You are on page 1of 31

1

Chapter 5 Signal-Space Analysis


Signal space analysis provides a
mathematically elegant and highly insightful
tool for the study of data transmission.
Po- Ning Chen@cm.nct u Chapt er 5- 2
5.1 Introduction
Statistical model for a genetic digital communication
system
Message source: A priori probabilities for information
source
Transmitter: The transmitter takes the message source
output m
i
and codes it into a distinct signal s
i
(t) suitable
for transmission over the channel. So:
M i m P p
i i
,..., 2 , 1 for ) ( = =
M i t s P m P p
i i i
,..., 2 , 1 for )) ( ( ) ( = = =
2
Po- Ning Chen@cm.nct u Chapt er 5- 3
5.1 Introduction
s
i
(t) must be a real-valued energy signal (i.e., a
signal with finite energy) with duration T.
Channel: The channel is assumed (in this text) linear
and with a bandwidth wide enough to pass s
i
(t) with no
distortion.
Zero-mean additive white Gaussian noise (AWGN)
is also assumed (to facilitate analysis).
. ) (
0
2
< =

T
i i
dt t s E
Po- Ning Chen@cm.nct u Chapt er 5- 4
5.1 A mathematical model
We can then simplify the previous system block diagram to:
Upon the receipt of x(t) for a duration of T, the receiver makes
the best estimate of m
i
. (We havent defined what the best is.)
3
Po- Ning Chen@cm.nct u Chapt er 5- 5
5.1 Criterion for best decision
Best = Minimization of the average probability of symbol
error.
This is optimum in the minimum probability of error
sense.
Based on this criterion, we can begin to design the
receiver that can give the best decision.

=
=
M
i
i i i e
m m m P p P
1
) | (
Po- Ning Chen@cm.nct u Chapt er 5- 6
5.2 Geometric representation of signals
Signal space concept
Vectorization of the (discrete or continuous) signals
removes the redundancy in the signals, and provides a
compact representation for them.
Determination of the vectorization basis
Gram-Schmidt orthogonalization procedure
4
Po- Ning Chen@cm.nct u Chapt er 5- 7
5.2 Gram-Schmidt orthogonalization procedure
( ) . basis l orthonorma an form , , , Then ) step (
.
|| ||
Set
. ) ( ) ( ) ( Let
,..., 4 , 3 For ) step (
.
|| ||
Set . ) ( ) step (
.
|| ||
Let ) step (
for them? basis l orthonorma an find how to , , , , Given
2 1
'
'
1 1 2 2 1 1
'
'
2
'
2
2 1 1 2 2
'
2
1
1
1
2 1
k
i
i
i
i i i i i i i i i
k
u u u iv
u
u
u
u u v u u v u u v v u
i iii
u
u
u u u v v u ii
v
v
u i
v v v
r
L
r r
r
r
r
r r r
L
r r r r r r r r
r
r
r r r r r r
r
r
r
r
K
r r
=
=
=
= =
=

. || || || || || || : inequality triangle (vii)
others. of n combinatio
linear a as d represente be can none if t, independen linearly (vi)
1. norm indivual and 0, product inner if al, orthosnorm v) (
|| || : norm (iv)
0. product inter if , orthogonal (iii)
@ : product inner (ii)
) ( : vector (i)
: Properties
2 1 2 1
2 2
1
1
2 1 2 1
1
v v v v
v v v
v v v v
,v , v v
n
n
i
i i
n
r r r r
L
r
r r
K
r
+ +
= =
+ + =
=
=
=

=
Po- Ning Chen@cm.nct u Chapt er 5- 8
5
Po- Ning Chen@cm.nct u Chapt er 5- 9
| |
. A of solution
: eigenvalue w.r.t. rs eigenvecto xiii) (
. 0 I - A det of solution
: A matrix w.r.t. s eigenvalue xii) (
.
: A matrix w.r.t. on ansformati Matrix tr xi) (
. || || || || || ||
, orthogonal If : property an Pythagore x) (
. 2 || || || || || ||
: square norm xi) (
. if holds, equality ith w
|| || || || | |
: inequality Schwartz Cauchy viii) (
2 1
2
2
2
1
2
2 1
2 1
2
2
2
1
2
2 1
2 1
2 1 2 1
v v v
v A v
v v v v
v v v v v v
v a v
v v v v
r r r
r r
r r r r
r r r r r r
r r
r r r r



=
=
=
+ = +
+ + = +
=

Po- Ning Chen@cm.nct u Chapt er 5- 10
5.2 Signal space concept for continuous functions
. || ) ( || || ) ( || ) ( ) ( : inequality triangle (vii)
others. of n combinatio
linear a as d represente be can none if t vectors, independen linearly (vi)
1. norm indivual and 0, product inner if l, orthonorma v) (
| ) ( | ) ( : norm (iv)
0. product inter if , orthogonal (iii)
. ) ( ) ( ) ( ), ( @ : product inner (ii)
) ( : signal alued) (complex v (i)
functions continuous for Properties
2
*
t z t z t z t z
dt t z t z
dt t z t z t z t z
t z
b
a
b
a
+ +
= =
=
=
>= <

6
. ) ( ) ( ) , ( as d represente be can ) , ( and
d ) ( ) , ( ) ( of )} ( { and solutions
: ) ( function continuous a w.r.t. ions eigenfunct and s eigenvalue (xii.a)
.) (Recall ) ( ) , ( ) (
: ) ( function a w.r.t. tion Transforma xi) (
. || ) ( || || ) ( || || ) ( ) ( || , orthogonal If : property an Pythagore x) (
. ) ( ), ( ) ( ), ( || ) ( || || ) ( || || ) ( ) ( ||
: square norm xi) (
number.) complex a is ( ). ( ) ( if holds, equality ith w
|| ) ( || || ) ( || ) ( ), (
: inequality Schwartz Cauchy viii) (
1
1
2
1
1
2 2 2
2 2 2




k k
k
k
k
b
a
k k k k k
i
n
i
jn j
b
a
t t C t C
t C t t
t, C
v a v d z t C t z
t, C
t z t z t z t z
t z t z t z t z t z t z t z t z
a t z a t z
t z t z t z t z
=
=
= =
+ = +
> < + > < + + = +
=
> <

=
=
Po- Ning Chen@cm.nct u Chapt er 5- 11
Po- Ning Chen@cm.nct u Chapt er 5- 12
. } { of choices all for ) ( ) ( ) ( that possible is it
then space, the span not does )} ( { set l orthonorma If (xii.c)
. ) ( ) ( here w
, 0 ), ( ) (
Then ). ( span can that )} ( { basis l orthonorma
of set a and )} , 0 [ ), ( { function tic determinis a Give (xii.b)
1
0
1
0
0
1
K k k
K
k
k k
K k k
k
T
k
k
k k
k k
a t s t a t s
t
dt t t s a
T t t a t s
t s t
T t t s
<
=
<

=
<
=
=
< =

7
Po- Ning Chen@cm.nct u Chapt er 5- 13
Problem : How to minimize the energy of
? ) ( ) ( ) ( t s t s t e =
( )
| |
. ) ( ) ( . 0 2 ) ( ) ( 2
) ( ) ( 2 ) ( ) ( 2
) ( ) ( ) ( 2
] ) ( ) ( [
) (

, ) ( minimize that } { ts coefficien select the To
1
1
2
1
2
2


= = + =
+ =

|
.
|

\
|

=

dt t t s a a dt t t s
dt t t a dt t t s
dt t t a t s
a
dt t a t s
a
dt t e
dt t e a
j j j j
K
k
j k k j
j
K
k
k k
j
K
k
k k
j
k


Po- Ning Chen@cm.nct u Chapt er 5- 14


) (t s
) (
1
t
) (
2
t
) ( t s
) ( ), (
2 2
t t s a =
) ( ), (
1 1
t t s a =
) (t e
0 ) ( ) ( ) ( ), ( Hence, = =


dt t s t e t s t e
Interpretation
a
j
is the projection of s(t) onto the
j
(t)-axis.
(a
j
)
2
is the energy-projection of s(t) onto the
j
(t)-axis.
8
Po- Ning Chen@cm.nct u Chapt er 5- 15
) (t s
) (
1
t
) (
2
t
) ( t s
) ( ), (
2 2
t t s a =
) ( ), (
1 1
t t s a =
) (t e


=
=
|
.
|

\
|
=
= = =
= =
K
k
k
k
K
k
k
K
k
k k
a dt t s
dt t t s a dt t s
dt t s t a dt t s
dt t s t s t s dt t s t e dt t s t e
dt t s t e dt t s t e dt t s t s t e dt t e
1
2 2
1
2
1
2
2
) (
) ( ) ( ) (
) ( ) ( ) (
) ( )] ( ) ( [ ) ( ) ( 0 ) ( ) (
) ( ) ( ) ( ) ( )] ( ) ( )[ ( ) (

. ) ( Notably,
1
2 2


=
K
k
k
a dt t s
Po- Ning Chen@cm.nct u Chapt er 5- 16
5.2 Signal space concept for continuous functions
Completeness
If every finite energy signal satisfies
Example. Fourier series
set. l orthonorma a is )} ( {
, ) (
1
1
2 2
complete t
a dt t s
K k k
K
k
k

. ] , 0 [ over defined
signals for set l orthonorma complete a is

2
sin
2
,
2
cos
2

0
T
T
kt
T T
kt
T
k
)
`

|
.
|

\
|
|
.
|

\
|

9
Po- Ning Chen@cm.nct u Chapt er 5- 17
5.2 Gram-Schmidt orthogonalization procedure
( ) . basis l orthonorma an form ) ( , ), ( ), ( Then ) step (
.
|| ) ( ||
) (
) ( Set
). ( ) ( ), ( ) ( ) ( ), ( ) ( ) ( Let
,..., 4 , 3 For ) step (
.
|| ) ( ||
) (
) ( Set ). ( )) ( ), ( ( ) ( ) ( ) step (
.
|| ) ( ||
) (
) ( Let ) step (
for them? basis l orthonorma an find how to ), ( , ), ( ), ( Given
2 1
'
'
1 1 1 1
'
'
2
'
2
2 1 1 2 2
'
2
1
1
1
2 1
t u t u t u iv
t u
t u
t u
t u t u t v t u t u t v t v t u
i iii
t u
t u
t u t u t u t v t v t u ii
t v
t v
t u i
t v t v t v
k
i
i
i
i i i i i i
k
L
L
K
=
=
=
= =
=

Po- Ning Chen@cm.nct u Chapt er 5- 18
5.2 Geometric representation of signals
M i T t t f s t s
N
j
j ij i
,..., 2 , 1 , 0 , ) ( ) (
1
= < =

=
N j M i
dt t f t s s
T
j i ij
,..., 2 , 1 , ,... 2 , 1
, ) ( ) (
0
= =
=

l orthonorma } {
1
N
i i
f
=
10
Po- Ning Chen@cm.nct u Chapt er 5- 19
5.2 Geometric representation of signals
Through the signal space concept, s
i
(t) (where 1 i M)
can be unambiguously represented by an N-dimensional
signal vector (s
i1
, s
i2
,, s
iN
) over an N-dimensional signal
space.
The design of transmitters becomes the selection of M
points over the signal space, and the receivers make a guess
about which of the M points was transmitted.
In the N-dimensional signal space,
length of the vector = energy of the signal
angle between vectors = energy correlation between
signals

=
= =
N
j
ij
T
i i
s dt t s t s
1
2
0
2 2
) ( || ) ( ||
|| ) ( || || ) ( ||
) ( ), (
) cos(
t s t s
t s t s
k i
k i
ik

=
Po- Ning Chen@cm.nct u Chapt er 5- 20
5.2 Geometric representation of signals
the angle between vectors is independent of the basis
used.
From this view,
the transmitter may be viewed as a synthesizer, which
synthesizes the transmitted signal by a bank of N
multipliers.
the receiver may be viewed as an analyzer, which
correlates (product-integrate) the common input into
individual informational signal.
11
Po- Ning Chen@cm.nct u Chapt er 5- 21
5.2 Geometric representation
of energy signals
Illustration the geometric
representation of signals for the
case when N = 2 and M= 3
Po- Ning Chen@cm.nct u Chapt er 5- 22
5.2 Euclidean distance
After vectorization, we can then calculate the Euclidean
distance between two signals:

=
= =
N
j
kj ij k i
T
k i
s s t s t s dt t s t s
1
2 2
0
2
) ( || ) ( ) ( || )) ( ) ( (
. ) ( ), ( means lity orthonorma say that may We : ns Applicatio
, 0
, 1
: function delta Kronecker
ij j i
ij
t t
j i
j i

=
=
12
Po- Ning Chen@cm.nct u Chapt er 5- 23
Example 5.1 Schwarz inequality
Cauchy-Schwarz inequality and angle between signals
Cauchy-Schwarz inequality said that
Also, angles between signals give that
Hence, Cauchy-Schwarz inequality is equivalently
stated as:
). ( ) ( if holds equality with || ) ( || || ) ( || ) ( ), (
2 1
2
2
2
1
2
2 1
t cs t s t s t s t s t s =
|| ) ( || || ) ( ||
) ( ), (
) cos(
2 1
2 1
12
t s t s
t s t s

=
or 0 if holds equality with 1 | ) cos( |
12
2
12
=
Po- Ning Chen@cm.nct u Chapt er 5- 24
5.2 Basis
The (complete) orthonormal basis for a signal space is not
unique!
So the synthesizer and analyzer for the transmission of
the same informational messages are not unique!
One way to determine a set of orthonormal basis is the
Gram-Schmidt orthogonalization procedure.
Try and practice Example 5.2 yourself!
13
Po- Ning Chen@cm.nct u Chapt er 5- 25
5.3 Conversion of the continuous AWGN channel
into a vector channel
Influence of the AWGN noise to the signal space concept
where w(t) is zero-mean AWGN with PSD N
0
/2.
After the correlator at the receiver, we obtain:
Or equivalently,
Notably, there is no information loss
by the signal space representation.
) ( ) ( ) ( t w t s t x
i
+ =
.
j ij j
w s x + =
) ( ), ( ) ( ), ( ) ( ), ( t f t w t f t s t f t x
j j i j
+ =
) (t x
) (t f
N
) (
1
t f
N
x
,
,
1
x
Po- Ning Chen@cm.nct u Chapt er 5- 26
5.3 Conversion of the continuous AWGN channel
into a vector channel
Statistics of {w
j
}
Since {s
ij
} is deterministic, the distribution of x is a mean-
shift of that of w.
Observe that w is Gaussian distributed because w(t) is
AWGN. The distribution of w can therefore be determined
by its mean vector and covariance matrix.
(
(
(

+
(
(
(

=
(
(
(

N iN
i
N
w
w
s
s
x
x
M M M
1 1 1
14
Po- Ning Chen@cm.nct u Chapt er 5- 27
5.3 Conversion of the continuous AWGN channel
into a vector channel
Mean
Covariance
| | 0 ) ( )] ( [ ) ( ) ( ] [
0 0
= = =

T
j
T
j j
dt t f t w E dt t f t w E w E
( )( ) | |
ij
T
j i
T T
j i
T T
j i
T
j
T
i j i
N
dt t f t f
N
dsdt t f s f t s
N
dsdt t f s f t w s w E
dt t f t w ds s f s w E w w E

2
) ( ) (
2
) ( ) ( ) (
2
) ( ) ( )] ( ) ( [
) ( ) ( ) ( ) ( ] [
0
0
0
0 0
0
0 0
0 0
= =
=
=
=




Po- Ning Chen@cm.nct u Chapt er 5- 28
5.3 Conversion of the continuous AWGN channel
into a vector channel
As a result, [w
1
, w
2
, , w
N
] are zero-mean i.i.d. Gaussian
distributed with variance N
0
/2.
This shows that x is independent Gaussian distributed with
common variance N
0
/2 and mean vector s
i
= [s
i1
, s
i2
, , s
iN
].
Equivalently,

=
(

=
N
j
ij j i
s x
N N
f
1
2
0 0
) (
1
exp
1
) | (

s x
15
Po- Ning Chen@cm.nct u Chapt er 5- 29
5.3 Conversion of the continuous AWGN channel
into a vector channel
Remainder term in noise
It is possible that
However, it can be shown that w(t) is orthogonal to s
i
(t)
for 1 i M. Hence, w(t) will not affect the decision
error rate on message i.
0 ) ( ) ( ) ( '
1
=

=
N
i
i i
t f w t w t w
1. y probabilit with 0 ) ( ), ( ' = t s t w
i
Po- Ning Chen@cm.nct u Chapt er 5- 30
5.4 Likelihood functions
An equivalent signal-space channel model
The best decision function d( ) that minimizes the decision
error is:
This is the maximum a posteriori probability (MAP)
decision rule.
} ,..., { ) ( ) ( 1 ,
1 M i
m m d m m c M i m m = + = = = x w s x s
} | { max arg
1 all for } | { } | { if , ) (
} ,..., {
1
x
x x x
m P
M k m P m P m d
M
m m m
k i i

=
=
16
Po- Ning Chen@cm.nct u Chapt er 5- 31
5.4 Likelihood functions
With equal prior,
{ }
{ } ) | ( ),..., | ( ), | ( max arg
) (
) ( } | {
,...,
) (
) ( } | {
,
) (
) ( } | {
max arg
/ 1
) ( } | {
,...,
/ 1
) ( } | {
,
/ 1
) ( } | {
max arg
} | { },..., | { }, | { max arg
} | { max arg ) (
2 1
2
2
1
1
2 1
2 1
} ,..., {
1
M
M
M
M
M
m m m
m f m f m f
m P
f m P
m P
f m P
m P
f m P
M
f m P
M
f m P
M
f m P
m P m P m P
m P d
M
x x x
x x x x x x
x x x x x x
x x x
x x
=
)
`

=
)
`

=
=
=

f(x|m
i
) is named the likelihood function given that m
i
is transmitted
Hence, the above rule is named the maximum-likelihood decision rule.
Po- Ning Chen@cm.nct u Chapt er 5- 32
5.4 Likelihood functions
MAP rule = ML rule, if equal prior is assumed.
In practice, it is more convenient to work on the log-
likelihood functions, defined by
Why log-likelihood functions are more convenient? The
decision function becomes sum of Euclidean distances in
AWGN channel.
{ }
{ } ) | ( log ),..., | ( log ), | ( log max arg
) | ( ),..., | ( ), | ( max arg ) (
2 1
2 1
M
M
m f m f m f
m f m f m f d
x x x
x x x x
=
=
17
2
1
1
2
1
1
2
0
0
1
1
2
0 0
1
1 1
|| || min arg
) ( min arg
) (
1
log
2
1
max arg
) (
1
exp
1
log max arg
) | ( log max arg ) | ( log max arg ) (
i
M i
N
j
ij j
M i
N
j
ij j
M i
N
j
ij j
M i
i
M i
i
M i
s x
s x
N
N
s x
N N
f m f d
s x
s x x x
=
=
|
|
.
|

\
|
=
(

=
= =

=

=

=

Po- Ning Chen@cm.nct u Chapt er 5- 33


Upon receipt of received signal point x, find the signal
point s
i
that is closest in Euclidean distance to x.
Po- Ning Chen@cm.nct u Chapt er 5- 34
5.5 Coherent detection of signals in noise:
Maximum likelihood decoding
Signal constellation
Set of M signal points in the signal space
Example. Signal constellation for 2B1Q code
decision region
for s
1
decision region
for s
2
decision region
for s
3
decision region
for s
4
18
Po- Ning Chen@cm.nct u Chapt er 5- 35
5.5 Coherent detection of signals in noise:
Maximum likelihood decoding
Decision regions for
N = 2 and M= 4
Po- Ning Chen@cm.nct u Chapt er 5- 36
5.5 Coherent detection of signals in noise:
Maximum likelihood decoding
Usually, s
1
, s
2
, , s
M
are named the message points.
The received signal point x then wanders about the
transmitted message point in a Gaussian-distributed random
fashion.
19
Po- Ning Chen@cm.nct u Chapt er 5- 37
5.5 Coherent detection of signals in noise:
Maximum likelihood decoding
Constant-energy signal constellation
The ML decision rule can be reduced to an inner-
product.
( )
( )
constant. is if , , max arg
, 2 min arg
|| || , 2 || || min arg
|| || min arg ) (
1
1
2 2
1
2
1
i i
M i
i i
M i
i i
M i
i
M i
E
E
d
s x
s x
s s x x
s x x




=
+ =
+ =
=
Po- Ning Chen@cm.nct u Chapt er 5- 38
5.6 Correlation receiver
If signals do not have equal energy, we can use
to implement the ML rule.
The receiver is coherent because the receiver requires to
be in perfect synchronization with the transmitter (more
specifically, the integration must begin at the right time
instance).
.
2
1
, max arg ) (
1
|
.
|

\
|
=

i i
M i
E d s x x
20
demodulator
or detector
decision maker
Po- Ning Chen@cm.nct u Chapt er 5- 39
N product-
integrators or
correlators
Correlation receiver
1
x
2
x
N
x
matched filter
decision maker
Po- Ning Chen@cm.nct u Chapt er 5- 40
N product-
integrators or
correlators
1
x
2
x
N
x
21
Po- Ning Chen@cm.nct u Chapt er 5- 41
5.6 Equivalence of correlation and matched filter
receivers
The correlator and matched filter can be made equivalent.
Specifically,



= = d T h x dt t t x x
i
T
i i
) ( ) ( ) ( ) (
0
. 0 outside zero is ) ( implicitly and ) ( ) ( if T t t t T t h
i i i
=
Po- Ning Chen@cm.nct u Chapt er 5- 42
5.7 Probability of symbol error
Average probability of symbol error
{ }


=
=

=
=
=
=
= =
= = =
M
i
Z
i
M
i
i j
i j M j
i
M
i
i i
M
i
i i i c e
i
d f
M
m
M
m m d P
M
m m d P m P P P
1
1
2
, 1
2
1
1
) | (
1
1
ed transmitt || || min || || Pr
1
1
ed) transmitt | ) ( (
1
1
ed) transmitt | ) ( ( ) ( 1 1
x s x
s x s x
x
x
{ }. || || min || :|| where
2
, 1
2
j
i j M j
i
N
i
Z s x s x x =

22
Po- Ning Chen@cm.nct u Chapt er 5- 43
5.7 Invariance of probability of symbol error
Probability of symbol error is invariant with respect to basis
change (i.e., rotation and translation of the signal space).
Specifically, SER (symbol error rate) only depends on the
relative Euclidean distances between the message points.
{ }

=

=
M
i
i j
i j M j
i e
m
M
P
1
2
, 1
2
ed transmitt || || min || || Pr
1
1 s x s x
Po- Ning Chen@cm.nct u Chapt er 5- 44
5.7 Invariance of probability of symbol error
Specifically, if Q is a reversible transform (matrix), such as
rotation, then
If a signal constellation is rotated by an orthonormal
transformation, where Q is an orthonormal matrix, then
the probability of symbol error P
e
incurred in maximum
likelihood signal detection over an AWGN channel is
completely unchanged.
{ }
{ }
2
, 1
2
2
, 1
2
|| || min || :||
|| || min || :||
j
i j M j
i
N
j
i j M j
i
N
s x s x x
s x s x x
Q Q Q Q =



23
Po- Ning Chen@cm.nct u Chapt er 5- 45
5.7 Invariance of probability of symbol error
A pair of signal constellation for illustrating the principle of
rotational invariance.
Po- Ning Chen@cm.nct u Chapt er 5- 46
5.7 Invariance of probability of symbol error
The invariance in SER for translation can be likewise
proved.
Is the transmission power the same for both
constellation?
24
Po- Ning Chen@cm.nct u Chapt er 5- 47
5.7 Minimum energy signals
Since SER is invariant to rotation and translation, we may
rotate and translate the signal constellation to minimize the
transmission power without affecting SER.
But Q does not change the norm (i.e., transmission
power). Thus, we only need to determine the right a.

=
=
M
i
i i g
p E
1
2
|| || s
minimized. is || ) ( || ) , ( that such and Find
1
2

=
=
M
i
i i g
p E a s a a Q Q Q
Po- Ning Chen@cm.nct u Chapt er 5- 48
5.7 Minimum energy signals
Determine the optimal a.
( )
2
1 1
2
1
2 2
1
2
|| || 2 || ||
|| || 2 || ||
|| || ) (
a s a s
a s a s
a s a
+
|
.
|

\
|
=
+ =
=

= =
=
=
M
i
i i
T
M
i
i i
M
i
i
T
i i
M
i
i i g
p p
p
p E

=
=
M
i
i i
p
1
optimal
s a
2
1 1
2
optimal
|| || ) ( and

= =
=
M
i
i i
M
i
i i g
p p E s s a
25
Po- Ning Chen@cm.nct u Chapt er 5- 49
5.7 Minimum energy signals
So (a) has minimum average energy.
Po- Ning Chen@cm.nct u Chapt er 5- 50
5.7 Union bound on the probability of error
Union bound
{ }
( )
{
( )
( )
{
( )
{ }

= =
=

=
=

>

>
>
=

|
.
|

\
|
=
=
M
i
M
i j j
i j i
M
i
i
M i
i j
i
M
i
i
M i
i j
i
M
i
M
i
i j
i j M j
i e
m
M
m
M
m
M M
m
M
P
1 , 1
2 2
1
2 2
2
1
2
1
2 2
2
1
2
1
1
2
, 1
2
ed transmitt || || || || Pr
1
ed transmitt
|| || || ||
|| || || ||
Pr
1
ed transmitt
|| || || ||
|| || || ||
Pr
1
1
1
ed transmitt || || min || || Pr
1
1
s x s x
s x s x
s x s x
s x s x
s x s x
s x s x
L
L
L
L
) ( ) ( ) ( B P A P B A P + U
26
Po- Ning Chen@cm.nct u Chapt er 5- 51
5.7 Union bound on the probability of error
( )
( )
( )

>
>
>
2
4
2
1
2
3
2
1
2
2
2
1
|| || || ||
|| || || ||
|| || || ||
s x s x
s x s x
s x s x
{ }
2
2
2
1
|| || || || s x s x > { }
2
3
2
1
|| || || || s x s x > { }
2
4
2
1
|| || || || s x s x >
Po- Ning Chen@cm.nct u Chapt er 5- 52
5.7 Union bound on the probability of error

= =

M
i
M
i j j
j i e
P
M
P
1 , 1
2
) , (
1
s s
{ }
. mean with d distribute Gaussian is ed, transmitt given Notably,
. ed transmitt || || || || Pr ) , ( where
2 2
2
i i
i j i j i
m
m P
s x
s x s x s s > =
i
s
j
s
w
j i
s s
2
1
27
Po- Ning Chen@cm.nct u Chapt er 5- 53
5.7 Union bound on the probability of error
Hence,
For N = 1,
{ }
. ) exp(
2
erfc(u) where ,
2
erfc
2
1
| | where , exp
1
| |
2
1
Pr
ed transmitt || || || || Pr ) , (
2
0
2 /
0
2
0
2 2
2

=
|
|
.
|

\
|
=
=
|
|
.
|

\
|
=
)
`

> =
> =
u
ij
j i ij
d
j i
i j i j i
dz z
N
d
s s d dv
N
v
N
s s w
m P
ij

s x s x s s
Po- Ning Chen@cm.nct u Chapt er 5- 54
5.7 Union bound on the probability of error
For N = 2,
The same formula is valid for any N.
{ }
. ) exp(
2
erfc(u) where ,
2
erfc
2
1
exp
1
|| || where , care t don' and
2
1
Pr
ed transmitt || || || || Pr ) , (
2
0
2 /
0
2
0
2
2 1
2 2
2

=
|
|
.
|

\
|
=
|
|
.
|

\
|
=
=
)
`

= > =
> =
u
ij
d
j i ij ij
i j i j i
dz z
N
d
dv
N
v
N
d w d w
m P
ij

s s
s x s x s s
28
Po- Ning Chen@cm.nct u Chapt er 5- 55
5.7 Union bound on the probability of error
Consequently, the union bound for symbol error rate is:
The above bound can be further simplified when additional
condition is given.
For example, if the signal constellation is circularly
symmetric in the sense that {d
i1
, d
i2
, , d
iM
} is a
permutation of {d
k1
, d
k2
, , d
kM
} for i k, then

= = = =
|
|
.
|

\
|
=
M
i
M
i j j
ij
M
i
M
i j j
j i e
N
d
M
P
M
P
1 , 1
0
1 , 1
2
2
erfc
2
1 1
) , (
1
s s

=
|
|
.
|

\
|

M
i j j
ij
e
N
d
P
, 1
0
2
erfc
2
1
Po- Ning Chen@cm.nct u Chapt er 5- 56
5.7 Union bound on the probability of error
Another simplification of union bound
Define the minimum distance of a signal constellation
as:
Then by the strict decreasing property of erfc function,
ij
j i M j M i
d d

=
, 1 , 1
min
min
|
|
.
|

\
|

|
|
.
|

\
|
0
min
0
2
erfc
2
erfc
N
d
N
d
ij
|
|
.
|

\
|

=
|
|
.
|

\
|

|
|
.
|

\
|


= = = =
0
min
1 , 1
0
min
1 , 1
0
2
erfc
2
1
2
erfc
2
1 1
2
erfc
2
1 1
N
d M
N
d
M N
d
M
P
M
i
M
i j j
M
i
M
i j j
ij
e
29
Po- Ning Chen@cm.nct u Chapt er 5- 57
5.7 Union bound on the probability of error
We may use the bound for erfc function to realize the
relation between SER and d
min
.
Conclusion: SER decreases exponentially as the squared
minimum distance.
608131 . 0 for
) exp(
) ( erfc
2
>

u
u
u

. 47929 . 1 if ,
4
exp
2
1
2
erfc
2
1
0
2
min
0
2
min
0
min
N d
N
d M
N
d M
P
e
>
|
|
.
|

\
|

|
|
.
|

\
|

Po- Ning Chen@cm.nct u Chapt er 5- 58


5.7 Relation between BER and SER
The information bits are transmitted in group of log
2
M bits
to form an M-ary symbol.
This gives the result that a large symbol error rate (SER)
may not cause a large bit error rate (BER).
For example, a symbol error (for large M) may be due to
only 1 bit error.
Optimistically, if every symbol error is due to a single
bit error, then (assuming n symbols are transmitted)
.
) ( log
SER
) ( log
SER
BER
2 2
M M n
n
=


=
|
|
.
|

\
|
.
) ( log
SER
BER general, In
2
M
30
Po- Ning Chen@cm.nct u Chapt er 5- 59
5.7 Relation between BER and SER
Pessimistically, if every symbol error causes log
2
M bit
errors, then (assuming n symbols are transmitted)
Summary:
. SER
log
SER log
BER
2
2
=


=
M n
M n
( ) . SER BER general, In
SER BER
log
SER
2

M
Po- Ning Chen@cm.nct u Chapt er 5- 60
5.7 Relation between BER and SER
If the statistics for number of bit error patterns causes one
symbol error is known, we can then determine the exact
relation between BER and SER.
M n
P n
M
j
j j
2
1
1
log
) ( ) ( # SER
BER

=
b b
pattern. bit log of n permutatio binary one represents and
, in s 1' of number ) ( # where
2
M
j
j j
b
b b =
Here, a 1s in b
j
means a bit error is occurred in the corresponding position;
hence, all-zero pattern is excluded because it represents no symbol error.
31
Po- Ning Chen@cm.nct u Chapt er 5- 61
5.7 Relation between BER and SER
Example. If all bit error patters are equally likely, then
SER
1
2 /
SER
log ) 1 (
2 log
.) 2 (Note
log
log ) 1 (
SER
log
)) 1 /( 1 ( ) ( # SER
log
) ( ) ( # SER
BER
2
1 log
2
1
1
log
1
2
2
2
1
1
2
1
1
2
2
|
.
|

\
|

=


=
=
|
|
.
|

\
|

|
|
.
|

\
|


=

=

= =

=


M
M
M M
M
k
u
k
u
u
M
u
M M
M
M
M n
P n
M
k
k
u
M
u
M
j
j
M
j
j j
b b b