of two distinct
real-valued signals
i
b
1
( ) s t and
2
( ) s t of duration
with finite energy
b
T
1
E and
2
E , respectively.


2
1 1
0
2
2 2
0
0 ( )
0 ( )
b
b
T
T
E s t dt
E s t dt
< = <
< = <

Model of a Binary Digital Communication

System
Channel

The channel is linear and distortionless with
bandwidth much larger than the message signal.

The signal ( ), 1,2
i
s t i = is perturbed by zero-
mean, stationary, additive white Gaussian noise
(AWGN) process ( ) W t with sample function
( )
. w t

Received signal is expressed as ( ) r t

( ) ( ) ( )
i
r t s t w t = + 1,2 k = , 0
b
t T .
Model of a Binary Digital Communication
System

Receiver

The receiver estimates the transmitted bit in
the bit interval based on the observation it
makes on the received signal

b
( )
. r t

Since the received signal is corrupted with
noise, the estimated bit will be in error leading to
average probability of bit error
( )
r t
e
p

( )
e i
p P b b =
Geometric Representation of Signals

A set of energy signals
{ }
( )
i
s t , 1,2 i = can be
represented as a linear combination of a set of
2 N orthonormal basis functions { ( )}
j
t


1
( ) ( )
N
i ij j
j
s t s t
=
=

1,2 i = 0
b
t T

Orthonormality of ( )
j
t implies that
0
1
( ) ( )
0
b
T
m j
m j
t t dt
m j

=

Geometric Representation of Signals

Coefficients of expansion
ij
s are given by


0
( ) ( )
b
T
ij i j
s s t t dt =

1,2
1,2
i
j
=
=


( )
i
s t is generated from 1,2 i =
ij
s 1,2 j = by a bank
of multipliers followed by a summer.


ij
s , are generated from 1,2 j = ( )
i
s t , 1,2 i = by a
bank of correlators (multiplication followed by
integration).
2

Geometric Representation of Signals
( )
1
t












0
b
T
dt
1 i
s

( )
i
s t
0
b
T
dt

2 i
s
( )
2
t
2 i
s
1 i
s
( )
i
s t
( )
1
t
( )
2
t
Geometric Representation of Signals

Gram-Schmidt Orthogonalization Procedure

A set of orthonormal basis functions } { ( )
j
t
1,2 j = is obtained as follows:

Define
1
( ) s t (first signal is arbitrarily selected) as


( ) ( )
1 11 1 12 2 12
( ) where 0 s t s t s t s = + = , 0
b
t T

Obtain
1
( ) t by squaring and integrating

1
1
1
(
( )
) s t
t
E
=
1
E is the energy of
1
( ) s t
Geometric Representation of Signals


Define
2
( ) s t as


2 21 1 22 2
2 21 1 22 2
( ) ( ) ( )
( ) ( ) ( )
s t s t s t
s t s t s t


= +
=
, 0
b
t T

Obtain
2
( ) t by squaring both sides and
integrating from 0 to
b
T


2 21 1
2
2
2 21
( ) ( )
( )
s t s t
t
E s

2
E is the energy of
2
( ) s t

Geometric Representation of Signals

Define the correlation coefficient as

1 2
0
1 2
1
( ) ( )
b
T
s t s t dt
E E
=

21 2
s E =
Thus

( )
2 1
2
2
2 1
1 ( ) ( )
( )
1
s t s t
t
E E

=







Geometric Representation of Signals

The coefficients of expansion
11 12 21 22
, , , s s s s are
given by

11 1
s E = ,
12
0 s =

21 2
s E = ,
2
22 2
1 s E =

Each signal in the set
{ }
( )
i
s t , 1,2 i = , can be
uniquely determined by the signal vector
i
s

1
2
i
i
s
s

=


i
s
Geometric Representation of Signals


Each signal in the set
{ }
( )
i
s t 1,2 i = can be
represented geometrically by a point associated
with the signal vector in an N-dimensional
signal space to form a signal space diagram or
signal constellation.
i
s








( )
2
t
21
s
2
s
1
s
12
d
11
s
22
s
( )
1
t
Geometric Representation of Signals

( Energy of the signal )
i
s t is equal to the squared-
length of its vector

2
2
2 2
0
1
( )
b
T
ij i i
j
s s t dt E
=
= = =

i
s (Prove)

Energy of
1 2
( ) ( ) s t s t is equal to the square of the
Euclidian distance between their vectors

( )
2
2
2
2 2
12 1 2 1 2 1 2
0
1
[ ( ) ( )]
b
T
j j
j
d s s s t s t dt
=
= = =

s s
(Prove)
Geometric Representation of Signals

Example: Consider the signal set
{ }
1 2
( ), ( ) s t s t of
orthogonal signals below. Determine the
orthonormal signals
{ }
1 2
( ), ( ) t t and the signal
space diagram (constellation) for the signal set.









( )
1
s t
A
t
b
T
A
t
A
0
0
b
T
( )
2
s t
Optimum Receiver Design

Received signal for the AWGN channel is
given by
( ) r t
( ) ( ) ( )
i
r t s t w t = +
0
1,2
b
t T
i

=


: sample function of the received random
( ) R t
( ) r t
process .
( )
i
s t : transmitted signal.
( ) w t : sample function of an AWGN process
of zero-mean and PSD ( ) W t
0
/2. N


Optimum Receiver Design

Signal point associated with the signal vector
represents the transmitted signal ( )
i
s
i
s t .

Signal point associated with observation vector
r represents the received signal
( )
. r t

, 1,2. i = + =
i
r s w

r and are samples of random vectors and w R W


1 1 1
2 2 2
i
i
s w r
r s w
+

= =

+

r
Optimum Receiver Design

Each
2
r is a sample value of a Gaussian
random variable
2
R R , respectively.
1
and r
and
1

Mean values of R are
1 2
and R


[ ] [ ]
1 2
1 1 2 2
, , 1,2
R i R i
m E R s m E R s i = = = = =

Variances of R are
1 2
and R


1 2
2 2
0
2
R R
N
= = (Prove)

Optimum Receiver Design

1 2
R are mutually uncorrelated, hence are
statistically independent
and R

[ ] ( )( )
1 2 1 1 2 2
Cov 0
i i
R R E R s R s = = (Prove)

and R

Conditional pdfs of
2
R given
1
( )
i
s t

( ) ( )
1
2
1 1 1
0
0
1 1
| ( ) exp
R i i
f r s t r s
N
N

=



( ) ( )
2
2
2 2 2
0
0
1 1
| ( ) exp
R i i
f r s t r s
N
N

=



Optimum Receiver Design

Conditional probability density function of given
( )
R
i
s t (likelihood function)


( )
( )
( )
( )
( )
( )
1 2
1 2
| | | t
i R i R i
f s t f r s t f r s =
R
r 1,2 i =

( )
2
2
1
j
j
r s
=

( )
1
0
0
1
exp
j i
N
N

Detection Problem:

Perform mapping from to an estimate of , with
minimum probability of bit error .
r
i
b

b
e
p
Optimum Receiver Design

Probability of bit error

( ) ( )
, 1 sent|
e i i
p b P b = r r

( )
sent|
i
P b r is the a posteriori probability of the
binary integer .
i
b

Maximum a posteriori (MAP) probability optimum
decision rule is

1
1 2

set if
( sent| ) ( sent| )
b b
P b P b
=
r r

Optimum Receiver Design

Using Bayes Rule

( )
( )
( )
|
sent|
i i
i
p f b
P b
f
=
R
R
r
r
r


The MAP decision rule becomes

( ) ( )
1
1 1 2 2

set if
| sent | sent
b b
p f b p f b
=

R R
r r




Optimum Receiver Design

Substituting for
( ) ( )
1 2
| sent and | sent f b f b
R R
r r and
simplifying yield

( ) ( )
( ) ( )
2 2
1 11 2 12
2 2
1 21 2 22
r s

+
1
2
1

set if
1
exp
1
exp
o
o
b b
p
N
p
N
r s r s
r s
=

Optimum Receiver Design

Taking the on both sides and simplifying log
e

1
2 2
1
2

set if
ln
o
b b
p
N
p
=
< +
1 2
r s r s


For a case where
1 2
p p p = = , the decision rule
becomes (Maximum Likelyhood - ML Rule)

1
2 2

set if b b =
<
1 2
r s r s

Optimum Receiver Design

Physical interpretation: The optimum ML receiver
determines the distance from the received signal
point to signal points and and selects the
signal
r
1
s
2
s
( )
i
s t with minimum distance.



( )
1 2
, r r = r




( )
2
t
( )
1
t
( )
1 11 12
, s s = s
( )
21 22
, s s =
2
s
2
d
1
d
Optimum Receiver Design

The expansion of the MAP decision rule leads to

( ) ( )
1
1 2 1 11 21 2 12 22
2
2 2 ln
o
p
E E r s s r s s N
p
< + +

The line partitions the signal space diagram into
two decision regions and , respectively.
1
Z
2
Z

If falls in , the receiver decides in favor of bit
0, otherwise it decides in favor of bit 1. If r falls
on the boundary, the receiver flips a coin in
making a decision.
r
1
Z
Optimum Receiver Design

Example: Consider the signal set
{ }
1 2
( ), ( ) s t s t of
orthogonal signals below. Determine and show
the decision regions for the case where (a)
2
p p > (b)
2
p p
1 1
< (c)
1 2
p p = .









A
t
b
T
A
t
A
0
0
b
T
( )
1
s t
( )
2
s t
Optimum Receiver Implementation

Determine the received signal vector elements
.
1 2
and r r

Compute the decision rule based on the received
signal vector elements
2
and r r and vector
elements of two signals ,
1
1,2; 1,2
ij
s i j = = .


1 1 2 2
ln
2 2
i o
i i i
E N
r s r s p + + , 1,2. i =

Choose the largest.

Optimum Receiver Implementation

Correlation Receiver


























( )
1
t
1
r
2
r
b
t T =
b
t T =
( )
2
2 2 2
ln
o
N E
p

i
r s


Choose
the
largest

b
( )
1
1 2 2
ln
o
N E
p
( )
r t
( )
2
t
0
b
T
dt

0
b
T
dt

Optimum Receiver Implementation
Example: Consider the signal set
{ }
1 2
( ), ( ) s t s t of
orthogonal signals below. Design an optimum
receiver for this signal set assuming an AWGN
channel and (a)
1 2
0.4, 0.6 p p = = (b)
1 2
p p = .









A
t
b
T
A
t
A
0
0
b
T
( )
1
s t
( )
2
s t

Optimum Receiver Implementation

Matched Filter Receiver

Output
( )
j
y t of a linear filter with impulse response
( )
and input
j
h t
( ) ( ) ( )
i
r t s t w t = + is given by

( ) ( ) ( ) ( ) ( )
j j j
y t r t h t r h t d

= =

Substitute
( )
j b
T t for
( )
j
h t , output
( )
j
y t becomes

( ) ( ) ( ) ( ) ( )
j j b j b
y t r t T t r T t d

= = +

Optimum Receiver Implementation

Matched Filter Receiver

Output
( )
j
y t at
b
t T = is given by

( ) ( ) ( )
j b j j
y T r d r

= =

The filter with the impulse response
( ) ( )
j j b
h t T t =
is called a matched filter.

A receiver that uses matched filters in place of
correlators is called a matched filter receiver.

Optimum Receiver Implementation

Matched Filter Receiver




















( )
1 b
T t
( )
r t
1
r
2
r
( )
2
2 2 2
ln
o
N E
p
b
t T =
b
t T =

i
r s
( )
2 b
T t


Choose
the
largest

b
( )
1
1 2 2
ln
o
N E
p